关键词: 图像识别, 景观资源评价, 乡村风景道, 选线规划, 乡村振兴

Abstract: Previous studies on rural scenic byways have depended on spatial remote sensing data and emphasized spatial analysis but have not considered the complicated practical environmental conditions in rural areas. This study concerned natural landscape resources that can highlight rural regionality and scale in water network areas as well as human landscape resources that represent local functionality and heterogeneity. The visual information of the field landscape of rural scenic byways was recognized by machine learning image semantic recognition technology, while invalid landscape resources that could not be truly perceived offline were eliminated. On this basis, a method of route selection for scenic byways that could adapt to the complicated environment in rural areas was obtained, to provide a reference for rural scenic byway construction in water network regions.
　　In this study, the settlement landscape quality, degree of concentration of human landscape resources, diversity of landscape spatial patterns, closeness of symbolic natural landscape elements, proportion of natural landscape resources, and proportion of visible landscape area were chosen as evaluation indexes of landscape resources. The weights of these indexes were determined by an analytic hierarchy process (AHP). With reference to view image acquisition methods in existing studies about scenic byways, scenery images of both sides of the road that contained the positional information were acquired. Landscape elements in images surrounding the roads, including plants, buildings, water, and mountains, were recognized using image semantic recognition technology and were labelled. The area proportions of these landscape elements on both sides of the roads were calculated. The optimization coefficient of route selection for the scenic byway was calculated based on the visual area of different elements and the route selection was optimized in combination with the evaluation results of the landscape resources of roads.
　　Wujiang District in Suzhou City is located in the core belt of the Yangtze River Delta urban agglomeration. It is a typical water network area. In this study, the overall layout of rural scenic byways in Wujiang District, Suzhou City was ultimately built, comprising one horizontal and one vertical axis, four interwoven rings, and multiple segments in series through landscape resource evaluation and route selection optimization. The east-to-west principal line extended along the Taipu River, forming the horizontal main axis that spanned the whole Wujiang District. The south-to-north axis pointed to Wuzhen in the south and the scenic byway area around Taihu in the north, connecting several important nodes including Beimayang and Changyang.
　　In site selection for scenic byways, accurate evaluation and the reasonable use of surrounding landscape resources are crucial to guarantee the maximization of line selection and the development of regional resource values, thereby strengthening overall experiences and the attraction of scenic byways. During the weighting of natural elements (e.g., landscapes, lakes, farmlands, and forests) and cultural elements (e.g., historical sites and landmark buildings), multiple influencing factors, such as data accuracy, regional differences, complicated rural environments, subjective preferences and service demands—need to be continuously considered and adjusted to optimize the evaluation standards and methods. 
　　Image semantic recognition technology based on machine learning can evaluate rural survey data effectively and objectively. The batching processing ability increases planning efficiency and enhances evaluation accuracy from an objective perspective. The route selection scheme after image semantic optimization decreases the difficulties and error rate in practical implementation. Analysis of the area proportion of landscape elements in images can provide an accurate comprehension of the practical environmental conditions on both sides of the road and reveal rural environmental problems influencing landscape resource values. This optimized route design can avoid cost increases caused by a poor environment.
　　The route selection method in the present study is more applicable to water network areas; its applicability to mountainous areas and other landform types requires further verification. In future, the application range of the proposed method could be expanded by more diversified data sources like digital footprints. Furthermore, relevant indexes and parameter settings could be continuously improved to extend its application in rural scenic byway planning and construction.

Key words: image recognition, evaluation of landscape resources, rural scenic byways, route selection planning, rural revitalization