Method for detecting pollutant concentration through deep ultraviolet Raman spectrum based on deep learning

Abstract

The invention discloses a method for detecting pollutant concentration based on deep ultraviolet Raman spectrum of deep learning, and belongs to the technical field of spectral analysis. According to the method, aiming at the problems of spectral line overlapping interference, weak trace signals, strong background noise, nonlinear response and the like when a traditional deep ultraviolet Raman spectrum technology is used for detecting pollutants, high-precision quantitative analysis is realized by constructing an SSA-CNN-LSTM-Attention fusion model and combining a deep ultraviolet Raman spectrum pretreatment technology. The model adopts 1D-CNN to extract spectral local features, LSTM to capture sequence long-range dependence, an Attention mechanism to enhance feature peak response weight, SSA to optimize hyper-parameters, and a concentration prediction value is output through a weighted loss function optimization model. And meanwhile, by combining a synthetic data expansion strategy and environment matrix standard sample library training, the adaptability of the model to a complex scene is improved. According to the method, the anti-interference capability and generalization of new pollutant detection are effectively improved, and the method has important application value in the fields of environmental monitoring, industrial safety and the like.

Images