Plant factories have a great potential for mitigating the contradiction between the world’sgrowing population and food scarcity. During the process of its automatic production,machine vision plays a significant role....Plant factories have a great potential for mitigating the contradiction between the world’sgrowing population and food scarcity. During the process of its automatic production,machine vision plays a significant role. This technique almost covers every production linkfrom raising seedlings, transplanting, management, and harvesting to fruit grading. To provide references and a starting point for those who are committed to studying this issue. Inthis paper, the application prospects of machine vision in plant factories were analyzed,and the present researches were summarized from the fields of plant growth monitoring,robot operation assistance, and fruit grading. The results found that although the existingmethods have solved some practical problems at low cost, high efficiency and precision,some challenges still are faced by machine vision. Firstly, the changing lighting, complexbackgrounds, and color similarity within plant different parts cause the commonly usedimage segmentation algorithms to fail. The shortage of standard agricultural datasets alsokeeps deep learning and unsupervised classification algorithms from making progress.Secondly, there are some theoretical knowledge gaps for machine vision application in aparticular environment of plant factories, which seriously contains its application effect.Thirdly, the lack of special image acquisition devices and supporting facilities resulted inpoor image quality. All these factors hinder machine vision application in plant factories.Nevertheless, it is still a powerful tool and irreplaceable at present. We believed that thistechnique would promote plant factory development greatly with more robust, efficient,and reliable algorithms are developed in the future.展开更多
针对高地隙植保机底盘玉米田间植保作业压苗严重的现象,该研究提出了基于车轮正前方可行走动态感兴趣区域(Region of Interest,ROI)的玉米行导航线实时提取算法。首先将获取的玉米苗带图像进行像素归一化,采用过绿算法和最大类间方差法...针对高地隙植保机底盘玉米田间植保作业压苗严重的现象,该研究提出了基于车轮正前方可行走动态感兴趣区域(Region of Interest,ROI)的玉米行导航线实时提取算法。首先将获取的玉米苗带图像进行像素归一化,采用过绿算法和最大类间方差法分割玉米与背景,并通过形态学处理对图像进行增强和去噪;然后对视频第1帧图像应用垂直投影法确定静态ROI区域,并在静态ROI区域内利用特征点聚类算法拟合作物行识别线,基于已识别的玉米行识别线更新和优化动态ROI区域,实现动态ROI区域的动态迁移;最后在动态ROI区域内采用最小二乘法获取高地隙植保机底盘玉米行间导航线。试验表明,该算法具有较好的抗干扰性能,能够很好地适应较为复杂的田间环境,导航线提取准确率为96%,处理一帧分辨率为1920像素×1080像素图像平均耗时97.56 ms,该研究提出的算法能够为高地隙植保机车轮沿玉米垄间行走提供可靠、实时的导航路径。展开更多
文摘Plant factories have a great potential for mitigating the contradiction between the world’sgrowing population and food scarcity. During the process of its automatic production,machine vision plays a significant role. This technique almost covers every production linkfrom raising seedlings, transplanting, management, and harvesting to fruit grading. To provide references and a starting point for those who are committed to studying this issue. Inthis paper, the application prospects of machine vision in plant factories were analyzed,and the present researches were summarized from the fields of plant growth monitoring,robot operation assistance, and fruit grading. The results found that although the existingmethods have solved some practical problems at low cost, high efficiency and precision,some challenges still are faced by machine vision. Firstly, the changing lighting, complexbackgrounds, and color similarity within plant different parts cause the commonly usedimage segmentation algorithms to fail. The shortage of standard agricultural datasets alsokeeps deep learning and unsupervised classification algorithms from making progress.Secondly, there are some theoretical knowledge gaps for machine vision application in aparticular environment of plant factories, which seriously contains its application effect.Thirdly, the lack of special image acquisition devices and supporting facilities resulted inpoor image quality. All these factors hinder machine vision application in plant factories.Nevertheless, it is still a powerful tool and irreplaceable at present. We believed that thistechnique would promote plant factory development greatly with more robust, efficient,and reliable algorithms are developed in the future.
文摘针对高地隙植保机底盘玉米田间植保作业压苗严重的现象,该研究提出了基于车轮正前方可行走动态感兴趣区域(Region of Interest,ROI)的玉米行导航线实时提取算法。首先将获取的玉米苗带图像进行像素归一化,采用过绿算法和最大类间方差法分割玉米与背景,并通过形态学处理对图像进行增强和去噪;然后对视频第1帧图像应用垂直投影法确定静态ROI区域,并在静态ROI区域内利用特征点聚类算法拟合作物行识别线,基于已识别的玉米行识别线更新和优化动态ROI区域,实现动态ROI区域的动态迁移;最后在动态ROI区域内采用最小二乘法获取高地隙植保机底盘玉米行间导航线。试验表明,该算法具有较好的抗干扰性能,能够很好地适应较为复杂的田间环境,导航线提取准确率为96%,处理一帧分辨率为1920像素×1080像素图像平均耗时97.56 ms,该研究提出的算法能够为高地隙植保机车轮沿玉米垄间行走提供可靠、实时的导航路径。