In the study of model parameter based transfer learning, both the sample distribution discrepancy between two domains and the co-adaptation between convolutional layers of the source model impact performance of model. In response to these problems, a Multi-Convolution Adaptation (MCA) deep transfer framework was proposed and applied to the grade classification of typhoons in satellite cloud images, and a CE-MMD loss function was defined by adding the improved L-MMD (Maximum Mean Discrepancy) algorithm as a regular term to the cross-entropy function and applying the linear unbiased estimation to the distribution of the samples in Reproducing Kernel Hilbert Space (RKHS). In the back propagation process, the residual error and the distribution discrepancy between the samples in two domains were used as common indexes to update the network parameters, making model converge faster and have higher accuracy. Comparison experimental results of L-MMD and two measurement algorithms-Bregman difference and KL (Kullback-Leibler) divergence on the self-built typhoon dataset show that the precision of the proposed algorithm is improved by 11.76 percentage points and 8.05 percentage points respectively compared to those of the other two algorithms. It verifies that L-MMD is superior to other measurement algorithms and the MCA deep transfer framework is feasible.

The traditional testing process does not specifically consider the system performance. With the wide application of parallel testing method, more attention was paid to the system performance and data throughput capacity. Optimizing the software design with multithreading technology becomes an effective way to improve the performance of automatic test system. By modeling testing pipeline process, using asynchronous pipeline design patterns and combining task-oriented concepts, an available test system programming model was proposed. The experiment results prove that the model can significantly shorten the average test time in the ideal case of random input of test tasks. Applying this model to an instance of measuring characteristic parameters of Alternating Current (AC) contactor, the results further indicate that this model can significantly increase the flexibility of test configuration and avoid the complexity of multi-threaded programming.

Abstract: To overcome the shortcoming that random measurement matrix is hard for hardware implementation due to its randomly generated elements, a new structural and sparse deterministic measurement matrix was proposed by studying the theory of measurement matrix in Compressed Sensing (CS). The new matrix was based on parity check matrix in Quasi-Cyclic Low Density Parity Check (QC-LDPC) code over finite field. Due to the good channel decoding performance of QC-LDPC code, the CS measurement matrix based on it was expected to have good performance. To verify the performance of the new matrix, CS reconstruction experiments aiming at one-dimensional signals and two-dimensional signals were conducted. The experimental results show that, compared with the commonly used matrices, the proposed matrix has lower reconstruction error under the same reconstruction algorithm and compression ratio. The proposed method achieves certain improvement (about 0.5-1dB) in Peak Signal-to-Noise Ratio (PSNR). Especially, if the new matrix is applied to hardware implementation, the need for physical storage space and the complexity of the hardware implementation should be greatly reduced due to the quasi-cyclic and symmetric properties in the structure.

Concerning the key problems of selecting threshold function in wavelet threshold denoising, in order to address the discontinuity of conventional threshold function and large deviation existing in the estimated wavelet coefficients, a continuous adaptive threshold function in the whole wavelet domain was proposed. It fully considered the characteristics of different sub-band coefficients in different scales, and set the energy of sub-band coefficients in different scales as threshold function's initial weights. Optimal weights were iteratively solved by using interval advanced-retreat method and golden section method, so as to adaptively improve approximation level between estimated and decomposed wavelet coefficients. The experimental results show that the proposed method can both efficiently reduce noise and simultaneously preserve the edges and details of image, also achieve higher Peak Signal-to-Noise Ratio (PSNR) under different noise standard deviations.

Dynamic tree rendering plays an important role in the natural scenery simulation. In this paper, by using Cook-Torrance lighting model and pre-computed translucency texture, rendering scattering and translucency of the leaf were implemented. Using the polynomial fitted from tapered circular beam model expression and length correct method, the speed of calculation deform was boosted. By introducing the hierarchical branches texture with index, branches deform could be calculated in Graphic Processing Unit (GPU). Using pre-compaction and GPU helps to balance the reality and real-time in the simulation. The experiments show that the proposed method can render the dynamic tree scene vividly and rapidly.

For the requirements of building Project Information Portal (PIP) data center based on a unified data model, a manageable entity object-oriented data model was proposed. The project data were treated as a series of managerial entity based on management workflows which were decomposed according to the whole life cycle. The conceptual layer data model was designed. The project data could be naturally represented and recorded by using this model. The data organization method was presented based on MongoDB (document-oriented database technology). The cluster storage architecture for PIP was also addressed. The experiments show that it has efficient performance in data writing and querying. It also has high availability and storage capacity scalability.

Considering the problems for nonlinearnoise processing and taking account of that higher-order statistics of the signal and unitary space can be a good deal with non-Gaussian noise，the noise processing algorithm based on Max Correntropy Criteria (MCC) in the time domain and the unitary space was proposed. Combining the MCC and gradient descent algorithm, a nonlinearnoise filtering algorithm in the time domain was designed. At the same time, extending the algorithm to the noise processing in the unitary space, the unitary space filtering algorithm based on the MCC was put forward. The simulation study shows that the algorithm based on the MCC algorithm has significant advantages compared with the traditional Least Mean Square (LMS) based filtering algorithm, which means it can achieve more complete signal characteristics by faster convergence.

To protect the security of social network, ensure the availability of social network after perturbation, the paper proposed perturbation method of social network based on the signless Laplacian matrix and the social network non-randomness. In the perturbation process, this method controlled the social network spectral radius and the social network non-randomness by certain constraints, thus ensuring the usability and improving the privacy protection degree of the social network. The paper analyzed the security of this method in theory, and provided corresponding algorithm. At last, the experimental results on comparison of harmonic mean of the shortest distance of the social network, subgraph centrality and the social network non-randomness of change, show that the proposed method effectively protects the structural feature of social network and improving the availability of the social network.

The major procedures of text classification such as uniform text format expression, training, testing and classifying based on Naive Bayesian text classification algorithm were implemented using MapReduce programming mode. The experiments were given in Hadoop cloud computing environment. The experimental results indicate basically linear speedup with an increasing number of node computers. A recall rate of 86% was achieved when classifying Chinese Web pages.

An automatically-controlled power transformation system was studied and a distributed control scheme for the system was proposed and implemented in order to improve the safety and efficiency of power systems. Advanced software and hardware techniques, system architectures and safety requirements of power transformation systems were all taken into account. Real-time event and disturbance logs and control commands were communicated over a local power transformation system based on the embedded operating system of VxWorks and the hardware platform of PowerPC860 CPU. The distributed control scheme for power transformation systems realized supervisory control, data acquisition and logic functions, and has been validated by industrial experiments.