Technology Research Of Forestry Information Text Classification Based On Gauss Mixture Model

Forestry information text classification is established according to the classifier, let the computer of forestry information for a given set of text classification process is valuable both in theory and in practice. This paper introduces in detail the process of feature extraction of forestry information text, forestry information text classification use feature matrix to classify, a detailed analysis of Gauss mixture model and the principle of neural network algorithm based on weighted Gauss Newton, classification for the forestry information text provides a new train of thought.The main conclusions of this study:(1) The Gauss mixture model algorithm is introduced to study the forestry information text classification. The parameters are estimated by the parameter estimation algorithm of Gauss mixture model, because the EM parameter estimation algorithm gradually converge to the maximum, but the choice of initial value for the EM algorithm convergence effect ultimately play a significant influence, so this paper puts forward the K-means algorithm to estimate the initial results which are assigned to the EM parameter estimation algorithm, the accuracy of the parameters estimation of EM algorithm has been greatly improved. Because the K-means algorithm is a clustering algorithm, when initial value input Gauss mixture model, the forestry information types match the sample group, so there is a limitation to the sample, the choice of forestry information text is balanced here.(2) Based on weighted Gauss-Newton neural network algorithm is introduced into the research of forestry information text classification, Based on weighted Gauss-Newton neural network algorithm(RW-GN) is the algorithm of improved BP neural network algorithm, is to optimize the parameters of neural network training algorithm, the proposed algorithm improves the rate of correct classification. The algorithm is based on the neural network algorithm, so this algorithm has high stability and suitable for the classification of uneven forestry information text, the experimental results show that, Based on weighted Gauss-Newton neural network algorithm for uneven forestry information text classification obtained a high accuracy rate.(3) Two kinds of algorithms have been presented in this paper are applied in forestry information of balanced and uneven samples. Sample selection is based on the advantages of algorithm, experimental results will be compared between the forestry information text classification algorithm and common algorithm, such as: BP neural network, support vector machine, Bayes, the decision tree algorithm. The experimental results show that Gauss mixture model algorithm fits for balanced forestry information text classification, based on weighted Gauss-Newton neural network algorithm fits for uneven forestry information text classification, the two algorithms can obtain a higher accuracy, have high practical value.