Nondestructive Detection and Quality Grading System of Walnut Using X-Ray Imaging and Lightweight WKNet

Area of Science:

• Agricultural Engineering
• Computer Science
• Food Science

Background:

• Internal quality detection is crucial for agricultural products like walnuts.
• Existing methods for walnut quality assessment face challenges in accuracy and efficiency.

Purpose of the Study:

• To develop a comprehensive method for walnut internal quality detection using X-ray imaging and deep learning.
• To address the limitations of time consumption and parameter redundancy in current detection models.

Main Methods:

• Designed an X-ray machine vision system and constructed a walnut kernel detection (WKD) dataset.
• Developed an effective walnut kernel detection network (WKNet) by integrating Transformer, GhostNet, and criss-cross attention (CCA) modules into the YOLO v5s model.
• Evaluated WKNet performance using mAP, precision, and recall metrics, and measured inference time.

Main Results:

• The WKNet achieved high performance with an mAP_0.5 of 0.9869, precision of 0.9779, and recall of 0.9875 for walnut kernel detection.
• The model demonstrated a fast inference time of only 11.9 ms per image.
• Comparative experiments confirmed WKNet's superiority over state-of-the-art deep learning models.

Conclusions:

• The innovative combination of X-ray imaging and the proposed WKNet offers a highly effective solution for walnut internal quality detection.
• This approach has significant implications for improving walnut quality control processes.