Research On Remote Sensing Image Object Detection Algorithm Based On Deep Learning

Remote sensing image object detection is to obtain image data using remote sensing technology and realize automatic detection of target objects contained in the image with the help of deep learning algorithm,which is widely used in natural resources reconnaissance,military object reconnaissance,urban and rural development statistical monitoring.From the perspective of current research methods,remote sensing image object detection mainly uses deep learning algorithm to carry out automatic feature extraction.However,in general,remote sensing image object detection results are poor due to complex scene,large change of object scale and noise interference and other special factors.Aiming at these problems,this paper conducts an in-depth study on remote sensing image object detection.Based on the single-stage object detection algorithm,two new algorithms suitable for remote sensing image object detection are designed as follows:(1)An object detection algorithm based on one-stage deep channel attention mechanism in remote sensing images is proposed,which is called Deep CA-Net.Aiming at the problem of insufficient extraction of small object semantic features,the algorithm adjusts the residual structure,increases the diversity of small object semantic features,and improves the detection accuracy of small object.Secondly,due to the different importance of different channels in the feature graph in the process of convolution and pooling,this model introduced the channel attention mechanism module after the additional feature extraction layer of the deep channel to realize the attention mechanism of the deep channel.This makes it possible to focus on effective information more accurately and quickly when extracting high-level semantic features of object in complex scenes,thus saving computing resources and improving the accuracy of object detection.(2)An object detection algorithm based on depth feature pyramid cross-module fusion(DC-YOLOv7)is proposed.First,by introducing the convolutional attention module,the algorithm enhances the attention to the features useful for object detection,especially in the image containing noise interference,with strong accuracy.Then,a bottom-up path enhancement module is added after the feature pyramid,which improves the resolution and semantic information difference of the feature map,effectively solves the problem of object scale change,and enhances the characterization ability of the feature pyramid.Finally,by fusing the Extended Efficient Layer Aggregation Networks(E-ELAN)in the feature pyramid with the E-ELAN features in the path enhancement module,we take advantage of the feature information at each level.Improved correlation between front and back features.In order to verify the effectiveness of the two remote sensing image object detection methods proposed in this paper,experiments were carried out on the RSOD and NWPU VHR10 remote sensing data sets respectively,and each module of each algorithm was evaluated.In RSOD,the results of the two algorithms were 95.59% m AP and 96.27% m AP,and in NWPU VHR-10 dataset,the results of 96.25% m AP and 97.16% m AP were obtained,respectively.According to these results,the method proposed in this dissertation solves the problem of remote sensing image object detection and achieves good results,which has certain reference value.