结合遗传算法的优化卷积神经网络学习方法

发布时间：2018-08-24 08:33

【摘要】：经典卷积神经网络学习方法采用最陡下降算法进行学习,学习性能受卷积层和全连接层的初始权重设置的影响较大。采用遗传算法生成多组初始权重,经过选择、交叉和变异操作得到最优权重;采用这些权重作为卷积神经网络的初始权重,其学习性能优于最陡下降算法随机选择的初始权重;采用遗传算法生成的多组权重训练多个卷积神经网络分类器,由其构建联合分类器进行分类,可进一步提高分类正确率。实验结果表明,与经典卷积神经网络方法以及常用的支持向量机、随机森林、后向传播神经网络和极速学习机相比,该方法的分类正确率更高。
[Abstract]:The steepest descent algorithm is used in classical convolution neural network learning. The learning performance is greatly affected by the initial weight setting of convolution layer and full connection layer. Genetic algorithm is used to generate multiple groups of initial weights, and the optimal weights are obtained by selecting, crossover and mutation operations, and using these weights as initial weights of convolutional neural networks, their learning performance is better than the initial weights randomly selected by steepest descent algorithm. Multiple convolutional neural network classifiers are trained by genetic algorithm, which can be used to construct a combined classifier for classification, which can further improve the accuracy of classification. The experimental results show that the classification accuracy of this method is higher than that of classical convolution neural network, support vector machine, random forest, backward propagation neural network and extreme learning machine.
【作者单位】：商丘学院计算机工程学院;甘肃农业大学信息科学技术学院;
【基金】：国家自然科学基金项目(034031122)
【分类号】：TP18

【参考文献】