Ultrafast on-site adulteration detection and quantification in Asian black truffle using smartphone-based computer...

Area of Science:

• Food science and technology
• Artificial intelligence in agriculture
• Computational biology

Background:

• Asian black truffle (Tuber sinense) is a valuable edible fungus.
• Truffle adulteration poses a significant economic and quality control challenge.
• Existing identification methods can be slow, destructive, or lack automation.

Purpose of the Study:

• To develop a fast, non-destructive, automatic, and intelligent method for identifying Asian black truffle (BT).
• To create a lightweight convolutional neural network (FastBTNet) for efficient smartphone deployment.
• To enable absolute quantification of adulteration in BT.

Main Methods:

• Development of a novel lightweight convolutional neural network (FastBTNet) incorporating knowledge distillation.
• Integration of a fast object location technique for adulteration quantification.
• Utilizing Grad-CAM for model interpretability and a greenness assessment.

Main Results:

• FastBTNet achieved 99.0% classification accuracy for BT identification.
• The model demonstrated an 8.5% root mean squared error in predicting adulteration levels.
• Prediction of 1024 samples took only 5.3 seconds, showcasing high efficiency.

Conclusions:

• The FastBTNet model offers a highly accurate and efficient solution for on-site BT identification and adulteration prediction.
• Deployment in a smartphone app ('Truffle Identifier') facilitates rapid, real-time analysis.
• This technology addresses the critical need for reliable truffle authentication in the food industry.