Improved Character Detection of Car Brake Disc Based on YOLOv8

To address the high false negative and false positive rates in automotive brake disc character detection, we propose an improved YOLOv8 model for brake disc character detection. The RFCBAMConv convolution is used to replace the standard convolution, alleviating the limitations of parameter sharing in traditional convolutions, such as insufficient adaptability to multi-scale targets and difficulty in recognizing curved or irregular-shaped objects, thereby enhancing the model's sensitivity to key features. A new C2f_ RC module is designed, which integrates receptive field attention convolution and the CBAM attention mechanism within the C2f module to improve the model's feature representation capability. The CARAFE operator is used to replace the nearest neighbor upsampling operator in the original YOLOv8, expanding the receptive field and enhancing the ability to aggregate contextual information. Experimental results show that, compared to the original algorithm, the average precision increased by 0.6%, mAP50 improved by 0.2%, and mAP50-95 improved by 2.4%. Compared to YOLOv7 and YOLOv9, the proposed algorithm achieved a 3.5% and 4.3% increase in mAP50-95, respectively.