杨振宇, 张文强, 李 伟, 陈 英, 宋 鹏. 利用单目视觉获取钵苗移栽适合度信息的方法[J]. 农业工程学报, 2014, 30(3): 112-119. DOI: 10.3969/j.issn.1002-6819.2014.03.015
    引用本文: 杨振宇, 张文强, 李 伟, 陈 英, 宋 鹏. 利用单目视觉获取钵苗移栽适合度信息的方法[J]. 农业工程学报, 2014, 30(3): 112-119. DOI: 10.3969/j.issn.1002-6819.2014.03.015
    Yang Zhenyu, Zhang Wenqiang, Li Wei, Chen Ying, Song Peng. Information acquisition method of potted-seedling transplanting fitness using monocular vision[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2014, 30(3): 112-119. DOI: 10.3969/j.issn.1002-6819.2014.03.015
    Citation: Yang Zhenyu, Zhang Wenqiang, Li Wei, Chen Ying, Song Peng. Information acquisition method of potted-seedling transplanting fitness using monocular vision[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2014, 30(3): 112-119. DOI: 10.3969/j.issn.1002-6819.2014.03.015

    利用单目视觉获取钵苗移栽适合度信息的方法

    Information acquisition method of potted-seedling transplanting fitness using monocular vision

    • 摘要: 钵苗在穴盘中的生长状态各异,钵苗移栽是育苗过程的关键环节。为提高移栽后钵苗成活率,解决出苗不齐和断苗等问题,需对被移栽钵苗的直立度和高度等适合度信息进行综合评价,筛选出符合移栽要求的钵苗。图像采集系统中的顶杆能够顶起穴盘中的钵苗旋转90°,相机采集钵苗转动90°前后2幅图像,图像RGB各通道灰度值概率直方图存在灰度集中区域。首先,采用24位RGB源图像转8位灰度图、中值滤波和灰度拉伸算法对图像进行预处理;然后,使用细化、水平膨胀和垂直腐蚀等图像处理算法,获取钵苗主茎秆的特征;最后,采用标准差为0.65的3×3高斯模板Harris角点检测算法提取每株钵苗主茎秆上的关键点信息,对其加权最小方差直线拟合获取拟合直线,直线的斜率换算后作为钵苗直立度的判定值,以每一株钵苗的全部角点y坐标最大差值的110%作为钵苗高度的判定值。只有钵苗的直立度和高度都满足各自评价指标要求才被认为适合移栽。试验获取了12幅图像共计30株辣椒钵苗在0和90°位置的直立度和高度,每幅图像处理算法平均耗时0.35 s。按直立度(45°<α<135°)和高度(H>105 mm)评价指标判定,5株钵苗不适合移栽;2株钵苗的视觉检测结果与人工测量结果相反,视觉检测结果与人工测量结果之间的偏差率为6.67%。出现检测偏差的原因主要是钵苗叶对茎秆的遮挡和移栽机的振动使得钵苗在转动90°的前后与顶杆的相对位置发生了变化影响了人工和视觉测量的精度。该方法能够满足钵苗移栽机实时筛选工作的需求。

       

      Abstract: Abstract: The potted-seedling growth state varies in the tray. Transplanting is the key in the nursery seedling process. In order to improve the survival rate of a transplanted potted-seedling, and solve the emergence problems such as seedlings growth, or irregular or broken seedlings, the fitness information of the perpendicularity and height of transplanted potted seedlings was comprehensively evaluated, and filtered out potted seedlings meeting the transplanting requirements. The potted-seedling jacked up by the push rod of the image acquisition system can be rotated 90°. A camera captured two images of the potted-seedling before and after it was rotated 90°. The gray-value probabilities histogram of the grayscale images RGB channels existing concentrated areas. At first,the image was pre-processed by 24-bit RGB source image converting to 8 bit grayscale algorithm, median filtering algorithm and gray stretching algorithm. Then, the trunk features of the potted-seedling were extracted by the image processing algorithms of thinning, horizontal expansion, and vertical erosion. Finally, the key points of every potted-seedling trunk were extracted by a Harris corner detection algorithm of 3×3 Gauss templates of the standard deviation was 0.65, and the fitting line was obtained by the weighted least-squares linear fitting with the key points, and the converted line slope was used as the identification parameter of the potted-seedling upright, and found out the maximum y-coordinate difference of all corners coordinates in each strain of potted-seedling, 110% of the difference was used as the identification parameters of the potted-seedling height. The perpendicularity and height of potted-seedlings had to meet evaluation requirements before being considered suitable for transplant. The perpendicularity and height of 30 strains of a pepper potted-seedling in twelve images were distinguished by machine vision technology at the 0 and 90°position, and the average time consuming of each image being processed was 0.35 s. According to evaluating indicators of the perpendicularity (45°<α<135°) and height (H>105 mm), it was determined that five potted-seedlings were suitable for transplanting, and wherein the measurement results of two potted-seedlings were converse using the vision measurement method or the manual measuring method. The error rate was 6.67% between the machine vision recognition results and the manual measurement results. The main reason for the deviation was that the manual measuring accuracy was affected by the leaves of the potted-seedlings obscuring the stems and the vibrations of the transplanting machine causing location changes during the process of rotating 90°.

       

    /

    返回文章
    返回