Research On Signal Peak Preservation And Smoothing Algorithm Based On Segmentation Classification

Signal smoothing is a classical problem in the field of signal processing.Signal smoothing usually leads to the passivation of peaks in the signal,which will affect the qualitative and quantitative analysis of the signal.Therefore,the peak-preserving smoothing of signals is a challenging task.In order to protect the peak shape,the high-order Savitzky-Golay method is usually used for smoothing in the time domain,and in order to eliminate the noise to the maximum extent,the low-order Savitzky-Golay method is usually used for smoothing.In order to solve this contradiction,this paper uses the idea of segmented classification,combined with collaborative smoothing and deep learning methods to propose two effective peak-preserving smoothing algorithms.The main contents of this thesis are organized as follows:(1)A peak-preserving smoothing algorithm based on the Savitzky-Golay method and collaborative smoothing is proposed.By extracting the similar segments of the signal,using the inter-segment information of the similar segments and selecting the appropriate filtering method,the best smoothing of the signal segment is realized.Firstly,the similar segments of each segment are extracted,and the similar segments of each segment are arranged into a twodimensional array,which is recorded as X-direction and Y-direction respectively.Then,the twodimensional array is smoothed.In the X-direction,the Savitzky-Golay method with different orders is used to smooth according to the curvature of the signal.The Y-direction is constructed by the similarity of the signal segments,which is fitted by a low-order polynomial.Here,the cooperative processing of the two directions is called cooperative smoothing.Finally,the processed data segment is restored to the original position,and the overlapped part is weighted average to reconstruct the final smooth signal.The advantage of this method is that it makes use of the collaborative smoothing of similar segments.Besides,the Savitzky-Golay method with different orders can be used for segments with different features,which not only increases the flexibility of the method but also protects the detailed features of the signal,such as the peak shape.(2)A peak-preserving smoothing algorithm based on deep learning and collaborative smoothing is proposed.Using deep learning instead of collaborative smoothing,the goal of processing different signal segments by different smoothing methods can be achieved to a greater extent.With the learning ability of the network model,a convolutional autoencoder neural network is used to realize the adaptive smoothing of the signal segment.First,a large number of peak signals are generated randomly,then the noise signal and the corresponding real signal are segmented,and then the convolutional autoencoder neural network is used to train the signal segment,and then the trained convolutional autoencoder network model is used to process the test signal to get the smooth signal segment.Finally,these signal segments are reconstructed to get the final smooth signal.Experimental results show that the proposed peak-preserving smoothing algorithm has better peak preserving smoothing performance than the traditional signal smoothing algorithm.The peak-preserving smoothing performance is further improved by replacing the collaborative smoothing ring with deep learning.