基于机器视觉的农田作物行检测方法研究

发布时间：2018-03-22 14:34

本文选题：作物行检测　切入点：视觉导航　出处：《河南理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：本课题的研究内容为基于机器视觉的农田作物行检测。以农田作物为研究对象,根据其垄行种植的特点,利用数字图像处理技术提取作物行直线,实现农事中耕管理期间农业机器人自动导航的目标。本文的主要研究内容如下:(1)研究农田作物图像的灰度转换和分割过程。通过对不同彩色模型下的农田作物图像进行分析,提出采用RGB模型研究图像。综合考虑了程序的运行时间及处理效果后,提出采用加强绿色分量(G),抑制蓝色(B)和红色分量(R)的颜色特征因子2G-R-B对图像进行灰度转换,并利用Otsu法对经过灰度转换的图像进行二值化处理分割作物区域和背景区域。(2)研究农田作物图像中作物行特征点的提取方法。根据农田作物垄行种植的特点,提出了基于移动窗的作物行特征点提取方法和改进的基于移动窗的作物行特征点提取方法。第一种方法通过构造一个矩形窗口,采用每隔五行逐列扫描二值图像的方法,提取若干能够代表作物行中心的特征点。考虑到相邻两条水平扫描线之间,同一条作物行的特征点所在的列的位置变化不会很大这样一个事实,提出了另一种改进的特征点提取方法,首先利用移动窗口在第一条水平扫描线上逐列扫描提取若干作物行特征点,然后在后续扫描线上,基于前一次扫描获取的特征点所在的列给出一个小的改变,进行邻近区域扫描的方法提取剩余特征点。(3)研究作物行直线检测方法。针对Hough变换峰值检测困难的问题,提出了一个基于Hough变换与消隐点的作物行检测算法。首先利用Hough变换对上一步获取的特征点图像进行直线拟合,提取出所有可能的候选作物行,然后利用K-means聚类算法确定作物行直线形成的消隐点的位置,最后利用消隐点原理消除伪执作物行,识别出正确的作物行。该算法与其他的基于Hough变换的方法相比,不仅能够避免峰值寻找的困难,而且具有速度快、更简单、易于实现的特点。(4)对提出的作物行检测算法进行试验验证与分析。结果表明,本文提出的作物行检测算法,能够有效地检出作物行直线,检测率达到90%以上。
[Abstract]:The purpose of this paper is to detect crop lines based on machine vision. According to the characteristics of ridge planting, the line line of crop is extracted by digital image processing technology. The main research contents of this paper are as follows: 1) the grayscale conversion and segmentation process of crop images are studied. Through the analysis of cropland crop images under different color models, the main contents of this paper are as follows: (1) the main contents of this paper are as follows: (1) the main contents of this paper are as follows: 1. In this paper, RGB model is used to study images. After considering the running time and processing effect of the program, it is proposed that the color feature factor (2G-R-B), which is used to strengthen the green component and suppress the blue and red components, is used to transform the gray level of the image. The method of extracting crop row feature points from farmland crop image is studied by using Otsu method to segment crop region and background area by binary processing. According to the characteristics of crop ridge planting, the paper studies the method of extracting crop row feature points from field crop image. In this paper, a method of crop row feature point extraction based on moving window and an improved method of crop row feature point extraction based on moving window are proposed. Extract a number of feature points that represent the center of a crop row. Taking into account the fact that the position of the column in which the feature points of the same crop row do not change significantly between two adjacent horizontal scan lines, In this paper, another improved feature point extraction method is proposed. Firstly, a moving window is used to extract a number of crop row feature points on the first horizontal scan line, and then on the subsequent scanning line. Based on the list of feature points obtained from the previous scan, a small change is given, and the residual feature points are extracted by scanning the adjacent region.) the method of crop line line detection is studied. Aiming at the problem of Hough transform peak detection, it is difficult to detect the peak value of crop line. A crop row detection algorithm based on Hough transform and blanking point is proposed. Firstly, the feature point images obtained in the previous step are fitted by Hough transform, and all possible candidate crop rows are extracted. Then, K-means clustering algorithm is used to determine the location of the blanking point formed by the straight line of the crop line, and finally, the false crop row is eliminated by using the blanking point principle, and the correct crop row is recognized. The algorithm is compared with other methods based on Hough transform. It can not only avoid the difficulty of finding the peak value, but also has the characteristics of fast, simpler and easy to realize. The proposed crop row detection algorithm is tested and analyzed. The results show that the crop row detection algorithm proposed in this paper is simple and easy to implement. The linear line of crop can be detected effectively, and the detection rate is over 90%.
【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】：TP391.41

【相似文献】