With the acceleration of industrialization, the demand for high quality and reliability in the manufacturing of power electronic devices has become increasingly prominent. This paper proposes a small object detection algorithm based on YOLO-DHGC, specifically designed for defect detection in such devices. By integrating a high-resolution backbone network, DenseHRNet, with a dual-stream structure based on edge-gated branches, the algorithm captures richer detail information and enhances the network’s ability to perceive the boundaries and shape features of small objects. Additionally, the use of a sub-pixel convolution upsampling module reduces the loss of semantic information during feature propagation, thereby improving the reconstruction and localization accuracy of small objects. Experimental results indicate that the algorithm achieves high detection accuracy, robustness, and generalization capability on the PKU-Market-PCB small object defect dataset.