Enhancing Efficientnet-Yolov4 For Integrated Circuit Detection On Printed Circuit Boards

• Main Author: Tay, Shiek Chi
• Format: Thesis
• Language: English
• Published: 2024
• Subjects: QA75.5-76.95 Electronic computers. Computer science
• Online Access: http://eprints.usm.my/63597/

Automated visual inspection of printed circuit boards (pcbs) is vital for ensuring the quality and functionality of pcbs throughout the manufacturing process. Accurately detecting integrated circuits (ics) on pcbs presents a significant challenge in automated inspection due to the wide range of component sizes and types, as well as various printing and markings on the pcb, which complicate object detection. This thesis addresses these intricacies by proposing an improved algorithm, efficientnet-yolov4. The research methodology combines the high-performance feature extraction capabilities of efficientnet as the backbone network with the precise object localisation capabilities of yolov4, a dual advantage unique compared to other methods that may rely on less sophisticated localisation algorithms. To ensure the model's generalisation ability, various data augmentation techniques, such as blur, grid distortion, and random brightness adjustments, were employed to simulate real-world variations. Extensive experiments and evaluations demonstrate the proposed algorithm's effectiveness and robustness in complex pcb layouts, as well as its adaptability to varying colour and brightness randomness, surpassing the performance of other pcb inspection models.