Published Paper Details:

ABSTRACT -- We present a technique for identifying surface flaws in railroad tracks using the YOLOv8 model, which is designed to overcome the challenges of detecting small and hidden targets. To improve the model's attention mechanism, we replace the original folding of YOLOv8n with the SPDCONV block while maintaining the backbone network and the overall architecture of the original model. We also integrate the EMA Atones mechanism into the neck component, enabling the model to leverage data from various attributes and enhancing its feature representation capabilities. Additionally, we replace the original CIOU loss function with YOLOv8's Focus SIOU loss function, which adjusts the weights of positive and negative samples to better penalize difficult examples. This modification significantly enhances the model's ability to detect challenging instances, ensuring that each target receives more focused attention from the network. As a result, the model's overall performance and effectiveness are improved. Experimental results provide compelling evidence of the improved algorithm's enhanced accuracy and recall. Compared to the original YOLOv8n model, the extended version achieves average accuracy, recall, and precision rates of 93.9%, 93.7%, and 91.1%, respectively, reflecting improvements of 3.6%, 5.0%, and 5.7% . Notably, these gains are achieved without increasing the number of parameters or the size of the model. The improved method demonstrates high effectiveness in detecting surface flaws on railroad tracks.