基于Bagging-PNN算法的树叶分类方法优化

Optimization of Leaf Classification Method Based on Bagging-PNN Algorithm

不同类型的树叶有不同的形状特征,依据叶片的这些特征可以简单而有效地区分不同种类的叶片。该文在Matlab平台上从二值化的树叶图片中提取了13维特征指标,包括长宽比、矩形度、圆形度等,依据这些特征指标,使用概率神经网络(PNN)在83种树叶的数据集上进行实验,识别结果的平均准确率约为86.3%,使用集成学习(Bagging)对分类算法进行改进,使用PNN作为弱分类器,将多个PNN分类器的投票结果作为最终分类结果输出,相比于传统的PNN算法,该文使用的Bagging-PNN算法对于叶片识别准确率提高到了90.3%。

Different kinds of leaves can be distinguished explicitly and effectively according to the leaf shape characteristics which are unique from different types of leaves.In this paper,from binary graph of leaves on the Matlab platform,we extracted 13 dimensional characteristic indices,including vertical and horizontal axis ratio,rectangular,circular,etc.From an experiment based on characteristic data sets of 83 kinds of leaves,we found that the average recognition accuracy is about 86.3%when using probabilistic neural network(PNN).Besides,we used the integrated learning(Bagging)to improve the classification algorithm and took PNN as weak classifier by which determines the voting results of multiple PNN classifiers as the output.Compared with that of the traditional PNN algorithm,the recognition accuracy of Bagging-PNN increased to 90.3%.