Improved deep convolutional algorithm for intelligent maintenance and detection of cables in complex background environments

To enhance the accuracy, robustness, and operational efficiency of cable anomaly recognition and defect localization within complex background environments, this study introduces a multi-scale residual deep convolutional neural network. This model integrates multi-scale convolution to improve feature representation for various defect sizes and incorporates residual connections to optimize gradient propagation while retaining low-level texture information. Furthermore, channel compression and joint classification–localization optimization are utilized to facilitate collaborative learning for image-level anomaly classification and pixel-level defect localization. Experimental validation was performed using the publicly available CableInspect-AD dataset, comparing the proposed model against conventional supervised convolutional networks and established industrial anomaly detection methods. The results demonstrate that the proposed model achieves superior overall performance in classification accuracy, anomaly region localization, and stability across repeated trials, evidenced by an F1-score of 0.971 and an image-level Area Under the Receiver Operating Characteristic Curve of 0.973. Controlled perturbation experiments further confirm the model's stable detection performance amidst occlusion, blur, noise, illumination variations, and image compression. Efficiency analysis indicates that the proposed model offers low inference latency and moderate memory consumption, without compromising its robust anomaly recognition and defect localization capabilities. These findings collectively suggest that the proposed method strikes an advantageous balance between detection accuracy, environmental adaptability, and computational efficiency, thereby offering significant technical support for intelligent cable inspection and defect detection in challenging environments.