Rural Landscape Features Extraction And Analysis Based On UAV Imagery

The issues of agriculture,rural areas and farmers are fundamental issues concerning the national economy and the people’s livelihood.Hence,the Chinese government put forward implementing the rural revitalization strategy in 2018.One of the most important principles for the "three rural problems" and implementation of the rural revitalization strategy is to adhere to the harmonious coexistence of man and nature,which requires strengthening the construction of ecological civilization in rural areas,carrying out in-depth governance of rural human residential environment and construction of beautiful livable rural areas,and promoting the green development of rural areas.In recent years,as a tool for acquiring remote sensing data,consumer UAV(unmanned aerial vehicle)has been increasingly applied in various fields.Meanwhile,OBIA(Object-Based Image Analysis)technologies has been widely accepted in the field of remote sensing classification.However,references to the uses of UAV and OBIA to map rural environmental and characterize landscape features are still scarce in the field of research.In this paper,the UAV is used to acquire aerial images of the study area.The DOM(Digital Orthophoto Map)of the study area is segmented in multiple scales.ESP(estimation of scale parameters)tool is used to determine the optimal segmentation scale for different ground objects.Then,on the determined segmentation scale,the extraction methods of living factor information(buildings,road,solar water heater,hardened ground,shadow,rural bare land and rural greening)and rural environment natural factor information(forest land,water body,arable land and rape field)in rural residential areas were studied in detail.Finally,based on the results of the optimal classification of rural environment,the landscape characteristics of rural environment are analyzed with the basic concepts and theories of landscape ecology.The main conclusions are as follows:(1)The ESP tool was used to determine the optimal segmentation scale parameters of ground objects.Finally,the solar water heater was extracted from the segmentation layer with Scale=30,Shape parameter=0.1 and Compactness parameter=0.5.The segmentation layer with Scale=50,Shape parameter=0.2 and Compactness parameter=0.5 was used to extract the information of Wucun residential area.The segmentation layer with Scale=80,Shape parameter=0.2 and Compactness parameter=0.5 was used to extract the information of Wucun natural elements.(2)The rule-based method had a good effect on the extraction of solar water heaters,buildings and rural greening in the rural residential areas.The monitoring of solar water heater is conducive to the promotion and planning of rural solar energy.In this study,the template matching method combined with threshold rule was used to extract the solar water heater,and the PA(Producer’s Accuracy)values reached 92%.By analyzing the impact of training samples and features on the classification results of RF(Random Forest),SVM(Support Vector Machine)and K-NN(K-Nearest Neighbor)classifier,the results indicated that the training set ratio can be set below 10%for high-resolution UAV images(resolution<0.1 m).RF classifier had a more stable performance in the extraction of rural residential areas.Among all the extraction results of rural residential areas,RF classifier had the best result in scheme C(Training samples=440,Number of Features=32),with OA(Overall Accuracy)values of 91.34%and Kappa coefficient of 0.89.Therefore,the RF classifier is preferred in the information extraction of living elements in rural residential areas.(3)Based on rules,RF,SVM and K-NN methods,the OA values were 94.27%,96.8%,95.72%and 96.4%,respectively,and the Kappa coefficients were 0.89,0.93,0.917 and 0.92,respectively.The results indicating that the rule sets and feature sets constructed in this study had a good effect on the extraction of rural natural elements information.RF method had the best performance,compared with SVM,K-NN and rule-based method,the OA values was 1.08%,0.4%and 2.53%higher,and Kappa coefficient was 1.3%,1%and 4%higher,respectively.RF classifier is preferred to extract information of natural elements of rural environment.(4)The classification results of rural environment were analyzed.Natural element land in the study area accounts for 90.57%of the total land area in the study area,among which there are a large number of dominant patches and obvious aggregation in the cultivated land.In Wucun,a small number of farmers grow rape,with a small planting area and scattered distribution in the research area.The woodland in Wucun area was rich in resources.The woodland in the study area was planar and widely distributed,covering an area second only to cultivated land in the study area.For the land used for living factors in the research area,the impervious surface area accounts for 38.22%of the land used for living factors,indicating that the research area had good connectivity,the ground reaches a high hardening level,and the villagers have a good living place to use.The residential density of wu village is at a low level,and the housing area only accounts for 32.02%of the land area for living elements.The greening area of the village accounts for 17.15%of the land area for living elements,indicating that the greening area in the village needs to be further improved.The land utilization rate of Wucun residential area is high,but the corner area of the village has not been effectively used.The land use matrix of the land group in the study area is farmland and forest land,and the combination coefficient value is 448.87,which indicates that the urbanization development of the study area is slow,rural residential areas have not formed a scale,and the main land use mode is still traditional agriculture and forestry.