油菜籽粒千粒重图像测定方法

Determination method on thousand-seed weight of rapeseed based on image processing

为快速、自动、精准地测定油菜籽粒千粒重,建立一种基于图像处理技术,对不同品种的甘蓝型油菜籽粒进行分样、称取质量、获取图像并经图像处理得到表征籽粒面积的像素数,建立籽粒面积与质量之间的相关性模型。利用选择性极限腐蚀算法获得每个籽粒的核,并标记在籽粒的梯度图像上,再利用分水岭算法对标记的梯度图像进行分割。提取一次分割后仍然粘连的籽粒,利用距离变换和求极大值的方法获得籽粒内部的核,再利用分水岭算法进行二次分割。在分割后的图像中随机选取1000粒提取其面积,通过籽粒面积与质量之间的相关性模型得到千粒重。测定的3个品种千粒重相对误差均不超过3.16%。测试结果完全满足千粒重测试国家标准的精度要求。

In order to resolve low automation and timeconsuming in determination of thousand-seed weight of rapeseeds, a method based on image processing technology was proposed. Pixels number representing rapeseed(Brassica napus)seed area was obtained by image processing. Correlation models were established between seeds area and seed mass. Kernel of each seed was obtained by selective limit erosion algorithm and then was labeled on gradient image of seed. Then watershed algorithm was used to segment rapeseeds on the labeled gradient image. The rape seeds which were still adhesive after first segmentation were extracted, and their kernel inside the seed was obtained by method of maximum value after distance transformation.Then the watershed algorithm was used to segment seeds again. In the segmented image, 1000 seeds were randomly selected and their area was extracted. 1000-seed weight was calculated by the correlation model between seeds area and mass. Results showed that the relative error of 1000-seed weight of the 3 varieties was less than 3.16%. This accuracy could fully meet the requirements of national standard.