Search

Diverse pose estimation of 3D object in the whole view-space is a problem perplexed many researchers. In this paper we propose an algorithm extended from LLE which can estimate the arbitrary pose of 3D object in the whole view space. First, we compute the eigen-images of training set by introducing the idea of PCA using the low-dimensional embedding coordinate deduced from LLE. For a new sample we can compute its projection to the eigen-images, and the nearest training images from the new sample are the estimation poses. Next, we set different weight for different projection direction depends on its eigen-value when computing the distance between the new sample and the training images. Experimental results obtained demonstrated that the performance of the proposed method could estimate the diverse pose of 3D object efficiently and precisely, also our algorithm can be extended to real-time pose estimate, is of a potential future.

To overcome the limitations of traditional text classification approaches based on bag-of-words representation and to effectively incorporate linguistic knowledge and conceptual index into text vector space representation, based on WordNet thesaurus and latent semantic indexing (LSI) model, combinative method of them is presented to realize naive Bayes text classification and simple vector distance text classification, and five groups of contrastive experiments are made respectively. The results show that the accuracy rates of the two text classification methods are both gradually advanced along with more and more in-depth semantic analysis, which indicates that semantic mining is very important and necessary to text classification. The comparative analysis of the related work is also given

To more advance classification accuracy of English texts, naive Bayes method based on base noun phrase (BaseNP) identification is presented. The rising maximum entropy model is applied to the identification. Firstly, use training corpus and user-defined feature templates to generate candidate features. Secondly, the feature selection algorithm computing feature gains is applied to select features. Finally, at the parameter estimation stage, the improved iterative scaling (IIS) algorithm is adopted. The experimental results show that this technique achieved precision and recall rates of roughly 93% for BaseNP identification and the classification accuracy is remarkably improved on this basis. It indicates that shallow parsing of high accuracy is very helpful to text classification

A noise insensitive SVM multi-class classifier is proposed. The algorithm is used to analyze data characteristic in the high-dimension data set. Firstly a noise insensitive SVM two-class classifier is built to tackle the noise problem. On the basis of standard SVM, constraint distance is also considered to determine the optimal separating hyperplane. According to these, the noise insensitive SVM multi-class classifier is designed with edited SVM, confidence interval and one-against-one method.

A rough set based fuzzy neural network algorithm is proposed to solve weather prediction. In order to avoid arriving at local minimum value, the least square algorithm (LSA) is used in the learning process of fuzzy neural network to obtain global convergence. Because structure of fuzzy neural networks, the numbers of rules and the initial weights are difficult to be determined, here the rough sets method is introduced to decide the numbers of rules and original weights. Finally, the proposed algorithm through standard data set is testified to have better rationality and availability than BP algorithm. Experiment results show the algorithm for weather prediction may get better effect.

This paper proposes a sentence space model, which expresses a text by sentence units and keeps the structure of the original text. It accomplishes the TC task by the sentence-class contribution and the double-voting method. Comparing with the VSM, this model has a higher classification accuracy in many data sets.

The battlefield environment affects maneuvers and the war situation. We choose the battle position through a training support vector machine (SVM). We give firstly the method of two-class classification by means of SVM. Choosing of battle position is a problem of multi-class classification, so we adopt a "divide and conquer" approach to train the data set. We use recursively SVM until the result meets the condition of battle position. In the course of training the data, we use one-against-one and decision tree methods. Using the schemes, we select the suited battle position and get better effect.

The classical K-nearest neighbor (K-nn) algorithm in case-based reasoning (CBR) has been used widely. But in real situations, cases often have kinds of features that the classical K-nn algorithm cannot tackle well. Other fuzzy K-nn algorithms may apply well to these perspective systems, but do not adapt to weather prediction. In this paper, we propose a novel fuzzy K-nn algorithm. Because weather is continuous, dynamic and chaotic, in our algorithm, the time function as an adjustable factor is introduced to the similarity-measuring function. Fuzzy logic is used in the retrieval of cases. Experimental results show the efficacy of the algorithm.

In previous attribute reduction researches, the criteria of reduction are intended that the numbers of attributes are the least, the last rules are the simplest or amount of reduction is the most. But in database theory, the criteria are that the redundancy of attributes and dependency of attributes are as few as possible. According to these, authors propose the rough set based attribute reduction algorithm. The decision table is judged firstly whether or not it is consistent. To the complete consistent table, using the knowledge of rough set and information theory, authors get attribute reduction set by discernibility matrix, and compute relevance of attributes through conditional entropy. The best attribute reduction is the set which value is the minimum of average of attribute relevance. To the complete inconsistent table, authors make directly the decision rules with rough operator. The experiment shows it can get better effect. Reduction results of UCI databases are gotten through using the algorithm.