YOLOv3 Target Detection Algorithm Based on Channel Attention Mechanism
Daxiang Li, Chao Huang, Ying Liu · 2021
Aiming at the problems of real-time target detection algorithm YOLOv3, such as insufficient positioning of bounding boxes and difficulty in distinguishing overlapping objects, a YOLOv3 detection algorithm with attention mechanism is proposed. First, a new channel attention module is proposed, which can automatically obtain the importance of each feature channel; then it is combined with the residual module of YOLOv3 to form a new skeleton network, which solves the imbalance of the detection frame distribution in the edge area Question: Finally, in order to improve the detection speed, the loss function of the original YOLOv3 detection network is further optimized. Comparing experiments on the Pascal VOC2007 data set, the experimental results show that compared with mainstream target detection models, the performance of this model is greatly improved, and it also has a good detection effect on overlapping objects. The average accuracy of the model on the test set is as good as The detection speed reached 93.50% and 47.58FPS, meeting the requirements of real-time detection.