To address the issues of missed detection and false alarms in small-target defect identification during the construction of cable intermediate joints, this study proposes an enhanced detection algorithm based on an improved YOLO11s model. First, a panoramic imaging system is established via multi-camera collaborative acquisition and SIFT (Scale-Invariant Feature Transform) image stitching technology, effectively resolving the issue of defect omission in single-view images. Second, the algorithm incorporates DCNv2 (Deformable Convolutional Networks v2) modules into the YOLO11s backbone network to enhance deformation modeling capabilities for detecting semi-conductive layer misalignment defects. Third, LSKA (Large Separable Kernel Attention) mechanisms are integrated to improve global feature perception of structural defects in cable joints. Finally, a P2 small-target detection layer is added to strengthen localization of contaminants in the main insulation layer, surface scratches, and crimping burrs. Experimental results show that the improved algorithm achieves an 80.3% detection accuracy and 70.3% mAP@0.5 for four typical defect types, representing improvements of 4.1% and 16.6% over the original YOLO11s. The proposed method offers an effective technical enhancement for quality inspection during cable intermediate joint construction.