Abstract: In the micro-assembly process of microwave modules, traditional algorithms such as standard image comparison and gray scale binaryzation used in automatic optical inspection (AOI) for wire bonding face challenges, including difficulties in quality inspection of manual wire bonding, cumbersome programming for complex image detection, and the need for extensive manual reevaluation. In this paper an intelligent inspection model for wire bonding quality in machine vision based on deep learning is proposed. By leveraging intelligent recognition and classification of micro-scale targets and rotating object detection with attention mechanisms, the model achieves solder joint defect detection, amorphous wire inspection, and adaptive defect recognition for wire bonding. Compared with traditional methods, the deep learning-based AOI model reduces the false alarm rate from 4.97% to 1.92% and decreases manual reevaluation workload by 62%. Furthermore, with the iterative accumulation of defect data and continuous model optimization, the proposed approach is expected to further enhance detection accuracy and operational efficiency for wire bonding quality control in industrial applications.