Search

As a new information acquisition and processing technology, wireless sensor networks can achieve tasks with complex large-scale monitoring and tracking in wide range of applications, but the localization for node itself is the basis of most applications. APIT is a major localization algorithm of ranged free algorithms, but the algorithm has the problems of big localization errors and low coverage rate, it also has high dependence on the number of anchor nodes. In order to overcome these disadvantages, this paper propose a mixed localization algorithm based on RSSI and APIT with fitness analysis and optimization, According to the experiment and simulation, the new algorithm has smaller localization errors, higher coverage rate, and improves the overall stability of nodes' localization.

We propose a new remote new remote-sensing image fusion method based on IHS transformation and MQPSO(Mutation Quantum-Behaved Particle Swarm Optimization) algorithms for the fusion of multispectral images and panchromatic images. IHS transformation may lose much spectral information when fusing multispectral images and panchromatic images. In our method, we resolve the problem of the optimum transformation for spectral intensity component using MQPSO algorithm. The experiments results show that our method is better than tradition IHS transformation in both keeping the spectral information and enhancing the spatial detail information expressive ability.

To enhance the versatility, robustness, and convergence rate of automatic classification of images, an ant-colony-based classification model is proposed in this paper. According to the characteristics of the image classification, this model adopts and improves the traditional Ant-Colony algorithm. It defines two types of ants that have different search strategies and refreshing mechanisms. The stochastic ants identify new categories, construct the category tables and determine the clustering center of each category. The Intellectual ants classify the image pixels using their search advancing strategies, with the guidance of the information provided by stochastic ants. Comparing with the traditional ant colony algorithms, this algorithm provides a more effective and accurate approach for automatic image classification.

In recent years, some nonlinear dimension reduction methods, named manifold learning, have been proposed and widely used in data visualization and pattern recognition. Of them, Isomap is a representative, which can project data from high-dimensional space into low-dimensional space with local structure preserved perfectly. However, Isomap suffers from the topological stability and is sensitive to noise. Moreover, it can only run in a batch mode, so cannot be directly used in pattern classification. In this paper, firstly, an improved Isomap based on image distance, namely IMD-Isomap, is proposed. Because spatial information of images is considered in image distance, as our experiments will show, IMD-Isomap outperforms Isomap for data visualization especially when noise is added. Then, combining IMD-Isomap and generalized regression neural network, which has a good ability for approximation, a classification method is proposed. Experimental results showed that our methods are robust to noise for data visualization or image classification when compared with KNN, Isomap or eigenface.

Several novel methods for nonlinear dimensionality reduction, named as manifold learning, have been proposed recently and widely used in pattern recognition and machine learning. In this paper, we present three face recognition methods based on kernel Isomap, which is a representative manifold learning method using kernel trick. Considering the class label by adjusting the Euclidean distance using weight factor w, both SK-Isomap-I and SK-Isomap-II are supervised and perform better than original K-Isomap. Unlike SK-Isomap-I, SK-Isomap-II utilizes nearest class center instead of KNN to determine class label of a test data. The experimental results showed that SK-Isomap-II performed the best in three of them and the error rate of SK-Isomap-II was only about 50% of k-Isomap.

Recently, a series of methods called manifold learning have been developed to visualize the convex but intrinsically flat manifolds such as Swiss roll. Isomap is a representative of them, which can easily discover low dimensional manifolds from high dimensional data but its computation complexity is quadratic. To speed up Isomap, L-Isomap was proposed to reduce the complexity by using landmark points. But how to select landmarks is an open problem. In this paper, we present an extension of Isomap focusing on the suitable selection of landmarks even the number of landmarks is quite small. In our method, each data point is assigned a weight according to the distance between it and its neighbors and point with a higher weight has a larger probability to be selected as a landmark point. The selection of landmarks falls into two phases. In 1st phase, n' candidate landmarks are selected only by the weights of data points. And in 2nd phase, n landmarks are refined from the candidates by maximizing the sum of distances between all pariwise landmarks. Experimental results showed that our method was more stable than L-Isomap and outperformed L-Isomap especially when the number of landmark points is quite small.

Telecommunication service providers must not only maintain a good understanding of the present state of the network, but also be able to forecast the future as accurately and precisely as possible. We focus on making traffic forecasting using neural network method. Two multi-layer perceptron neural network models are proposed respectively to make traffic prediction for two goals - one day's 24-hour load shape prediction at a time and peak-load prediction of a day. The data collected from the real world is applied for the performance tests and the results show that our two models are both effective.