基于视觉的玉米苗期作物识别与定位方法研究

Research on the method of crop recognition and location in maize seedling stage based on vision

基于视觉的苗期作物目标识别技术逐渐成熟,对作物进行准确识别和精准定位,是实现株间除草的技术关键和难点。本文以苗期玉米为研究对象,提出了一种苗期作物的识别与定位方法,通过作业车辆的图像采集装置来实时获取田间作物的苗期图像,基于HSV色彩空间对输入图像进行预处理,根据作物与土壤背景颜色差异,选取固定取值范围的三通道阈值,通过二值化处理去除土壤背景,再通过深度开运算来去除杂草噪声,对得到的苗期作物提取轮廓信息,经过骨架提取算法后得到作物骨架,并以此确定作物茎秆位置坐标,从而实现对作物的精准定位。作物图像中幼苗的识别率为98.3%,定位误差距离在10 mm以内的定位精准度为85.9%,基本可以满足智能除草机器人实时除草作业要求。

The technology of crop target recognition based on vision is becoming more and more mature.It is the key and difficulty to identify and locate crops accurately.In this paper,a method of identifying and locating seedling maize is proposed.Through the image acquisition device of the working vehicle,the seedling images of the field crops can be acquired in real time.The input image is preprocessed based on HSV color space.According to the color difference between crop and soil background,the three channel threshold value with fixed value range is selected,and the soil background is removed by binary processing.Then through the depth operation to remove the weed noise,the contour information of the seedling crops is extracted,the crop skeleton is obtained after the skeleton extraction algorithm,and then the position coordinates of the crop stem are determined,so as to achieve the accurate positioning of the crops.In the crop image,the recognition rate of seedlings is 98.3%,and the positioning accuracy is 85.9%when the positioning error distance is within 10 mm,which can basically meet the requirements of real-time weeding operation of intelligent weeding robot.