A Lightweight Algorithm for Detecting Steel Defects in Small Samples:RBSI-YOLOv8n

Aiming at the current problems of low efficiency and poor accuracy of steel surface defect detection with few samples, as well as the challenges of large number of existing defect detection model parameters and low detection accuracy, a lightweight small-sample steel defect detection algorithm based on YOLOv8n (RBSI-YOLOv8n) is proposed. The method introduces the receptive field attention RFCBAMConv module, and accordingly optimizes the design of RF-C2f module instead of C2f, which effectively retains the spatial and channel features and suppresses the redundant information in order to improve the detection accuracy; changes the Neck to the BCAFPN-SE structure, which reduces the complexity of the model and better handles the cross-scale feature information, thus improving the detection performance; optimizes the loss function, and designs the The Inner-Shape-IOU loss function is optimized and the loss function is designed to improve the convergence speed and precision of the algorithm under small-sample detection objects Experimentally verified on the NEU-DET steel defect dataset, compared with the YOLOv8n algorithm, the RBSI-YOLOv8n precision rate is improved by 8.6%, the recall rate is improved by 0.7%, and the mAP is improved by 3.4%. It effectively reduces the misdetection rate and leakage rate during the actual detection. And the number of parameters is only 66.7% of the original model, and the computational amount is 95% of the original model, the model is more lightweight and meets the needs of more scenarios. The results show that the algorithm achieves a better balance between detection accuracy and lightweighting under the steel small sample detection object, and provides a reliable reference for edge terminal devices with high accuracy and low computing power deployment.