【摘要】 为了提高神经网络对手写数字的识别率,基于Adaboost思想改进Adaboost-BP二分类算法,实现用于多分类的Adaboost-BP算法,提高了神经网络对手写数字的识别率。改进了"弱"分类器权重值的计算公式,将权重值归一化处理的步骤放到"弱"分类器迭代训练完成之后,"强"分类器的构成不使用符号函数而是直接计算分类结果。实验数据采用MNIST手写数据库,实验结果显示改进的Adaboost-BP算法构造出的"强"分类器分类结果正确率明显高于"弱"分类器。改进的Adaboost-BP算法可明显提高手写数字识别正确率。更多还原

【Abstract】 In order to improve the recognition rate of handwritten digits by neural network, the Adaboost-BP binary classification algorithm is improved based on the idea of Adaboost, and the Adaboost-BP algorithm for multi-classification is realized, which improves the recognition rate of handwritten digits by neural network. In this paper, the calculation formula of the weight value of the "weak" classifier is improved. The step of normalizing the weight value is put after the iterative training of the "weak" classifier, and the composition of the "strong" classifier is calculated without using the symbol function but directly computes the classification results. The experimental data is based on MNIST handwritten database. The experimental results show that the correct rate of the "strong" classifier constructed by the improved Adaboost-BP algorithm is obviously higher than that of the "weak" classifier. The improved Adaboost-BP algorithm can obviously improve the accuracy of handwritten digit recognition.更多还原