YOLOV8X-HICAUPS: AN ATTENTION BASED APPROACH FOR ACCURATE PCB DEFECT DETECTION

Existing PCB inspection methods are becoming outdated due to their limitations in detecting errors. Computer vision technology offers a more effective, automated solution for maintaining PCB quality by improving problem area identification. In this paper, we introduce YOLOv8x-HICAUps (YOLOv8x-HorNetCBAM-Attention based Up-sampling), an innovative method for PCB defect detection that enhances earlier developments while tackling significant challenges in practical applications. The imperfections in the surface of Printed Circuit Boards (PCBs) during the production negatively impact product quality, directly affecting the stability and dependability of devices. Precisely identifying small flaws on the compact PCB surface amidst intricate backgrounds continues to be a major challenge. Earlier models showed encouraging outcomes but continue to encounter limitations, such as insufficient real-world validation, elevated computational expenses for training, and an emphasis on identifying only minor flaws, which limits their adaptability. YOLOv8x-HICAUps includes multiple crucial innovations to tackle these difficulties, featuring the use of the HorNet backbone, CBAM-driven attention mechanisms, attention based up-sampling, and enhanced detection heads. The HorNet backbone is optimized to capture intricate features from high-density PCB designs and improves the capacity to identify even tiniest flaws. The integration of CBAM will improve spatial and channel attention so that defects like "mouse bites" and "spur" can be accurately identified in intricate PCB environments. Attention-based Up-sampling promotes better preservation of defect characteristics, preventing blurriness and improving overall detection precision. Finally, the simple design of the detection head stresses small-object detection, improving efficiency for immediate deployment on embedded systems. Harmonizing a lightweight design with exceptional accuracy, YOLOv8x-HICAUps is fine-tuned for resourcelimited settings and has enhanced generalizability and effectiveness in practical PCB manufacturing situations. Extensive tests on PCB defect datasets confirm the capabilities of YOLOv8x-HICAUps, demonstrating superior detection accuracy of 98.3% and reliability in comparison to current models, thereby establishing it as a valuable resource for practical PCB quality assessment.