Nondestructive Detection of Soluble Solids Content in Apples Based on Multi-Attention Convolutional Neural Network...

Area of Science:

• Agricultural Science
• Computer Vision
• Spectroscopy

Background:

• Soluble solids content (SSC) is a key indicator of apple quality and market value.
• Accurate and non-destructive methods for SSC detection are essential for the fruit industry.

Purpose of the Study:

• To develop and evaluate a deep learning model for detecting soluble solids content (SSC) in 'Fuji' apples.
• To combine hyperspectral imaging with a novel multi-attention convolutional neural network (MA-CNN) for enhanced feature extraction.

Main Methods:

• Acquisition and preprocessing of hyperspectral images from 570 apple samples.
• Development of a MA-CNN model incorporating channel attention (CA) and spatial attention (SA) modules.
• Hyperparameter optimization using the Bayesian optimization algorithm (BOA).

Main Results:

• The MA-CNN model achieved high prediction accuracy for SSC detection.
• Rp2 value of 0.9602 and RMSEP value of 0.0612 °Brix were obtained with the MA-CNN model.
• MA-CNN outperformed other evaluated models, including CA-CNN, SA-CNN, and mainstream approaches.

Conclusions:

• MA-CNN combined with hyperspectral imaging provides an effective and rapid method for detecting apple SSC.
• The attention mechanisms in MA-CNN enhance the adaptive focus on relevant spectral-spatial features.
• This technology holds significant potential for non-destructive apple quality assessment.