Early detection of Blossom-End Rot in green peppers using fluorescence and normal color images in visible region

Abstract

Green peppers, a globally renowned vegetable known for their nutritional value, frequently encounter Blossom-End Rot (BER), a prevalent issue causing postharvest losses. Despite its significance, limited research has addressed the classification of BER-affected fruits from healthy ones. This study introduces a fluorescence imaging-based sensing approach designed to detect the presence of BER areas within captured fluorescence and color images of green peppers. Conducted across a two-year span in three Japanese greenhouses, this research focused on gathering BER-affected green pepper samples and characterizing their features. Utilizing a spectrofluorometer and fluorescence microscope, distinct characteristics of BER areas were estimated in comparison to healthy regions. A total of 774 image sets, each comprising images captured under various light sources, were analyzed. The findings revealed pronounced cell death and chlorophyll degradation within the BER areas, notably manifesting considerable changes in the G channel of images illuminated by white and 365 nm light sources. A classification model trained on these paired images achieved a 94.9% accuracy, including 86.2% for early-stage BER detection. Compared to the conventional ResNet50-based method, which struggled to detect subtle early-stage BER symptoms, the proposed method demonstrated superior performance. This dual-channel sensing approach offers a cost-effective and practical solution for accurately detecting BER at all stages, providing valuable applications for quality control in grading facilities and on farms.