矮化密植枣园视觉导航路径检测方法的研究

发布时间：2018-08-31 14:34

【摘要】：农业装备智能化自主导航技术能够促进和提高农业机械化生产效率。新疆是我国红枣主产区之一,主要采用矮化密植种植模式,为视觉导航的应用提供了有利的应用条件。本文以矮化密植枣园田管时期和收获时期机械作业环节为研究对象,重点探讨了田管时期和收获时期两个阶段的枣园环境目标像素分布特征,目标特征离散点群的检测算法及视觉导航路径的拟合方法。本文主要研究内容如下:(1)搭建矮化密植枣园视觉导航路径检测系统和图像采集方案。软件系统由目标像素分布规律分析模块,图像预处理模块,路径提取模块,视频检测模块构成,导航路径性能参数提取模块。硬件系统图像采集设备选用佳能IXUS870相机,图像处理硬件平台采用处理器AMD Athlon(tm)II X2 240 Processor,主频:2.8GHz,内存:2GB,系统类型:32位WINDOWS 7操作系统的计算机。(2)田管时期及收获时期机械化生产过程的视觉导航候补点群的检测算法研究。针对田管时期,将图像分为两个小类来处理,第一类是行间有杂草,施肥和剪枝环境,第二类环境为中耕作业环境。针对第一小类,采用|R-B|色差法灰度化,第二类采用传统灰度化方法,两类灰度化均采用大津法分割图像,而后采用梯形扫描算法、面积法、垂直投影等方法去噪。针对收获时期图像,将图像归为晴天、阴天、逆光,顺光、背景多元叠加等五个类型进行研究。采用B分量图像进行灰度化,利用基于行扫描自适应方法分割目标背景,通过行间扫描方法、灰度垂直投影、形态学处理去噪。通过引入趋势线方法描述行间边缘走势,采用点到直线的距离公式提取离散点候补点群,结果证明,提取目标点群符合枣树行列分布特征。(3)采用最小二乘法对离散候补点群拟合边缘线,而后求取两条边缘线中线作为视觉导航路径。试验表明,本文研究算法具有比较高的准确率和鲁棒性,检测到的特征点与各个时期的特征分布吻合。在本文搭建的软件系统下,图像的宽乘高采用230×168,田管时期静态图像检测:平均每张图像路径检测平均时间小于14.0s,检测准确率高于78.3%,视频检测准确率达80%以上,每帧图像检测平均耗时低于2.3s;收获时期静态图像检测单一工况条件下准确率高于83.4%,每张图像检测平均时间小于9.2s,多工况条件为45%,每张图像平均检测时间低于9.4s,视频检测单一工况准确率达81.3%以上,每帧检测平均耗时低于1.7s,多工况条件检测准确率为42.3%,每帧平均检测耗时为1.0s。
[Abstract]:Intelligent autonomous navigation technology of agricultural equipment can promote and improve the efficiency of agricultural mechanization production. Xinjiang is one of the main producing areas of red jujube in China, which mainly adopts dwarf and dense planting pattern, which provides favorable conditions for the application of visual navigation. In this paper, the characteristics of pixel distribution of environmental targets in jujube orchard during the two stages of field management and harvest were discussed, taking the mechanical operation links of the dwarf and dense jujube orchard as the research object. The detection algorithm of discrete point group of target features and the fitting method of visual navigation path. The main contents of this paper are as follows: (1) build a vision navigation path detection system and image acquisition scheme for dwarf and dense jujube garden. The software system consists of object pixel distribution analysis module, image preprocessing module, path extraction module, video detection module, navigation path performance parameter extraction module. Canon IXUS870 camera is used for image acquisition in hardware system. The hardware platform of image processing adopts the computer of processor AMD Athlon (tm) II X240 Processor, main frequency: 2.8 GHz, memory of 2 GBs, system type of 32 bit WINDOWS 7 operating system. (2) Research on the detection algorithm of visual navigation alternate point group in mechanized production process in field management period and harvest period. In view of the field management period, the image was divided into two subgroups, the first was the interrow weed, fertilization and pruning environment, and the second was the intercropping environment. For the first small category, R-B color difference method is used to grayscale, the second kind is traditional grayscale method, the two kinds of grayscale method are used to segment image, then trapezoidal scanning algorithm, area method and vertical projection method are used to remove noise. In this paper, the image of harvest period is classified into five types: sunny, overcast, reverse light, shinning and background multivariate superposition. B component image is used to grayscale and line scan adaptive method is used to segment the background of the target. The method of interline scanning gray vertical projection and morphological processing is used to remove noise. The trend line method is introduced to describe the interline edge trend, and the distance formula from point to line is used to extract the alternate point group of discrete points. (3) the least square method is used to fit the edge line of the discrete alternate point group, and then two central lines of edge line are selected as the visual navigation path. The experimental results show that the proposed algorithm has high accuracy and robustness, and the detected feature points are consistent with the feature distribution of each period. In the software system built in this paper, the width multiplication and height of the image is 230 脳 168. The static image detection during the field management period: the average time of each image path detection is less than 14.0s, the detection accuracy is higher than 78.3%, the video detection accuracy is more than 80%. The average time of image detection per frame is less than 2.3 s, the accuracy of static image detection in harvest period is more than 83.4 under a single working condition, the average time of each image detection is less than 9.2 s, the multi-working condition is 45, the average detection time of each image is less than 9.4 s, and the average detection time of each image is less than 9.4 s. The accuracy rate of frequency detection in a single working condition is more than 81.3%. The average detection time per frame is less than 1.7 s, the detection accuracy of multi-working conditions is 42.3 and the average detection time per frame is 1.0 s.
【学位授予单位】：石河子大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S665.1;TP391.41

【参考文献】