Identification of soil heavy metal pollution by constructing 2D plane using hyperspectral index

This study proposes a method for the rapid identification of elements in soil heavy metal pollution using spectral indices. Set up a simulation experiment of planting crops in soil polluted by multi-gradient Cu and Pb. The obtained polluted soil spectral data was initially pre-processed to obtain the original spectrum (OR), the continuum removed spectrum (CR), and the first-order differential spectrum (FOD). Then the preliminary model of soil heavy metal pollution index (SHMPI) was constructed. Using the correlation optimal algorithm, the maximum median distance algorithm, and the maximum average distance algorithm to select the optimal bands corresponding to the OR, CR, and FOD. The optimal bands selected by each algorithm were substituted into the SHMPI. Each algorithm obtains three indices, and two of them were selected as the x-axis and y-axis to form a two-dimensional pollution identification plane. Nine two-dimensional planes can be obtained by three algorithms and three combinations of OR, CR, and FOD. Support vector machine classifier was used to classify the Cu and Pb polluted samples in the planes, and nine classification models to distinguish Cu and Pb pollution in soil were constructed. The results show that using the correlation optimal algorithm to extract the optimal bands, and using OR and CR to construct SHMPI, the accuracy of the classification line model of Cu and Pb pollution obtained was 93% in the training group and 86% in the validation group. This method can stably and effectively identify the types of heavy metal pollution in soil, and can also effectively identify whether the soil is polluted by heavy metals, which is expected to guide the rapid and non-destructive identification of heavy metal pollution in polluted areas, and provide new ideas for the identification of other types of heavy metals in soil.