Aiming at the problem of the jump existed in the first frame of human motion synthesis method based on Recurrent Neural Network （RNN）， which affects the quality of generated motion， a human motion synthesis method with hidden state initialization was proposed. The initial hidden state was used as independent variable， the objective function of the neural network was used as optimization goal， and the gradient descent method was used to optimize and solve the problem to obtain a suitable initial hidden state. Compared with Encoder-Recurrent-Decoder （ERD） model and Residual Gate Recurrent Unit （RGRU） model， the proposed method with initial hidden state estimation reduces the prediction error of the first frame by 63.51% and 6.90% respectively， and decreases the total error of 10 frames by 50.00% and 4.89% respectively. Experimental results show that the proposed method is better than the method without initial hidden state estimation in both motion synthesis quality and motion prediction accuracy. And the proposed method accurately estimates the hidden state of the first frame of RNN-based human motion model， which improves the quality of motion synthesis and provides reliable data support for action recognition model in real-time security monitoring.

Aiming at the problems of low computational efficiency and insufficient accuracy in the privacy-preserving neural network based on homomorphic encryption， an efficient Homomorphic Neural Network （HNN） under three-party collaborative supporting privacy-preserving training was proposed. Firstly， in order to reduce the computational cost of ciphertext-ciphertext multiplication in homomorphic encryption， the idea of secret sharing was combined to design a secure fast multiplication protocol to convert the ciphertext-ciphertext multiplication into plaintext-ciphertext multiplication with low complexity. Then， in order to avoid multiple iterations of ciphertext polynomials generated during the construction of HNN and improve the nonlinear calculation accuracy， a secure nonlinear calculation method was studied， which executed the corresponding nonlinear operator for the confused plaintext message with random mask. Finally， the security， correctness and efficiency of the proposed protocols were analyzed theoretically， and the effectiveness and superiority of HNN were verified by experiments. Experimental results show that compared with the dual server scheme PPML （Privacy Protection Machine Learning）， HNN has the training efficiency improved by 18.9 times and the model accuracy improved by 1.4 percentage points.

To solve the problem that many improved ant colony algorithms are not efficient to solve the problem of multi-constrained Quality of Service Routing (QoSR), such as slow convergence and local optimization, an Ant Colony Algorithm with Gradient Descent (ACAGD) was proposed. The gradient descent method was introduced into the local search of ant colony, and combined with residual pheromone, the next-hop selection strategy of ants was synthetically determined. Ant colony not only search for the next hop according to the pheromone concentration with certain probability, but also search for the next hop according to the gradient descent method with certain probability, which reduced the possibility that the traditional ant colony algorithm was easy to fall into the local optimum. The Waxman network model was used to randomly generate the network topology with different number of routing nodes. The experimental results show that compared with other improved ACO algorithms, the ACAGD can obtain the route with relatively low comprehensive cost while the convergence rate is not affected, and the stability of the algorithm is better.

To solve the load balance problem of multi-engine intrusion detection system, a dynamic load regulation algorithm of detection engine was proposed. Firstly, load was calculated by monitoring each engine node. Then, the scheduling of the heavy load node was performed by scheduling the overload or no-load node as a scheduling opportunity, and the nodes were traversed to adjust the load balancing. As the session for the scheduling unit, the algorithm was not the absolute average load for the purpose, just to ensure that the engine node does not appear overload or no load to achieve the basic goal. The KDD cup99 data set was used to simulate experiment. The experimental results show that compared with average load allocation algorithm and secure load allocation, the proposed algorithm has a significant effect on session-based load balancing, the running cost is lower, and the packet loss rate under heavy load are lower, which improves the detection rate of intrusion detection system.

In cluster, it is a conventional method to increase attack efficiency for intruder by attacking the specific target, so it is effective to improve the detection efficiency by scheduling the computing resource contrapuntally. A frequency self-adjusting algorithm for Network Intrusion Detection System (NIDS) based on target prediction, named DFSATP, was proposed. By detecting and analyzing the collected data packets, the data packets sent to potentially attacked targets were marked as high risk data and the other packets were marked as low risk data. The efficiency of NIDS was improved by high frequency detection of high risk data packets and low frequency detection of low risk packets, thus the detection rate of abnormal data was also increased to some extent in limited computing resource circumstances. The simulation results show that the detection rate of abnormal data packets is increased because of the detection frequency adjustment of NIDS by using DFSATP.

The original New Edge-Directed Interpolation (NEDI) algorithm is of high complexity, difficult for hardware implementation, and the interpolated images may suffer from blurring edges around edge area. To achieve a better subjective quality, an improved NEDI algorithm was proposed in this paper. In the new algorithm, a circular window was adopted, and the interpolation coefficient calculation was calculated only once, which could be reused in interpolating the center-pixels, thus the errors introduced by iterative computation were avoided and the interpolation time was saved. As to non-center pixels, six original neighbors were involved to estimate local covariance characteristics at high resolution. In comparison with the results of bi-cubic interpolation and the traditional NEDI, the experimental results indicate that proposed algorithm can eliminate the sawtooth of the interpolated picture in large-scale, and decrease the computational complexity.

No method could be used to get the exact numbers of these variables in the original Koschmieder model, whose solution is an ill-posed problem. Thus, we propose a novel Koschmieder model whose solution is much easier, while the proposed model has something in common with Atmosphere Degradation Model. Then a novel method for fast image dehazing is proposed based on the proposed model, compared with the dehazing method proposed by He, the experimental results yields that out method could realize fast dehazing, while could keep the scene’s color constancy and get the same or even better contrast promotion

After analyzing and researching the performance and deficiencies of TZSearch algorithm adopted by multi-view video, concerning the sequences of multi-view video obtained by parallel cameras, a new multi-view video coding optimization algorithm was put forward. The optimization was suggested mainly from the following three aspects: selection of search model, search strategy and adaptive threshold setting, so as to reduce the computational complexity of the algorithm. Tests were given on software testing platform for multi-view video named JMVC4.0. The experimental results show that: in ensuring the reconstruction video quality within tolerance, and under the premise of controlling the coding overhead, the optimized algorithm has reduced the average encoding time about 75% compared with the original algorithm, greatly improves the real-time performance of coding.