Deep Learning and Hyperspectral Images Based Tomato Soluble Solids Content and Firmness Estimation

Area of Science:

• Agricultural Science
• Spectroscopy
• Computer Vision

Background:

• Soluble solids content (SSC) and firmness are critical quality indicators for cherry tomatoes (Solanum lycopersicum).
• Current quality assessment methods are often destructive and time-consuming.
• Non-destructive techniques are needed for efficient and accurate quality evaluation.

Purpose of the Study:

• To develop and validate non-destructive techniques for assessing cherry tomato SSC and firmness.
• To utilize hyperspectral imaging and deep learning for predicting these quality traits.
• To compare the proposed method against existing state-of-the-art techniques.

Main Methods:

• Acquisition of hyperspectral reflectance images (400-1,000 nm) from over 200 cherry tomato samples.
• Image correction and spectral information extraction.
• Development of a novel one-dimensional convolutional ResNet (Con1dResNet) deep learning regression model.

Main Results:

• The Con1dResNet model achieved significant improvements in predicting SSC and firmness.
• The technique demonstrated a 26.4% improvement for SSC and 33.7% for firmness compared to state-of-the-art methods.
• The study highlights the effectiveness of hyperspectral imaging combined with deep learning for quality assessment.

Conclusions:

• Hyperspectral imaging coupled with a Con1dResNet model offers a promising non-destructive approach for cherry tomato quality evaluation.
• This method provides a new, efficient option for detecting SSC and firmness, crucial for fruit quality.
• The findings support the broader application of hyperspectral imaging in agricultural product quality control.