NIR hyperspectral imaging for the identification of damage caused by Halyomorpha halys on pears: development of classification models

Halyomorpha halys, the Brown Marmorated Stink Bug (BMSB), has become a major threat for pear orchards, causing severe economic losses throughout Southern Europe. Italy, one of the leading European pear producers, has experienced a dramatic reduction in yields of economically relevant cultivars over the past decade due to extreme climate events and outbreaks of pests and pathogens. To address this issue, Near Infrared Hyperspectral Imaging (NIR-HSI) is proposed as a non-destructive technique to automatically discard damaged fruits. The study was performed on organic pears (cv. Abate Fétel and cv. Williams) harvested over two consecutive years. Following an automated annotation procedure that identified punctured and sound areas of the fruits, representative spectra of both Punctured and Sound classes were extracted to build pixel-level classification models. Classification models were developed using Soft Partial Least Squares Discriminant Analysis (Soft PLS-DA), a hybrid approach that combines the advantages of both discriminant and class modelling techniques, and its sparse variant (s-Soft PLS-DA), which incorporates variable selection to achieve a parsimonious solution more suitable for online post-harvest sorting systems. The prediction images were elaborated considering an additional decision criterion based on the minimum detectable defect area, enabling image-level classification. Besides being advantageous for automated sorting systems, image-level classification allowed a comprehensive assessment of models’ classification performances towards modelled and unmodelled (i.e., damage not ascribable to BMSB) classes.