AWUCD-Net: The Armored Wire Umbilical Cable Surface Defect Detection Algorithm Based on Improved YOLOv7

Abstract

When armored wire umbilicals are used for launching and recovering the seafloor drill, they need to be detected for defects to prevent accidents. Existing wire rope non-destructive detection methods are susceptible to interference, low efficiency, high detection costs, and other problems. Therefore, we propose a network algorithm for defect detection on armored wire umbilical cables based on YOLOv7, abbreviated as AWUCD-Net. Replacing the YOLOv7 backbone with the EfficientViT-M4 backbone reduces computing costs. To enhance the accuracy of image feature extraction and detection in the neck’s ELAN-W module, deformable convolution is employed. To mitigate the impact of bounding box regression imbalance across high- and low-quality samples, WIoU is utilized as a loss function. We created an armored wire umbilical cable dataset and conducted experiments using AWUCD-Net. The results indicate that, in contrast to YOLOv7, precision and recall rates increased by 5.6% and 4%, the mAP@0.5 increased by 4.7%, and the number of parameters and floating point operations decreased by 13.7% and 56%. On the VOC2012 dataset, the mAP@0.5 and the mAP@0.5:0.95 enhanced by 4.4% and 4%, respectively, validating the effectiveness of defect detection on armored wire umbilical cables.