基于分层卷积深度学习系统的植物叶片识别研究

Leaf image recognition based on layered convolutions neural network deep learning

深度学习已成为图像识别领域的研究热点。本文以植物叶片图像识别为研究对象,对单一背景和复杂背景图像分别给出了优化预处理方案;设计了一个8层卷积神经网络深度学习系统分别对Pl@ant Net叶片库和自扩展的叶片图库中33 293张简单背景和复杂背景叶片图像进行训练和识别,并与传统基于植物叶片多特征的识别方法进行了比较分析。实验证明:本文提供的CNN+SVM和CNN+Softmax分类器识别方法对单一背景叶片图像识别率高达91.11%和90.90%,识别复杂背景叶片图像的识别率也能高达34.38%,取得了较好的识别效果。利用本文实现的分层卷积深度学习识别系统在数据量大而无法做出更多优化的情况下,叶片图像的识别率更高,尤其是针对复杂背景下的叶片图像,取得了极佳的识别效果。

Deep learning has been recently becoming research hotspot in the field of image recognition. In this study, plant leaf images were used as recognition objects. Plant leaf images were divided into single and complex background and were treated by using image segmentation method. We design a deep learning system which includes eight layers of Convolution Neural Network （CNNs） to identify leaf images. And then the deep learning system was tested with 33 293 leaf sample images which come from PI@ antNet libraries and our extending leaf libraries for image training and recognition. Compared with the traditional identification methods, the classifier provided in this paper has achieved better recognition effect. For CNN ＋ SVM and CNN ＋ Softmax recognition system, the simple background leaf recognition rate reaches 91.11% and 90.90% , and the recognition rate of complex background reaches 34. 38%. The system has higher recognition rate for the large amount of leaf images with no more optimization, and has a higher recognition rate especially for the recognition of complex background images.