YOLOv4 Object Detection Algorithm with Efficient Channel Attention Mechanism

Abstract

Channel attention mechanism has been widely used in object detection algorithms because of its strong feature representation ability. The real-time object detection algorithm YOLOv4 has fast detection speed and high accuracy, but it still has some shortcomings, such as inaccurate bounding box positioning and poor robustness. Therefore, we introduced channel attention mechanism into the YOLOv4 algorithm to enhance the feature representation ability of images, and proposed a object detection algorithm with channel attention mechanism. This module firstly carries out global average pooling operation on the features extracted by YOLOv4, and then carries out local cross-channel interactive operation on the feature channels through one-dimensional convolution to enhance the correlation between the features of channels, so as to improve the positioning accuracy of YOLOv4. Our method has achieved good results in the PASCAL VOC dataset. Compared with the original YOLOv4 algorithm, the mAP of this algorithm in the PASCAL VOC test set is improved by 0.62%.