Efficient and Non-Invasive Grading of Chinese Mitten Crab Based on Fatness Estimated by Combing Machine Vision and...

Area of Science:

• Aquaculture
• Computer Vision
• Deep Learning

Background:

• The Chinese mitten crab (Eriocheir sinensis) is a valuable seafood facing increasing market demand.
• Current manual quality grading methods are inefficient, labor-intensive, and costly.
• There is a need for precise, non-destructive grading systems in the aquaculture and food industries.

Purpose of the Study:

• To develop an efficient, automated quality-grading system for Chinese mitten crabs using computer vision and deep learning.
• To quantify key physiological traits including sex, carapace dimensions, and fatness for quality classification.
• To compare the performance of a novel deep learning model against existing methods.

Main Methods:

• Utilized a YOLOv5-seg model integrated with an SE attention mechanism for image analysis.
• Trained the model on 2282 RGB images of crabs, with data augmentation techniques applied.
• Developed an improved conditional factor K for fatness assessment and quality grading.

Main Results:

• Achieved 100% accuracy in sex recognition for the crabs.
• Obtained a mean Average Precision (mAP) of 0.995 for carapace segmentation, outperforming other variants.
• Demonstrated 100% consistency between the proposed automated grading method and manual grading.

Conclusions:

• The developed computer vision and deep learning model provides a precise and non-destructive method for grading Chinese mitten crabs.
• This technology can significantly enhance the efficiency and accuracy of quality assessment in the seafood industry.
• The findings support the implementation of advanced AI solutions for aquaculture and food processing.