Remote Sensing Image Target Detection Based On Multi-scale Feature Fusion And Visual Attention Mechanism

Target detection is one of the classic tasks in the field of computer vision,and it has a broad application space in unmanned vehicle driving,traffic control,agriculture,and military.Target detection algorithms can be divided into traditional target detection methods based on hand-designed features and target detection methods based on deep learning to automatically extract features.At present,the latter is divided into single-stage algorithms based on regression and two-stage algorithms based on candidate regions.In the field of remote sensing image target detection,the main challenges faced include messy background information,large target scale gap,many small targets,and target deformation.Therefore,this article has conducted research and exploration to address these challenges,and the main research results are as follows.Fristly,aiming at the characteristics of remote sensing images with a lot of background information and interference,a remote sensing image target detection method based on residual channel attention network is proposed.The residual channel attention mechanism is introduced,combined with multi-scale feature fusion module,through weight The parameter space weights the original channel,increases the weight of the important target area channel,and improves the detection effect of the detection model on the remote sensing image.Secondly,aiming at the characteristics of remote sensing images with large target scales and many small targets,a remote sensing image target detection method based on multi-scale feature fusion network is proposed,which uses a multi-scale feature fusion method to include deep layers with rich semantic features.Features are combined with shallow features that contain many details,and a spatial channel attention module is added,so that the network can better learn the features of small targets,and improve the detection effect of the detection model.Thirdly,aiming at the situation where there are many small targets in remote sensing images and deformed,a remote sensing image target detection method based on deformable full convolutional network is proposed.The remote sensing image target detection method based on full convolutional network is equipped with deformable convolution And deformable pooling,to improve the accuracy of feature extraction of the target,so that the network can learn more features of the target,and reduce the learning of useless background information.Finally,sufficient experiments have been carried out on the remote sensing image data set NWPU VHR-10 data set and UCAS-AOD data set.The experimental results show that the algorithm proposed in this paper is compared with the baseline model and classic algorithm in the NWPU VHR-10 data set.,UCAS-AOD data set has been consistently improved,which verifies the effectiveness of the algorithm.