The manufacturing industry faces a big problem. Detecting surface defects on products is crucial. But current methods often fall short. They struggle with different types of defects and complex backgrounds. This is especially true for small defects that vary in size. These issues lead to poor detection performance. To tackle this, a new algorithm has been developed. It is called EPSC-YOLO. This algorithm aims to boost the efficiency and accuracy of defect detection.

EPSC-YOLO introduces several new features. First, it uses multi-scale attention modules. These modules help identify defects of different sizes. Two new pyramid convolutions are also added to the backbone network. This improves the detection of multi-scale defects. Another key feature is the use of Soft-NMS. This replaces the traditional NMS. Soft-NMS reduces information loss. It also improves the accuracy of detecting multiple targets. This is done by smoothing and suppressing the scores of overlapping boxes.

A new convolutional attention module, CISBA, is also designed. This module enhances the detection of small targets in complex backgrounds. The effectiveness of EPSC-YOLO has been tested on two datasets: NEU-DET and GC10-DET. The results are promising. Compared to YOLOv9c, EPSC-YOLO shows significant improvements. It increases

Spotlight on Surface Defect Detection: A New Approach

Actions