基于并行残差卷积神经网络的多种树叶分类

Multiple types of leaves′ classification based on parallel residual convolution neural network

树叶分类识别对于鉴定新的或者稀缺树种至关重要,采用卷积神经网络算法可以实现对树叶图像特征的自动提取,减少繁琐的人工成本,实现使用人工智能的方法来分类树叶。实验采用一种并行残差卷积神经网络和一种加入残差学习的传统Alexnet网络在制作的30种分类树叶的数据集上测试效果并作对比。以上两种方式分别比传统Alexnet网络提高了15.36%和9.36%,而且使网络更轻量化,最高准确率为90.67%,为树种识别研究提供了有效的分类方法。

The leaf classification and identification are of great importance for identifying new or scarce tree species. The convolution neural network algorithm can be used to automatically extract leaf image features,reduce fussy labor costs and classify leaves with the artificial intelligence method. In the experiment,a parallel residual convolution neural network and a traditional Alexnet network with residual learning are used to test the 30 kinds of classified leaves in the produced data set,and the testing effects of the two networks are contrasted. The accuracy obtained with the above two methods are respectively 15.36% and 9.36% higher than that obtained with the traditional Alexnet network,and the highest accuracy reaches 90.67%,which makes the network lightweight. Therefore,it provides an effective classification method for the research on tree species identification.