Traditional deep learning shadow removal methods often change the pixels in non-shadow areas and cannot obtain results with smooth boundary transition. In order to solve these problems, a new multistage shadow removal framework based on Generative Adversarial Network (GAN) was proposed. Firstly, shadow mask and shadow matte of the input image were generated by multitask driven generator via shadow detection subnet and shadow matter generation subnet respectively. Secondly, under the guidance of shadow mask and shadow matte, an umbra module and a penumbra module were designed respectively to remove different types of shadows successively. Thirdly, a new compose loss function dominated by least squares loss was created to obtain a better result. Compared with state-of-the-art shadow removal methods based on deep learning, the proposed method has the Balanced Error Rate (BER) averagely reduced by 4.39%, the Structural SIMilarity index (SSIM) averagely improved by 0.44%, and the Root Mean Square Error (RMSE) averagely reduced by 13.32%. Experimental results show that the boundary transition of shadow removal result of the proposed method is smoother.

Cross-domain discrepancy frequently hinders deep neural networks to generalize to different datasets. In order to improve the robustness of shadow detection, a novel unsupervised domain adaptive shadow detection framework was proposed. Firstly, in order to reduce the data bias between different domains, a multi-level domain adaptive model was introduced to align the feature distributions of source domain and target domain from low level to high level. Secondly, to improve the model ability of soft shadow detection, a boundary-driven adversarial branch was proposed to guarantee the structured shadow boundary was also able to be obtained by the model on the target dataset. Thirdly, the entropy adversarial branch was combined to further suppress the high uncertainty at shadow boundary of the prediction result, so as to obtain an accurate and smooth shadow mask. Compared with the existing deep learning-based shadow detection methods, the proposed method has the Balance Error Rate (BER) averagely reduced by 10.5% and 18.75% on ISTD dataset and SBU dataset respectively. The experimental results demonstrate that the shadow detection results of the proposed algorithm have better boundary structure.

Shadow in an image is important visual information of the projective object, but it affects computer vision tasks. Existing single image shadow removal methods cannot obtain good shadow-free results due to the lack of robust shadow features or insufficiency of and errors in training sample data. In order to generate accurately the shadow mask image for describing the illumination attenuation degree and obtain the high quality shadow-free image, a single image shadow removal method based on attenuated generative adversarial network was proposed. Firstly, an attenuator guided by the sensitive parameters was used to augment the training sample data in order to provide shadow sample images agreed with physical illumination model for a subsequent generator and discriminator. Then, with the supervision from the discriminator, the generator combined perceptual loss function to generate the final shadow mask. Compared with related works, the proposed method can effectively recover the illumination information of shadow regions and obtain the more realistic shadow-free image with natural transition of shadow boundary. Shadow removal results were evaluated using objective metric. Experimental results show that the proposed method can remove shadow effectively in various real scenes with a good visual consistency.

In order to solve spatio-temporally incoherent problem of traditional shadow removal methods for videos captured by free moving cameras, a shadow detection and removal approach using region matching guided by illumination transfer was proposed. Firstly, the input video was segmented by using Mean Shift method based on Scale Invariant Feature Transform (SIFT), and the video shadow was detected by Support Vector Machine (SVM) classifier. Secondly, the input video was decomposed into overlapped 2D patches, and a Markov Random Field (MRF) for this video was set up, and the corresponding lit patch for every shadow patch was found via region matching guided by optical flow. Finally, in order to get spatio-temporally coherent results, each shadow patch was processed with its matched lit patch by local illumination transfer operation and global shadow removal. The experimental results show that the proposed algorithm obtains higher accuracy and lower error than the traditional methods based on illumination transfer, the comprehensive evaluation metric is improved by about 6.23%, and the Root Mean Square Error (RMSE) is reduced by about 30.12%. It can obtain better shadow removal results with more spatio-temporal coherence but much less time.

A new voting protocol was proposed to solve the problem that most of the existing voting protocols require a trusted management authority. This protocol comprehensively makes use of homomorphic encryption, threshold cryptography, blind signature, ring signature, zero knowledge proof, and so on, to resolves the coexistence problem between robustness and the absence of trusted third party under the assumption that no one abstains from voting or the authority does not cheat conspiracy with other voters when one voter abstains from voting, at the same time, anonymity, eligibility, robustness, verifiability and no trusted third party are also satisfied.

The network replication technology has become the cornerstone of the evaluation platform for network security experiments and the system for network emulation. Facing the requirements of fidelity and scalability of network replication, a multi-scale network replication technology based on cloud platform for the fusion of lightweight virtualization, full virtualization and digital simulation was proposed. The architecture of the seamless fusion of these three scales was introduced at the beginning; And then the network construction technology based on the architecture was studied. The emulation experimental results show that the emulation network which is built with the construction technology has the characteristics of flexibility, transparency and concurrency; in addition, the construction technology is capable of emulating networks with high extensibility. At last, communication tests for a variety of protocols and simple network security experiments on the large-scale emulation network were conducted to verify the availability of this large-scale emulation network. The extensive experimental results show that the multi-scale network replication technology for the fusion of virtualization and digital simulation can be used as the powerful support for creating large-scale emulation networks.

Recent incidents such as the Stuxnet worm have shown that cyber attacks can cause serious physical damage of Cyber-Physical System (CPS). Aiming at this problem, a risk assessment method based on attack graph was proposed. Firstly, the attack behavior of CPS was analyzed and the result showed that the vulnerabilities in physical devices such as Programmable Logic Controller (PLC) were the keys of cross-domain attack. Then the utilization modes and impact of vulnerabilities were described. Secondly, the risk assessment model was proposed as well as the successful-attack-probability index and the attack-impact index. Furthermore, the successful-attack-probability index was calculated considering the intrinsic characteristics of vulnerabilities and the ability of attacker. The attack-impact index was calculated considering the host importance and the utilization mode of vulnerabilities. The method was developed to assess the cyber layer and physical layer as a whole system and the impact of multiple cross-domain attacks on system risk was considered. The numerical examples show that the risk of combined attack is five times the risk of a single attack and the risk value obtained is more accurate.

To resolve the problem that the existing adaptive slicing algorithm in 3D printing cannot retain effectively model characteristics, a new adaptive slicing method for recognizing and retaining model characteristics was proposed. Firstly, the definition of model characteristic was extended, and the concept of loss and offset of model characteristic was introduced. Secondly, a characteristic recognition method was proposed, the key point of which is to make use of the fact that the surface complexity and number of contours must change around the model characteristics. Finally, based on existing adaptive slicing algorithms, this algorithm retained model characteristics by slicing the model with minimum layer thickness near the model characteristics. On the self-developed software Slicer3DP, the following algorithms were implemented: the uniform slicing algorithm, the adaptive slicing algorithm and the proposed slicing algorithm. By comparing these algorithms, it is found that the proposed slicing algorithm resolves effectively the loss and offset of model characteristics while maintaining both slicing precision and efficiency. The result shows that the proposed method can be used for 3D printing with high precision requirement.

The generalization of Support Vector Machines (SVM) will decline when the training data sets get imbalanced distribution. A modification method of the optimal hyperplane based on average divergence ratio according to Fisher within-class scatter was proposed to solve the problem. The normal vector of the optimal hyperplane was got after SVM training. The Fisher within-class scatter was introduced to evaluate the distribution of the two classes. On this basis, the optimal hyperplane was modified by the ratio of the average distribution scatter that was obtained according to the number of samples. The experimental results on benchmarks data sets show that the proposed method improves the classification accuracy of the class with less training data, so as to improve the SVM's generalization.

To resolve the peculiar problem of the Jacobian matrix for a special class of nonlinear equations, an improved Newton mtheod was proposed based on the dual space. This paper proposed an explicit formula to compute the dual space of an ideal in a point through polynomial multiplication, and constructed augmented equations using the dual space. Meanwhile, the Jacobian matrix of augmented equations at initial point was full rank, and then the algorithm recovered quadratical convergence of Newton's iteration. The experimental results show that after three iterations, the accuracy of computation can achieve 10^(-15). The proposed method further enriches the theories of the dual space of ideal in algebra geometry and provides a new method for the numerical calculation in engineering applications.

To improve the survivability of networks, a strategy selection model based on stochastic game was proposed. According to the damage impact on system made by the attacker behavior, the game process was divided into the resistance phase, the recognition phase, and the recovery phase. The relationship between the transition of system state and strategy selection was depicted from a specific view, based on it, a strategy selection analytical method was presented. A representative network example was provided to verify the feasibility and validity of the proposed method in the prediction of attacker behavior and decision of optimal strategy.

To deal with the problems of low accuracy and local convergence in conventional Particle Swarm Optimization (PSO) algorithm, the chaos algorithm and cloud model algorithm were introduced into the evolutionary process of PSO algorithm and the chaos cloud model particle swarm optimization (CCMPSO) algorithm was proposed. The particles were divided into excellent particles and normal particles when CCMPSO was in convergent status. To search the global optimum location, the cloud model algorithm as well as excellent particles was applied to local refinement in convergent area, meanwhile chaos algorithm and normal particles were used to global optimization in the outside space of convergent area. The convergence of CCMPSO was analyzed by eigenvalue method. The simulation results prove the CCMPSO has better optimization performance than other main PSO algorithms.