Application of aerosol fluorescence spectrum recognition based on deep learning

The high-risk pathogenic microorganisms in the air pose a great threat to human society, but the traditional monitoring methods cannot accurately identify and classify the microorganisms in the air. Therefore, based on the principle of laser-induced fluorescence technology and single photon detector as the core device, an efficient fluorescence spectrometer was designed and built for the identification and classification of high-risk pathogenic microorganisms in the air, and the spectrometer could predict the concentration of microorganisms, which was of great significance to environmental safety. For the data collected by the spectrometer, the two input forms of one-dimensional vector and two-dimensional matrix were used to realize the identification and classification of fluorescence spectra, and the identification and classification effects of deep learning networks such as principal component analysis network, convolutional neural network and full convolutional network were studied and compared. The experimental results show that the identification and classification accuracy of convolutional neural network model with matrix input reaches 98.05% in the test set, and the prediction accuracy of microorganisms concentration of full convolutional network model with matrix input reaches 98.97% in the test set.