ATR-FTIR spectral identification study of bioaerosols based on improved LeNet-5 modeling

Bioaerosols have a certain impact on environmental quality and human health. In order to quickly identify bioaerosols and carry out timely and effective prevention and treatment. This study will optimize and improve the classical convolutional neural network LeNet-5 model, and combine attenuated total reflection Fourier transform infrared spectroscopy ( ATRFTIR ) technology to classify 15 bioaerosols. By changing the two-dimensional convolution kernel into one-dimensional convolution kernel to adapt to one-dimensional spectral data, the convolution layer and pooling layer are added to optimize the network structure, and the dropout algorithm, batch normalization and ReLU activation function are used to enhance the generalization ability and robustness of the network. The results show that the accuracy of the model reaches 100 % when the training times reach 80 times. Compared with the nearest neighbor ( KNN ) and Partial least squares Discriminant Analysis( PLS-DA ), the accuracy of the improved LeNet-5 model on the test set is significantly improved, showing its excellent performance in the rapid and accurate identification of bioaerosols, which provides new ideas and methods for the rapid and accurate identification of bioaerosols.