Abstract: With the rapid development of the integrated circuit industry, bonding leads, as key components connecting internal circuits in chip packaging, their quality directly affects the reliability and stability of electronic products.The traditional methods for detecting bonding lead defects mainly include three categories:mechanical parameter detection, electrical parameter detection and morphological feature detection.Although they have certain applicability and representativeness, they generally have problems such as insufficient detection accuracy, low degree of automation and difficulty in meeting the requirements of complex defect identification.In recent years, with the rapid development of machine vision and deep learning technologies, automated defect detection methods based on image processing and neural networks have gradually become a research hotspot.Firstly, systematically sort out various advanced detection methods such as 2D deep learning, 3D point cloud analysis, and image-point cloud fusion.Secondly, compare the detection performance and applicable scenarios of typical algorithms.Finally, the application potential of the YOLO series models and the 3D deep learning architecture in actual detection is analyzed emphatically, aiming to provide a systematic technical overview and reference for relevant researchers and promote the in-depth application of high-precision and intelligent defect detection technology in semiconductor packaging.