Research On Target Detection Algorithm For Remote Sensing Aircraft Images Based On Improved YOLO

With the improvement of resolution and quality of images obtained from remote sensing satellites,remote sensing images have become one of the data sources for object detection.However,remote sensing images also have some problems,including complex backgrounds,size differences,imbalanced scales,and susceptibility to background interference,all of which can lead to low detection accuracy.In this paper,the YOLOv5 model is optimized for aircraft target detection in remote sensing images to improve the effectiveness of target detection.The specific work is as follows:(1)We propose an improved network model ES-CB-YOLOv5 based on the YOLOv5 algorithm.Firstly,this article optimizes the feature fusion structure of YOLOv5 by fusing features from different levels,thereby improving the expression ability and robustness of features;Secondly,we replace the bilinear interpolation up sampling operation with the CARAFE up sampling module,which can adaptively adjust the size of the Receptive field and retain more details,thus improving the accuracy of target detection;In addition,the attention mechanism is also introduced.Using the ECANet attention network module,the accuracy and computational efficiency of the model can be improved,and the Semantic information of features can be strengthened,thus improving the accuracy of target detection;Finally,replace the SPPF module with the SPPFCSPC module,delete the C3 module after the SPPF module,perform feature fusion and information compression,and further improve the performance of the model.(2)The image sources of the dataset used in this article are mainly the NWPU VHR-10 dataset from Northwestern Polytechnical University and aircraft images from Google satellite imagery.After the images are acquired,the labeling work and data enhancement of the dataset are carried out.Finally,2146 images were obtained for training the improved network model.(3)After 300 training sessions,on a dataset containing 2146 optical remote sensing images with aircraft target information,the improved ES-CB-YOLOv5 model achieved an average detection accuracy of 94.80%and a detection speed of 32.50 frames per second.Compared with Faster R-CNN,SSD,YOLOv5,and YOLOv7,the mAP increased by 6.58%,11.46%,2.61%,and 0.33%,respectively.The experimental results show that the improved ES-CB-YOLOv5 model can accurately detect aircraft targets in remote sensing images.