A Mechanical Part Inspection Method Based on Improved YOLOv8

In the field of industrial visual inspection, traditional horizontal detection box approaches often encounter challenges such as inadequate feature extraction for small-scale components, localization inaccuracies, and reduced detection efficiency when applied to densely arranged micro-mechanical assemblies. This study presents a novel enhancement of the YOLOv8 framework, introducing an algorithm specifically designed to address the precision inspection requirements of industrial environments. Firstly, a globally receptive field-enhanced spatial pyramid pooling module (GRF-SPPF) is proposed, which significantly improves the model’s ability to capture fine-grained features of micro-screw heads—such as cross-groove depth and hexagonal profiles—through multi-scale feature fusion. Secondly, a coordinate attention (CA) mechanism is integrated, combining spatial and channel attention via coordinate decomposition. This mechanism enhances the model’s joint spatial-channel representation capability while simultaneously reducing the number of parameters and computational cost. Thirdly, to effectively manage rotation-sensitive targets, the conventional intersection over union (IoU) metric is replaced with an angle-constrained Skew IoU (SIoU) metric, which substantially improves convergence speed and spatial localization accuracy. Empirical validation using real-world production line datasets demonstrates that the optimized model, with a compact size of 6.4 MB, achieves a 1.5 percentage point increase in mean Average Precision at 0.5 IoU ([email protected]), reaching 89.6%. These results confirm the model’s ability to meet the dual demands of high detection accuracy and real-time performance in industrial inspection scenarios.