Abstract: improved YOLOv8-based mechanical part detection method. First，the original SPPF module in YOLOv8 is replaced with the AIFI module, which has stronger feature extraction capabilities. Then，a dynamic detection head (Dyhead) module is introduced in the detection head to enhance the modelʹs feature representation ability and generalization performance，while the original C2f module is optimized into the C2f-MLCA module，further improving feature fusion. Finally，a new Wise-Inner-Shape IoU loss function is used in the design of the loss function to effectively improve the localization accuracy of mechanical parts. Experimental results show that the improved YOLOv8 algorithm achieves significant results in mechanical part detection tasks，with Precision，Recall，and mAP@50 reaching 90.8%，84.2%, and 91.7%，respectively. The average precision improved by 2.5 percentage points compared to the base model，with both accuracy and recall also improved，fully verifying the superiority and practicality of the algorithm in complex mechanical part detection tasks.