【摘要】 为提高花卉图像的识别率,实现良好的花卉分类效果,提出一类改进型LeNet-5卷积神经网络模型。将原LeNet-5卷积神经网络模型的S4层与C5层之间的连接方式改为全连接,将S2层、S4层的池化操作分别设置为均值池化、最大池化。在此基础上采用随机梯度下降方法和Dropout防止过度拟合的方法相结合的算法,对Oxford-17花卉数据集进行仿真实验。实验结果表明,改进型LeNet-5卷积神经网络有效且可行,该模型对花卉图像的识别率高达96.5%,与未改进的LeNet-5卷积神经网络模型相比,识别率提高了6.5%。更多还原

【Abstract】 To improve the recognition rate of flower images and achieve a good flower classification result,an improved LeNet-5 convolution neural network model was proposed.The connection mode between the S4 layer and the C5 layer of the original LeNet-5 convolutional neural network model was changed to full connection,and the pooling operations of the S2 layer and the S4 layer were respectively set to mean pooling and maximum pooling.Based on this,a random gradient descent method and a Dropout method to prevent overfitting were adopted and simulation experiments were carried out on the Oxford-17 flower dataset.Results of experiments show that the improved LeNet-5 convolutional neural network is effective and feasible.The model has a recognition rate of 96.5%for flower classification,and the recognition rate is increased by 6.5%compared with the unimproved LeNet-5 convolutional neural network model.更多还原