Object Detection Algorithm for UAV Images Based on Lightweight YOLO

Abstract

In the application of YOLOv7 network to the object detection scene of high-resolution UAV aerial images, the problem of slow detection speed has been exposed.Based on the YOLOv7 network, this paper proposes a lightweight improved model by designing a new module called GSConv and Depthwise Separable Conv(DSC) Combined Module(GSDSCCM) which based on DSC and GSConv to replace the E-ELAN module in backbone and integrating the VoVGSCSP module into the head of the network. Experimental results show that the improved network reduces the number of network parameters (Params) and floating point operations (GFLOPs) by 32.8% and 38.6% respectively and leads to precision loss of 3.9% in [email protected] compared to the base model, achieving a significant improvement in detection speed at a lower cost of precision loss.