A comprehensive soil pollution monitoring system based on convolutional recursive sequence network and terahertz spectroscopy

Abstract

Background

Accurate analysis of the total amount of heavy metals and their distribution patterns in soil can aid in determining the extent of soil pollution and iake appropriate remediation measures.Traditional detection methods are inefficient, and standard machine learning approaches are insufficiently sensitive in detecting the range when dealing with several contaminants of heavy metals. In this study, the contents of three heavy metals, cadmium (Cd), nickel (Ni), and zinc (Zn), in soils of different pollution types and concentrations were detected through terahertz spectral curve analysis.

Results

A convolutional recursive sequence network (CRST) with cascaded convolutional kernel and cyclic sequence structure was designed to further improve the soil heavy metal detection accuracy by capturing and integrating the sequence dependencies in the continuous terahertz bands. We compared common machine learning models with feature extraction algorithms, and the experimental results show that the CRST method can accurately identify soil heavy metal contamination species under the combined model of all data. The CARS-CRST combination achieved a classification accuracy of 99.54 % when the soil was contaminated with a single heavy metal. The CARS-CRST combination achieved a classification accuracy of 95.26 % when the soil was contaminated with a mixture of heavy metals. The CRST method can also accurately determine the heavy metal content in soil for Cd, Ni, and Zn, achieving optimal R² and RMSE values when the prediction model is constructed for these elements.

Significance

Based on terahertz spectroscopy and machine learning technology, this study conducted both qualitative and quantitative research on soil heavy metals, providing robust technical support for the scientific management and precise monitoring of soil.The results show that terahertz technology and machine learning technology can bring new ideas and prospects for soil pollution detection.