Application Of Convolution Neural Network In Remote Sensing Image Recognition

Remote sensing image recognition is an important research topic in the field of remote sensing image processing.Its main task is to detect the target candidate region from the remote sensing image and give the category of the candidate region.Traditional remote sensing image recognition algorithms can achieve good results in simple scenes,but in complex background,often get poor results.With the development of depth learning,deep convolution neural network can extract the deep semantic features of remote sensing images,which greatly improves the accuracy of remote sensing image recognition.In this paper,a feature selection algorithm based on multi-objective bare-bones particle swarm optimization(MOBPSO)is proposed.The algorithm is applied to Faster R-CNN,and the candidate region features extracted from Faster R-CNN algorithm are selected,which improves the accuracy of remote sensing image recognition.The main tasks include the following three aspects:(1)To solve the problem of redundancy in remote sensing image features extracted by convolutional neural network,a feature selection algorithm based on MOBPSO is proposed.Compared with the two classical multi-objective feature selection algorithms on 12 standard datasets,the accuracy and running time of the MOBPSO based feature selection algorithm are better than the other two algorithms on most datasets,which can effectively remove redundant features.(2)Deep convolution neural network model is used to extract features from remote sensing images,and feature selection algorithm based on MOBPSO is used to select features from the extracted remote sensing images.Compared with the other two classical feature selection algorithms,the proposed feature selection algorithm based on MOBPSO can achieve smaller classification error rate in remote sensing images.(3)Aiming at the redundancy of the classification features used in Faster R-CNN algorithm,the Faster R-CNN algorithm is improved.Before classification and regression network classifies the target region,referring to the MOBPSO feature selection algorithm,the target region features extracted by Faster R-CNN algorithm are selected,and the selected features are used as the feature input of classification and regression layer.The improved Faster R-CNN algorithm is applied to remote sensing image recognition.Theexperimental results show that the accuracy of the remote sensing image recognition algorithm based on the improved Faster R-CNN is about 3% higher than that of the non-improved Faster R-CNN algorithm.