Research on mechanical automatic food packaging defect detection model based on improved YOLOv5 algorithm.

Area of Science:

• Computer Vision
• Machine Learning
• Industrial Automation

Background:

• Traditional manual inspection methods for food packaging defects are inefficient and error-prone.
• Modern industrial automation demands high efficiency and precision in defect detection.
• Existing automated systems struggle with identifying small targets and subtle defects.

Purpose of the Study:

• To develop an enhanced YOLOv5-based model for accurate and efficient food packaging defect detection.
• To improve the identification of small targets and subtle defects in food packaging.
• To enhance the automation and precision of the food packaging inspection process.

Main Methods:

• Integrated a Convolutional Block Attention Module (CBAM) to enhance feature prioritization.
• Employed pyramid and aggregation networks for multi-scale feature fusion to capture diverse defects.
• Optimized the YOLOv5 backbone with a streamlined YOLOv5s model and Adaptive Spatial Feature Fusion (ASFF) for improved feature blending.

Main Results:

• The enhanced YOLOv5 model achieved superior performance with Accuracy (Ac)=0.96, Recall (Re)=0.94, and F1 score=0.94.
• Outperformed original YOLOv5 (Ac=0.82, Re=0.85, F1=0.88), YOLOv5+CBAM (Ac=0.88, Re=0.9, F1=0.89), and YOLOv5+ASFF (Ac=0.94, Re=0.95, F1=0.94).
• The combined model (CBAM+FPN+PANet+ASFF) demonstrated performance comparable to related research works.

Conclusions:

• The proposed enhanced YOLOv5 model significantly improves automated food packaging defect detection.
• The integration of CBAM, multi-scale fusion, and ASFF effectively addresses limitations in detecting small and subtle defects.
• The developed system offers a robust solution for enhancing automation and accuracy in food production quality control.