Offensive speech has a serious negative impact on social stability. Currently， automatic detection of offensive speech focuses on a few high?resource languages， and the lack of sufficient offensive speech tagged corpus for low?resource languages makes it difficult to detect offensive speech in low?resource languages. In order to solve the above problem， a cross?language unsupervised offensiveness transfer detection method was proposed. Firstly， an original model was obtained by using the multilingual BERT （multilingual Bidirectional Encoder Representation from Transformers， mBERT） model to learn the offensive features on the high?resource English dataset. Then， by analyzing the language similarity between English and Danish， Arabic， Turkish， Greek， the obtained original model was transferred to the above four low?resource languages to achieve automatic detection of offensive speech on low?resource languages. Experimental results show that compared with the four methods of BERT， Linear Regression （LR）， Support Vector Machine （SVM） and Multi?Layer Perceptron （MLP）， the proposed method increases both the accuracy and F1 score of detecting offensive speech of languages such as Danish， Arabic， Turkish， and Greek by nearly 2 percentage points， which are close to those of the current supervised detection， showing that the combination of cross?language model transfer learning and transfer detection can achieve unsupervised offensiveness detection of low?resource languages.

The advantages of simplicity and easy implementation of Particle Swarm Optimization (PSO) algorithm have been validated in science and engineering fields. However, the weaknesses of PSO algorithm are the same as that of other evolutionary algorithms, such as being easy to fall into local minimum, premature convergence. The causes of these disadvantages were analyzed, and an improved algorithm named Cosine PSO (CPSO) was proposed, in which the inertia weight of the particle was nonlinearly adjusted based on cosine functions and the learning factor was symmetrically changed, as well as population diversity was maintained based on bacterial chemotaxis. Therefore, CPSO algorithm is better than the Standard PSO (SPSO) in a certain degree. Simulation comparison of the three algorithms on five standard test functions indicates that, CPSO algorithm not only jumps out of local optimum and effectively alleviates the problem of premature convergence, but also has fast convergence speed.

To get all the accurate solutions to Nonlinear Systems of Equations (NSE), the algorithm with global convergence was constructed for solving NSE based on the characteristics of Cellular Automata (CA). In the algorithm, the theoretical search space of NSE was divided into the discrete space, the discrete space was defined as the cellular space; each point in the discrete space was a cell in the cellular space, and each cell was a trial solution of NSE; a cellular state consisted of position and increment of position. The cellular space was divided into many nonempty subsets, and states evolution of all cells from one nonempty subset to another realized the search of the cellular space on the theoretical search space. During evolution process of all cells, each cells transition probability from one position to any another position could be simply calculated; each state of cells during evolution corresponded to a state of a finite Markov chain. The stability condition of a reducible stochastic matrix was used to prove the global convergence of the algorithm. The case study shows that the algorithm is efficient.

This paper proposed a method for skin segmentation based on the similarity of skin-tone given by strengthened classifier which constructed via Real AdaBoost algorithm and dynamic threshold.Based on the clustering property of skin-tone distribution in YCrCb chrominance space, a set of weak classifiers in Look-Up-Table (LUT) type using circle-like features was trained via Real AdaBoost to form a strengthened classifier. Firstly, gray scale images indicated the skin-tone similarity of the pixels was created by processing the images with the strengthened classifier. Then skin segmentation was implemented according to the dynamic threshold selected through Da-Jing method. The experimental results show that the strengthened classifier has an outstanding ability for describing the distribution of skin-tone color in the YCrCb space. The method is robust, and efficient.

The denoising model based on Partial Differential Equation (PDE) model cannot eliminate impulse noise and low-order PDE will produce blocky effect. In order to solve these problems, a denoising model combining PDE and adaptive median filtering was proposed. Through analyzing the image gradient, this model used second order model to denoise at the region with obvious gradient change and the region with tiny gradient change. At the smooth region, fourth order model was used to denoise. The region of the impulse noise was localized by making use of the characteristic that the gradient of the impulse noise is far bigger than the gradient of the edge. At this region, the adaptive median filtering was used to eliminate impulse noise. This method can eliminate impulse noise and protect the image edge effectively. It also can overcome the blocky effect and improve the denoising efficiency. The experiments prove the validity of the model.