Research on non-destructive detection model of tomato fruit quality based on electrical properties and machine...

Area of Science:

• Agricultural Science
• Machine Learning
• Sensory Science

Background:

• Traditional methods for assessing internal tomato quality are destructive and time-consuming.
• There is a need for rapid, non-destructive, and accurate methods for agricultural product quality assessment.

Purpose of the Study:

• To develop a novel predictive method for non-destructive detection of internal tomato quality indicators.
• To integrate Long Short-Term Memory Autoencoder (LSTMAE) and XGBoost for enhanced prediction accuracy.

Main Methods:

• Collected electrical parameters (capacitance, resistance, quality factor) from 300 tomato samples.
• Performed physicochemical analysis for vitamin C, soluble sugar, soluble protein, and titratable acidity.
• Developed a LSTMAE-XGBoost model using electrical and physicochemical data for non-destructive quality prediction.

Main Results:

• The LSTMAE-XGBoost model achieved high prediction accuracy, with coefficients of determination of 0.805 (vitamin C), 0.945 (soluble sugar), 0.838 (soluble protein), and 0.845 (titratable acidity).
• The model demonstrated superior performance compared to traditional machine learning models and improved feature extraction by up to 14.3%.
• The model efficiently predicts all four internal quality indicators simultaneously.

Conclusions:

• LSTMAE-XGBoost is an effective ensemble model for non-destructive detection of internal tomato quality indicators.
• This method offers significant advancements for fruit quality assessment in the horticultural industry.
• The proposed model provides efficient technical means for agricultural product quality evaluation.