DL-YOLOv8: Improved YOLOv8 for Cable Sheath Defect Detection

In the task of cable sheath defect detection, traditional detection methods face challenges of low detection accuracy due to multi-scale transformations of damaged targets and susceptibility to background interference. To address this issue, this paper proposed DL- YOLOv8, an algorithm specifically designed for cable sheath defect detection. Firstly, since cable sheath defects exhibit distinct edge features, a Gradient Attention (GA) module is constructed to focus the model's attention on the damaged target areas. Secondly, a new loss function, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\beta_{-}\text{IoU}$</tex>, is designed to optimize the model's localization performance for elongated damaged targets by using a parameter <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\beta$</tex> to assign different sensitivities to the length and width penalty terms in the E IoU loss function. On our self-built dataset, compared to the original YOLOv8 model, DL- YOLOv8 improves the performance metrics AP50 and AP75 for cable sheath defect detection by 4.7% and 11.8%, respectively. The improved algorithm demonstrates higher accuracy and better localization, making it more suitable for the task of cable sheath defect detection.