Ju-LiteMobileAtt: A lightweight attention network for efficient jujube defect classification.

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Surface defect detection is vital for organic jujube quality control.
• Traditional machine vision struggles with diverse defect types.
• Deep learning models present a trade-off between performance and computational cost for edge deployment.

Purpose of the Study:

• To develop a high-precision, lightweight network for real-time edge-based organic jujube surface defect detection.
• To overcome the limitations of existing methods in adaptability and computational efficiency.

Main Methods:

• Proposed Ju-LiteMobileAtt, a novel lightweight network based on MobileNetV2.
• Introduced the Efficient Residual Coordinate Attention Module (EfficientRCAM) for multi-scale feature capture.
• Implemented the Cascaded Residual Coordinate Attention Module (CascadedRCAM) for efficient feature refinement.

Main Results:

• Ju-LiteMobileAtt achieved a 1.72% accuracy improvement over the baseline.
• The proposed network significantly reduced model parameters.
• Demonstrated effective real-time defect detection capabilities on edge devices.

Conclusions:

• Ju-LiteMobileAtt offers a superior solution for organic jujube surface defect detection.
• The lightweight and high-precision network is suitable for practical edge deployment in quality assessment.
• The novel attention modules enhance feature representation and model efficiency.