Vehicle Detection Technology For Remote Sensing Satellite Images Based On Deep Perception

With the rapid development of satellite remote sensing technology,the resolution of remote sensing images is improving day by day,and small remote sensing targets such as vehicles can be clearly shown in large scale remote sensing images.As a major means of transportation,vehicles have important applications in both civil and military fields,and are the key targets for remote sensing image target detection.Compared with natural images,the scale of remote sensing images is larger and the scenes are complex and changeable,which increases the difficulty of remote sensing target detection;in the detection of vehicle targets in remote sensing images,the pixel ratio of targets is small,and the scale of vehicle targets in remote sensing images of different resolutions is also different;and there are often scenes of dense parking of vehicle targets.The existing target detection methods based on natural scenes are difficult to effectively detect the vehicle targets in remote sensing images.Therefore,based on the current theoretical knowledge of deep learning image target detection,we propose a vehicle detection technology for remote sensing satellite images based on deep perception to address the difficulties related to vehicle target detection in remote sensing images.First,for the characteristics of complex and diverse remote sensing image scenes and small percentage of vehicle target pixels,this thesis proposes to introduce a channel attention mechanism,construct the connection between each feature channel,use learning to get the importance degree of each feature channel,and then optimize the features that are not useful for the current task,so as to adjust the feature response value of each channel and make the extracted features more directional,thus improving the detection effect.Second,to address the problem of large differences in the scale of vehicle targets in remote sensing images with different resolutions,this thesis proposes that a method based on weighted bidirectional feature pyramid fusion is used in the feature fusion stage.By constructing top-down and bottom-up bi-directional channels in addition to forward propagation,the feature information from different scales of the backbone network is fused,thus effectively solving the problem of large scale differences of vehicle targets in remote sensing images with different resolutions.Again,for the problem that vehicle targets often appear in remote sensing images with dense parking scenes,this thesis proposes a method based on discrete angle classification.By converting the regression problem of rotation angle into a classification problem to obtain more accurate detection results.In addition,using a rectangular box with rotation angle to describe the target position can reduce the overlapping of bounding boxes caused by dense parking of vehicle targets,which in turn effectively improves the detection performance of such scenes.Finally,based on the above key technology development,a vehicle detection technology for remote sensing satellite images based on deep perception algorithm is formed,and the effectiveness of the proposed method is verified by conducting qualitative and quantitative analysis experiments using the vehicle subset of the large publicly available remote sensing dataset DOTA and the collected vehicle dataset from Goole Earth.