UAV Object Detection Model Based on Improved Ghost Module

UAVs with computer vision functions have been widely used in various fields. However, due to the limited memory and computing power of embedded devices, the real-time scene analysis of detection algorithm running on UAV platform is still a challenge. To face the challenge, in this paper, we based on the YOLOV3 detection model by changing the convolution method and using fewer parameters to generate more feature maps. We evaluate our detector on the VisDrone2020 dataset. Compared with the original YOLOV3, we achieved impressive results, including a reduction of 76.38% of the parameter size, 77.95% of floating-point operations, and a considerable accuracy rate and faster inference speed.