"Defect Detection Dataset"

"Defect Detection Dataset Information: This paper constructs a dedicated dataset for PCB defect detection, comprising 654 raw images collected from actual production lines. These images comprehensively cover five common types of electronic component defects: cold solder joints, component shift, missing parts, insufficient solder, and short circuits. A total of 2,451 defect instances have been carefully annotated using bounding boxes to ensure high-quality ground truth data. To enhance the diversity and robustness of the dataset, various data augmentation techniques such as rotation, flipping, brightness adjustment, and noise injection were applied. The augmented dataset was subsequently split into training, validation, and test sets following a strict 8:1:1 ratio. A distinctive characteristic of this dataset is that each image contains at least two different types of defects, which introduces greater complexity and mimics real-world scenarios where multiple defects often coexist on a single board. This makes the dataset particularly suitable for developing and evaluating multi-label defect detection algorithms in complex industrial environments."