A Label Embedding Projective and Incremental Dictionary Learning Regression Method for COD Detection

Spectral analysis technology is widely used in the detection of chemical oxygen demand (COD) and other water quality parameters. The general practice is to use historical spectral data to build a detection model. However, the high dimension of the spectral data may cause important information to be obscured, resulting in the “dimensional curse” and affecting the accuracy of the detection model. At the same time, the variation of the components of the measured water body can cause the mismatch of the detection model and seriously affect the detection accuracy. To address these issues, we propose a label embedding projective and incremental dictionary learning regression method (LEPIDLR) for COD detection. First, using the spectral data of COD target substances to construct a graph model with embedded label information. Projective dimensionality reduction matrix, dictionary and regressor are jointly learned to optimize the feature input space, extract the optimal detection features, and establish a baseline model for COD detection. Second, based on the baseline model, a small amount of actual water sample spectral data is used to solve the model mismatch problem by combining the dictionary structure incoherence constraint and incremental learning to separate detection features and irrelevant interference features. Finally, the experiments demonstrates the effectiveness of the proposed method in improving the accuracy of COD detection.