基于高分辨率光谱图像采集及混合模型的植物病害检测方法

发布时间：2018-12-16 23:27

【摘要】：基于数字化的无损伤的植物病害识别已成为作物病害诊断的趋势,论文针对植物病害症状复杂、现有仅通过单一的图像对比识别的诊断技术检测效率低等问题,综合利用光谱成像、光谱分析、光谱数据库技术以及颜色科学等诸多领域的知识,开展对植物叶片主要病害的快速、无损检测方法研究,并在此基础上建立了病害的快速检测模型和病害诊断系统。通过对光谱立方体的构造的深入研究,基于高分辨率光谱图像格式,提出了一种通用的尤其适用于高分辨率图像光谱数据快速处理的数据结构;与SQL Server数据库结合,基于上述高分辨率光谱成像系统,对树木等园艺类植物的上百株样本的培养、采集和分析,获得了园艺类病害的大量的图像和光谱数据,为找到不同浓度,不同发病时段的各种作物病害的光谱规律以及色彩差异提供了数据依据。本论文的研究结果为植物病害的快速诊断提供了新的手段,并为计算机技术、信息技术以及光谱技术在农业中的应用提供了实例,具有重要的理论和实践意义。文章从课题背景入手,在阐述成像光谱技术、植物病害检测相关理论的基础上,首先从光谱成像立方体的数据结构入手,介绍光谱成像技术的原理以及常见的光谱成像技术,并搭建基于病害检测光谱成像技术及LCTF的光谱成像硬件系统;接着讨论光谱图像数据的本地和网络的数据存储结构,并将提出一种针对高分辨率光谱图像数据的数据库模型以及本地化存储模型,对LCTF光谱成像系统的数据处理及相关的软件进行设计;最后通过比较主成分分析法、线性判别分析、神经网络方法等几种常用的特征提取方法并总结了各种分析对其在病害判别分析中使用的适宜程度进行了分析,提出了一种基于Fisher方法的逐步判别模型进行光谱降维,并使用RBF神经网络对降维前后的光谱分类能力进行检验。
[Abstract]:It has become the trend of crop disease diagnosis based on digitized and undamaged plant disease recognition. Aiming at the complex symptom of plant disease and the low detection efficiency of the existing diagnosis technology based on single image contrast recognition, etc. Comprehensive use of spectral imaging, spectral analysis, spectral database technology, color science and other fields of knowledge, to develop rapid, non-destructive detection of major plant leaf diseases, On this basis, the rapid detection model and the disease diagnosis system are established. Based on the high resolution spectral image format, a universal data structure is proposed, which is especially suitable for fast processing of spectral data of high resolution image. Combined with SQL Server database, based on the above high resolution spectral imaging system, hundreds of samples of horticultural plants such as trees were cultured, collected and analyzed. A large number of images and spectral data of horticultural diseases were obtained. The spectral law and color difference of various crop diseases in different time periods provide data basis. The results of this paper provide a new method for the rapid diagnosis of plant diseases, and provide examples for the application of computer technology, information technology and spectral technology in agriculture, which has important theoretical and practical significance. This paper begins with the background of the subject, on the basis of expounding the related theories of imaging spectrum technology and plant disease detection, starting with the data structure of spectral imaging cube, introduces the principle of spectral imaging technology and the common spectral imaging technology. The hardware system of spectral imaging based on disease detection and LCTF is built. Then, the local and network data storage structures of spectral image data are discussed, and a database model and localized storage model for high-resolution spectral image data are proposed. The data processing and related software of LCTF spectral imaging system are designed. Finally, by comparing principal component analysis (PCA), linear discriminant analysis (LDA), neural network (Ann) and other commonly used feature extraction methods, the suitable degree of their application in disease discriminant analysis is analyzed. A stepwise discriminant model based on Fisher method is proposed for spectral dimensionality reduction and RBF neural network is used to test the spectral classification ability before and after dimensionality reduction.
【学位授予单位】：辽宁科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：S43;TP391.41

【相似文献】