Data Analysis Of Near Infrared Spectroscopy Using Deep Learning

Near-infrared Spectroscopy(NIRS)analysis technology has been widely used in industry,agriculture,pharmaceutical industry and other fields because of its advantages such as simple sampling,fast detection speed,nondestructive and pollution-free.With the increase of NIRS data complexity,the traditional methods have some problems such as insufficient feature extraction ability and weak generalization ability.Based on the deep learning algorithm,this paper proposes effective features from the complex NIRS data and constructs qualitative and quantitative models of NIRS by using its strong ability to extract deep feature tables.The main contents are as follows:(1)Aiming at the problems of high dimensionality of NIRS data and complex internal information features,a classification model of one-dimensional deep convolutional neural network(1-DDCNN)suitable for NIRS data is proposed based on the Le Net-5 model.The deep network model with six layers of convolutional pooling can automatically extract feature information helpful for classification from the original NIRS data,making NIRS data analysis more intelligent.At the same time,a 1-DDCNN regression model based on multi-task learning is proposed,which uses the correlation between multiple tasks to improve the performance of each task and the generalization ability of the model.The example data set verifies that the method has high accuracy and robustness.(2)Aiming at the problem of limited NIRS data samples and imbalanced categories,this paper proposes a one-dimensional deep convolutional generation confrontation network(1-DDCGAN)data enhancement method based on the 1-DDCNN model for NIRS data.The network includes two networks,a one-dimensional generating network and a onedimensional discriminant network,and the adversarial learning method is used to train the model.This method generates data that is highly similar to the real sample data based on randomly generated noise vectors,and expands the original data set of small samples.Then,an example data set verifies the stability and effectiveness of the method.(3)Aiming at the problem of insufficient feature extraction ability of deep network models in the face of small samples and difficulty in training,a NIRS analysis method based on deep transfer learning is proposed.This method first uses a large number of labeled data sets to train the pre-trained model,and locks all the parameters of the convolutional layer and the pooling layer,then redesigns the fully connected layer and the output layer,and finally uses the target domain data set to fine-tune the model parameters.This method not only improves the prediction accuracy of the model,but also accelerates the convergence speed of the deep network model,and improves the efficiency of the NIRS data analysis model.Compared with traditional machine learning algorithms,this method solves the problems of difficult network convergence and long training time.The instance data set verifies the effectiveness of the method.(4)Developed an intelligent detection system for NIRS data based on a cloud platform.The secondary development based on the DLP NIR-Scan Nano module of Texas Instruments realizes NIRS data collection and wireless transmission.The cloud server receives the data and calls the established detection model for data analysis,and the mobile client displays the detection results.The experiment verifies the feasibility and stability of the detection system.