Hyperspectral Imaging-Based Deep Learning Method for Detecting Quarantine Diseases in Apples.

Area of Science:

• Agricultural science
• Plant pathology
• Spectroscopy

Background:

• Manual inspection of apples for quarantine diseases is slow and limits automation.
• Distinguishing similar disease symptoms in RGB images is challenging for customs.
• A need exists for rapid, non-destructive detection systems at ports.

Purpose of the Study:

• To develop a fast, non-destructive system for detecting common apple quarantine pathogens.
• To differentiate subtle disease signatures using hyperspectral imaging.
• To enable automated import-export control of apples.

Main Methods:

• Hyperspectral images (400-1000 nm) were acquired using a close-range camera.
• Reflectance curves were analyzed for apples with varying disease stages.
• A specialized convolutional neural network (CNN), HSC-Resnet, was designed for hyperspectral data.

Main Results:

• HSC-Resnet achieved a precision of 95.51% in detecting apple quarantine diseases.
• The system effectively separated subtle spectral differences between pathogens.
• Stage-dependent spectral variations were quantified.

Conclusions:

• Hyperspectral imaging combined with CNN offers a promising solution for rapid, non-destructive disease detection.
• This technology can significantly enhance apple import-export management and biosecurity.
• Accurate identification of quarantine pathogens is achievable at the port scale.