摘要
基于稻穗几何形态特征和在穗籽粒数二者之间的映射关系,提出基于稻穗图像形态学特征机器学习的在穗籽粒测量新方法.首先,利用图像处理方法提取一次枝梗的面积、骨架长度、周长、骨架距离均值等形态特征.其次,针对一次枝梗识别,提出基于局部距离方差的提取方法,获取一次枝梗骨架.最后,使用改进的支持向量机构建稻穗几何形态特征和在穗籽粒数两者之间的映射关系.实验结果表明,用以上特征训练的分类器,预测稻穗籽粒数的相对误差平均值为6.72%,可以有效解决测量在穗籽粒数时遇到的遮挡和粘连问题.研究结果表明,稻穗形态学特征与在穗籽粒数存在确定性内蕴映射关系,该映射能够被多分类集成支持向量机训练策略描述,且识别精度高于现有回归方法.
The seed number of panicle is an important index of crop breeding. Manual measurement is with a low efficiency, and the automatic counting method based on the traditional image segmentation algorithm is difficult to solve the occlusion and adhesion problems of panicle seeds. Based on the mapping relationship between panicle morphological features and panicle seed numbers, this paper puts forward a new method to measure panicle seed numbers using image morphological features and machine learning algorithm. Firstly, the image processing method is used to extract the morphological features of the primary branch, such as area, perimeter, skeleton length, mean distance on skeleton. Secondly, aiming to the primary branch recognition, this paper puts forward a method based on the local distance deviation. Thirdly, a mapping relationship is constructed between the seed numbers and panicle morphological features using an improved support vector machine(SVM) classifier. The experimental results show that the classifier trained by above features gives 6.72% average relative error for panicle seed numbers prediction, hence the method can effectively solve the occlusion and adhesion problems. The results show that the panicle morphology features and the panicle seed numbers exists deterministic intrinsic mapping relationship which can be described by multi-class SVM training strategy, and the recognition accuracy is higher than the existing regression method.
作者
马志宏
贡亮
林可
毛雨晗
吴伟
刘成良
MA Zhihong;GONG Liang;LIN Ke;MAO Yuhan;WU Wei;LIU Chengliang(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
2019年第2期239-246,共8页
Journal of Shanghai Jiaotong University
基金
国家自然科学基金(51775333)
上海市农业委员会(沪农种字(2015)第20号)资助项目
关键词
稻穗
一次枝梗
形态学特征
支持向量机
panicle
primary branch
morphological features
support vector machine(SVM)
