Search

Objectives: In order to ensure the reliability of underwater gravity matching navigation,a suitability analysis method of gravity matching navigation based on multiple attribute decision making theory is proposed, which is to solve the defect of incomplete evaluation results caused by a single feature index as a suitability analysis criterion. Methods: First, six main characteristic parameters, such as standard deviation of gravity field, roughness, correlation coefficient, slope standard deviation, skew coefficient and differential entropy of gravitational anomalies, are used as attribute sets for weighted normalization, so that the decision index is obtained to rank the suitability of the matching region, and the most excellent suitability area is evaluated by constructing the relative fit.Second, in the same conditions, we compare the simulation results of gravity matching in each region to be matched. Results: The results show that the matching effect is more stable in the higher decision-making index area, and the most excellent suitability area could be directly screened, The matching degree of decision results with navigation results is as high as 80%, the decision- making results of the method have good consistency with the accuracy of the matching experiment. Conclusions: The reliability of the method is high, so it could solve the problem of less information of a single feature index, and provide an important guarantee for the long-term voyage of underwater submersibles. [ABSTRACT FROM AUTHOR]

To improve the real-time estimation accuracy of vigilance, this paper developed a vigilance estimation method based on differential entropy (DE), improved moving average and two-directional two-dimensional principal component analysis (TD-2 DPCA). First, it decomposed the total frequency band into multiple sub-bands with a certain bandwidth, and extracted DE from the sub-band. Then combined with the temporal dynamic characteristics of vigilance, it improved the traditional moving average and used improved moving average to smooth the DE. After that, it used TD-2 DPCA to reduce the dimensionality of DE, and established a regression model between the feature matrix and vigilance with least squares-support vector machine (LS-SVM) to achieve accurate real-time estimation of vigilance. Finally, this paper used the SEED-VIG for experimental verification. The results show that the improved moving average and TD-2 DPCA can improve the accuracy of vigilance estimation and reduce the estimation time. Using 1 Hz or 2 Hz bandwidth to decompose the total frequency band within 0-35 Hz to extract DE and perform vigilance estimation, the result has the highest accuracy. The Pearson correlation coefficient is about 0.91, and the RMSE is about 0.09, which is better than the existing vigilance estimation methods. [ABSTRACT FROM AUTHOR]

In order to improve the accuracy rate of the emotional recognition of EEG signals in multi-classification, the SEED dataset published by SJTU is selected as the sample of EEG dataset. The original EEG signal is divided into five frequency bands, and their features are extracted. After the features of the differential entropy, the differential asymmetry and the rational asymmetry of EEG datasets are smoothed by linear dynamic system, the classification effect is compared with the feature of power spectral density. Then, the method of the long short-term memory neural network is used to classify emotion. It is concluded that the classification of the differential entropy feature is effective. Finally, compared with other machine learning methods, the recognition rate is improved, and the average accuracy of emotion recognition reaches 95.045 9%. [ABSTRACT FROM AUTHOR]

The relationship between EEG and emotion recognition has attracted wide attention, however, the partial accuracy of emotion recognition is low. For improving the accuracy rate, after filtered the original EEG signal to 5 bands, this paper then extracted 4 features, the differential entropy, power spectral density, wavelet entropy and approximate entropy. Finally it selected the features and used the support vector machine, K-nearest neighbor, naive Bayesian model and multi-layer perceptron for classification learning. It trained DE feature to get the higher accuracy with 4 different electrode placement methods by support vector machine. The results show that the accuracy rate reaches 91. 99% of all EEG band energy in the case of 12 channels, and gets the average accuracies up to 97. 59% with SVM that using genetic algorithm to acquire the optimization parameters. [ABSTRACT FROM AUTHOR]