BFL-YOLOv8: surface defect detection algorithm for titanium alloy plates based on improved YOLOv8

Defect detection in titanium alloy plates is a crucial step in industrial production, but surface defects are often too minute and subject to lighting, leading to missed detections in industrial environments. To address this issue, an improved defect detection algorithm, BFL-YOLOv8, based on the you only look once v8 (YOLOv8) model, is proposed to detect common surface defects on titanium alloy plates, such as cracks, spots and scratches. The system incorporates the BiFormer attention mechanism to enhance the ability of the model to extract and process image features, integrates the bidirectional feature pyramid network (BiFPN) for weighted fusion and bidirectional cross-scale connections and further uses the regularisation flow technique of the LLFlow algorithm to eliminate the interference of highlights and shadows in the dataset. The experimental results show that the BFL-YOLOv8 achieves a mean average precision (mAP) of 93.8% on the titanium alloy plate defect dataset, an 8.6% improvement over the original YOLOv8 model, and balances detection accuracy and speed well. It demonstrates excellent detection abilities compared to other similar target detection models and can be applied to defect detection tasks for titanium alloy plates in various complex environments.