基于概率神经网络和分形的植物叶片机器识别研究

Leaf recognition for plant based on Probabilistic Neural Networks and fractal

【目的】提出一种将植物叶片的形状特征与其纹理特征相结合的综合特征识别方法,有效解决了传统的植物机器识别分类特征单一且识别率较低的问题,为植物的快速机器识别提供技术参考。【方法】提取植物叶片样本的综合特征信息,以概率神经网络(PNN)为分类器对所得的特征信息进行训练,训练好的网络用来识别植物叶片的类别,从而确定相应植物的种类。【结果】有效提取了含有8个分量的植物叶片综合特征向量,通过对PNN分类器的训练,实现了30种植物叶片的快速机器识别,平均识别率达98.3%。比较测试表明,若去掉叶片纹理特征,单以其形状特征作为识别依据,平均识别率仅为93.7%。【结论】植物叶片综合特征识别方法有效弥补了传统单特征识别方法的不足,使识别效率得到了较大的提高。

[Objective]This paper gave a recognition approach, which combined the shape feature and the texture feature of the plant leaf, to effectively solve the problem that the classification features of traditional plant recognition were usually not synthetic and the recognition rate was always slightly low,hoping to provide technical reference for rapid machine recognition for plant. [Method] Firstly, the synthetic features of leaf were extracted; secondly the values of synthetic features of leaf, which had been extracted, were put into a classifier, which is Probabilistic Neural Networks（PNN）, to be trained; finally, the PNN trained well could work to classify the plant leaves. [Result] The synthetic feature vector of plant leaf, which included 8 elements, were extracted efficiently. By training the classifier PNN, the rapid recognition for thirty kinds of plant leaves was realized and the average recognition rate reached 98.3 %. Comparison tests demonstrated that if the shape features of plant leaf was solely used as the recognition features without the texture features, the average recognition rate just reached 93.7%. [Conclusion] The synthetic feature recognition method has effectively made up for the drawbacks of traditional recognition method to-wards unitary feature, strongly advancing the recognition rate.