基于分水岭和统计矩的大豆籽粒形态参数测量方法  

Measurement Method of Soybean Seed Morphological Parameters Based on Watershed and Statistical Moment

作　　者：丁琦 徐伟 李蒙[2] 王秀成 卢伟[1] 盖钧镒[2] 王玲[1] 邢光南[2] DING Qi;XU Wei;LI Meng;WANG Xiu-cheng;LU Wei;GAI Jun-yi;WANG Ling;XING Guang-nan(College of Engineering/Engineering Laboratory of Modem Facility Agriculture Technology and Equipment,Nanjing Agricultural University,Nanjing 210031,China;Soybean Research Institute/National Center for Soybean Improvement/MOA Key Laboratory for Biology and Genetic Improvement of Soybean(General)/State Key Laboratory for Crop Genetics and Germplasm Enhancement/Jiangsu Collaborative Innovation Center for Modem Crop Production,Nanjing Agricultural University,Nanjing 210095,China)

机构地区：[1]南京农业大学工学院/现代设施农业技术装备工程实验室,江苏南京210031 [2]南京农业大学大豆研究所/国家大豆改良中心/农业部大豆生物学与遗传改良重点实验室(综合)/作物遗传与种质创新国家重点实验室/江苏现代作物生产协同创新中心,江苏南京210095

出　　处：《大豆科学》2019年第6期960-967,共8页Soybean Science

基　　金：国家重点研发计划项目子课题(2016YFD0100201-22);国家自然科学基金(31571694);中央高校基本科研业务费专项资金(KYT201801);长江学者和创新团队发展计划(PCSIRT_17R55);教育部111项目(B08025);农业部国家大豆产业技术体系(CARS-04);江苏省优势学科建设工程专项;江苏省JCIC-MCP项目;扬州市科技计划(YZ2018038);江苏省农机三新工程(SZ120170036)

摘　　要：为促进大豆考种及育种工作的高效进行,本研究设计一种大豆籽粒计数和面积、周长、粒长、粒宽等形态参数的测量软件。使用高拍仪采集大豆籽粒和标定板彩色图像,基于"Otsu"阈值法去除图像背景,获得大豆籽粒二值图像,基于分水岭变换法分割二值图中的粘连大豆籽粒,获取了一系列单粒、多粒大豆连通域,对极少数多粒连通域进行籽粒计数校正,实现了大豆籽粒的计数;使用单粒连通域的白色像素点总和来计算大豆籽粒面积,基于freeman链码算法的校正公式计算大豆籽粒周长,使用二阶统计矩求取大豆籽粒的主轴方向并将大豆扭到水平方向,继而计算大豆边界点横、纵坐标的极差来获取粒长与粒宽。用3份不同尺寸的大豆材料来验证本软件的精确性,试验结果表明,大豆籽粒计数准确率可达100%,软件测量与人工测量大豆籽粒的平均粒长和平均粒宽的误差普遍为0.01~0.04 cm,相对误差为3.8%~9.7%,平均相对误差为5.6%。该软件及相应的图像采集装置可以有效地满足农业科学研究工作的准确性要求,同时具有成本低廉,工作效率高的特点。In order to promote soybean cultivation and breeding, this study designed a software that can count accurately and measure the morphological parameters of soybean seed, such as area, perimeter, length and width. The color image of soybean seed and calibration plate were collected by high-speed photographic apparatus. The image background was removed based on the ’Otsu’ threshold method to obtain the binary image of soybean seed. Based on the watershed transformation method, the adhesive soybean seeds in the binary map were segmented, and a series of single-seed and multi-seeds soybean connected domains were obtained. The seed count was corrected for a few multi-seeds connected domains, and then soybean seed count was achieved. The sum of white pixels in each connected domain is used to calculate the soybean seed area. The seed perimeter was calculated by using the correction formula based on the freeman chain code algorithm. The second-order statistical moment is used to obtain the main axis direction of the soybean seed and the soybean seed was twisted to the horizontal direction, and then we calculated the seed length and width by the extreme difference between the horizontal and vertical coordinates of the boundary point. The accuracy of the software was verified by three different sizes of soybean seed materials. The results showed that the accuracy of soybean seed count can reach 100%. Compared with manual measurement, the error of average seed length and average seed width of soybean seeds is generally 0.01-0.04 cm, the relative error was 3.8%-9.7%, and the average relative error is 5.6%. The software and the corresponding image acquisition device can effectively meet the accuracy of agricultural scientific research work with low cost and high work efficiency.

关 键 词：大豆籽粒图像 分水岭 计数 形态测量 软件开发 

分 类 号：S51[农业科学—作物学]