Crops ›› 2024, Vol. 40 ›› Issue (6): 242-248.doi: 10.16035/j.issn.1001-7283.2024.06.033

Previous Articles     Next Articles

Research on Refined Selection Method for Maize Seeds Based on Machine Vision Technology

Han Xiaowei1(), Zhou Jiangming1, Gao Yingbo2, Tian Xuehui1, Li Mingjun1, Hao Yanjie1, Li Wei1, Li Shubing1, Liu Shuze1()   

  1. 1Binzhou Academy of Agricultural Sciences, Binzhou 256600, Shandong, China
    2Maize Research Institute, Shandong Academy of Agricultural Sciences / National Engineering Laboratory of Wheat and Maize /Key Laboratory of Biology and Genetic Improvement of Maize in Northern Huang-Huai-Hai Plain, Ministry of Agriculture and Rural Affairs, Jinan 250100, Shandong, China
  • Received:2023-07-11 Revised:2023-10-17 Online:2024-12-15 Published:2024-12-05

RichHTML

3

PDF (PC)

180

Abstract:

In order to further improve the germination rate of maize seeds, the more suitable selection methods and parameters were discussed. Based on this, Zhengdan 958 was used as the test material in this experiment. The physical parameters of single maize seed were obtained by Seed Identification software, and the single seed germination test was carried out to study the correlation between maize seed vigor index and its morphological and physical parameters, so as to screen the optimal selection index; At the same time, the single index classification method, binary logistic regression model and multi-layer perceptron neural network model were used to predict the seed germination rate to determine the best selection method. The bud length, root length and fresh weight of seedlings were significantly correlated with the physical parameters of R, A, S and B3. According to the single index of 170≤R≤190, 10≤A≤20, 16≤S≤24, 71≤B3≤79, the germination rate increased from 66.0% to 72.1%, 73.7%, 75.0% and 73.6% respectively, and the selection rate was 56.8%, 63.6%, 52.3% and 50.8%, respectively; The seed germination rate of the binary logistic regression model method was increased to 80.9%, the seed germination selection rate was 88.4%, and the model stability rate was 97.3%; The seed germination rate of the multi-layer perceptron neural network model method was increased to 82.9%, the seed germination selection rate was 89.5%, and the model stability rate was 97.7%. In conclusion, the physical indexes R, A, S and B3 values can be used as the selection parameters of maize seeds; Compared with single index and binary logistic regression model, the multi-layer perceptron neural network model has strong advantages in predicting seed germination rate, selection rate and stability, and can be determined as the best selection method.

Key words: Maize seed, Refined selection method, Physical parameters, Machine vision technology, Multi-layer perceptron neural network model

Fig.1

Scan of maize seed"

Fig.2

Scan of pre-treated maize seed"

Fig.3

Automated seed identification software collects physical parameters related to maize seeds"

Fig.4

Multi-layer perceptron network topology X represents the original physical index parameter, and Y represents the new parameter obtained by multi-layer linear transformation and nonlinear activation functions."

Table 1

Correlation analysis between bud length, root length and fresh weight of maize seeds and their physical indicators"

物理指标
Physical indicator		 极小值
Minimal value		 极大值
Maximum value		 芽长
Bud length		 根长
Root length		 鲜重
Fresh weight		 变异系数
Coefficient of variation
R 148.8 189.6 0.624** 0.599** 0.624** 0.06
G 132.7 199.1 -0.037 -0.085 -0.055 0.11
B 138.5 179.0 -0.140 -0.171 -0.134 0.06
L 57.2 76.7 0.081 0.030 0.064 0.08
A -18.1 16.9 0.312* 0.347* 0.331* 1.08
B2 -8.6 20.6 0.233 0.203 0.209 -29.42
H 63.6 351.5 0.147 0.135 0.142 0.35
S 6.3 24.0 0.632** 0.607** 0.608** 0.29
B3 59.2 78.1 0.418** 0.369** 0.401** 0.07
灰度Gray scale 138.8 183.4 0.105 0.055 0.090 0.07
宽度Width (mm) 7.2 9.6 0.118 0.058 0.104 0.08
长度Length (mm) 9.0 11.7 0.011 -0.052 -0.008 0.07
投影面积Projection area (px) 4892.6 7504.0 0.095 0.028 0.077 0.13

Table 2

Maize seed screening by R, A, S and B3 indicators"

精选指标
Selection indicator		 区间
Interval		 总数
Total		 发芽数
Number of
germination		 未发芽数
Number of
ungerminated		 发芽率
Germination
rate (%)		 获选率
Selection
rate (%)		 平均鲜重
Average fresh
weight (g)		 简易活力指数
Simple vitality
index
原始对照Original control － 200 132 68 66.0 1.0 0.269 0.178
R 140≤R<150 9 5 4 55.6 3.8 0.025 0.014
150≤R<160 27 16 11 59.3 12.1 0.058 0.034
160≤R<170 60 36 24 60.0 27.3 0.132 0.079
170≤R<180 56 38 18 67.9 28.8 0.342 0.232
180≤R≤190 48 37 11 77.1 28.0 0.520 0.400
170≤R≤190 (s) 104 75 29 72.1 56.8 0.424 0.306
A A<0 13 8 5 61.5 6.1 0.143 0.088
0≤A<5 22 8 14 36.4 6.1 0.016 0.006
5≤A<10 51 32 19 62.7 24.2 0.032 0.020
10≤A<15 65 43 22 66.2 32.6 0.357 0.236
15≤A≤20 49 41 8 83.7 31.1 0.546 0.457
10≤A≤20 (s) 114 84 30 73.7 63.6 0.438 0.323
S 4≤S<8 13 7 6 53.8 5.3 0.024 0.013
8≤S<12 44 22 22 50.0 16.7 0.036 0.018
12≤S<16 51 34 17 66.7 25.8 0.176 0.117
16≤S<20 65 46 19 70.8 34.8 0.436 0.309
20≤S≤24 27 23 4 85.2 17.4 0.540 0.460
16≤S≤24 (s) 92 69 23 75.0 52.3 0.467 0.350
B3 59≤B3<63 24 15 9 62.5 11.4 0.031 0.019
63≤B3<67 36 21 15 58.3 15.9 0.032 0.019
67≤B3<71 49 29 20 59.2 22.0 0.306 0.181
71≤B3<75 42 30 12 71.4 22.7 0.364 0.260
75≤B3≤79 49 37 12 75.5 28.0 0.441 0.333
71≤B3≤79 (s) 91 67 24 73.6 50.8 0.405 0.299

Table 3

Maize seed selection index validation results"

精选指标
Selection indicator		 参数
Parameter		 发芽数
Number of germination		 未发芽数
Number of ungerminated		 发芽率
Germination rate
(%)		 平均鲜重
Average fresh
weight (g)		 简易活力指数
Simple vitality
index
原始对照Original control － 132 68 66.0 0.269 0.178
R 170≤R≤190 109 41 72.7 0.432 0.314
A 10≤A≤20 112 38 74.6 0.433 0.323
S 16≤S≤24 113 37 75.3 0.461 0.347
B3 71≤B3≤79 105 45 70.0 0.401 0.281

Table 4

Binary logistic regression model classification"

实测
Actual test		 选定案例预测Selected case projection 未选定案例预测No selected case projection
0 1 正确百分比Correct percentage (%) 0 1 正确百分比Correct percentage (%)
0 29 17 63.0 13 9 59.1
1 9 80 89.9 5 38 88.4
总体百分比Overall percentage (%) 80.7 78.5

Table 5

Classification for multi-layer perceptron neural network model"

样本
Sample		 实测
Actual test		 预测Projection
0 1 正确百分比
Correct percentage (%)
训练
Train		 0 29 17 63.0
1 6 88 93.6
总体百分比 (%) 25.0 75.0 83.6
测试
Test		 0 15 7 68.2
1 4 34 89.5
总体百分比 (%) 31.7 68.3 81.7

Table 6

Comparison of three maize seed selection methods %"

方法
Method		 物理指标
Physical indicator		 发芽率
Germination rate		 获选率
Selection rate		 稳定性
Stability
单一指标精选Selection single indicator 170≤R≤190 72.1 56.8 －
10≤A≤20 73.7 63.6 －
16≤S≤24 75.0 52.3 －
71≤B3≤79 73.6 50.8 －
二元逻辑回归模型Binary logistic regression model － 80.9 88.4 97.3
多层感知器神经网络模型Multi-layer perceptron neural network model － 82.9 89.5 97.7
[1] 赵久然, 王帅, 李明, 等. 玉米育种行业创新现状与发展趋势. 植物遗传资源学报, 2018, 19(3):435-446.
[2] 赵晴月, 许世杰, 张务帅, 等. 中国玉米主产区土壤养分的空间变异及影响因素分析. 中国农业科学, 2020, 53(15):3120- 3133.
doi: 10.3864/j.issn.0578-1752.2020.15.012
[3] 徐田军, 吕天放, 陈传永, 等. 种植密度和植物生长调节剂对玉米茎秆性状的影响及调控. 中国农业科学, 2019, 52(4):629- 638.
doi: 10.3864/j.issn.0578-1752.2019.04.005
[4] 高婷婷, 李洋, 王秀芬, 等. 基于冷浸法测定玉米种子活力的研究. 分子植物育种, 2020, 18(23):7879-7884.
[5] 崔敏嘉. 玉米种子理化性状与种子活力关系的研究. 沈阳:沈阳农业大学, 2016.
[6] 徐江, 谭敏, 张春庆, 等. 电晕场与介电分选提高水稻种子活力. 农业工程学报, 2013, 29(23):233-240.
[7] 杨冬风. 基于软X-射线造影和机器智能的玉米种子活力检测方法研究. 作物杂志, 2013(3):136-140.
[8] 展慧, 李小昱, 周竹, 等. 基于近红外光谱和机器视觉融合技术的板栗缺陷检测. 农业工程学报, 2011, 27(2):345-349.
[9] 柴玉华, 毕文佳, 谭克竹, 等. 基于高光谱图像技术的大豆品种无损鉴别. 东北农业大学学报, 2016, 47(3):86-93.
[10] 郝建平, 杨锦忠, 杜天庆, 等. 基于图像处理的玉米品种的种子形态分析及其分类研究. 中国农业科学, 2008, 41(4):994- 1002.
[11] 朱荣胜, 闫学慧, 陈庆山. 基于图像识别和卷积神经网络的大豆优良籽粒筛选研究. 大豆科学, 2020, 39(2):189-197.
[12] Kara M, Sayinci B, Elkoca E, et al. Seed size and shape analysis of registered common bean (Phaseolus vulgaris L.) cultivars in Turkey using digital photography. Tarim Bilimleri Dergisi, 2013, 19(3):219-234.
[13] Chen X, Xun Y, Wei L, et al. Combining discriminant analysis and neural networks for corn variety identification. Computers & Electronics in Agriculture, 2010, 71(S1):48-53.
[14] Torkashvand A M, Ahmadipour A, Khaneghah A M. Estimation of kiwifruit yield by leaf nutrients concentration and artificial neural network. The Journal of Agricultural Science, 2020, 158(3):185-193.
[15] 龙满生, 欧阳春娟, 刘欢, 等. 基于卷积神经网络与迁移学习的油茶病害图像识别. 农业工程学报, 2018, 34(18):194-201.
[16] 周亮, 慕号伟, 马海姣, 等. 基于卷积神经网络的中国北方冬小麦遥感估产. 农业工程学报, 2019, 35(15):119-128.
[17] Kurtulmuş F, Aliba L, Kavdir I. Classification of pepper seeds using machine vision based on neural network. International Journal of Agricultural and Biological Engineering, 2016(1):9.
[18] 叶凤林, 李琳, 杨丽明, 等. 基于机器视觉的黄芩种子精选技术研究. 种子, 2016, 35(11):100-104.
[19] 王润涛, 张长利, 房俊龙, 等. 基于机器视觉的大豆籽粒精选技术. 农业工程学报, 2011, 27(8):355-359.
[20] 吴尚智, 周运, 王欢欢, 等. 利用粗糙集和双隐层BP神经网络的小麦籽粒品种分类. 沈阳农业大学学报, 2020, 51(5):576- 585.
[21] ElMasry G, Wang N, Vigneault C. Detecting chilling injury in red delicious apple using hyperspectral imaging and neural networks. Postharvest Biology and Technology, 2009, 52(1):1-8.
[22] Tu K L, Li L J, Yang L M, et al. Selection for high quality pepper seeds by machine vision and classifiers. Journal of Integrative Agriculture, 2018, 17(9):1999-2006.
[23] 赵晓东, 吴依榕, 毛瑞琳, 等. 低温干燥贮藏包衣和非包衣玉米种子活力及生理差异性研究. 玉米科学, 2020, 28(3):105-110.
[24] Vilar W T S, Aranha R M, Medeiros E P, et al. Classification of individual castor seeds using digital imaging and multivariate analysis. Journal of the Brazilian Chemical Society, 2014, 26(1):102-109.
[25] 杨红云, 黄琼, 孙爱珍, 等. 基于卷积神经网络和支持向量机的水稻种子图像分类识别. 中国粮油学报, 2021, 36(12):144- 150.
[26] Bishaw Z, Struik P C, Van Gastel A J G. Farmers' seed sources and seed quality: 1. Physical and physiological quality. Journal of Crop Improvement, 2012, 26(5):655-692.
[27] 贾佳, 王建华, 谢宗铭, 等. 计算机图像识别技术在小麦种子精选中的应用. 中国农业大学学报, 2014, 19(5):180-186.
[28] 郝奇慧. 玉米种子理化性状与种子活力的关系研究. 沈阳:沈阳农业大学, 2018.
[29] Gupta P K, Rustgi S, Kumar N. Genetic and molecular basis of grain size and grain number and its relevance to grain productivity in higher plants. Genome, 2006, 49(6):565-571.
pmid: 16936836
[30] 刘旭欢, 得拉·努尔兰, 贾永红, 等. 籽粒成熟度与穗部位置对春小麦种子活力的影响. 西北农业学报, 2014, 23(10):71-75.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!