Deep convolutional neural networks are generally large in scale and complex in computation, which limits their application in high real-time and resource-constrained environments. Therefore, it is necessary to optimize the compression and acceleration of the existing structures of convolutional neural networks. In order to solve this problem, a hybrid compression method combining pruning and stream merging was proposed. In the method， the model was decompressed through different angles, further reducing the memory consumption and time consumption caused by parameter redundancy and structural redundancy. Firstly, the redundant parameters in each layer were cut off from the inside of the model. Then the non-essential layers were merged with the important layers from the structure of the model. Finally, the accuracy of the model was restored by retraining. The experimental results on the MNIST dataset show that the proposed hybrid compression method compresses LeNet-5 to 1/20 and improves its running speed by 8 times without reducing the accuracy of the model.

In view of the problem that the channel estimation error of the current Singular Value Decomposition-Linear Minimum Mean Square Error (SVD-LMMSE) algorithm was relatively large, an improved adaptive Linear Minimum Mean Square Error (LMMSE) channel estimation algorithm in Discrete Wavelet Transform (DWT) domain based on Empirical Mode Decomposition-Singular Value Decomposition (EMD-SVD) difference spectrum was proposed. The DWT was used to quantify the threshold of the signal high frequency coefficients after Least Square (LS) channel estimation and pre-filtering. Then, combined with the adaptive algorithm based on EMD-SVD difference spectrum, the weak signal was extracted from the strong noise wavelet coefficients, and the signal was reconstructed. Finally, the corresponding threshold was set based on Cyclic Prefix (CP) inside and outside the noise's variance of the mean, and the noise of the cyclic prefix length was handled to reduce the further influence of noise. The Bit Error Rate (BER) and the Mean Squared Error (MSE) performances of the algorithm was simulated. The simulation results show that the improved algorithm is better than the classcial LS algorithm, the traditonal LMMSE algorithm and the more popular SVD-LMMSE algorithm and can not only reduce the influence of noise, but also improve the accuracy of channel estimation effectively.

In view of the problem that the traditional channel estimation algorithm based on Discrete Cosine Transform (DCT) does not eliminate the noise in the cyclic prefix length, a new method of Orthogonal Frequency Division Multiplexing (OFDM) system channel estimation based on wavelet de-noising and DCT interpolation was proposed. First the method of Least Squares (LS) was used to preliminarily estimate channel for received pilot signal, then the results estimated by LS method were processed through discrete wavelet thresholding denoising, finally the noise of the cyclic prefix length was handled again by DCT interpolation algorithm to further reduce the influence of noise. The simulation on Matlab 2012 platform, compared with the traditional channel estimation algorithm based on DCT, under the conditions of the same Bit-Error-Rate (BER), the Signal-to-Noise Rate (SNR) performance of the proposed algorithm improved about 1 dB; under the conditions of the same Mean-Square-Error (MSE), the SNR performance of the proposed algorithm improved about 2 dB.The simulation results show that the proposed algorithm can not only reduce the influence of Additive White Gaussian Noise (AWGN), but also improve the accuracy of channel estimation effectively, and the proposed algorithm has better performances than the channel estimation algorithm based on DCT.

Traditional Linear Minimum Mean Square Error (LMMSE) channel estimation was required to know the statistical characteristics of the channel. However, these characteristics are usually unknown in practical applications. Aiming at the uncertainty of wireless channel statistics, taking the time-domin channel sparsity of the energy distribution into consideration, this article proposed an improved LMMSE channel estimation algorithm based on Least Squares (LS) estimation. The algorithm began with the highest confidence degree subcarrier, making the adjacent subcarrier channel estimation value as the current subcarrier real response to compute the weighting coefficient, then to complete channel response of the multiple channels by the method of weighted average. This algorithm avoided the complicated operation of the matrix inversion and decomposition, and might be done effectively and easily. The experimental results show that the performance of the improved algorithm is better than LS and the SVD-LMMSE (Singular Value Decomposition-Linear Minimum Mean Square Error) channel estimation, and the Bit Error Ratio (BER) is close to traditional LMMSE algorithm.

The existing Web services discovery method has limitations in time cost and accuracy because it does not make full use of the user context. Firstly, the clustering of similar user context was implemented to greatly reduce retrieval range of Web services. Secondly, based on this, making use of the current users' preference information and the history users' QoS-aware data, a method of Web services discovery based on history users' QoS-awareness was proposed. Finally, the comparison to other Web services methods indicates that this method is better than several other methods both in time cost and accuracy of Web services discovery.

The document images scanned may be skew somehow. Severe image skew makes image segmentation difficult and lowers character recognition accuracy. A new approach of skew detection based on Hough transform was presented. In order to overcome the heavy computing burdens of Hough transform,the method selected the subfield with part representation and extracted the horizontal edge from images in the first place, then performed two-stage Hough transform on the edge extracted. Experiment results show that it corrects the skew document images more rapidly and accurately than general Hough method and cross relation method.