Real-time discrimination of contamination source composed of multiple pollutants in surface water based on deep learning and UV–Vis spectral abundance estimation methodology

Abstract

Water resources are one of the most important strategic resources for human survival and development. At present, surface water pollution incidents occur frequently, most of which are caused by enterprises' over-discharge, stolen discharge, and other activities to evade supervision. Automatic and rapid determination of pollution source types is conducive to further targeting pollution-causing enterprises and realizing scientific accountability in law enforcement. The existing method mainly adopts the pattern recognition method for pollution discrimination, which is only suitable for the situation of a single source of pollutant, and cannot identify the pollution for multiple pollution sources mixed surface water. To solve the problem of identification of mixed chemical pollutants in surface water pollution sources and identification of simultaneous emission of multiple pollution sources, a total pollution source analysis method based on spectral unmixing is proposed in this paper, which is a radial basis bilinear mixing model automatic encoder algorithm. The unsupervised autoencoder neural network method was used to solve the proportion of water pollution types by using the spectral database of water pollution sources to realize the identification function of water pollution types and determine the types of pollutant discharge enterprises. In this paper, surface water was collected as experimental samples, mixed with domestic sewage, industrial sewage, agricultural sewage, and other pollution sources, and simulated experiments were carried out to estimate the type and proportion of water pollution. Experimental results show that the detection accuracy of the proposed algorithm is significantly improved compared with the traditional algorithm. Among them, the accuracy of judging whether there is industrial sewage in the mixed experiment of three types of pollution is as high as 95.2%. This method provides an important basis for pollution source investigation and accountability.