Efficient Multi-Receptive Pooling YOLOv5 with Coordinate Attention Module for Object Detection on Drone

Object detection is the most basic and significant research in computer vision in images, and it is a study to discriminate the position and class of an object. This operation has been continuously researched for the past few years. Object detection performance based on accuracy is gradually improving due to the recent development of hardware such as GPU computing power and cameras. Object detection operations grafted in drones can be implemented in many domains. To perform object detection algorithms in real-time in drones, the applied network must be lightweight. For an algorithm capable of real-time operation on low-cost devices, this paper proposes Efficient Multi-Receptive Pooling YOLOv5 with Coordinate(CAM). Efficient Residual Bottleneck and Efficient Multi-Receptive Pooling make the model lighter by reducing the number of parameters, and the CAM improves the object detection rate of the model. The model is trained using the VisDrone dataset, and the mAP value increased by about 19% to 20.6 mAP, and the number of parameters decreased by about 6% to 1,663,599.