Feature Extraction And Attribute Recognition Of Particle Light Scattering Signals Based On Machine Learning

Atmospheric environmental pollution has become a limiting factor for the sustainable development of social and economic development,and the identification and analysis technology of particulate matter attributes has important scientific significance for pollution prevention and control.In response to the development needs of feature extraction and analysis of particulate matter,research has been conducted on feature extraction and attribute recognition of atmospheric particulate light scattering signals based on machine learning technology.The main research content of this paper is as follows:(1)Around the light scattering theory,three approximate models of light scattering theory and their application conditions are introduced.By selecting standard spherical particles with different particle sizes and refractive indices for simulation calculations,the effects of particle size and refractive index on scattered light intensity were studied.In order to further study the influence of particle properties on light scattering characteristics,common particles in atmospheric components were selected for simulation research,and the differences in the distribution of scattered light intensity at different angles were analyzed.This provides a theoretical basis for subsequent feature extraction and attribute recognition of particle light scattering signals combined with machine learning.(2)Based on the multi angle detection light scattering signal system for particulate matter,the collection of light scattering signals has been designed and implemented.To reduce the impact of stray light and noise on the detection system,a method based on EMD and ICA is proposed to denoise the particle light scattering signal.The method is used to reconstruct the virtual noise channel and preprocess the particle light scattering signal,providing a prerequisite for subsequent signal processing.(3)According to the principles of machine learning,research is conducted on the feature extraction and attribute recognition methods of particle light scattering signals based on Generalized Regression Neural Network(GRNN)and Probabilistic Neural Networks(PNN).Extract the time-domain,frequency-domain,and information entropy features of particle light scattering signals separately,and use the ReliefF algorithm to obtain the optimal feature vector to achieve neural network model training and attribute classification recognition.The results show that the PNN network has an accuracy of 91.67%in identifying the attributes of six typical particulate matter types,including sodium chloride,silicon dioxide,dioctyl diacetate,glycerol,lime,and paraffin oil,and its recognition effect is better than that of GRNN network.(4)To further improve the effect of attribute recognition,the research on feature extraction and attribute recognition methods of particle light scattering signal based on Bayesian Optimization(BO),Wavelet Scattering Transform(WST)and Long Short-Term Memory(LSTM)neural network was focused.The adaptive feature extraction of signals is completed through WST,the optimization of LSTM neural network hyperparameter is realized by BO,the optimal hyperparameter combination is selected to construct LSTM neural network,and the classification and recognition of particles with different attributes are realized by probability classification.Select accuracy,precision,recall,and F1 value as performance evaluation indicators for the neural network,and compare the BO-WST-LSTM method with traditional classification models.The results show that this method can improve the attribute recognition performance of the six typical particulate matter types mentioned above,with an accuracy rate of 98.83%.