Research On Remote Sensing Image Target Detection Method Based On YOLOv4 Network

With the rapid development and progress of science and technology,more and more remote sensing images with high quality spatial resolution and spectral resolution can be obtained,which provides an extremely important research basis and significance for the application of remote sensing image target detection in agriculture,urban and military fields.However,remote sensing image target detection still suffers from many constraints and limitations of conditions,such as the problem of different shooting angles of targets in remote sensing images,the problem of inconsistent target scales and oversized network models,etc.These problems put forward higher detection requirements and capabilities for remote sensing image target detection technology.To address these problems,the main research work of this paper is as follows.（1）For the problem of difficulty in extracting the features of remote sensing targets by the network due to the specificity and inconsistent scale of the shooting viewpoint in remote sensing images,this paper combines the SE attention mechanism network and CBAM attention mechanism network in the base module of the backbone network respectively.By adding the attention mechanism network in the base module,the feature extraction capability of the backbone network and the focus of the network on the key regions of remote sensing targets are effectively enhanced,and the detection capability of the network on remote sensing targets is improved.（2）For the problem that the multi-scale of targets in remote sensing images leads to the difficulty of detecting different targets at different scales,this paper designs a twoway dense feature pyramid module,which not only improves the utilization of the underlying feature information and reduces the loss of feature information,but also enhances the repeated utilization of the backbone network feature information and improves the network’s feature extraction capability for remote sensing targets,thus improving the network’s detection capability for multi-scale remote sensing targets.（3）For the problem of large number of parameters,slow detection speed and large model weight value of multi-scale remote sensing image target detection network,this paper designs a lightweight structure of remote sensing image target detection network,replacing the complex backbone network with MobileNet-V2 network to make the model lightweight;meanwhile,ECA module and Re Zero network are introduced to improve the target detection capability of the network for remote sensing images without increasing the number of parameters of the network,which achieves a better balance between model lightweight and remote sensing target detection accuracy.