Abstract Anthracnose disease, caused by the fungal pathogen Colletotrichum spp., constitutes a major postharvest challenge to the quality and commercial viability of ‘Namdokmai Sithong’ mangoes. Traditional methods for disease detection, which often rely on subjective visual inspection, are only effective once symptoms are externally visible. This study was conducted to develop a rapid, objective and non-destructive method for the early classification of this latent disease using near-infrared spectroscopy (NIRS). Mango fruits at commercial maturity (100–105 days after flowering) with export-grade quality from a certified orchard were used to simulate natural infection. Each fruit was half-dipped horizontally for 1 min in a 1.4 × 10 5 spores/ml suspension of Colletotrichum spp. Near-infrared reflectance spectra (800 – 2500 nm) were subsequently acquired from 52 inoculated and 48 uninoculated mango fruits at 24 h over a 4-day period. An artificial neural network (ANN) classifier with k -fold cross-validation ( k = 5) achieved a precise classification result through the first-derivative (1D) spectra at the early stage of inoculation (24 h). A key finding was the absence of any false positive predictions from the 1D-ANN model during the crucial initial incubation period (24, 48 and 72 h), which is critical for preventing the unnecessary rejection of healthy fruit. While the overall accuracy of most ANN models exhibited a slight decline at 96 h, the 1D-ANN classifier maintained a perfect accuracy of 100% without generating any false negatives. This study conclusively demonstrates the feasibility of using reflectance spectroscopy for the early and accurate detection of anthracnose disease, offering a valuable and reliable tool for effective postharvest disease management and mitigating significant financial losses within the Thai mango industry. Significance of the study What is already known on this subject? Anthracnose disease, a significant postharvest disease caused by the fungal pathogen Colletotrichum spp., typically leads to quality degradation and postharvest losses in various fruits globally. The traditional method for disease detection generally relies on visual inspection, which is ineffective for latent infections, as it often fails to identify the disease before external symptoms appear. For this reason, near-infrared spectroscopy (NIRS) has been considered as a promising non-destructive technique, performed by measuring a sample’s light absorption and scattering to reveal its chemical composition. The resulting spectral data are subsequently analysed using artificial neural networks (ANN), the powerful classifiers for identifying and analysing complex data to enable the early detection of the disease at a non-visible stage. What are the new findings? This study confirms that early and non-invasive detection of anthracnose disease is possible using near-infrared spectral information along with an ANN as a classifier. The research found that the first-derivative (1D) data pre-processing method was crucial for achieving high classification accuracy, as it enhanced the subtle spectral alterations that occurred at early infection. The developed 1D-ANN model was highly robust and reliable, giving 100% accuracy and providing no false negatives for up to 96 h after inoculation. Furthermore, the ANN classifier also provided no false positive predictions, which is a significant advancement over current visual inspection methods and is critical for preventing the unnecessary rejection of healthy fruit. What are the expected impacts on horticulture? Implementing early anthracnose disease classifiers can significantly reduce postharvest losses in mango export and import markets. The NIRS-ANN classification model allows the selection of infected fruit at packinghouses, aiding operators in managing the commodity efficiently. It also prevents the further spread of fungal pathogens through the supply chain. This objective and quantitative method for assessing fungal infestation in mango also leads to improved quality control, providing more consistent product quality and building consumer trust in competitive global markets.
Sonthiya et al. (Thu,) studied this question.