Research On Change Detection Method Of Geological Heritage Based On Unmanned Aerial Vehicle Images

With the development and popularization of Unmanned Aerial Vehicles(UAVs),UAVs photography can help to monitor and protect natural environment.Among them,UAVs patrol monitoring is usually adopted to solve the problems of long patrol cycle and unreachable dangerous areas in manpower monitoring.The task of detecting geological heritage can use UAV s to easily obtain geological heritage images with high temporal,spatial,and spectral resolution,which is conducive to promptly discover changes in geological heritage and taking management and protection measures.A large number of geological heritage images acquired by UAVs require geologists to employ change detection methods to find suspicious changes.However,the overall and partial details of geological heritage captured by UAVs show that the texture details of geological heritage are quite comple x,so the background of the changes is complex,and the details of the objects in the image vary greatly.Existing traditional change detection methods and change detection models based on image classification can no longer meet the requirements of complex background image change detection under high resolution.In order to alleviate above problems,this thesis proposes an improved change detection network model based on dual generators,which is an end-to-end change detection network.The image translation structure based on the conditional generation adversarial network is used to solve geological heritage change detection task;and the architecture of two generators and one discriminator is designed to explicitly force the network to capture various modes to improve the stability of the training of the conditional generation adversarial network;add the class activation mapping network to enhance the change information by the class activation maps which can distinguish change regions.In order to verify the rationality and superiority of the improved model,this dissertation simulates the changes of geological heritage in the second period based on the images taken by the UAVs in the first period,constructing a geological heritage change detection dataset,and evaluate the proposed approach on this dataset.Experiments show that the model proposed in this thesis outperforms traditional change detection methods and image classification-based change detection models in the geological heritage change detection task of UAVs aerial imagery.A higher Recall and lower FNR guarantee low FPR value.The experimental results show the superiority of the proposed model.