基于CapsNet神经网络的树叶图像分类模型

A Neural Network-Based Leaf Image Classification Model

对树木研究的基础是对其进行分类处理.本文结合CapsNet神经网络模型,以提高树叶分类的准确率为目的,使用实验室拍摄的10种树叶图片建立树叶分类模型.考虑到模型效率和图像大小,在原有CapsNet上与传统卷积神经网络相结合,通过优化动态路由算法对CapsNet进行改进,得到了E-CapsNet网络模型,同时与经典的神经网络模型AlexNet和Inception V3模型进行对比.经过50次epoch的训练,模型训练准确率最高达到99.15%,验证集的准确率为98.51%,测试集准确率为98.63%,对比原CapsNet网络,测试集准确率提高了2.51%.实验结果表明,改进后的E-CapsNet模型实现了更高的精度.

The basis of research of trees is to classify them.Combined with the CapsNet neural network model,10 kinds of leaf pictures taken in the laboratory are used to establish a leaf classification model so as to improve the accuracy of leaf classification.Considering the efficiency and image size of the model,an E-CapsNet network model is obtained by combining the original CapsNet with the traditional convolutional neural network and optimizing the dynamic routing algorithm to improve the CapsNet.At the same time,the E-CapsNet network model is compared with the classical neural network model AlexNet and InceptionV3 model.After 50-epoch training,the highest accuracy of model training is 99.15%,the accuracy of verification set is 98.51%,and the accuracy of test set is 98.63%.Compared with that of the original CapsNet network,the accuracy of the test set is improved by 2.51%.The experimental results show that the improved E-CapsNet model achieves higher accuracy.