Lightweight YOLOv4 Algorithm for Remote Sensing Image Detection

Abstract

Remote sensing images have the characteristics of complex backgrounds, high resolution, and small targets. Although the existing object detection algorithms can improve the detection accuracy, there are generally problems such as a large number of model parameters, high computational cost, and poor real-time performance. Aiming at the above problems, this paper designs a lightweight object detection algorithm GSC-YOLO based on YOLOv4 to achieve fast and accurate detection of remote sensing images. First, Ghostnet is used as the feature extraction network of GSC-YOLO to reduce the number of parameters and improve the detection speed; Secondly, the improved shuffle attention mechanism is introduced in the prediction head to make the model pay attention to important information and improve the detection accuracy; Finally, the Confidence Propagation Cluster algorithm CP-Cluster is used to post-process the prediction frame to improve the object recognition. Taking the preprocessed DOTA dataset as the experimental object, the experimental results show that the GSC-YOLO algorithm has a detection accuracy of 93.44%, a detection speed of 58 frames per second, and a model size of 43.65MB. Compared with the remote sensing image object detection algorithm based on YOLOv4, the detection accuracy is increased by 3.93%, the detection speed is increased by 1.87 times, and the model size is reduced by 5.62 times, which is more suitable for deployment on devices with limited resources.