基于边缘检测和区域定位的玉米根茎导航线提取方法

Extraction Method of Corn Rhizome Navigation Lines Based on Edge Detection and Area Localization

基于玉米根茎图像信息,提出一种基于边缘检测和区域定位的玉米根茎导航线提取方法。首先,利用最大类间方差法自动分割2G-R-B灰度图像,得到二值化图像,结合形态学处理、位置/面积去噪方法提高二值化图像质量,对去噪图像按列累加得到列像素累加曲线;针对传统方法得到的特征点中伪特征点较多的问题,引入高斯滤波器平滑累加曲线,并运用极值法减少玉米根茎伪特征点的干扰,在提取玉米茎秆边缘直线时,提出基于最远茎秆成像宽度的双侧边缘判别思路,通过扫描每条边缘直线的四边形封闭邻域有效剔除伪边缘直线;最后,根据边缘直线二次定位玉米的根茎区域范围,并剔除伪特征点,采用最小二乘线性拟合方法准确提取导航线。试验表明,本文算法处理一幅1280像素×720像素图像耗时约236 ms,特征点拟合准确率为92%。与传统方法相比,本文算法精度高、实时性好,在缺苗、杂草较多和株距不标准的情况下仍具有较强的鲁棒性。

Based on the corn rhizome image information,a corn field navigation line extraction method combining edge detection and area localization was proposed.Firstly,the 2G-R-B grayscale image was segmented and the binary image was obtained by using the maximum between-class variance.The morphological processing was combined with position/area denoising methods to improve the quality of the binary image and reduce the noise.The images were accumulated in columns to obtain the column pixel accumulation curve.The traditional method needed to set the distance threshold when extracted the feature points.Gaussian filter was used to smooth the accumulation curve and extreme value method was used to reduce the interference of pseudo feature points in maize roots and stems.When extracted the straight lines of corn stalk edges,a two-sided edge discrimination method was proposed based on the image width of the furthest stalk,and the pseudo-edge straight lines were effectively eliminated by scanning the closed quadrilateral neighborhood of each edge line.Finally,based on the straight line of the edge,the local area of the corn rhizome was relocalized and the false feature points were eliminated.The least-squares linear fitting method was used to accurately extract the navigation lines.The experimental results showed that the algorithm took about 236 ms to process a 1280 pixels×720 pixels image,and the accuracy of feature point fitting was 92%.Compared with the traditional methods,the algorithm had the characteristics of high accuracy and good real-time performance.The algorithm was still more robust in the case of lack of seedlings,more weeds,and non-standard plant spacing.It can provide visual navigation for intelligent agricultural machinery to control corn diseases and insect pests.