Improved YOLOv5 Algorithm Based on CBAM Attention Mechanism
Ruixiang Fan, Qiu Zhongpan · 2022
Due to the presence of tiny targets that have a high incidence of missed detection and false detection as well as the occlusion of cars and people, object recognition in road scenes is difficult. We put up a better YOLOv5 object identification model to address this problem. First, to improve the extraction of significant characteristics from cars and pedestrians while suppressing the detection of generic features, we added the CBAM attention module to the YOLOv5 backbone network. Second, we integrate two hyperparameters into the focal loss function to regulate the weight ratio of positive and negative samples and difficult and easy samples, respectively, in order to maximize the positive and negative samples in the data set and solve the issue of imbalance between difficult and easy samples. The experiment is run on the KITTI public dataset, and the mAP value is utilized as the evaluation metric. The experimental findings demonstrate that, when compared to the previous YOLOv5 model, the suggested model enhanced mAP by 1.1%, demonstrating the new model's efficacy.