EFEN-YOLOv8: Surface defect detection network based on spatial feature capture and multi-level weighted attention.

Area of Science:

• Computer Vision
• Machine Learning
• Industrial Automation

Background:

• Surface defects significantly impact industrial product aesthetics, quality, and operational efficiency.
• Current deep learning models often struggle with inadequate feature extraction in industrial defect detection scenarios.

Purpose of the Study:

• To introduce EFEN-YOLOv8, a novel framework for enhanced industrial surface defect detection.
• To improve detection accuracy through efficient feature extraction and robust model representation.

Main Methods:

• Developed EFEN-YOLOv8 incorporating a novel [Formula: see text]-FEIoU loss function for improved discrimination and sample balancing.
• Integrated Shallow Attention Convolution (SAConv) for early layer feature localization.
• Employed Large Separable Kernel Attention (LSKA) to expand receptive fields and enhance processing efficiency.
• Utilized Weighted Atrous Spatial Pyramid Pooling (WASPP) for multi-scale feature fusion and richer abstract information capture.

Main Results:

• EFEN-YOLOv8 achieved a 7.4% mAP improvement on the NEU-DET dataset and a 3.3% enhancement on the GC10-DET dataset compared to baseline models.
• Demonstrated consistent performance across 8:2 and 9:1 train-test configurations, validating robust generalization capacity.
• Experimental validation and statistical significance testing confirmed superior performance over existing methods.

Conclusions:

• EFEN-YOLOv8 effectively addresses limitations in industrial surface defect detection by prioritizing efficient feature extraction.
• The framework shows significant potential for real-world industrial applications requiring accurate and reliable defect identification.
• Code and datasets are publicly available to facilitate further research and application.