Chromatographic peak recognition method based on convolutional neural network

Abstract

In view of the low accuracy of current chromatographic peak-peak recognition methods in identifying multi-peak overlapping feature points, this paper proposed a chromatographic peak recognition method combining the convolutional neural network and the improved first and second derivative methods. In the convolutional layer, the method obtains the output value by interacting the input data set with the non-convolution kernel, and extracts the features in each small part of the chromatographic peak. At the pooling layer, the number of training parameters is reduced by reducing the dimension of the feature vector output by the convolutional layer, the overfitting phenomenon is reduced, and the noise transmission is reduced. Finally, a classifier with the required number of classes is generated using the full connection layer to identify the peaks in the chromatogram. At the same time, the improved first derivative and second derivative methods are combined to further identify other characteristic points of chromatogram. With zhejiang university wisdom of N2000 chromatogram workstation after processing the data, using the proposed method and the method of first order and second order derivative and Clarity chromatography work-station software comparison experiment, the experimental results show that the proposed chromatographic peak recognition method recognition accuracy is by far the most commonly used method of first order and second order derivative of the recognition accuracy rate rose by 31.26%, and the starting point of intersect for two peak, the recognition accuracy is better than that at the end of N2000 chromatographic workstation and Clarity.