Hyperspectral imaging for detection of codling moth infestation in GoldRush apples

• Published in: Postharvest biology and technology 2017-07, Vol.129, p.37-44
• Main Authors: Rady, A., Ekramirad, N., Adedeji, A.A., Li, M., Alimardani, R.
• Format: Article
• Language: English
• Subjects: Apple infestation; Apples; Artificial neural networks; Butterflies & moths; Classification; Codling moth; Decision analysis; Decision trees; Detection; Discriminant analysis; Hyperspectral imaging; I.R. radiation; Image acquisition; Image classification; Imaging; Infestation; Infrared imaging; Insects; Neural networks; Nondestructive testing; Noninvasive; Reflectance; Wavelengths
• ISSN: 0925-5214; 1873-2356
• DOI: 10.1016/j.postharvbio.2017.03.007

•Potential of using VIS/NIR hyperspectral imaging for detecting codling moth in apple was investigated.•Several multivariate analysis and classification techniques were applied.•Wavelengths selection was conducted using sequential methods.•Feasible application of monitoring codling moth in GoldRush apples was shown. Effective detection of insect infestation is important for preserving quality of fresh fruits like apples. The objective of this research was to study the effectiveness of visible (VIS)/near-infrared (NIR) hyperspectral imaging (HSI) (400–900nm) in the diffuse reflectance mode for detecting and classifying codling moth (CM) infestation in apples. A pushbroom HSI was implemented to acquire hyperspectral images for GoldRush apples of fresh-picked and stored at 4, 10, 17, and 27°C, s for 4 months. Mean reflectance spectra (MRS) were calculated for the images and several classification techniques were applied including linear discriminant analysis (LDA), decision trees, K-nearest neighbor (Knn), partial least squares discriminant analysis (PLSDA), feed forward artificial neural networks (FFNN) in addition to majority voting. The most influential wavelengths were determined using the sequential forwards selection (SFS) approach. The highest classification performance was obtained using decision trees at 5 selected wavelengths (434.0nm, 437.5nm, 538.3nm, 582.8nm and 914.5nm) with overall classification rate of 82% for the test set, and 81% and 86% for healthy and CM-infested apples. This study shows the potential of using VIS/NIR hyperspectral imaging as a non-destructive method for detecting CM infestation in apples.