Introduction Monitoring and real-time detection of soil heavy metal pollution are crucial to ensuring the safety of agricultural food products. Conventional methods for determining heavy metal concentrations require substantial time and costs. This study investigates intelligent modeling approaches for online detection of heavy metal contamination in soils. Based on Internet of Things communication, large‐scale near-infrared (NIR) spectral data were collected from distributed sensors for federated analysis. Methods In chemometric studies, the improved binary firefly algorithm (IBFA) is proposed for evolutionary variable selection, and the modified method of maximum information coefficient (MMIC) is designed to estimate the nonlinear correlation of unevenly distributed samples. Experimental soil data are collected from the Karst geology on the north side of Guangxi ZAR, China. The NIR calibration model is established by fusion of the IBFA and the MMIC methods (denoted as IBFA + MMIC). The fusion model is applied for quantitative prediction targeting of four heavy metals in the Karst soil samples. Results The IBFA + MMIC model can observe correlation coefficients higher than 0.9 and the lowest prediction errors during model training. It is tested with the correlations very close to 0.9, while the testing errors are acceptably low. These results outperform the counterpart models established by other cross combinations of firefly algorithm (FA)/IBFA and maximum information coefficient (MIC)/MMIC. Discussion The proposed modeling methodology is effectively validated for quantitative NIR analysis of different heavy metals in Karst soil data. Meanwhile, it provides critical technical support for the federated analytical performance of distributed sensing data, thereby facilitating precision soil management practices.
Hong et al. (Tue,) studied this question.