| 本文已被：浏览 1006次 下载 471次 |
码上扫一扫！ |
| 玉米自动化考种过程的粘连籽粒图像分割 |
|
张新伟1,2, 易克传1,2, 刘向东3, 赵学观4, 程昕昕5,2, 高连兴6
|
|
|
| (1.安徽科技学院 机械工程学院, 安徽 凤阳 233100;2.玉米育种安徽省工程技术研究院, 安徽 凤阳 233100;3.新疆大学 科学技术学院, 新疆 阿克苏 843000;4.北京农业智能装备技术研究中心, 北京 100097;5.安徽科技学院 农学院, 安徽 凤阳 233100;6.沈阳农业大学 工程学院, 沈阳 100866) |
|
| 摘要: |
| 针对玉米自动化考种过程籽粒粘连导致穗粒数统计准确率低的问题，提出一种遗传算法（GA）与改进脉冲耦合神经网络（PCNN）相结合的分割方法（GA+改进PCNN），对粘连玉米籽粒图像的分割问题进行研究。采用数学形态学和wiener滤波对待分割图像去噪，基于小波变换进行多图像融合得到新图像；利用遗传算法寻找改进PCNN模型中参数β、αE和VE的最优解并进行图像分割。结果表明：1）本研究方法对粘连玉米籽粒的分割准确率为98%；2）本研究方法的交叉熵、区域内部均匀性、形状测度维和区域对比度指标依次为0.079 4、0.975 4、0.878 5和0.869 2，总体优于OTSU、改进分水岭、迭代法全局阈值和未改进PCNN分割算法；3）本研究方法的单幅图像处理时间为22.07 s，用时长于各比较算法，但分割效果最理想。 |
| 关键词: 玉米籽粒 自动化考种 籽粒粘连 图像分割 |
| DOI：10.11841/j.issn.1007-4333.2018.10.18 |
| 投稿时间：2018-02-28 |
| 基金项目:安徽省科技攻关项目（1501031095）；安徽科技学院自然科学研究项目（ZRC2016481）；安徽省高校自然科学研究重点项目（KJ2018A0542、KJ2018A0543） |
|
| Image segmentation of adhesive corn seeds during automatic seed test |
|
ZHANG Xinwei1,2, YI Kechuan1,2, LIU Xiangdong3, ZHAO Xueguan4, CHENG Xinxin5,2, GAO Lianxing6
|
| (1.School of Mechanical Engineering, Anhui Science and Technology University, Fengyang 233100, China;2.Institute of Corn Breeding Engineering and Technology of Anhui Province, Fengyang 233100, China;3.School of Science and Technology, Xinjiang University, Akesu 843000, China;4.Beijing Research Center of Intelligent Equipment for Agriculture, Beijing 100097, China;5.School of Agriculture, Anhui Science and Technology University, Fengyang 233100, China;6.College of Engineering, Shenyang Agricultural University, Shenyang 110866, China) |
| Abstract: |
| In order to solve the problem of segment difficulty caused by seeds adhesion during automatic seeds test, a new segmentation method was proposed in this study. Image segmentation algorithms combing genetic algorithms (GA) and improved pulse coupled neural net (PCNN) model were adopted. To improve contrast between the adhesive corn seeds and image background,adhesive corn seeds were pretreated to remove noise by using mathematical morphology and wiener method, multi-scale decomposed based on wavelet transform, and reconstruct images by using data fusion. The optimal parameters of β、αE and VE in the PCNN model were obtained by genetic algorithm. Small distracter of the segmented image were removed with the PCNN's automatic porter sign. The adhesion grain segmentation image was finally obtained. The results showed that:1) The accuracy of image segmentation algorithm by combing GA and adaptive pulse coupled neural net (PCNN) was 98% (the highest); 2) The cross entropy, regional uniformity, shape measure dimension and region contract indexes proposed in this study were respectively 0.079 4, 0.975 4, 0.878 5 and 0.869 2, which were superior to OTSU, improved watershed algorithm, iteration method global threshold algorithm and not improved PCNN algorithm; 3) The running time of GA+improved PCNN algorithm was 22.07 s, which was longer than the rest, but the segmentation effect of GA+improved PCNN algorithm program was the best. |
| Key words: corn seeds automatic seeds test grain adhesion image segmentation |
