Zokor Mounds Retrieval From UAV Multispectral Image Based On Vegetation Index

Grassland is not only an important natural resource for agriculture and animal husbandry in China,but also an important ecological barrier,so it is of vital importance to protect the ecological environment of grassland and maintain the sustainable development of grassland resources.However,in recent years,grassland rodent infestation in China has been rising,and this trend is worrying.Grassland rodent infestation not only aggravates soil erosion and desertification,but also causes plague.The zokors have caused serious damage to the grasslands due to their life under camps,massive gnawing,storage of underground plant roots and extensive mound-building activities.Traditional methods of counting zokor populations,such as manual exhaustion and mound counting,are time-consuming and laborious to operate,costly to survey,and can only be used for small area surveys,making it difficult to meet the requirements of large area implementation and monitoring.Conducting real-time and dynamic monitoring of the distribution of the number of mounds is an important preliminary basis for effectively formulating rodent control measures and preventing the occurrence of plague.Based on the UAV remote sensing image acquisition technology,this experiment was conducted to monitor the mounds of Myospalax aspalax in Xilinguole grassland and Myospalax psilurus in Hulunbuir grassland with low-altitude remote sensing monitoring with multi-height and multi-spectral imaging.The optimal vegetation index,optimal supervised classifier,and optimal aerial height were selected based on the calculation of vegetation index and separation degree within the rodent-infested areas of two different grassland types in spring.The results of the study showed that:(1)According to the analysis of the mean values of vegetation indices,the best differentiation between the mounds and non-mounds of the northeastern zokor is the NDVI vegetation index,and the best differentiation between the mounds and non-mounds of the grassland zokor is the OSAVI vegetation index.For the remaining three indices,the differentiation was not obvious and easily confused.(2)For the classification of aerial images of Myospalax psilurus mound in spring,the overall classification accuracy of support vector machine(SVM)was 99 %,the mapping accuracy was 93 %,and the Kappa coefficient was0.90.The overall classification accuracy of neural network(NN)processing was 99 %,the mapping accuracy was 81 %,and the Kappa coefficient was0.77.In the aerial image classification of spring Myospalax aspalax mound,the overall classification accuracy of support vector machine(SVM)processing was 99 %,the mapping accuracy was 89 %,and the Kappa coefficient was 0.74.The overall classification accuracy of neural network(NN)processing was 99 %,the mapping accuracy was 79 %,and the Kappa coefficient was 0.66.(3)In the study of the spring hazard investigation of Myospalax aspalax and Myospalax psilurus mound by multispectral UAV,it is suggested that the optimal classification processor is support vector machine,and the optimal aerial height is 30m-50 m.