Abstract: Aiming at the problems of low precision and slow speed of navigation line extraction caused by environmental factors such as stubble, soil and light on wheat harvest edge, a method of wheat harvesting edge navigation line extraction based on horizontal projection and gradient descent was proposed to realize precise wheat harvesting operations. Firstly, the image was segmented by LAB threshold segmentation and morphological filtering, then the horizontal projection was performed to extract the pseudo-feature points of wheat harvesting edge, the pseudo-feature points were least squares fitted to obtain the ROI region where the edge feature points were located, and the edge feature points were extracted by Canny edge detection of the region, and finally the wheat harvesting edge navigation line was fitted by the gradient descent algorithm, thus it solved the problems of low accuracy and slow fitting speed of the navigation line encountered in traditional algorithms. The experimental results showed that the average time required to process an image with a resolution of 640×360 pixels was 163 ms with a low contrast between harvested and unharvested areas of wheat, and the success rate of the generated navigation baseline was as high as 95%, which provided a reliable and real-time navigation method for autonomous walking of intelligent agricultural machinery in wheat fields.

Key words: agricultural machinery, machine vision, navigation, horizontal projection, gradient descent

摘要： 针对小麦收割边缘受到麦茬、土壤、光照等环境因素影响导致的导航线提取精度低、运行速度慢等问题,为实现小麦精准化收获作业而提出一种基于水平投影和梯度下降的小麦收获边缘导航线提取方法。首先通过LAB阈值分割、形态学滤波等进行图像分割,然后进行水平投影以提取出小麦收获边缘伪特征点,将伪特征点进行最小二乘拟合从而获得边缘特征点所在的ROI区域,并对该区域进行Canny边缘检测来提取出边缘特征点,最后利用梯度下降算法拟合出小麦收获边缘导航线,从而解决传统算法中所遇到的导航线拟合精度低、拟合速度慢等问题。试验结果表明:在小麦已收割和未收割区域对比度很低的情况下,处理一张分辨率为640像素×360像素的图像平均耗时163 ms,生成的导航基准线成功率高达95%,为智能农业机械在麦田中的自主行走提供一种可靠的、实时的导航方法。

关键词: 农业机械, 视觉, 导航, 水平投影, 梯度下降