Search

Purposes Speech emotion recognition allows computers to understand the emotional information contained in human speech, and is an important part of intelligent human-computer interaction. Feature extraction and fusion are key parts in speech emotion recognition systems, and have an important impact on recognition results. Aiming at the problem of insufficient emotional information contained in traditional acoustic features, a deep feature extraction method based on multi-task learning for optimization of acoustic features is proposed in this paper. Methods The proposed acoustic depth feature can better characterize itself and has more emotional information. Then, on the basis of the complementarity between acoustic features and spectrogram features, spectrogram features through convolutional neural network are extracted. Then, the multi-kernel principal component analysis method is used to perform feature fusion and dimension reduction on these two features, and the obtained fusion features can effectively improve the system recognition performance. Findings Experiments are carried out on the EMODB and the CASIA speech databases. When the DNN classifier is used, the multi-kernel fusion feature of the acoustic depth feature and the spectrogram feature achieve the highest recognition rates of 92.71% and 88.25%, respectively. Compared with direct feature splicing, this method increased the recognition rate by 2.43% and 2.83%, respectively.

Purposes Deep compressed sensing is the use of deep learning to solve the problems existing in traditional compressed sensing, such as the adaptability of observation matrix to traditional signal compression and the dependency on dictionary by reconstruction algorithm. Methods In this paper, the deep belief network (DBN) is used to adaptively compress the signal without destroying the randomness of observation matrix. At the same time, the stacked auto encoder (SAE) is used to train the reconstruction network end-to-end to get rid of the dependence of the reconstruction algorithm on sparse dictionary. According to the discrimination of the sparse representation of signal, a compressed sensing recognition model based on DBN and SAE is proposed (CS-DBN-SAE). Findings The results of four classification experiments on DEAP emotional EEG database show that the recognition rate of CS-DBN-SAE model is 83.29%, which is oven 4.3% higher than that of traditional compressed sensing recognition model.

To solve the problem that Dempster-Shafer (D-S) evidence theory gives combines results contrary to with facts when dealing with conflict evidence, a novel method of conflict evidence combination was proposed. First, the Spearman correlation coefficient is improved and the support between evidences is calculated, and original evidence is amended for the first time. Then, combined with the advantages of the interval distance based on fixed integrals, the uncertainty of evidence is analyzed, a new discount coefficient is determined, and original evidence body is secondarily revised. Finally, the fusion results are obtained by using the Dempster combination rule. The calculation and example analysis demonstrates that the proposed method can effectively fuse the evidence of conflict, and has a higher basic probability distribution than the classical improved algorithm. The proposed method was used for carbon/carbon composite sedimentation data, and a carbon/carbon composite sedimentation mass prediction model based on improved D-S evidence theory was established. Compared with several classical classifiers and the existing prediction model based on D-S evidence theory, this model is increased in accuraey by 5%~13%, which proves its validity.

In the quality analysis of polysilicon ingot batching, a two-stage genetic algorithm (NRS-SVM-GA) combining the NRS-SVM model with genetic algorithm (GA) was proposed to solve the problem of neighborhood radius and SVM parameter values in the processing of continuous data of polysilicon ingot batting with rough set-support vector machine (NRS-SVM) model. The first stage of the algorithm, by searching for a new neighborhood radius, a better reduction set is obtained. In the second stage the attribute reduction results from the first stage are adopted to, by searching the new parameters of SVM, train the classification model with higher accuracy. According to the purpose of each stage, the corresponding fitness function and termination conditions are put forward. The distinctive features of this method is to implement the NRS-SVM automatic feature extraction and classification prediction, and run the two stages independently to avoid the time consumption for the evaluation of reduction by classifiec. The experimental results of polysilicon ingot batching data set show that this method has shorter running time, stable output, less features, and higher classification accuracy than the standard genetic algorithm.