Abstract: Wire bonding, as a critical step in integrated circuit packaging, serves to interconnect various components and chips to ensure proper circuit functionality. Its quality inspection directly impacts product yield rates. To address the issues of low detection accuracy and efficiency in existing bond wire defect detection methods, this paper proposes a novel defect detection model: Depth-YOLO.First, the model reconstructs the input terminal of the YOLOv8 architecture, enabling it to process depth information from input images. Second, an input feature enhancement module is proposed to strengthen the model's capability in extracting depth information and texture features of bonding wires. Subsequently, the C2f module in the original YOLOv8 backbone network is replaced with a C2f_Faster module, effectively reducing model parameters and computational redundancy.Furthermore, a Multi-Dimensional Feature Attention (MDFA) mechanism is introduced to enhance the model's feature extraction capability for dense, complex, and irregular defects, thereby improving detection accuracy. Finally, the original CIoU loss function in YOLOv8 is replaced with WIoU to enhance the model's bounding box judgment accuracy and accelerate convergence speed.To address the lack of publicly available bond wire datasets in related research fields, we developed a specialized depth image dataset named DepthBondingWire. Experimental results on this custom dataset demonstrate that the Depth-YOLO model achieves a 7.2% improvement in mAP@0.5 compared to the original YOLOv8, reaching 98.6%. The proposed model shows superior detection accuracy compared with other mainstream object detection models.This methodology effectively enables high-precision automated inspection of semiconductor bond wires and can be extended to defect detection in other critical integrated circuit manufacturing processes. The DepthBondingWire dataset and experimental validation confirm that the proposed approach achieves 98.6% mAP@0.5, outperforming original YOLOv8 by 7.2% while maintaining advantages over other mainstream detection models. This research provides an effective solution for automated precision inspection in semiconductor bonding processes, with potential applications extending to quality control across various integrated circuit manufacturing stages.