Abstract:

A study of soybean row line extraction methods under high light conditions was conducted to address the problems of complex soybean field environment， high light intensity and poor adaptability of commonly used crop feature extraction methods in southern China. Using the 2G-R-B algorithm， this study extracted green features from soybean plants， and performed morphological processing， small area denoising of the grayscale image to remove the image background and noise from the image. This study extracted soybean field crop feature points based on the SUSAN corner point detection algorithm，and then used the location features of the feature points to automatically categorise the feature points， and the MLESAC algorithm was used to fit the discrete points to obtain the crop row lines. The results show that the algorithm had a strong noise immunity performance. Using this algorithm could accurately extract the feature points on the soya bean seedling strip， and use the location features of the feature points to automatically categorise the feature points and extract the soya bean seedling strip line on both sides of the monopoly. The accuracy of soya bean seedling strip line extraction was 93%， which could provide a research basis for field navigation operations.This results provided a reliable navigation method for autonomous field navigation operations such as weeding and fertilisation of soybean seedlings.

Key words: soybean, SUSAN corner point, MLESAC algorithm, lines

摘要：

针对中国南方大豆田间环境复杂，光照强度高，常用的作物特征提取算法适应性差的问题，对高光照条件下的大豆行线提取方法进行研究。通过超绿特征（2G-R-B）算法，提取大豆植株绿色特征；对灰度图进行形态学处理、小面积去噪，去除图像背景和图像中的噪声；根据SUSAN角点检测算法，提取出大豆田间作物特征点；利用特征点的位置特征自动归类特征点，采用MLESAC算法对离散点进行拟合，得到作物行线。结果表明，该算法具有较强的抗噪性能，可以准确地提取大豆苗带上的特征点，利用特征点的位置特征自动归类特征点，提取出垄上两侧的大豆苗带行线，大豆苗带行线提取准确率为93%，可以为田间导航作业提供研究基础。研究结果为大豆苗期除草、追肥等田间自主导航作业提供了一种可靠的导航方法。

关键词: 大豆, SUSAN角点, MLESAC算法, 行线