Mold Detection Algorithm Based on Improved YOLOv8n

To address the issues of low detection efficiency and accuracy, as well as the large parameters and slow speed of existing ceramic mold detection models in the ceramic product field, a lightweight ceramic mold detection algorithm based on YOLOv8n is proposed. Our improvements include: Firstly, data augmentation is performed using the GSES method on the 13KCeramics dataset. Secondly, StarNet is employed as the new backbone network to replace the original network in YOLOv8. Subsequently, the Agent attention module is added to some structures of the backbone and neck networks to enhance the model's expressive ability. Finally, Scale and Location Sensitive (SLS) loss is used to replace the original loss function. These improvements enable the model to better learn the scale and position information of ceramic molds during training, thereby further improving the accuracy of object detection. Results show that the improved algorithm achieves an increase of approximately 8.59%, 9.48%, and 0.07 in mAP@50, mAP@50-95, and F1 score, respectively, compared to the original YOLOv8n, reaching 98.7%, 98.12%, and 0.97. The FPS reaches 161 frames, an increase of approximately 13.8%, and the number of parameters is reduced by about 1.1M. These results demonstrate the significant effectiveness of the improved model.