基于机器视觉的叶菜类蔬菜菜垄识别算法

针对目前叶菜类蔬菜田间作业自动化程度低,缺乏适用自主导航技术等问题,提出一种基于机器视觉的叶菜类蔬菜菜垄识别算法。利用改进超绿算法(Gray=2Cg-Cr-Cb)对菜地图像进行灰度化,通过二值形态学变换和连通区域提取获得菜垄区域和边界,基于Huber损失函数进行边界曲线拟合,最终提取导航基准线。图像处理结果表明:1)本研究提出的识别算法在不同光照环境下具有较好的鲁棒性,自然综合光照条件下导航基准线提取成功率为97.5%;2)基于Huber损失函数获取到的导航基准线,平均均方根误差为0.668像素,比最小二乘法高72.5%,平均角度偏差为0.273°,比最小二乘法高72.6%,且处理速度与最小二乘法相似。试验证明本研究算法可实现在自然光照条件下对叶菜类蔬菜图像的菜垄识别和导航基准线提取。

Ridge recognition algorithm for leaf vegetables based on machine vision

To solve the problems of low automation and lack of autonomous navigation technology for leaf vegetables'' field operations, a machine vision based ridge recognition algorithm for leaf vegetables is proposed in this study. The improved super green algorithm(Gray=2Cg-Cr-Cb)is used for gray processing the vegetable map image. The edge of vegetable ridge and boundary are extracted by binary morphological transformation combined with the extraction of connected area. The boundary curve is fitted based on Huber loss function and the navigation datum line is finally obtained. The results of image processing show that: 1)The recognition algorithm proposed in this study has good robustness in different illumination environments, and the success rate of navigation datum extraction under natural integrated illumination is 97. 5%. 2)The navigation datum line obtained based on Huber loss function has an average root mean square error of 0. 668 pixels, and the recognition accuracy is 72. 5% higher than the least square method. The average angle deviation is 0. 273 degree, and the recognition accuracy is 72. 6% higher than the least square method. At the same time, the processing speed of this method is similar to that of the least square method. It is proved that the proposed algorithm can realize ridge recognition and navigation datum extraction of leaf vegetable images under natural illumination conditions.