Algorithm for Workpiece Recognition Under Complex Conditions Based on Enhanced YOLOv8

In response to the issue of poor object recognition accuracy in existing deep learning-based object detection algorithms due to factors such as changes in light intensity, complexity of image environment, and movement of capturing devices, the study introduces an enhanced YOLOv8 object detection algorithm. The proposed modification incorporates a Triplet Attention Mechanism module, enabling the model to engage with more comprehensive dimensional feature information. In an innovative approach, the conventional UpSample module is substituted with a DySample module, facilitating an increment in accuracy without imposing additional computational costs. Moreover, the integration of Adaptive Spatial Feature Fusion (ASFF) empowers the algorithm to autonomously select optimal feature combinations, while the introduction of the WIoU loss function further refines detection efficacy. Experimental results show that the improved YOLOv8 algorithm achieves a detection accuracy of 92.7% (mAP) on the T-LESS datasets, which is 2.6% higher than the baseline model. The algorithm achieves higher detection accuracy without sacrificing much detection speed, effectively balancing the accuracy and speed of the algorithm.