5.1 Case Background
5.1.1 BP_Adaboost model
Thought Adaboost algorithm is to combine a plurality of "weak" output of the classifier to produce an effective classification. The main steps are as follows: Firstly, weak learning algorithm and the sample space ($ X $, $ Y $), to find the $ m $ training data set from the sample space, the weight of each weight training data is $ \ frac {1 } {m} $. Then pay more attention to these individuals when training with a weak learning algorithm iteration $ T $ times, each time after the operation in accordance with the classification results are updated training data weight distribution for categorical failure of individual training given greater weight, the next iteration. Weak classifier obtained Category Function Sequence $ {f_1}, {f_2}, ..., {f_T} $, each category assigned a weighting function by iteration, function better classification results, which corresponds to the greater weight. After $ T $ iterations, the final strong classifiers obtained by the function $ F $ function weighted weak classifiers. BP_Adaboost the BP neural network model as a weak classifier, repeated training BP neural network prediction sample output, the strong classifier to obtain a plurality of weak classifiers BP neural network composed by Adaboost algorithm.