基于叶片图像的植物识别方法研究

发布时间：2018-05-19 20:00

本文选题：植物叶片识别 + 图像分割　；参考：《河南理工大学》2016年硕士论文

【摘要】：植物识别在维护植物物种多样性、植物资源开发与利用等领域均有着举足轻重的作用。传统的植物识别方法要求操作者拥有扎实的专业知识,并且往往存在着工作量大、工作效率低等问题。利用数字图像处理技术进行植物物种的自动识别成为植物识别领域的研究热点。植物叶片作为植物的重要器官,它具有二维结构且易于进行数字图像采集。因此,在植物自动识别方法的研究之中,叶片常常被作为首选的参考器官。论文从叶片图像预处理、特征提取和分类器设计三个方面对基于叶片图像的植物识别方法进行了研究与改进。在叶片图像预处理阶段,详细介绍了植物叶片图像的灰度化处理、植物叶片图像的分割、植物叶片图像的形态学处理、植物叶片的边界提取及位置归一化方法。为解决具有复杂背景的植物叶片图像分割,本文采用了一种半自动交互式的图像分割方法。相较于传统的阈值分割方法,半自动交互式图像分割方法对具有复杂背景的植物叶片图像能够取得理想的分割效果。在植物叶片特征提取阶段,利用植物叶片的几何特征与纹理特征相组合的方法对植物叶片进行特征提取操作。其中植物叶片的区域几何特征由不变矩特征和若干个叶片几何描述参数共同构成,纹理特征则利用灰度共生矩阵进行提取。在分类器设计阶段,将集成学习算法应用于植物叶片分类器的训练。针对植物叶片识别这个多类别分类问题,提出了一种新型的神经网络集成分类器算法。在Flavia叶片数据库中选取20类叶片图像,每类30张,共计600张图片进行实验。基于神经网络集成的植物叶片识别方法的平均识别精度为91%。与其它分类器算法相比,试验结果表明集成学习算法可以提高植物叶片识别的精度。在论文的最后,设计了一种基于随机森林算法的植物叶片识别软件,且在校园采集的9类植物叶片样本上进行了试验,取得了良好的识别效果。
[Abstract]:Plant identification plays an important role in the maintenance of plant species diversity and the development and utilization of plant resources. Traditional plant recognition methods require operators to have solid professional knowledge, and there are often problems such as heavy workload and low work efficiency. Automatic recognition of plant species using digital image processing technology has become a hotspot in the field of plant recognition. As an important organ of plants, plant leaves have two-dimensional structure and are easy to collect digital images. Therefore, in the research of plant automatic recognition, leaf is often used as the preferred reference organ. In this paper, the plant recognition method based on leaf image is studied and improved from three aspects: leaf image preprocessing, feature extraction and classifier design. In the stage of leaf image preprocessing, the grayscale processing of plant leaf image, the segmentation of plant leaf image, the morphological processing of plant leaf image, the edge extraction of plant leaf and the normalization of position are introduced in detail. In order to solve the problem of plant leaf image segmentation with complex background, a semi-automatic interactive image segmentation method is proposed in this paper. Compared with the traditional threshold segmentation method, semi-automatic interactive image segmentation method can achieve an ideal segmentation effect for plant leaf images with complex background. In the stage of plant leaf feature extraction, the feature extraction operation of plant leaf is carried out by combining the geometric feature and texture feature of plant leaf. The regional geometric features of plant leaves are composed of invariant moment features and several geometric description parameters of leaves, while texture features are extracted by gray level co-occurrence matrix. In the stage of classifier design, the integrated learning algorithm is applied to the training of plant leaf classifier. A new neural network ensemble classifier algorithm is proposed to deal with the multi-class classification problem of plant leaf recognition. In the Flavia leaf database, 20 kinds of leaf images were selected, 30 images per category, a total of 600 images were tested. The average recognition accuracy of the plant leaf recognition method based on neural network integration is 91. Compared with other classifier algorithms, the experimental results show that the integrated learning algorithm can improve the accuracy of plant leaf recognition. At the end of the paper, a plant leaf recognition software based on stochastic forest algorithm is designed, and the experiments are carried out on 9 kinds of plant leaf samples collected on campus, and good recognition results are obtained.
【学位授予单位】：河南理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：Q94;TP391.41

【相似文献】