Research On Human Disturbance Activity Detection Method Of High-resolution Image Based On Deep Learning

Human disturbance activity detection of remote sensing image is to analyze and obtain the process of surface changes caused by human factors by comparing two remote sensing images of the same area in different periods,which is also the change detection of remote sensing image in essence.With the progress of science and the development of remote sensing technology in China,it is very easy to obtain high-resolution remote sensing images.Change detection of high-resolution remote sensing images also plays an important role in land use renewal,disaster relief,environmental monitoring,remote sensing supervision,dynamic monitoring and other fields.In this paper,semantic segmentation model in deep learning is used to detect human-disturbed activity changes in two periods of high-resolution remote sensing images.The main contents of this paper are as follows:(1)Based on the selected aeronautical building data set and the self-made Long nan data set,artificial disturbance detection experiments are carried out on different deep learning models,and the experimental results are analyzed by using the evaluation indexes in the classification algorithm.The results show that the F1 values of UNet model are 0.805,0.792,0.763 and 0.652 respectively on the two sets of data sets.The missed detection rate is low,the prediction graph is closer to the change reference graph,and the detection performance is better than FCN,Seg Net and Siam UNet models.(2)In view of the problem that deep neural network is easy to produce model degradation and UNet model is easy to ignore the spatial relationship between pixels,which leads to fuzzy edge details in the generated change detection graph,the UNet model is improved.In this paper,aggregation residual module and attention module are introduced into the UNet model to form a new detection model--Res X-Att-UNet model.Two sets of data sets are used to detect human-disturbed activity changes,and the experimental results of FCN,Seg Net,Siam UNet and UNet are compared and analyzed.The analysis results show that the detection accuracy of Res X-Att-UNet model proposed in this paper is 89.5%,and the calculated F1 value is improved by12.8%,10.7%,19.3% and 5.6%,respectively.The detected anthropogenic disturbance area is more complete,and the change detection accuracy of remote sensing image is improved.There are 41 figures,12 tables and 77 references.