Developing a rapid COD detection method based on the fusion strategy of multi-depth hyperspectral data

Abstract

COD is a key indicator for evaluating the concentration of organic pollutants in wastewater as well as water quality monitoring. At present, COD detection methods have shortcomings such as long detection time, complicated detection process and high consumption of chemical agent. In this paper, a rapid detection method for COD in wastewater is proposed based on a multi-depth hyperspectral data fusion strategy (using hyperspectral data from sampling sites at multiple depths). In the method, the triangular prism acquisition method is proposed to acquire multi-depth hyperspectral data, and the bilateral filtering algorithm is introduced to reduce noise. The hyperspectral data acquired from sampling sites with three water depths (5 mm, 10 mm, and 15 mm) is analyzed by the random forest algorithm (RF), and two data fusion strategies are applied at the data level (the low-level fusion) and the feature level (the mid-level fusion). The results demonstrate that the modeling performance of the fused hyperspectral data is superior to that of the non-fused hyperspectral data. The lower-level fused data, combined with the competitive adaptive reweighted sampling (CARS) algorithm, produced a model capable of accurately predicting COD concentrations ($R_p^2$= 0.9312, RMSEP= 526.9, RPD= 3.54). This approach provides an environmentally friendly and efficient method for quantitative COD detection in wastewater.