In modern industrial environments, target detection and target tracking play an important role. When dealing with complex industrial scenes, the current popular target tracking algorithms have the problem of insufficient tracking accuracy for objects with the same features and low robustness under occlusion. This paper introduces TA-You Only Look Once version 8 small (TA-YOLOv8s). This model introduces the Triplet Attention module in the YOLOv8s framework, captures cross-dimensional interactions through three branch structures, and contains channel information and spatial information at the same time. It overcomes the problems of channel and space separation calculation and failure to consider cross-dimensional interactions and dimensionality reduction in common attention methods, and can better express network features. At the same time, a backtracking algorithm (Recall Track) is proposed. To solve the problem of tracking loss caused by occlusion. Experimental results show that our method achieves an accuracy of 90.5% for target tracking when the occlusion ratio is 36%, surpassing the existing technology.