Nondestructive Detection of Codling Moth Infestation in Apples Using Pixel-Based NIR Hyperspectral Imaging with...

Area of Science:

• Agricultural Science
• Food Science
• Remote Sensing Technology

Background:

• Codling moth (CM) (Cydia pomonella L.) infestation poses a significant threat to global apple production and marketability.
• Early, non-destructive detection of CM infestation is crucial for minimizing postharvest losses and enhancing apple quality.

Purpose of the Study:

• To apply near-infrared (NIR) hyperspectral reflectance imaging for pixel-level detection of CM infestation in Gala, Fuji, and Granny Smith apple cultivars.
• To develop robust and accurate classification models using machine learning and optimal wavelength selection.

Main Methods:

• Utilized NIR hyperspectral imaging (900-1700 nm) for data acquisition.
• Implemented region of interest (ROI) procedures and various machine learning algorithms.
• Employed sequential stepwise selection for optimal wavelength identification to create multispectral models.

Main Results:

• Achieved up to 97.4% classification accuracy for infested vs. healthy apples at the pixel level using a gradient tree boosting (GTB) classifier.
• Identified optimal wavelengths, reaching 91.6% accuracy with only 22 selected wavelengths.
• Demonstrated successful classification across three different organic apple cultivars.

Conclusions:

• NIR hyperspectral imaging shows high potential for non-destructive, early detection and classification of latent CM infestation in apples.
• This technology can significantly aid in quality control and reduce economic losses in the apple industry.
• Optimized multispectral models using selected wavelengths offer a faster alternative for CM detection.