Non-destructive detection of apple fungal infection based on VIS/NIR transmission spectroscopy

• Published in: Journal of food composition and analysis 2024-09, Vol.133, p.106469, Article 106469
• Main Authors: Jiang, Xiaogang, Ge, Kang, Li, Bin, Ouyang, Aiguo, Liu, Yande, Jiang, Nan, Liu, Hongfei
• Format: Article
• Language: English
• Subjects: Alternaria; Soft Voting model; Transmission spectroscopy; VIS/NIR
• ISSN: 0889-1575; 1096-0481
• DOI: 10.1016/j.jfca.2024.106469

The study is dedicated to developing a precise and efficient detection method specifically for identifying infections in apples caused by fungi. This type of infection not only damages the quality and value of apples but also poses potential risks to human health. By integrating four machine learning algorithms (LDA, KNN, LSSVM, and RF) into a Soft Voting model, as well as combining VIS/NIR transmission spectroscopy with CARS and SPA feature wavelength screening methods, this study significantly improves the accuracy of detecting fungal infections in apples. The research results demonstrate that the Soft Voting model achieves 93.33 % TPR, 100 % TNR, 100 % precision, and 98.75 % accuracy, which has far-reaching implications for safeguarding food safety, reducing economic losses, and enhancing the reputation of fruit farmers and merchants. •Study spectral changes over time to understand the evolution of fungal infections.•CARS and SPA were used to select the characteristic wavelength, and then LDA, KNN, LSSVM, RF and Soft voting strategies were applied to establish the discrimination model.•Analyze the characteristic spectra of fungus-infected apples based on the selected wavelengths.