Aiming at the prevalent defect issues in the packaging process of integrated circuit chips, a defect detection system based on deep learning for chip packaging is designed. By constructing an efficient image acquisition system, defect detection software based on the YOLOv7 algorithm is developed. Moreover, a set of high-speed turret testing and sorting device suitable for various types of small packaged chips and a control system based on a PC host computer are innovatively designed. This system achieves high-speed, high-precision online detection and recognition of chip packaging defects, while simultaneously completing the automatic rejection of defective chips, effectively enhancing the accuracy and timeliness of product quality information feedback. Experimental results demonstrate that the detection system achieves a recognition accuracy rate exceeding 90%, a detection speed of over 22.5FPS, and operates stably in actual production environments, meeting the requirements of efficient defect detection and rejection in integrated circuit chip packaging engineering.