Design And Implementation Of Reservoir Aerial Image Classification And Water-land Intersection Region Segmentation Method Based On Residual Network

With the continuous expansion of the scale of land and water resources utilization in my country,the planning and management of reservoirs and surrounding land is gradually increasing.In order to obtain geographic data information quickly,conveniently,and with high precision,most of the drone aerial photography technology is currently used to conduct specialized data processing based on the captured images,and finally achieve three-dimensional modeling.In the process of 3D modeling of UAV aerial images of reservoir scenes,the UAV image contains a large number of water images,which will interfere with the realization of 3D modeling and affect the final effect of 3D modeling.Therefore,it is very important to remove the water surface area of the water image and the water and land cross image.This paper uses deep learning-based image classification and semantic segmentation technology to automatically detect and remove the water surface area of the aerial image of the reservoir,and propose a method for the classification of the aerial image of the reservoir and the segmentation of the water and land cross area based on the residual network,and design the image based on the residual network Classification and image segmentation model.For images taken by drones,first obtain image feature information through the feature extraction network of the model,and input the obtained feature vectors into a classification network based on global average pooling for classification and recognition,which can achieve accurate image land and water classification.Then input the water and land intersection image obtained based on the classification network into the image segmentation network constructed for the segmentation task of the land and water intersection area,and use the multi-scale fusion method based on the cavity convolution to obtain the segmentation result,and realize the accurate segmentation of the water part of the land and water intersection image.Finally,the water and land classification and segmentation results of the classification and segmentation network are combined to complete the removal of the water surface area of the aerial image of the reservoir.