|
Feature extraction and attribute recognition of particulate light scattering signals are important in the fields of environmental monitoring, atmospheric science and industrial process control. However, due to the complexity of particles, it remains a challenge to accurately extract features from light scattering signals and perform attribute recognition. In this study, a deep learning-based feature extraction and attribute classification model for particulate light scattering signals is proposed. Firstly, the acquisition of particulate light scattering signal is accomplished by the multi-angle detection of light scattering signal experimental flat, secondly, a total of 20 statistical features in time domain, frequency domain and information entropy features of the signal are extracted to describe the local details of the signal at different frequencies and the feature weights of the signal are obtained by the ReliefF algorithm, so as to find the optimal feature vectors of the signal. Finally, the GRNN and PNN neural network algorithms are used to construct the particle attribute classification model, and the acquired optimal feature vectors are input into the model for attribute classification and recognition. The results show that the recognition accuracy of GRNN reaches 86.7%, while that of PNN reaches 91.67%. It is verified that the GRNN and PNN methods are able to effectively distinguish the above six particles with different attributes.
|
