Crack Detection Method Based on an Improved YOLOv8 Model

Crack detection techniques in practical applications often rely heavily on subjective human judgment, which limits their efficiency and consistency. In response to the slow speed and low accuracy of crack detection in industrial production processes, this paper proposes a crack detection algorithm based on an improved YOLOv8 model, developed after extensive experimentation. To enhance the dataset's universality and the model's generalization ability, data augmentation is applied during preprocessing. A small object detection layer is introduced to address the challenge of detecting fine cracks, significantly improving the model's accuracy. Additionally, an Adaptive Spatial Fusion (ASF) module is integrated to reduce the algorithm's computational complexity, optimizing the trade-off between performance and efficiency. The model's precision is further improved by modifying the loss function. Experimental results show that the accuracy of the proposed model on a crack dataset reaches 89.60%, representing a 4.5% improvement over the original YOLOv8 model. Meanwhile, the mean Average Precision (mAP) of this model is 4.4% higher than that of the original V9 model, and 4.9% higher than that of V11. Furthermore, the model achieves a 17.3% reduction in the number of parameters, confirming its effectiveness. These advancements demonstrate that the proposed method offers both high detection accuracy and improved computational efficiency, making it highly suitable for real-time crack detection in industrial settings.