基于叶片图像分析的葡萄品种识别方法研究

发布时间：2018-12-30 21:40

【摘要】：葡萄品种多且繁杂,识别较为困难,单纯依靠人力来进行品种识别,要求操作者拥有专业的葡萄分类知识和丰富的实践经验,并且工作量大,会影响识别的客观性。基于叶片图像分析的葡萄品种识别,可以帮助人们鉴定葡萄品种,缩短鉴定时间,普及葡萄品种知识,增加民众对葡萄品种的认识。本文以成熟期不同葡萄品种叶片作为研究对象,首先对其叶片图像进行预处理,然后对于人工设计特征和深度学习特征进行特征提取,最后将提取的特征向量作为参数输入,采用支持向量机分类器建立识别模型,实现对葡萄品种的识别,为葡萄品种识别提供一种新方法。本文的主要内容如下:(1)葡萄叶片样本图像预处理。对扫描的葡萄叶片样本图像进行分析,根据叶片图像和背景的像素差异较大,选择基于灰度阈值的方法进行分割并利用最小外接矩形提取叶片图像,使用形态学、灰度化和几何变化等技术对叶片图像进行预处理,为特征提取和识别打下基础。(2)基于人工设计特征和深度学习特征的葡萄叶片特征提取。分析葡萄叶片形态结构,利用人工设计的方向梯度直方图特征和基于卷积神经网络的深度学习特征表示叶片的形态特征。方向梯度直方图特征可以很好的表达叶片图像的边缘梯度信息和空间纹理信息等,基于卷积神经网络的深度学习特征可以自主学习葡萄叶片的特征,具有高维深度语义表达性,能够完整丰富地表达葡萄叶片属性。(3)根据提取的葡萄叶片特征,建立基于支持向量机葡萄的叶片品种识别模型。在Matlab2010b环境下进行实验,分别对人工设计特征(包括Hu不变矩、灰度共生矩、方向梯度直方图等特征)和基于卷积神经网络的深度学习特征进行建模实验,并对实验结果进行分析,实验结果表明,使用方向梯度直方图特征的识别率达到86.67%,使用深度学习特征的识别率达到88.33%。综上所述,通过对葡萄叶片形态特征的分析,本文提出了采用高维人工设计特征和深度学习特征表示葡萄叶片形态结构的方法,利用支持向量机模型进行葡萄叶片的品种识别,取得了较好的结果,为同种科目不同品种的植物叶片识别提供了一种借鉴方法,具有广阔的应用前景。
[Abstract]:Grape variety is more and more complicated, so it is difficult to identify grape varieties. Relying solely on manpower to identify grape varieties requires operators to have professional knowledge of grape classification and rich practical experience, and heavy workload, which will affect the objectivity of identification. The grape variety recognition based on leaf image analysis can help people to identify grape varieties, shorten the identification time, popularize grape variety knowledge and increase people's understanding of grape varieties. In this paper, the leaves of different grape varieties in mature period are used as the research object. Firstly, the leaf image is preprocessed, then the artificial design features and depth learning features are extracted, and the extracted feature vectors are input as parameters. Support vector machine classifier is used to establish recognition model to realize grape variety recognition and to provide a new method for grape variety recognition. The main contents of this paper are as follows: (1) sample image preprocessing of grape leaves. According to the pixel difference between leaf image and background, the grayscale threshold method is chosen to segment the leaf image and extract the leaf image by using the minimum external rectangle, and the morphology is used. Grayscale and geometric changes are used to preprocess the leaf image, which lays the foundation for feature extraction and recognition. (2) Grape leaf feature extraction based on artificial design feature and depth learning feature. The morphological structure of grape leaves was analyzed, and the morphological features of grape leaves were represented by artificial direction gradient histogram and depth learning feature based on convolution neural network. The directional gradient histogram features can well express the edge gradient information and spatial texture information of leaf images. The depth learning features based on convolution neural network can learn the characteristics of grape leaves independently and have high dimensional depth semantic expression. (3) based on the extracted characteristics of grape leaves, a vinifer-based vane variety recognition model based on support vector machine (SVM) was established. The experiments were carried out in Matlab2010b environment. The artificial design features (including Hu invariant moment, gray level co-occurrence moment, directional gradient histogram and so on) and the depth learning feature based on convolution neural network were modeled, respectively. The experimental results show that the recognition rate of directional gradient histogram is 86.67 and the recognition rate of depth learning is 88.33. To sum up, by analyzing the morphological characteristics of grape leaves, this paper proposes a method to express the morphological structure of grape leaves by using high-dimensional artificial design features and in-depth learning features, and the support vector machine model is used to identify the varieties of grape leaves. Good results have been obtained, which provides a reference method for the recognition of plant leaves of different varieties in the same subject, and has a broad application prospect.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】