基于叶片图像的植物识别方法

 基于计算机的植物自动识别是植物识别分类学的发展趋势，本文提出了一种基于植物叶片图像的植物自动识别方法。该方法在对叶片图像进行亮度校正、中值滤波和阈值分割等预处理后，计算植物叶片的偏心率、圆形性、圆形度指标、方向角、最小矩形宽轴/长轴、最佳椭圆短轴/长轴6个形状特征参数和植物叶片的二阶矩、对比度、相关、熵、逆差矩5个纹理特征参数，再使用径向基人工神经网络设计了植物自动识别的分类器。通过对3种植物的60个叶片图像进行实验，仅用植物叶片形状特征进行植物识别的平均正确识别率为70.83%，利用植物叶片形状特征和纹理特征进行植物自动识别的平均正确识别率为83.3%，并得到了径向基人工神经网络的参数。实验结果表明，植物叶片图像的纹理特征能够提高植物自动识别的平均正确率，基于植物叶片图像的植物自动识别是切实可行的，研究成果为深入研究植物自动识别分类系统奠定了一定的理论基础。

Plant Recognition Method Based on Leaf Images

This paper proposes a plant automatic recognition method based on the leaf images, to be used in the computer-based automatic recognition system of the plant taxonomy. The color image is first transformed into a grey-level and a binary image in pre-processing including brightness correction, median filter and threshold segmentation. Six relative shape parameters and 5 texture parameters are then calculated, respectively, from the binary image and the grey-level image. The 6 shape parameters are the eccentricity, roundness, roundness index, direction angle, width and length ratio of the smallest rectangle that can cover the leaf, and short and long axes ratio of the best ellipse that can cover the leaf; and the five texture parameters are the second moment, contrast, correlation, entropy, moment deficit. Finally, an automatic recognition classifier based on RBF neural network is designed to determine which type of plant the leaf is from with the image of the leaf being used as a sample. Then, 60 images of leaf from 3 plants are used as samples to test the performance of the automatic recognition classifier. The average correct recognition rate reaches 70.83% when only the 6 shape parameters are used as the input data of the classifier, and it reaches 83.33% when both shape parameters and texture parameters are used as the input data of the classifier. The results show that the texture features can improve the average correct rate, and that the plant automatic recognition based on the leaf images is feasible.