Research On Remote Sensing Image Target Detection Technology Based On Convolutional Neural Network

Remote sensing images with high spatial resolution and rich contents are constantly emerging,providing an important data source for remote sensing image target detection.Remote sensing image target detection can provide important usable information for military and civil fields and has become a research hotspot in recent years.Convolutional neural network has been widely used in target detection because of its powerful feature extraction ability.This thesis studies the target detection technology based on convolutional neural network.First of all,the main components of convolution neural network are introduced,the common convolutional neural network models are studied,and the two-stage target detection technology based on region and the single-stage target detection technology based on regression are analyzed.Then,from the perspective of optimizing the structure of the network model and reducing the parameters of the network model,a lightweight depth separable residual network is designed by combining depth separable convolution with residual module,and remote sensing image target detection based on the depth separable residual network is studied.The detection accuracy is 76.4% on NWPU VHR-10 dataset,which verifies the effectiveness of the depth separable residual network in image feature extraction.Finally,this thesis studies remote sensing image target detection based on Faster R-CNN.Aiming at the problem that the existing target detection technology cannot give attention to both detection speed and accuracy,this thesis optimizes Faster R-CNN and applies the optimized Faster R-CNN to remote sensing image target detection.The lightweight depth separable residual network is used as the backnone network of Faster R-CNN to reduce the number of parameters in the backnone network model.The multi-layer convolution features in the backnone network are fused after local response normalization to enhance the completeness of target feature information and improve the problem that small targets are easy to miss detection.The network model is trained by combining softmax loss function and center loss function to reduce the changes within categories and increase the differences between categories,so that the network model can learn more different target features.The proposed method is verified on NWPU VHR-10 dataset.Compared with the traditional Faster R-CNN,the detection accuracy of the proposed method is improved by 7.1%.