Infrared hyperspectral imaging integrated with an attentive spatial-spectral neural network for precise postharvest detection of Aspergillus flavus contamination and nutrient variations in peanut kernels

Guo, Z., Qin, Y., Shao, X., Cheein, F.A.A., Xia, L., Guo, Y., Sun, X. and Du, F. (2026) Infrared hyperspectral imaging integrated with an attentive spatial-spectral neural network for precise postharvest detection of Aspergillus flavus contamination and nutrient variations in peanut kernels. Infrared Physics & Technology, 154. ISSN 13504495

Full text not available from this repository.

Abstract

Peanut kernels are highly susceptible to Aspergillus flavus contamination, posing significant food safety risks due to aflatoxin B1 accumulation. This study applied infrared hyperspectral imaging, specifically covering visible-near infrared (VNIR, 400–1000 nm) and short-wave infrared (SWIR, 1000–2500 nm), to investigate the micro-interaction mechanisms between Aspergillus flavus and peanut kernels, focusing on spatio-temporal nutrient consumption and toxin accumulation. Infrared-based generalized two-dimensional correlation spectroscopy revealed a phased nutrient utilization strategy employed by Aspergillus flavus, identifying critical contamination phases at day 3 and day 5. An innovative attentive spatial-spectral synergy network (AS3Net) integrated with a novel bi-dimensional focus ripple module (BFRM), and significantly enhanced the prediction accuracy of moisture, protein, and oil contents in peanut kernels, achieving coefficient of determination of validation values of 0.932, 0.859, and 0.786, respectively. Ablation experiments highlighted that the combined use of spatial discovery, spectral insight modules, and dual fusion strategies improved model robustness, especially in predicting moisture content. Additionally, the AS3Net-BFRM framework provided a rapid, accurate classification of fungal-contaminated kernels with 100% accuracy. This advanced infrared hyperspectral imaging and deep learning approach presents a scalable, non-destructive, and efficient solution for real-time fungal contamination detection, which is crucial for enhancing food safety and managing aflatoxin risks in agricultural products.

Item Type:	Article
Additional Information:	Full text not available from this repository.
Divisions:	Engineering
Depositing User:	Mrs Susan Howe
Date Deposited:	16 Mar 2026 16:20
Last Modified:	16 Mar 2026 16:20
URI:	https://hau.repository.guildhe.ac.uk/id/eprint/18339

Actions (login required)

Edit Item