Next Article in Journal
A Review on Emerging Cementitious Materials, Reactivity Evaluation and Treatment Methods
Next Article in Special Issue
Benefits and Obstacles to the Adoption of Reality Capture Technologies in the U.S. Commercial and Infrastructure Construction Sectors
Previous Article in Journal
Study of Application of Fiber Reinforced Concrete in Anchorage Zone
Previous Article in Special Issue
SwarmL: A Language for Programming Fully Distributed Intelligent Building Systems
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Buildings
Volume 13
Issue 2
10.3390/buildings13020525
buildings-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

A Review: How Deep Learning Technology Impacts the Evaluation of Traditional Village Landscapes

by
Tao Wang
1,†,
Jingjing Chen
2,3,†,
Li Liu
4,* and
Lingling Guo
5,*
1
Zhijiang College, Zhejiang University of Technology, Shaoxing 312000, China
2
Department of Digital Urban Governance, Zhejiang University City College, Hangzhou 310011, China
3
School of Economics, Fudan University, Shanghai 200433, China
4
School of Economics and Social Welfare, Zhejiang Shuren University, Hangzhou 310015, China
5
College of Chemical Engineering, Zhejiang University of Technology, Hangzhou 310023, China
*
Authors to whom correspondence should be addressed.
†
These authors contributed equally to this work.
Buildings 2023, 13(2), 525; https://doi.org/10.3390/buildings13020525
Submission received: 31 December 2022 / Revised: 26 January 2023 / Accepted: 7 February 2023 / Published: 15 February 2023
(This article belongs to the Special Issue Application of Computer Technology in Buildings)
Versions Notes

Abstract

:
Recently, the deep learning technology has been adopted in the study of traditional village landscape. More precisely, it’s usually used to explore the representation of cultural heritage and the diversity of heritage information. In this study, we comprehensively reviewed these deep learning-related literatures for the evaluation of traditional village landscapes. E.g., the landscape image recognition led by the pixel-level semantic segmentation algorithm and image feature extraction technology enable user-centred exploration and make cultural heritage digitally and visually accessible. By suggesting a analytic framework using the pixel-level semantic segmentation algorithm and extracting image features, we attempted to identify the physical attributes and spatial characteristics of traditional village landscapes and further simulate the value perception thinking of experts and the public. Besides, we analysed the impact factors and correlation mechanism of spatial attributes to provide a scientific basis and technical support for the protection and utilization of traditional villages.

1. Introduction

In recent years, the use of computing and digital technology to record and assess cultural heritage has become a global trend in heritage conservation. UNESCO’s Charter for the Preservation of Digital Heritage [1] and the Vancouver Declaration on Digitization and Conservation [2] have expanded the traditional understanding of cultural heritage to include digital representation as an object of conservation. Digital and computing technologies offer unprecedented perspective, detail, and precision in observing and recording cultural heritage, which has revolutionised the workflow of heritage documentation and provided opportunities to explore innovative approaches to conservation and management. Moreover, advances in the digital technology of computers have made the acquisition, recording, and manipulation of visual three-dimensional (3D) data and image recognition technically possible [3]. Other computing techniques, such as reverse engineering and computer graphics, are used to analyse, study, conserve, and visualise cultural heritage assets [4]. Under this context, cultural heritage, as a domain field in preserving urban and rural landscapes, benefits significantly from using these computing technologies as users and practitioners can experience and understand cultural artefacts in a new way.
In terms of integrating computing technologies with cultural heritage, the concept of cultural computing has been developed. It is used to collect, record, analyse, organize, mine, express, relate, disseminate and display the information/knowledge contained in cultural heritage-intelligent information technologies such as big data, artificial intelligence and virtual reality [5,6]. Moreover, cultural computing can integrate other digitisation methods and artificial intelligence, such as computational analysis, historical simulation, virtual simulation, and immersive experience. Such integration may bring new opportunities for understanding and preserving cultural heritage [4]. In recent years, the development of deep learning algorithms, image recognition and evaluation applications has provided experience and technical support for cultural heritage in urban and rural contexts [7]. Furthermore, digitisation and artificial intelligence could promote the construction of digital villages [8], and such methods also provide a new perspective and technical approach for studying the traditional village landscape.
Image recognition and deep learning technologies help interpret recognition and classification in multiple aspects of processing large-scale image dataset, including image aesthetics quality evaluation, urban and rural environment recognition, etc. [3,9,10]. Furthermore, image recognition based on deep learning technology provides technical support for the mining, cognition, protection, and utilisation of traditional village features [11,12].
This study used the Web of Science (WOS) Core Collection (i.e., SCI-EXPANDED, SSCI, A&HCI) to obtain literature information including titles, authors, abstracts, keywords, citations and references by literature search method. The CiteSpace was used to perform the data analysis. For this study, the time span for filtering these literatures was from 1950 to 2020. We searched the databases by using keywords: “deep learning”, “machine learning”, “traditional village landscape”, “image recognition”, “landscape evaluation”, and a total of 238 papers were obtained, and by carefully checking the title and abstract of each paper, we removed these articles that were not closely relevant to the research of this paper, and finally 96 papers, which are closely related to the research and up-to-date ones, were selected.
The contributions in this article are summarized as:
(1) we investigate these literatures introducing the advancement of artificial intelligence and its application to the evaluation of traditional village landscape with artificial intelligence technology.
(2) we summarize the current advancements in this field, thereby revealing the main technical features of deep learning.
(3) we construct a cutting-edge analysis framework to analyse the characteristic interpretation and cognitive evaluation of traditional village landscapes based on deep learning.
The rest of this article is organized as follows: Section 2 introduces the methodologies for the evaluation of traditional village landscape; Section 3 details the insides of deep learning technology for landscape recognition and evaluation; Section 4 focuses on the evaluation of traditional village landscape based on deep learning; Section 5 concludes the study.

2. Methodologies for the Studies of Traditional Village Landscape

As sites of revitalised cultural heritage, traditional villages carry a considerable amount of historical memory, human ecology, architectural aesthetics, and social development [13]. However, the cultural heritage found in traditional village landscapes faces destruction due to globalization, modernisation, and urbanization, and its protection and utilisation have become an issue of great concern to all countries in the world [14]. Traditional villages are considered as heritage sites and offered legislative protection. The International Council of Antiquities and Heritage has issued documents, such as the “Tlaxcala Declaration on the Regeneration of Small Settlements” and “the Charter on Vernacular Architectural Heritage”, to provide a theoretical basis and technical support for the protection and utilization of traditional villages [15,16].
Research on rural landscapes was initially based on subjective and empirical research on the results of rural construction, focusing on the shaping of the landscape and the establishment of a theoretical framework [17,18]. Later, it gradually began to emphasize data analysis, focusing on objective data research and paying attention to the evolution, characteristics, evaluation system, prediction model, and construction mode of the rural landscape style, historical humanities, and ecological environment [19,20]. At present, studies on rural landscapes mainly rely on modern communication techniques to carry out comprehensive multidisciplinary and multi-method research. Visualisation technologies based on 3D graphics and integrated interactive studies using various methods for landscape study are manifested and combined with multiple natural, artificial, and human aspects [21,22].
In China, research on rural landscapes mainly focuses on analysing landscape features, remediation planning, evaluation and optimisation, and protection and management. Firstly, in terms of village site selection, architectural texture, and street space, current research aims to analyse the composition, development, evolution, essential characteristics, typical features, and problems of landscape elements [5,23,24]. Secondly, to explore the mode, strategy, and path of rural landscape planning and renovation based on a multidisciplinary perspective, current studies focus on creating rural revitalisation strategies and countryside construction plans [12,25,26]. Thirdly, with respect to the triadic evaluation of ecology, morphology, and mentality, current research integrates resident perception, environmental ecology, and morphological aesthetics and applies new technological methods, such as databases [27,28,29,30]. Finally, it is necessary to analyse current predicaments, problems, and causal mechanisms; explore the modes, methods, strategies, and paths of rural landscape protection and management; and emphasise public participation [7,31].
These researchers have gradually expanded their research on rural landscape scenery to multiple fields and have carried out comprehensive explorations, obtaining rich results with image recognition and deep learning technology, especially in image aesthetic quality assessment and town scenery recognition. Furthermore, although deep learning technology and image recognition have not been widely applied in traditional village landscape protection, the relevant literature already provides empirical references and technical support. Therefore, with computer-aided analysis and mathematical models, the identification, evaluation, and construction of models will be an important part of the research on the protection and use of traditional village landscapes in China.
As shown in Table 1, the application contexts include the value research of traditional villages, cultural landscape characteristics, integrated protection and development, the evolution of village spatial forms, and landscape planning and design, the development of the village landscape, the improvement of the tourism environment, and the improvement of the overall landscape, etc. The mainstream research methods include landscape genetic methods [32], space syntax [3], landscape patterns [33], ArcGIS analysis methods [34], least resistance models [35], and landscape sensitivity analysis methods [36]. The methods used in foreign countries are more mature, and mainly include ASEB raster analyses of tourists’ experiences [37], decision-making laboratory analysis methods [38], GIS analysis methods [39], etc.

3. Technical Details of Deep Learning Used for the Study of Traditional Village Landscape

The traditional village landscapes involving information technologies have been heavily studied in past decades, as shown in Table 2, we grouped these literatures (by searching http://scholar.google.com (accessed on 30 December 2020) with keywords: “traditional village landscape”, “evaluation”, “artificial intelligence”, “deep learning”) into three categories: (1) image recognition and evaluation based on deep learning, (2) application of image recognition in the field of urban and rural construction, (3) recognition, evaluation, and protection of rural landscapes.
In recent years, image recognition using deep learning has seen several breakthroughs and has surpassed that of the human eye. Moreover, compared with traditional machine learning, intelligent recognition based on deep learning and large-scale image dataset greatly improves the accuracy [40,48,49,57]. This technology is used in many fields, such as in the prediction of facial beauty [58], the image feature extraction of agricultural diseases and insect pests [50], and the intelligent recognition of plant species from images [51,52]. These methods help analyse the composition and characteristics of urban physical environment elements. For example, by identifying the urban elements in an image and the social environmental attributes [53,54]. In terms of economic and political aspects, these methods could provide information for predicting the non-visual attributes of a city based on the characteristics of the physical environment and link household income data and housing price data to realise an intelligent evaluation of the economic environment [55]. In terms of visual aesthetic and experience, it is critical to utilise image recognition based on the characteristics of the physical environment to associate aesthetic theories with people’s cognitive concepts and carry out an intelligent evaluation of urban style [9], visual quality [56].
In China, research on deep learning and its use in image recognition technology has gradually focused on the levels of streets and cities. At the street level, based on the large-scale street view images, deep learning has been used to identify and classify street space elements, measure the level of greening [59], openness [41], and walkability [60]. and it’s also used for the evaluation of street space quality, dynamic change measurements, and the factors influencing discrimination [61]. At the city level, the deep learning has been used to carry out a cognitive evaluation of urban image structures and types [62]: (1) the extraction of urban development types and their evolutionary routines [63]; (2) an essential evaluation and quantification of urban colours [64] and (3) the identification and evaluation of urban textures and an analysis of their economic relevance [65]. The custom image discrimination model has also been used to recognise and evaluate the aesthetic characteristics of the city skyline [66] and the city style [67]. In addition, research has also been conducted on remote sensing images for township building recognition, the intelligent classification of landscape elements, and the quality evaluation of landscape visualization.
At present, deep learning-based image recognition and evaluation techniques mainly involve convolutional neural networks [68], pixel-level semantic segmentation [1], image feature extraction techniques [69], image recognition [42], image evaluation [70], migration learning algorithms, and many other methods. The keys for deep learning-based image recognition is: (1) to establish the most effective fit between the selected model and the image feature elements; (2) to integrate and translate traditional cognitive language and image feature information data; (3) to determine the most effective elements to characterise the image through a set of procedures (e.g., image segmentation, feature extraction). Firstly, image pre-processing, such as geometric correction, image mosaic, image cropping, is employed for different categories of image dataset to eliminate irrelevant information. These methods help enhance the detectability of relevant information and improve the reliability of image feature extraction to some extent. Secondly, due to the large amount of information carried by various types of images, it is necessary to select the appropriate techniques for feature extraction and image segmentation. Thirdly, algorithms such as Histogram of Oriented Gradient (HOG), Maximally Stable Extremal Regions (MSER), and Scale-invariant Feature Transform (SIFT) are used to extract features from images to obtain elements that can fully characterize the features. The feature detection layer of the convolutional neural network learns automatically from the training data, avoiding explicit feature extraction, which is more effective for image recognition. The following subsections will introduce the technical details.

3.1. Image Feature Extraction

Feature extraction is an essential stage of image processing. Each image has characteristics that distinguish it from other images, some of which are natural and intuitively perceptible, such as brightness, edges, texture, and colour, while others require transformation or processing to observe, such as histogram moments and principal components. In 1988, Harris et al. [71] proposed the Harris corner detection algorithm to obtain an image’s corner points (as an image features) by judging the singular values in the image structure. In 1996, T. Ojala and others [72] proposed the Local Binary Pattern (LBP) algorithm, which generates a LBP value by comparing the local pixel value and the centre pixel value to extract the image texture. In 1999, David Lowe [73,74] proposed the SIFT algorithm, which was perfected and supplemented in 2004 [75]. The algorithm selects key feature points in different scaled spaces for description and generates image features. In 2001, Viola et al. [76] proposed the combination of the Haar feature extraction method and the AdaBoost classification algorithm to achieve face detection. In 2005, Dalal [77] proposed the HOG algorithm. This algorithm uses edge features for feature recognition combined with SVM classifiers to achieve pedestrian detection. In 2008, Bay et al. [78] proposed Speed-up Robust Features (SURF) as an improvement to the SIFT algorithm. It significantly reduces the program’s running time and increases the algorithm’s robustness. As shown in Table 3, these classical methods for feature extraction are compared.

3.2. Deep Learning-Based Workflow of Image Recognition & Evaluation

Deep learning simulates the neural connectivity structure of the human brain, discovers hidden structures and relationships within image data, and characterizes and interprets the data [79]. Typical deep learning models include convolutional neural networks (CNNs) [2], recurrent neural network (RNNs) [10], generative adversarial networks (GANs) [6], and CNN is the most widely used [80]. The essence of a CNN is the realization of an input-to-output mapping relationship, where no precise mathematical expressions are required between the input and output. A CNN contains convolutional calculations and a feedforward neural network with a deep structure, which has the ability to characterize learning. It can automatically learn features from large-scale data and generalise the results to anonymous data of the same type. The CNN extracts the local and global characteristics of the image by constructing a reasonable network model. Moreover, the evaluation model can learn the shallow features that humans can perceive while extracting deep features and laws that are otherwise difficult to discover. Therefore, the objective cognitive score given by a deep learning algorithm based on a CNN is more in line with the subjective score of people. As shown in Table 4, the classic CNN models are: LeNet-5, AlexNet, ZF-Net, VGGNet, GoogLeNet, ResNet, and DenseNet.
The LeNet is the first CNN structure [81]. However, the first well-known network was AlexNet [82]. AlexNet won the 2012 ImageNet competition with an absolute advantage of 10.9% over the runner-up, and convolutional neural networks (CNN) and deep learning (DL) have regained widespread attention [83]. Compared to LeNet, AlexNet has a deeper network, which uses ReLU as the activation function and uses dropout in training [84]. In addition, it randomly ignores some neurons to avoid overfitting the model. VGGNet was proposed by the Visual Geometry Group (VGG) of Oxford University [18], and it won first place in the positioning task and second place in the classification task in the 2014 ImageNet competition [85]. GoogLeNet defeated VGG-Net on the 2014 ImageNet classification task to win the championship [86]. GoogLeNet is different as it does not simply rely on deepening the number of network structure layers to improve network performance. By extracting information at different scales through multiple convolutional kernels and fusing them to better represent the image [43]. ResNet was launched by He in 2016 [87], and it is characterized as very deep, with more than a hundred layers. The main idea of ResNet is the addition of a direct connection channel to the network; that is, the idea of a highway network [88].
Table
Table 4. Comparison between CNN classic models.
Table 4. Comparison between CNN classic models.
ModelYearAuthorAchievement
LeNet-51994LeCun Y et al. [81]One of the earliest convolutional neural networks; promoted the development of deep learning.
AlexNet2012Krizhevsky A et al. [82]Won the 2012 ImageNet competition with an absolute advantage of 10.9% over second place.
ZF-Net2013Matthew D. Zeiler et al. [83]A network architecture with better performance than AlexNet; proposed a method of feature visualization to analyse and understand the network accordingly.
VGGNet2014Hull, R et al. [18]Based on AlexNet; an attempt to build a network with more layers and greater depth.
GoogLeNet2014Szegedy C et al. [86]Defeated VGG-Nets on the 2014 ImageNet classification task to win the championship.
ResNet2015Kaiming H et al. [87]Beat all players on ISLVRC and COCO to win the championship.
DenseNet2016Gao H et al. [89]The paper “Densely Connected Convolutional Networks” was selected as the best paper of CVPR 2017.
The image evaluation based on deep learning aims to use computers to simulate the perception and understanding of images and scenes by the human brain and automatically make objective evaluations about the “value” of the image content. The core idea is to use CNNs to learn a large number of sample images with labels to build a model and extract the hidden rules and mechanism of action. It usually consists of three steps: (1) image collection and pre-processing; (2) model training; and (3) evaluation applications. The first step is to use the extracted image features to train a regression model. This step aims to form a mapping relationship between the extracted features and the scoring values and then use this mapping relationship to make decisions, which can be implemented with mature feature parameters and algorithms, such as SVM, SVR, and random forest [90]. Secondly, to minimise the difference between predicted values and manual ratings, credibility weights are added based on the number of ratings of the image and the kurtosis of the rating distribution, the weight size is adjusted, and the weight calculation method is optimized. Third, to evaluate the score distribution prediction, one can choose algorithms such as LDSVR [91,92]. It can predict the multi-label distribution for any input instance, and overcomes the unobjective problems caused by using only a single point value. Finally, based on the individual scores of each characteristic factor, the overall cognitive score is increased to test and check the accuracy and objectivity of the evaluation. There are many evaluation indicators for image quality evaluation [93], e.g., PLCC, SROCC, KROCC, and RMSE.
Image recognition and evaluation extracts potential features, laws, and patterns from a considerable amount of labelled image dataset to realize the prediction and classification, which requires a large number of training samples. Currently, the datasets used to train image models include open-source label datasets and custom label training sets. The direct use of existing open-source image datasets brings greater limitations and errors, while the traditional method with manually labelling operation requires extensive labour and time-consuming. In order to overcome the problem of overfitting caused by too few labelled samples, small-sample deep learning techniques are proposed [8]. The typical technique is based on the transfer learning [94]. Its core idea is to establish a mapping relationship between the source dataset and the target dataset (e.g., SIFT Flow dataset, Pascal VOC dataset) with a large amount of rich annotation information using cross-domain migration algorithms.

4. Evaluation of Traditional Village Landscape Based on Deep Learning

As summarised in the previous sections, digital technology offers unprecedented perspective in observing and recording cultural landscape, and has revolutionised the workflow of landscape recording. To date, most digital cultural landscape projects have focused on technical challenges, such as methods and tools, to quickly obtain accurate and complete physical spatial information [4,95]. Such information includes the site selection pattern, spatial texture, streets and lanes, courtyards, and vernacular buildings. With it, the historical memory, wisdom of production and life, and crystallisation of the culture and art can be inherited [13,96]. However, the focus of digital landscape research is gradually expanding its attention to theoretical and ethical aspects [97]. Such elements include paying attention to integrating cultural, intangible, sensory, or experiential elements in the documentation of digital cultural landscape [11,98] and staying grounded in ethical principles that can inform and improve landscape documentation practices [98].

4.1. Interpretation of the Physical Characteristics of Traditional Village Landscape

Technologies enable the conceptual interpretation, connotation, and extension of traditional village landscapes, exploring the hierarchical structure and element types, from the macro site selection pattern and the meso-morphology to the microsite space, and summarising and analysing their influencing factors and formation mechanisms. To achieve such objectives, we suggest to adopt network data capture, drone aerial photography, departmental data collection, and other methods to obtain a considerable amount of image data reflecting the landscape of traditional villages. Then, the pixel-level semantic segmentation technology was adopted to build a deep learning model of the traditional village landscape style based on the image contents. Moreover, large-scale and refined style feature recognition can be used to analyse the spatial characteristics and physical representations of traditional village landscapes of different scales and types. Finally, these influencing factors and formation mechanism for protecting the traditional Chinese village landscape can be determined.
In terms of collecting image data for cognitive evaluation and deep learning processing, image recognition models can be used to identify, classify, statistically analyse, and compare various landscape elements. Furthermore, three levels of image recognition-morphology and fabric, streets and zones, and quarters and buildings-were organised by the measurable spatial features, visual patterns and appearances, and cultural characters from the image data to achieve a formal refinement quantitative description of these core features.
At the level of the morphology and fabric of the village landscape, image recognition measured two-dimensional plane attributes integrated with the data collected from unmanned aerial photography, satellite remote sensing, and GIS. The characteristics obtained at this level include the spatial distribution, size, shape, texture, and fractal dimension of traditional villages. Such characteristic elements could help summarise the spatial pattern and relationship of rhythm, form, contrast, change, etc. At the level of streets and zones, the various constituent elements and their three-dimensional spatial attributes, such as scale, quantity, shape, and volume, were identified. Digital methods, such as 360-degree panoramic photography and 3D landscape models, were manifested and processed with image recognition. Then, the characteristics of each element, such as the association and combination, proportional relationship, and influence mechanism, were deciphered for image recognition analysis. The last level was quarters and buildings. AR (augmented reality) and AV (augmented virtual) were employed to offer architectural details and texture information for specific buildings and small-scale places. Image data generated from the three levels were integrated with image recognition, processing representative characters and evaluations of the different types of traditional village landscape.
In addition to the image data collected at the research site, this study captured online image data from open-source database websites. Based on open-source database systems, such as traditional village databases and digital museums, rural tourism websites, forums, and other thematic websites, the Python language-based crawler technology was used to obtain a large amount of image data related to the traditional village cultural landscape space. The image data included social media images, Baidu Street View photos, satellite images, air-ground combination 360 panoramas, overall aerial views, characteristic node maps, and many other types. On-site image data collection of typical traditional villages was also employed. Standard samples of traditional villages were selected based on the principles of balanced distribution, comprehensive types, and strong representativeness.
Based on the collection and sorting of original image data, pixel-level semantic segmentation technology was applied to build an algorithm model based on the classic semantic segmentation convolutional neural network on the Caffe deep learning framework. The first step of processing the image data is to increase the effective information of model training and its generalisation performance through pre-processing and external assistance. It then applies a framework of deep learning techniques based on transfer learning algorithms across data domains. After that, the pixel-level semantic segmentation technology then explores the joint application of open-source image datasets and manually annotated image training sets, then builds a traditional village landscape recognition model based on deep learning. Finally, it summarises the matching algorithm model and its empirical parameters and structure between image attributes and features. Therefore, any complex image scene can be quickly decomposed and clustered under the model, enabling interpretation and quantitative analysis to be processed with image data. The analysis carries out the identification, classification, measurement, and quantification of traditional village landscape elements. It then analyses the spatial characteristics and physical properties of different scales and types through image pre-processing, feature extraction, semantic segmentation, and post-processing.

4.2. Cognitive Evaluation of Traditional Village Landscape

Evaluation Index of Traditional Village Landscapes

With regards to the regional characteristics and core attributes of traditional village landscapes, theoretical analysis and expert consultation based on image recognition and deep learning should be taken into consideration. In this way, we could determine the value system of conservation and utilisation based on image, associations, and symbols they evoked. Moreover, the combination of computing technologies and theoretical analysis could help overcome the shortcomings of traditional text questionnaires that “ignore the process” and integrate the ideas of “self-recall” and “self-experience”.
An essential prior to performing the intelligent cognitive evaluation of traditional village landscapes is to obtain a large number of high-quality image samples with score tags from different groups of people. A combination of online and offline large-scale public and small-scale expert samples was used to obtain a consensus accumulation of large-scale individual opinions from different groups of people, such as villagers, tourists, academics, designers, and government personnel. In addition, various methods, such as laboratory manual scoring, field questionnaire interview scoring, social media online scoring, and internet crowdsourcing platform scoring, were employed to collect and calibrate evaluation information. We obtained a cognitive evaluation of the traditional village landscape from people with diverse background attributes, including different regions, ages, educational backgrounds, and occupations. Then, the cognitive evaluation was used to construct a label image dataset with multi-dimensional value information, such as history, aesthetics, culture, usage, economy, and society, all of which were used to form training and testing samples for subsequent deep learning models. At the same time, based on the individual scoring of each feature factor, an overall cognitive evaluation of the image landscape features was added to check and verify its accuracy and objectivity, solving the quality and quantity problems of traditional village landscape image data at the source and ensuring the depth and breadth of subsequent processing.
Based on artificially labelled image sets and experts’ cognitive and situational processes, tourists and villagers were simulated to construct an intelligent image evaluation model. To address the potential overfitting problem caused by insufficiently annotated samples, this study explored a framework of small-sample deep learning techniques based on internal data augmentation and cross-domain transfer learning. Of the annotated images, 75% were selected for model training, optimisation, and tuning, and the remaining 25% were used for validation and testing. In summary, the sample selection and deep analysis was used to build a deep learning model capable of simulating human recognition for value assessment, enabling the evaluation and measurement of traditional rural landscape images.

4.3. The Protection and Utilisation of Traditional Village Landscapes

4.3.1. Utilisation Countermeasures Based on Physical Characterization

Based on the superposition, clustering, and correlation of information, including human historical background, spatial characteristics and attributes, value perception, and evaluation, a GIS-based map of traditional village landscape features was constructed from an authentic perspective, linking relevant historical images, real-world photos, and critical features and evaluation information; providing real-time queries and dynamic updates; and building a digital display platform in combination with signage, tour route systems, etc.
According to the type and distribution of the traditional village landscapes and their characteristic attributes, a core protection zone, characteristic demonstration zone, characteristic experience zone, and coordinated integration zone were designated from the perspective of adaptability, and region-wide protection and development zoning, a co-ordination mechanism, and a digital management platform were established. Such platforms aim to explore the control and continuation of morphological patterns, the reproduction and implantation of characteristic places for protection, and the enhancement and creation of typical landmarks. Finally, we combined the character map and conservation and development zoning, then identified proposals and directions for maintaining, enhancing, restoring, and creating various landscape features. As such, we proposed response strategies and technical support for overall protection, relevance protection, authenticity protection, adaptive development, etc., highlighting the control and continuation of the pattern of landscape forms and the reproduction and implantation of characteristic places.

4.3.2. Utilisation Countermeasures Based on Cognitive Evaluation

Firstly, this study explored the conservation model and implementation path from “elite conservation” to “public conservation”, from “government-led” to “people-initiated”, and from “top-down” to “bottom-up”. Secondly, it emphasised the transition from “absolute value quantification” to “subjective and objective compound value”, which highlights village democracies’ status and gives full play to the role of civil society organisations. Moreover, we adopted a multi-professional, multi-department, and multi-channel collaborative protection and utilisation mode. Thirdly, it respected the multi-dimensional perception and interpretation of the landscape features of traditional villages and the meaning and values they carry for the public and experts of different backgrounds and attributes. Finally, based on public perception and preference attributes, diversified media platforms, such as microblogging, WeChat, and television. They were selected to showcase the spatial characteristics and historical, cultural, and value attributes of traditional village landscapes; improve public awareness and education for all; and enhance public cultural self-awareness and cultural confidence.

4.3.3. Utilization Countermeasures Based on Multiple Appropriateness

Based on the spatial composition of cultural landscapes and the analysis of intelligent decision-making platforms, this study analysed the challenges, opportunities, and constraints faced by different villages in their future conservation and development by exploring conservation and development models and pathways that suit their characteristics and critical users. Moreover, it focused on liveable villages, highlighting the main position and actual needs of villagers, and building a bottom-up conservation and utilisation model to improve the living environment of villagers as the main body of construction and countermeasures. In relation to tourism, this study examined tourism expansion villages as emphasising the aesthetic values and tourism needs of tourists under the premise of conservation. Moreover, it highlights the improvement of facilities and tourism experience and promotes the sustainable conservation and use of village cultural landscape spaces through tourism development. Finally, in the construction and development of cultural heritage villages, it is necessary to fully consider the theoretical and cultural values of unique cultural landscape spaces and the policy support from the national and local governments when exploring appropriate modes and strategies of conservation and utilisation.

5. Conclusions

This study aims to provide the overview of the adoption of the artificial intelligence to the evaluation of traditional village landscapes. We investigated the state-of-the-art deep learning methods, and their applications in the field of traditional village landscape applied in recent years. We first introduce the basic concepts and history the studies of artificial intelligence and traditional village landscape. Then we analyse the core technical principles of artificial intelligence technology (e.g., deep learning), discuss the technical roles of deep learning help improve the study of traditional village landscape in different application contexts and processes. It reveals that the image recognition technology enabled by deep learning could significantly improve the efficiency of the study of traditional village landscape. We further suggest a cutting-edge analysis framework for applying deep learning technology into the evaluation of traditional village landscape, with which, the deep aggregation and knowledge output of multi-dimensional attribute information was achieved.

Author Contributions

Conceptualization, T.W. and J.C.; methodology, T.W.; investigation, J.C.; writing—original draft preparation, J.C.; writing—review and editing, L.L.; visualization, L.G. All authors have read and agreed to the published version of the manuscript.

Funding

The research is supported by the Ministry of Education humanities social sciences research project (20YJC760101), Zhejiang Provincial Natural Science Foundation of China (LQ20E080021), Major humanities and social sciences research program of colleges and universities in Zhejiang (2021QN053), Zhejiang cultural relics protection science and technology project (2021017), Scientific research project of Zhejiang Provincial Department of Education (Y202147553), China youth fund of national natural science foundation projects (52208057) and Zhejiang University of Technology Humanities and Social Sciences pre Research Fund Project (SKY-ZX-20210186).

Data Availability Statement

No new data were created or analyzed in this study. Data sharing is not applicable to this article.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Ahn, J.; Kwak, S. Learning pixel-level semantic affinity with image-level supervision for weakly supervised semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–23 June 2018; pp. 4981–4990. [Google Scholar]
  2. Amerini, I.; Li, C.-T.; Caldelli, R. Social network identification through image classification with CNN. IEEE Access 2019, 7, 35264–35273. [Google Scholar] [CrossRef]
  3. Bafna, S. Space syntax: A brief introduction to its logic and analytical techniques. Environ. Behav. 2003, 35, 17–29. [Google Scholar] [CrossRef]
  4. Baptista, M.L.V. Documenting a complex modern heritage building using multi image close range photogrammetry and 3d laser scanned point clouds. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2013, 5, W2. [Google Scholar] [CrossRef] [Green Version]
  5. Li, B.; Liu, M.; Liu, P.; Dou, Y. Research on the Features of Traditional Villages from the Perspective of Landscape Gene Information Chain—Take the above Gantang Village as an example. Hum. Geogr. 2020, 35, 40–47. [Google Scholar]
  6. Brock, A.; Donahue, J.; Simonyan, K. Large scale GAN training for high fidelity natural image synthesis. arXiv 2018, arXiv:1809.11096. [Google Scholar]
  7. Li, C.; Zhao, Z. Research on the Status Quo of Lingnan Style Villages and Their Protection Strategies—Take Bijiang Village as an example. Herit. Conserv. Res. 2019, 4, 27–31. [Google Scholar]
  8. Cheng, F.; Zhang, H.; Fan, W.; Harris, B. Image recognition technology based on deep learning. Wirel. Pers. Commun. 2018, 102, 1917–1933. [Google Scholar] [CrossRef]
  9. Cheng, L.; Chu, S.; Zong, W.; Li, S.; Wu, J.; Li, M. Use of tencent street view imagery for visual perception of streets. ISPRS Int. J. Geo-Inf. 2017, 6, 265. [Google Scholar] [CrossRef] [Green Version]
  10. Cho, K.; van Merriënboer, B.; Gulcehre, C.; Bahdanau, D.; Bougares, F.; Schwenk, H.; Bengio, Y. Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv 2014, arXiv:1406.1078. [Google Scholar]
  11. Connolly, C. Whose landscape, whose heritage? Landscape politics of ‘swiftlet farming’ in a World Heritage City. Landsc. Res. 2017, 42, 307–320. [Google Scholar] [CrossRef]
  12. Dong, H. Analysis of Material Characteristics and Component Elements of Villages and Towns in Chongming District, Shanghai. Small Town Constr. 2019, 10, 43–52. [Google Scholar]
  13. Zeng, Z.; Li, L.; Pang, Y. Analysis on climate adaptability of traditional villages in Lingnan, China--World Cultural Heritage Site of Majianglong Villages as example. Procedia Eng. 2017, 205, 2011–2018. [Google Scholar] [CrossRef]
  14. Hu, Y.; Chen, S.; Cao, W.; Cao, C. The concept and cultural connotation of traditional villages. Urban Dev. Stud. 2014, 21, 11–13. [Google Scholar]
  15. Njuguna, M.B.; Wahome, E.W.; Deisser, A.M. The role of the national construction authority in the conservation of vernacular architectural heritage. East Afr. J. Eng. 2020, 2, 23–32. [Google Scholar] [CrossRef]
  16. Li, W. Analysis of Countermeasures for the Protection of Cultural Relics and Historic Sites in Egypt and the Development of Tourism Resources. Cult. Ind. 2020, 21, 13–14. [Google Scholar]
  17. Falade, J. Amenity and open space planning in Nigeria. Land Use Policy 1989, 6, 162–171. [Google Scholar] [CrossRef]
  18. Hull, R.B., IV; Harvey, A. Explaining the emotion people experience in suburban parks. Environ. Behav. 1989, 21, 323–345. [Google Scholar] [CrossRef]
  19. Giupponi, C.; Ramanzin, M.; Sturaro, E.; Fuser, S. Climate and land use changes, biodiversity and agri-environmental measures in the Belluno province, Italy. Environ. Sci. Policy 2006, 2, 163–173. [Google Scholar] [CrossRef]
  20. Hietel, E.; Waldhardt, R.; Otte, A. Statistical modeling of land-cover changes based on key socio-economic indicators. Ecol. Econ. 2007, 62, 496–507. [Google Scholar] [CrossRef]
  21. Hart, A.K.; Milder, J.C.; Estrada-Carmona, N.; Declerck, F.A.; Harvey, C.A.; Dobie, P. Integrated landscape initiatives in practice: Assessing experiences from 191 landscapes in Africa and Latin America. Clim.-Smart Landsc. Multifunct. Practice. Nairobi World Agrofor. Cent. 2015, 7, 89–101. [Google Scholar]
  22. Milder, J.C.; Hart, A.K.; Dobie, P.; Minai, J.; Zaleski, C. Integrated landscape initiatives for African agriculture, development, and conservation: A region-wide assessment. World Dev. 2014, 54, 68–80. [Google Scholar] [CrossRef]
  23. Yang, G.; Wang, Z. Analysis of the Material Elements and Formation Modes of the Features of Traditional Villages—Take Shatan Village, Yutou Township, Huangyan District, Zhejiang Province as an example. Town Ctry. Plan. 2018, 1, 24–32. [Google Scholar]
  24. Liu, Y.; Zhang, D. Exploring the ancient villages in the deep mountains of Luzhong—Analysis on the style and features of Dieyu Village in Boshan District of Zibo. Collect 2019, 3, 80–84. [Google Scholar]
  25. Wang, M.; Hou, X.; Wang, J. Rural landscape planning under the dual-objective system of ecology and aesthetics: Conceptual framework and practical approach. Landsc. Gard. 2017, 6, 95–104. [Google Scholar]
  26. Li, T.; Gao, Z. Kangping County, Shenyang City: The Construction and Reconstruction Strategy of Rural Scenery under the Background of Rural Revitalization Strategy. Beijing Plan. Constr. 2020, 06, 3. [Google Scholar]
  27. Zhang, H. Residents’ Landscape Perception and Rural Landscape Development Patterns in the Midst of Urbanization. Urban. Intensive Land Use 2016, 01, 1–10. [Google Scholar]
  28. Xia, L.; Cheng, W.; Zhao, T. Research on Evaluation of Village Settlement Landscape in Severe Cold Regions with the Aid of Database. Chin. Gard. 2017, 11, 99–104. [Google Scholar]
  29. Zhang, L.; Zhang, J.; Liu, B. Research on the Perceived Value of Traditional Village Landscape Based on Tourist Behavior Preference. Chin. Gard. 2017, 33, 92–96. [Google Scholar]
  30. Shao, Y.; Liu, B. Analysis of visual perception of rural landscape. Chin. Gard. 2016, 32, 5–10. [Google Scholar]
  31. Jiang, W.; Jiang, H.; Mao, Y. Protection and Regeneration of Rural Style and Features of Water Network in the Context of Ecological Aesthetics—Take Giethoorn in the Netherlands as an example. Hous. Technol. 2020, 40, 50–56. [Google Scholar]
  32. van strien, M.J.; Keller, D.; Holderegger, R. A new analytical approach to landscape genetic modelling: Least—Cost transect analysis and linear mixed models. Mol. Ecol. 2012, 21, 4010–4023. [Google Scholar] [CrossRef] [PubMed]
  33. O’neill, R.; Krummel, J.; Gardner, R.E.A.; Sugihara, G.; Jackson, B.; Deangelis, D.; Milne, B.; Turner, M.G.; Zygmunt, B.; Christensen, S. Indices of landscape pattern. Landsc. Ecol. 1988, 1, 153–162. [Google Scholar] [CrossRef]
  34. Jiménez-Perálvarez, J.; Irigaray, C.; El Hamdouni, R.; Chacón, J. Building models for automatic landslide-susceptibility analysis, mapping and validation in ArcGIS. Nat. Hazards 2009, 50, 571–590. [Google Scholar] [CrossRef]
  35. Wei, C.; Zhang, J.; Cheng, M.; Geng, X. Research on the landscape security pattern of Wu’an City based on the minimum resistance model. Resour. Dev. Mark. 2017, 12, 1417–1421. [Google Scholar]
  36. Newham, L.T.; Norton, J.P.; Prosser, I.P.; Croke, B.F.; Jakeman, A.J. Sensitivity analysis for assessing the behaviour of a landscape-based sediment source and transport model. Environ. Model. Softw. 2003, 18, 741–751. [Google Scholar] [CrossRef]
  37. Qin, Y.; Huang, Y.; Luo, S.; Miao, Y. Research on Sustainable Development of Forest Park Tourism Based on ASEB Grid Analysis. In IOP Conference Series: Earth and Environmental Science; IOP Publishing: Bristol, UK, 2020; p. 012060. [Google Scholar]
  38. Seyed-Hosseini, S.M.; Safaei, N.; Asgharpour, M. Reprioritization of failures in a system failure mode and effects analysis by decision making trial and evaluation laboratory technique. Reliab. Eng. Syst. Saf. 2006, 91, 872–881. [Google Scholar] [CrossRef]
  39. Fotheringham, S.; Rogerson, P. Spatial Analysis and GIS; Crc Press: Boca Raton, FL, USA, 2013. [Google Scholar]
  40. Wu, R.; Yan, S.; Shan, Y.; Dang, Q.; Sun, G. Deep image: Scaling up image recognition. arXiv 2015, arXiv:1501.02876. [Google Scholar]
  41. Li, X.; Cai, Y.; Kaluo, L. Urban landscape research based on street images and deep learning. Collect 2018, 02, 20–29. [Google Scholar]
  42. Shih, J.-L.; Lee, C.-H.; Yang, C.-S. An adult image identification system employing image retrieval technique. Pattern Recognit. Lett. 2007, 28, 2367–2374. [Google Scholar] [CrossRef]
  43. Liu, C.; Wu, X.; Yu, X.; Tang, Y.; Zhang, J.; Zhou, J. Fusing multi-scale information in convolution network for MR image super-resolution reconstruction. Biomed. Eng. Online 2018, 17, 1–23. [Google Scholar] [CrossRef]
  44. Hayou, S.; Doucet, A.; Rousseau, J. On the impact of the activation function on deep neural networks training. In Proceedings of the International Conference on Machine Learning, Long Beach, CA, USA, 9–15 June 2019; pp. 2672–2680. [Google Scholar]
  45. Pan, H. A micro-exploration of the application practice of GIS information technology in geographic research. Resour. Inf. Eng. 2017, 1, 121–122. [Google Scholar]
  46. Lau, M.M.; Lim, K.H. Review of adaptive activation function in deep neural network. In Proceedings of the 2018 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES), Sarawak, Malaysia, 3–6 December 2018; pp. 686–690. [Google Scholar]
  47. Nicolae, A. PLU: The piecewise linear unit activation function. arXiv 2018, arXiv:1809.09534. [Google Scholar]
  48. Simonyan, K.; Zisserman, A. Very deep convolutional networks for large-scale image recognition. arXiv 2014, arXiv:1409.1556. [Google Scholar]
  49. Szegedy, C.; Liu, W.; Jia, Y.; Sermanet, P.; Reed, S.; Anguelov, D.; Erhan, D.; Vanhoucke, V.; Rabinovich, A. Going deeper with convolutions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 7–12 June 2015; pp. 1–9. [Google Scholar]
  50. Qin, F.; Liu, D.; Sun, B.; Ruan, L.; Ma, Z.; Wang, H. Image recognition of 4 alfalfa leaf diseases based on deep learning and support vector machine. J. China Agric. Univ. 2017, 22, 123–133. [Google Scholar]
  51. Zhang, X.; Chen, J.; Zhuge, J.; Yu, L. Fast plant image recognition based on deep learning. J. East China Univ. Sci. Technol. (Nat. Sci. Ed.) 2018, 44, 887–895. [Google Scholar]
  52. Li, X.; Zhang, C.; Li, W. Building block level urban land-use information retrieval based on Google Street View images. GIScience Remote Sens. 2017, 54, 819–835. [Google Scholar] [CrossRef]
  53. Yin, L.; Wang, Z. Measuring visual enclosure for street walkability: Using machine learning algorithms and Google Street View imagery. Appl. Geogr. 2016, 76, 147–153. [Google Scholar] [CrossRef]
  54. Porzi, L.; Rota Bulò, S.; Lepri, B.; Ricci, E. Predicting and understanding urban perception with convolutional neural networks. In Proceedings of the 23rd ACM International Conference on Multimedia, Brisbane, Australia, 26–30 October 2015; pp. 139–148. [Google Scholar]
  55. Glaeser, E.L.; Kominers, S.D.; Luca, M.; Naik, N. Big data and big cities: The promises and limitations of improved measures of urban life. Econ. Inq. 2018, 56, 114–137. [Google Scholar] [CrossRef]
  56. Liu, L.; Silva, E.A.; Wu, C.; Wang, H. A machine learning-based method for the large-scale evaluation of the qualities of the urban environment. Comput. Environ. Urban Syst. 2017, 65, 113–125. [Google Scholar] [CrossRef]
  57. Zhao, J.; Zhong, H. A demographic factor as a determinant of migration: What is the effect of sibship size on migration decisions? J. Demogr. Econ. 2019, 85, 321–345. [Google Scholar] [CrossRef] [Green Version]
  58. Mu, R.; Zeng, X. A review of deep learning research. KSII Trans. Internet Inf. Syst. 2019, 13, 1738–1764. [Google Scholar]
  59. Hao, X.; Long, Y. Street greening: A new walkability evaluation index. Shanghai Urban Plan. 2017, 01, 32–36. [Google Scholar]
  60. Long, Y.; Tang, J. Large-scale Quantitative Measurement of the Quality of Urban Street Space: The Research Progress. City Plan. Rev. 2019, 43, 107–114. [Google Scholar]
  61. Tang, J.; Long, Y. The Measurement of Street Space Quality in the Central District of Megacities—Take Beijing Second and Third Ring Road and Shanghai Inner Ring as examples. Planner 2017, 33, 68–73. [Google Scholar]
  62. Cao, H.; Yang, P.; Long, Y. Innovation and Expansion of Urban Image Cognitive Methods Based on Deep Learning—Take the main urban area of Chongqing as an example. Chin. Gard. 2019, 35, 90. [Google Scholar]
  63. Wu, Z. Artificial intelligence assisted city planning. Era Archit. 2018, 1, 6–11. [Google Scholar]
  64. Zhang, D.; Liu, L. Quantitative evaluation method of visual color based on deep learning—Take the color situation of visual architecture in Shanghai as an example. In Proceedings of the Sharing and Quality—2018 China Urban Planning Annual Conference Proceedings (10 Urban Video), Hangzhou, China, 24 November 2018. [Google Scholar]
  65. Yao, J.; Huang, C.; Liu, P.; Zhang, Y. Research on Recognition and Evaluation of Urban Texture Based on Artificial Intelligence. Hous. Technol. 2019, 11, 9–14. [Google Scholar]
  66. Song, J.; Zhu, T. Parametric architectural aesthetics. Archit. Cult. 2017, 6, 112–113. [Google Scholar]
  67. He, W.; Li, C.; Nie, G.; Yang, L.; Wang, C. Possible applications of deep learning in urban perception—Image Discriminant Analysis Based on Convolutional Neural Network. Int. City Plan. 2019, 34, 8–17. [Google Scholar]
  68. Yamashita, R.; Nishio, M.; Do, R.K.G.; Togashi, K. Convolutional neural networks: An overview and application in radiology. Insights Into Imaging 2018, 9, 611–629. [Google Scholar] [CrossRef] [Green Version]
  69. Nixon, M.; Aguado, A. Feature Extraction and Image Processing for Computer Vision; Academic Press: Cambridge, MA, USA, 2019. [Google Scholar]
  70. Zhang, Y.J. A survey on evaluation methods for image segmentation. Pattern Recognit. 1996, 29, 1335–1346. [Google Scholar] [CrossRef] [Green Version]
  71. Harris, C.; Stephens, M. A combined corner and edge detector. In Proceedings of the Alvey Vision Conference, Manchester, UK, 2 September 1988. [Google Scholar]
  72. Ojala, T.; Pietikäinen, M.; Harwood, D. A comparative study of texture measures with classification based on featured distributions. Pattern Recognit. 1996, 29, 51–59. [Google Scholar] [CrossRef]
  73. Jacobs, I.J.; Skates, S.J.; Macdonald, N.; Menon, U.; Rosenthal, A.N.; Davies, A.P.; Woolas, R.; Jeyarajah, A.R.; Sibley, K.; Lowe, D.G. Screening for ovarian cancer: A pilot randomised controlled trial. Lancet 1999, 353, 1207–1210. [Google Scholar] [CrossRef] [PubMed]
  74. Lowe, D.G. Object recognition from local scale-invariant features. In Proceedings of the seventh IEEE International Conference on Computer Vision, Kerkyra, Greece, 20–27 September 1999; pp. 1150–1157. [Google Scholar]
  75. Lowe, D.G. Distinctive image features from scale-invariant keypoints. Int. J. Comput. Vis. 2004, 60, 91–110. [Google Scholar] [CrossRef]
  76. Viola, P.; Jones, M. Rapid object detection using a boosted cascade of simple features. In Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001, Kauai, HI, USA, 8–14 December 2001; p. 1. [Google Scholar]
  77. Dalal, N.; Triggs, B. Histograms of oriented gradients for human detection. In Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05), San Diego, CA, USA, 20–25 June 2005; pp. 886–893. [Google Scholar]
  78. Bay, H.; Tuytelaars, T.; Gool, L. SURF: Speeded up Robust Features. Lect. Notes Comput. Sci. 2006, 7, 404–417. [Google Scholar]
  79. Too, E.C.; Yujian, L.; Njuki, S.; Yingchun, L. A comparative study of fine-tuning deep learning models for plant disease identification. Comput. Electron. Agric. 2019, 161, 272–279. [Google Scholar] [CrossRef]
  80. Sornam, M.; Muthusubash, K.; Vanitha, V. A survey on image classification and activity recognition using deep convolutional neural network architecture. In Proceedings of the 2017 Ninth International Conference on Advanced Computing (ICoAC), Chennai, India, 14–16 December 2017; pp. 121–126. [Google Scholar]
  81. Lecun, Y.; Bottou, L.; Bengio, Y.; Haffner, P. Gradient-based learning applied to document recognition. Proc. IEEE 1998, 86, 2278–2324. [Google Scholar] [CrossRef] [Green Version]
  82. Krizhevsky, A.; Sutskever, I.; Hinton, G.E. Imagenet classification with deep convolutional neural networks. Adv. Neural Inf. Process. Syst. 2012, 25. [Google Scholar] [CrossRef] [Green Version]
  83. Hinton, G.E.; Srivastava, N.; Krizhevsky, A.; Sutskever, I.; Salakhutdinov, R.R. Improving neural networks by preventing co-adaptation of feature detectors. arXiv 2012, arXiv:1207.0580. [Google Scholar]
  84. Zeiler, M.; Fergus, R. Visualizing and Understanding Convolutional Networks. In Computer Vision–ECCV 2014, Proceedings of the 13th European Conference, Zurich, Switzerland, 6–12 September 2014; Springer: Berlin, Germany, 2014. [Google Scholar]
  85. He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 27–30 June 2016; pp. 770–778. [Google Scholar]
  86. Goodfellow, I.J.; Shlens, J.; Szegedy, C. Explaining and harnessing adversarial examples. arXiv 2014, arXiv:1412.6572. [Google Scholar]
  87. He, K.; Zhang, X.; Ren, S.; Sun, J. Identity mappings in deep residual networks. In Computer Vision–ECCV 2016, Proceedings of the 14th European Conference, Amsterdam, The Netherlands, 11–14 October 2016; Springer: Berlin/Heidelberg, Germany, 2016; pp. 630–645. [Google Scholar]
  88. Srivastava, R.K.; Greff, K.; Schmidhuber, J. Highway networks. arXiv 2015, arXiv:1505.00387. [Google Scholar]
  89. Huang, G.; Liu, Z.; Laurens, V.; Weinberger, K. Densely connected convolutional networks. IEEE Comput. Soc. 2016. [Google Scholar] [CrossRef] [Green Version]
  90. Goodfellow, I.; Bengio, Y.; Courville, A. Machine Learning Basics in Deep Learning; MIT Press: Cambridge, MA, USA, 2016. [Google Scholar]
  91. Geng, X.; Hou, P. Pre-release prediction of crowd opinion on movies by label distribution learning. In Proceedings of the Twenty-Fourth International Joint Conference on Artificial Intelligence, Buenos Aires, Argentina, 25–31 July 2015. [Google Scholar]
  92. Davis, T.A. Algorithm 849: A concise sparse Cholesky factorization package. ACM Trans. Math. Softw. 2005, 31, 587–591. [Google Scholar] [CrossRef]
  93. Wang, Z.; Bovik, A.C. A universal image quality index. IEEE Signal Process. Lett. 2002, 9, 81–84. [Google Scholar] [CrossRef]
  94. Zhao, W. Research on the deep learning of the small sample data based on transfer learning. In AIP Conference Proceedings; AIP Publishing LLC: Melville, NY, USA, 2017; p. 020018. [Google Scholar]
  95. Kivilcim, C.; Duran, Z. A semi-automated point cloud processing methodology for 3D cultural heritage documentation. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2016, XLI-B5, 293–296. [Google Scholar] [CrossRef] [Green Version]
  96. Wang, L.; Wen, C. Traditional villages in forest areas: Exploring the spatiotemporal dynamics of land use and landscape patterns in enshi prefecture, China. Forests 2021, 12, 65. [Google Scholar] [CrossRef]
  97. Nancarrow, J.-H.; Yang, C.; Yang, J. An emotions-informed approach for digital documentation of rural heritage landscapes: Baojiatun, a traditional tunpu village in Guizhou, China. Built Herit. 2021, 5, 1. [Google Scholar] [CrossRef]
  98. Santana Quintero, M.; Awad, R.; Barazzetti, L. Harnessing digital workflows for the understanding, promotion and participation in the conservation of heritage sites by meeting both ethical and technical challenges. Built Herit. 2020, 4, 1–18. [Google Scholar]
Table
Table 1. Comparison of the methods applied in landscape research.
Table 1. Comparison of the methods applied in landscape research.
MethodApplication ContextsRepresentative Literatures
Landscape genetic methodTraditional settlement zoning, feature recognition, tourism planning, and other fields.Van Strien M J, 2012 [32]
Space syntaxUrban renewal, analysis of spatial structure changes, analysis of the city’s diachronic changes, etc.Bafna S, 2003 [3]
Landscape patternConservation of ecosystem, land utilization, industry change, etc. O’Neill R V, 1988 [33]; Wu J, 2004 [40]
ArcGIS analysis methodUsers create, browse, use, and share smart map information online.Jiménez-Perálvarez J D, 2009 [34]; Xiao Y, 2016 [41]
Least resistance modelLand ecological suitability evaluation, cost analysis, landscape protection, etc.Hultman K E, 1979 [35]
Landscape sensitivity analysis methodLandscape planning and design, landscape protection and utilization, area division, etc. Newham L T H, 2003 [36];
ASEB raster analysis methodDesign of Beijing Longcheng Garden, tourism product development, etc. Yang-lian LIU, 2013 [42]
Decision-making laboratory analysis methodRisk assessment of traditional architectural landscapes, application of efficiency curve, etc. Seyed-Hosseini S M, 2006 [38]; Liu H C, 2015 [43]
GIS analysis methodVarious types of spatial analysis, use of maps for tactical research and strategic decision making, etc. Hayou S, 2019 [44]; Nantomah K, 2019 [45]; Lau M, 2018 [46], Nicolae A, 2018 [47]
Table
Table 2. A list of these existing literatures by three categories.
Table 2. A list of these existing literatures by three categories.
CategoryResearch TopicsRepresentative Literatures
Image recognition and evaluation based on deep learningImage object classification, image segmentation, image recognition, face beauty prediction, agricultural pest and disease recognition, plant species recognition, medical imaging diagnosis, migration learning algorithms, etc.Simonyan K, 2014 [48]; Szegedy C, 2015 [49]; Wu R, 2015 [40]; Feng Q, 2017 [50]; Xue Z, 2018 [51]
Application of image recognition in the field of urban and rural constructionStreet greening, architectural features, urban form, walkability status, urban image, urban skyline, urban style, building recognition, intelligent classification of landscape elements, landscape visual quality evaluation, etc.Li X, 2017 [52]; Hu F, 2015 [14]; Yin L, 2016 [53]; Porzi L, 2015 [54]; Glaeser E L, 2018 [55]; Cheng L, 2017 [9]; Liu L, 2017 [56]
Recognition, evaluation, and protection of rural landscapesSubjective and empirical, objective data-based research, and multiple comprehensive research; focus more on the analysis of style features, style renovation planning, style evaluation optimization, style protection and management, etc.Hull IV R B, 1989 [18]; Falade J B, 1989 [17]; Giupponi C, 2006 [19]; Hietel E, 2007 [20]; Milder J C, 2014 [22]; Hart A K, 2014 [21]; Bo L, 2020 [5]; Yong L, 2019 [24]; Gui Y, 2018 [23]; Tie L, 2020 [26]
Table
Table 3. Comparison of image feature extraction techniques.
Table 3. Comparison of image feature extraction techniques.
ModelYearAuthorsAchievement
Harris corner detection algorithm1998C. Harris et al. [71] Obtains the corner points of an image by judging the singular values in the image structure to extract the image features.
Local binary pattern algorithm1996T. Ojala et al. [72] Generates the LBP value by comparing the local pixel value and the centre pixel value and extracts the image texture.
Scale-invariant feature transform algorithm1999David Lowe [73,74] Selects key feature points in different scaled spaces for description and generates image features.
The Haar feature extraction method2001Viola et al. [76] Combined with the AdaBoost classification algorithm to achieve face detection.
Histogram of oriented gradient algorithm2005Dalal et al. [77]Uses edge features for feature recognition combined with SVM classifiers to achieve pedestrian detection.
Speed-up robust features algorithm2008Bay et al. [78]An improvement to the SIFT algorithm, greatly reduces the running time of the program and increases the robustness of the algorithm.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Wang, T.; Chen, J.; Liu, L.; Guo, L. A Review: How Deep Learning Technology Impacts the Evaluation of Traditional Village Landscapes. Buildings 2023, 13, 525. https://doi.org/10.3390/buildings13020525

AMA Style

Wang T, Chen J, Liu L, Guo L. A Review: How Deep Learning Technology Impacts the Evaluation of Traditional Village Landscapes. Buildings. 2023; 13(2):525. https://doi.org/10.3390/buildings13020525

Chicago/Turabian Style

Wang, Tao, Jingjing Chen, Li Liu, and Lingling Guo. 2023. "A Review: How Deep Learning Technology Impacts the Evaluation of Traditional Village Landscapes" Buildings 13, no. 2: 525. https://doi.org/10.3390/buildings13020525

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop