| 摘要: |
| 为探索快速、高效检测甜玉米种子生活力的方法，利用机器视觉技术（Seed Identification）批量快速提取金菲甜玉米种子的Red（红基色）、Green（绿基色）、Blue（蓝基色）、Hues（色相）、Saturation（饱和度）、Brightness（亮度）、Light（明度）、a（红色至绿色的范围）、b（蓝色至黄色的范围）、灰度、宽度、长度和投影面积等物理特征参数，通过单粒发芽试验确定每粒种子的生活力，然后采用人工神经网络和二元逻辑回归结合主成分分析进行建模。结果表明：1）a值、b值、Saturation和投影面积与种子的活力均存在极显著或显著相关，且变异系数相对较大，其中当a ≤ 3时，发芽率可从72.7%提升至77.6%，获选率达到79.4%；投影面积≤ 77.31 mm2时，发芽率可提升至73.7%，获选率87.6%；2）用13个物理指标标准化后直接进行人工神经网络建模，双隐藏层（训练集：测试集=6：4）建模，模型整体预测正确率为74.2%，优质种子获选率达到93.8%，发芽率可提升至76.9%；3）经二元逻辑回归模型预测发芽率为74.5%，但神经网络模型稳定性优于二元逻辑回归建模。 |
| 关键词: 甜玉米 机器视觉技术 主成分分析 人工神经网络 二元逻辑回归 |
| DOI：10.11841/j.issn.1007-4333.2018.07.01 |
| 投稿时间：2017-07-22 |
| 基金项目:北京科委项目（Z151100001015004） |
|
| Seed viability testing model of sweet corn based on artificial neural network and binary logistic regression |
|
LIU Minjie1,2, XU Xuan1,2, WANG Jianhua2, SUN Qun2, XIANG Chunyang1
|
| (1.Tianjin Agricultural University, Tianjin 300384, China;2.College of Agronomy and Biotechnology/Beijing Innovation Center of Crop Seeds Full Technologies Research of Ministry/Beijing Key Laboratory of Crop Genetic Improvement, China Agricultural University, Beijing 100193, China) |
| Abstract: |
| In order to test the vitality of sweet corn seed quickly and efficiently,several physical characteristics (red,green,blue,hues,saturation,brightness,light,a, b,width,length and projected area) of a certain cultivar of sweet corn,Jinfei,were obtained in batch and quickly went through image recognition technology (Seed Identification software).Each seed viability was confirmed through single seed germination test,and a model for seed viability discrimination of Jinfei was established by applying principle component analysis,artificial neural network and binary logistic regression.The results showed that:1) a,b,saturation,width and projection area all had significant correlation with the seed vigor and relatively high variation coefficient.For the seeds with a ≤ 3,the germination percentage was increased from 72.7% to 77.6%,and the selected rate reached to 79.4%.For the seeds with projection area ≤ 77.31 mm2,the germination percentage can be increased to 73.7% and the selected rate reached 87.6%.2) Based on 13 standardized physical indicators,artificial neural network model with double hidden layers (training set:testing set=6:4) possessed an overall predicting accuracy at 74.2%,germination percentage also reached to 76.9%,with 93.8% quality seed selection.3) The germination percentage of binary logistic regression was 74.5%,but the model stability of Artificial Neural Network model was better than binary logistic regression.These results would provide references for further research and implementation of quality seed sorting technology. |
| Key words: sweet corn machine vision technology Principal Component Analysis Artificial Neural Network Binary Logic Regression |
