Early contamination warning of Aflatoxin B1 in stored maize based on the dynamic change of catalase activity and data fusion of hyperspectral images

•Dynamic change of CAT activity of mold was used to warn AFB1 contamination of stored maize.•The rapid increase of CAT activity preceded the rapid accumulation of AFB1 for 2–3 days.•Linkage-warning models of CAT-AFB1 under different storage environment were built.•Nondestructive prediction models of CAT activity were built based on the fusion of spectra and texture data.•UPLS method, we proposed, achieved better results for CAT activity prediction than PLS method. Aflatoxin B1 (AFB1) is a metabolite of mold with high toxicity, the early warning is of significance to prevent stored maize from AFB1 contamination. Catalase (CAT) is a precursor production before the synthesis of AFB1, which has a significant relationship with the accumulation of AFB1. Hence it is possible to assess the risk level of AFB1 contamination in stored maize according to the change rate of CAT activity, so that to help granary managers avoid AFB1 contamination by regulating storage environment. However, the the chemical detection procedure of CAT activity is complicated, a nondestructive and fast determination method of CAT activity would be more convenient for CAT activity detection and then help to provide the early warning of AFB1 contamination in stored maize. This study proposed to evaluate the contamination risk of AFB1 based on the linkage-warning models of CAT-AFB1 and the nondestructive prediction models of CAT activity. In terms of the dynamic response mechanism between CAT activity of mold and the AFB1 accumulation of maize, the effects of moisture content and storage temperature on the CAT activity of mold and AFB1 accumulation under different storage environment were analyzed, the linkage-warning models of CAT-AFB1 showed that CAT activity of mold and AFB1 content of maize increased exponentially with the change of storage time, proving that CAT activity of mold had the early warning ability for AFB1 contamination. In terms of nondestructive detection of CAT activity, the nondestructive prediction models of CAT activity of mold were developed based on the characteristic spectra and texture information using a novel algorithms namely union partial least squares, which achieved satisfying prediction results with RP2 = 0.924, RMSEP = 0.158, and RPD = 3.65 based on the fused characteristic spectra and texture information of visible-shortwave near-infrared region. This result obtained by this study will provides theoretical basis and technical guarantee for safe storage of mai