Research On High-resolution Remote Sensing Target Detection And Application Based On Mask R-CNN

Remote sensing image target detection is one of the core issues in remote sensing image processing.It has important research significance and application prospects in the construction of smart cities,the monitoring of urban roads,and the deployment of national defense security.In recent years,with the continuous improvement of the platform and sensor technology,the spatial resolution of remote sensing images has also increased,and the visual difference with natural images has also been decreasing.More and more deep learning methods have been applied to remote sensing targets.Detection.However,large-scale high-resolution remote sensing images are mixed with various complex backgrounds and small target objects.These problems are still the key problems that lead to low detection accuracy.In order to solve the above problems,this article is based on the Mask R-CNN model to carry out research.The specific work and research results are as follows:(1)This paper fine-tunes the Mask R-CNN model to adapt it to remote sensing image target detection,and selects a feature extraction network structure suitable for this paper.At the same time,the remote sensing data set required for this article is introduced.Aiming at the characteristics of wide range,high resolution and large scale of remote sensing images,a new cutting method is designed to preprocess the remote sensing images to ensure that the target in the remote sensing image will not be lost due to excessive scaling of the remote sensing image.More information also effectively avoids the image loss caused by the inability of the image size to accurately match the segmentation size,and improves the detection accuracy of the target.Experiments show that more targets can be detected in the high-resolution remote sensing image after segmentation.(2)Aiming at the difficult points of complex background and too small target objects in remote sensing images that are not easy to detect,this paper improves Mask R-CNN.The attention mechanism module is added to the Res Net-101 network,which is intended to obtain more detection target information during image feature extraction and suppress useless background information.At the same time,the pyramid network is improved,and a bottom-up path is added to integrate high-level information with low-level information,so as to solve the problem of complicated background and small targets caused by the excessive field of view of remote sensing images.Experiments show that after adding the attention mechanism and improving the pyramid,the detection effect of complex background targets in remote sensing images is improved,and the detection of small targets is enhanced.In the same remote sensing data set,the accuracy and other indicators have been improved compared to the original network.(3)Mask R-CNN can not only detect the target but also segment the target at the pixel level.In this paper,Mask R-CNN is used as the building extraction network to extract the binary image of the building,and combined with the U-Net model to detect the change of the building.To solve the problem of smoothing the edge contour of the segmented building image,the parameters of Mask R-CNN are adjusted,and the size of the mask is modified.Experiments have proved that in the building data set and change detection images,the accuracy and recall rate of the model in this paper perform well,and the segmentation accuracy of the adjusted model is higher.