Focusing on the issue of speech recognition in multi-noise environment, a new hierarchical speech recognition model considering environmental noise as the context of speech recognition was proposed. The proposed model was composed of two layers of noisy speech classification model and acoustic model under specific noise environment. The difference between training data and test data was reduced by noisy speech classification model, which eliminated the limitation of noise stability required in feature space research and solved the disadvantage of low recognition rate caused by traditional multi-type training under certain noise environment. Furthermore, a Deep Neural Network (DNN) was used for modeling of acoustic model, which could further enhance the ability of acoustic model to distinguish noise and speech, and the noise robustness of speech recognition in model space was improved. In the experiment, the proposed model was compared with the benchmark model obtained by multi-type training. The experimental results show that, the proposed hierarchical speech recognition model has relatively reduced the Word Error Rate (WER) by 20.3% compared with the traditional benchmark model. The proposed hierarchical speech recognition model is helpful to enhance the noise robustness of speech recognition.

Since the tourism text for destinations contains semantic noise and different scenic spots, which can not be displayed intuitively, a new scenic spots-topic model based on the probabilistic topic model was proposed. The model assumed that one document included several scenic spots with correlation, and a special scenic spot named "global scenic spot" was introduced to filter the semantic noise. Then Gibbs sampling algorithm was employed to learn the maximum a posteriori estimates of the model and get a topic distribution vector for each scenic spot. A clustering experiment was conducted to indirectly evaluate the effects of the model and analyze the impact of "global scenic spot" on the model. The result shows that the proposed model has better effect than baseline model such as TF-IDF (Term Frequency-Inverse Document Frequency) and Latent Dirichlet Allocation (LDA), and the "global scenic spot" can improve the modeling effect significantly. Finally, scenic spots association graph was employed to display the result visually.

Multiple Kernel Learning and Support Vector Machine (MKL-SVM) was applied to automatic music genre classification to choose the optimal kernel functions for different features, a method of conducting the optimal kernel function combination into the synthetic kernel function by weighting for music genre classification was proposed. Different optimal kernel functions were chosen for different acoustic features by multiple kernel classification learning, the weight of each kernel function in classification was obtained, and the weight of each acoustic feature in the classification of the genre was clarified, which provided a clear and definite result for the analysis and selection of the feature vector in the classification of music genre. The experiments on the dataset of ISMIR 2011 show that, compared with the traditional single kernel support vector machine classification, the accuracy of the proposed music genre automatic classification method based on MKL-SVM is greatly improved by 6.58%. And the proposed method can more clearly reveal the the different features' impacts on music genre classification results, the classification results has also been significantly improved by selecting features with larger effects on classification.

Considering the data security and integrity on the processing of HD video data stream transmission, acrypto DSP with special implementation for DES, SHA1, MD5, RSA are introduced. In order to improve the performance and decrease the cost, the DSP has 11 pipeline stages, and two parallel execution clusters (each cluster contains 3 function units). In order to improve throughput, special instructions are customized for complex operations. The methods of realizing the symmetrical encryption, public-key encryption and authentication algorithm based on such DSP are presented. In order to improve the throughput, The simulation experiment results show that the performance can well satisfy the requirement of real time HD video data stream applications.

In view of the difficult in management of airborne multi-sensor, a fuzzy decision tree algorithm is applied to multi-sensor intelligent management, combined with operational processes, and combat stage will be merging with the target type, different target attributes is built reasonable. The fuzzy decision trees model are established based on target type, then the airborne multi-sensor management model is established. Finally, the simulation results show that the method can compose battlefield objectives and tactical effectively, complete airborne multi-sensor management rapidly, and the method is reasonable and effective.