基于多颜色空间特征融合的作物识别方法研究

发布时间：2018-03-09 23:00

本文选题：阴影干扰　切入点：颜色空间　出处：《天津理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：基于视觉的农业机器人是农业智能化、精确化和机械化发展的趋势,其核心问题是怎样通过视觉对采集到农田图像进行快速、准确的分析。但是,由于植株相互遮挡以及太阳光入射角的影响,植株和杂草会被外界和自身间的阴影干扰,使得识别图像阴影区的植株和杂草比较困难,进而降低了图像目标识别的准确率,同时影响了机器人工作效率。目前图像处理中存在许多不足之处:算法复杂性、实时性差和阴影区域图像细节丢失和识别率不足等问题,所以对图像中阴影去除的研究有重要的意义。本文主要从光照阴影和分类识别率不高这两方面进行了深入研究。1)对于室外农田中环境下,采集图像受阴影干扰的问题,本文采用了一种逐步消除阴影的方法。该方法是将RGB图像转换到其他颜色空间中,分析利用不同颜色通道中阴影的特性,同时结合直方图、熵值、均方误差和峰值信噪比等得到仿真数据来分析颜色通道特性,最终构造出适合农田场景下的阴影去除的颜色空间,即UVI颜色空间。在该空间可以利用简单的算法实现较好的去除阴影,实现图像中前景农作物不受背景阴影干扰下的识别。2)对农作物与杂草的识别精度不高和识别时间较长问题,本文提出一种基于小波变换和主成分分析结合的多特征数据融合方法。该方法对在多颜色空间中提取的图像组合特征数据采用小波变换和主成分分析法进行融合,再利用支持向量机决策下实现农作物与杂草的高识别率。该方法解决了组合特征数据结构复杂和分类器输入维数过高的难点,提高了农作物与杂草的识别能力与速度。3)基于VS2012开发平台,设计了组合颜色空间阴影去除软件,能够有效的在界面上操作图像,提高图像处理速度。软件代码可以移植到机器人系统,具有一定的实用价值。
[Abstract]:The vision based agricultural robot is the development trend of agricultural intelligence, precision and mechanization. Its core problem is how to quickly and accurately analyze the farmland images collected by vision. However, Because of the mutual occlusion of plants and the influence of the incidence angle of the sun, the plants and weeds will be interfered by the outside and their own shadows, which makes it difficult to identify the plants and weeds in the shadow region of the image, thus reducing the accuracy of image target recognition. At the same time, it affects the efficiency of the robot. At present, there are many shortcomings in image processing, such as algorithm complexity, poor real-time performance, loss of image details in shadow area and insufficient recognition rate, etc. Therefore, the study of shadow removal in the image is of great significance. This paper studies the shadow interference in the environment of outdoor farmland from the two aspects of light shadow and low classification recognition rate. In this paper, we adopt a method of eliminating shadow step by step. The method is to convert RGB image into other color space, analyze and utilize the characteristics of shadow in different color channels, and combine histogram, entropy value. The simulation data such as mean square error and peak signal-to-noise ratio are obtained to analyze the color channel characteristics. Finally, the color space suitable for shadow removal in farmland scene is constructed. That is, UVI color space. In this space, a simple algorithm can be used to remove shadows and realize the recognition of foreground crops without background shadow. 2) the recognition accuracy of crops and weeds is not high and the recognition time is long. In this paper, a multi-feature data fusion method based on wavelet transform and principal component analysis is proposed, in which wavelet transform and principal component analysis are used to fuse the image combination feature data extracted in multi-color space. Using support vector machine (SVM) to realize the high recognition rate of crops and weeds, this method solves the difficulties of complex data structure and high input dimension of classifier. Based on the VS2012 development platform, the combined color space shadow removal software is designed, which can effectively operate the image on the interface. The software code can be transplanted to the robot system and has some practical value.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S126;TP391.41

【相似文献】