红提糖度和硬度的高光谱成像无损检测

高升; 王巧华; 付丹丹; 李庆旭

doi:doi:10.3788/AOS201939.1030004

光学学报, 2019, 39 (10): 1030004, 网络出版: 2019-10-09

红提糖度和硬度的高光谱成像无损检测下载： 1012次

Nondestructive Detection of Sugar Content and Firmness of Red Globe Grape by Hyperspectral Imaging

高升 ¹王巧华 ^1,2,*付丹丹 ¹李庆旭 ¹

作者单位

¹ 华中农业大学工学院, 湖北武汉 430070

² 农业部长江中下游农业装备重点实验室, 湖北武汉 430070

图 & 表

图 1. 三种放置模式下红提果粒的高光谱图像。(a)横放;(b)果柄侧朝下;(c)果柄侧朝上

Fig. 1. Hyperspectral images in three placement orientations. (a) Horizontal; (b) fruit stalk-side down; (c) fruit stalk-side up

下载图片查看原文

图 2. 高光谱图像中背景与红提区域的反射率

Fig. 2. Reflectivity of background and red globe grape area in hyperspectral images

下载图片查看原文

图 3. 红提果粒的高光谱图像处理。(a) 726.6 nm处高光谱的图像;(b)掩模板图像;(c)掩模后红提区域的图像

Fig. 3. Hyperspectral image processing of red globe grapes. (a) Hyperspectral image at 726.6 nm; (b) mask template image; (c) masked image of red globe grape area

下载图片查看原文

图 4. 红提样本的原始光谱

Fig. 4. Originalspectra of red globe grape samples

下载图片查看原文

图 5. 红提糖度的GA特征波长选取图。(a)GA筛选图;(b)RMSECV变化图

Fig. 5. GA characteristic wavelength extraction of sugar content of red globe grape. (a) GA-screened image; (b) change of RMSECV

下载图片查看原文

图 6. 红提糖度的SPA特征波长选取图。(a)RMSE变化图;(b)SPA选取的变量

Fig. 6. SPA characteristic wavelength extraction of sugar content of red globe grape. (a) Change of RMSE; (b) selected variables of SPA

下载图片查看原文

图 7. 红提糖度的CARS特征波长选取图。(a)采样变量数;(b) RMSECV;(c)回归系数路径

Fig. 7. CARS characteristic wavelength extraction of sugar content of red globe grape. (a) Number of sampled variables; (b) RMSECV; (c) paths of regression coefficients

下载图片查看原文

图 8. 红提糖度的UVE特征波长选取图

Fig. 8. UVE characteristic wavelength extraction of sugar content of red globe grape

下载图片查看原文

图 9. 基于GA-RF的红提糖度的最优模型

Fig. 9. Optimal model for sugar content of red globe grape based on GA-RF

下载图片查看原文

图 10. 基于MA-SPA-RF的红提硬度的最优模型

Fig. 10. Optimal model for firmness of red globe grape based on MA-SPA-RF

下载图片查看原文

表 1采用不同预处理方法得到的全波长PLSR预测模型

Table1. Full-band PLSR prediction model using different preprocessing methods

Index	Pretreatment	LVs	Calibration set		Prediction set
Index	Pretreatment	LVs	R_c	RMSEC	R_p	RMSEP
Sugar content	Raw	19	0.827	0.564	0.726	0.474
	SNV	14	0.808	0.597	0.712	0.493
	S_G	2	0.602	0.769	0.483	0.619
	MSC	8	0.811	0.595	0.665	0.515
	MA	8	0.479	0.782	0.373	0.881
	MC	18	0.825	0.617	0.701	0.503
Firmness	RAW	8	0.696	4.743	0.675	4.575
	SNV	6	0.605	5.015	0.684	4.365
	MSC	9	0.685	4.613	0.569	4.789
	MA	16	0.730	4.226	0.808	3.821
	MC	10	0.646	4.652	0.691	4.830

查看原文

表 2利用SPXY算法划分样本集的数据统计

Table2. Datastatistics of partitioning sample sets by SPXY algorithm

Number of samples	Index	Minimum	Maximum	Mean	Standard deviation
Calibration set (126 samples)	Sugar content /(° Brix)	13.875	18.625	16.109	0.971
Calibration set (126 samples)	Firmness /N	1.200	27.000	13.711	6.213
Prediction set (42 samples)	Sugar content /(° Brix)	15.000	17.5000	15.858	0.653
Prediction set (42 samples)	Firmness /N	2.700	23.500	12.774	6.264

查看原文

表 3不同放置位置下得到的全波长PLSR预测模型

Table3. Full-band PLSR prediction model with different placementorientations

Placement position	Index	LVs	Calibration set		Prediction set
Placement position	Index	LVs	R_c	RMSEC	R_p	RMSEP
Fruit stalk-side down	Sugar content	17	0.807	0.628	0.712	0.488
Fruit stalk-side down	Firmness	3	0.546	5.290	0.534	5.017
Fruit stalk-side up	Sugar content	13	0.792	0.659	0.705	0.492
Fruit stalk-side up	Firmness	6	0.646	4.854	0.606	4.780
Horizontal	Sugar content	19	0.805	0.631	0.690	0.497
Horizontal	Firmness	6	0.558	5.245	0.602	4.785
Whole fruit	Sugar content	19	0.827	0.564	0.726	0.474
Whole fruit	Firmness	19	0.730	4.226	0.808	3.821

查看原文

表 4基于红提特征波长建立的糖度和硬度预测模型的结果

Table4. Results of prediction model for sugar content and firmness based on characteristic wavelengths of red globe grape

Index	Modeling method	Extraction method	No. of wavelength	Calibration set		Prediction set
Index	Modeling method	Extraction method	No. of wavelength	R_c	RMSEC	R_p	RMSEP
Sugar content	PLSR	Raw	438	0.827	0.564	0.726	0.474
		GA	26	0.875	0.469	0.728	0.443
		SPA	17	0.862	0.492	0.745	0.429
		CARS	24	0.879	0.461	0.753	0.422
		UVE	47	0.863	0.490	0.729	0.444
	LSSVM	Raw	438	0.825	0.568	0.486	0.675
		GA	26	0.870	0.479	0.759	0.415
		SPA	17	0.864	0.489	0.752	0.426
		CARS	24	0.866	0.486	0.810	0.376
		UVE	47	0.875	0.470	0.749	0.426
	RF	Raw	438	0.954	0.260	0.873	0.402
		GA	26	0.969	0.266	0.928	0.254
		SPA	17	0.962	0.268	0.895	0.411
		CARS	24	0.946	0.296	0.890	0.406
		UVE	47	0.961	0.267	0.917	0.297
Firmness	PLSR	MA-Raw	438	0.730	4.226	0.808	3.821
		MA-GA	60	0.802	3.696	0.898	3.273
		MA-SPA	24	0.802	3.699	0.903	2.888
		MA-CARS	22	0.731	4.224	0.886	3.215
		MA-UVE	139	0.804	3.687	0.887	3.114
	LSSVM	MA-Raw	438	0.738	4.224	0.754	4.021
		MA-GA	60	0.795	3.788	0.901	3.023
		MA-SPA	24	0.741	4.183	0.870	3.578
		MA-CARS	22	0.746	4.163	0.893	3.288
		MA-UVE	139	0.833	3.444	0.921	2.674
	RF	MA-Raw	438	0.960	2.195	0.905	3.049
		MA-GA	60	0.950	2.132	0.918	2.031
		MA-SPA	24	0.961	2.119	0.932	1.634
		MA-CARS	22	0.948	2.199	0.911	2.053
		MA-UVE	139	0.959	2.120	0.921	1.893

查看原文

表 5糖度和硬度最优模型的特征波点列表

Table5. Characteristic wave points of optimal model for sugar content and firmness

Index	Modeling method	Selected variables (wavelength) /nm
Sugar content(26 points)	GA-RF	452.76, 456.53, 461.55, 600.98, 626.10, 627.36, 628.62, 631.13, 633.64, 639.92, 644.95, 646.20, 647.46, 648.71,651.23, 655.00, 859.75, 894.92, 918.78, 922.55, 927.58, 936.37, 941.40, 943.91, 945.16, 969.03
Firmness(24 points)	MA-SPA-RF	450.24, 451.50, 454.01, 464.06, 476.62, 489.19, 505.51, 557.02, 677.61, 688.91, 706.50, 825.83, 938.88, 947.68, 952.70, 958.98, 961.49, 962.75, 965.26, 969.03, 977.82, 990.38, 996.67, 997.92

查看原文

高升, 王巧华, 付丹丹, 李庆旭. 红提糖度和硬度的高光谱成像无损检测[J]. 光学学报, 2019, 39(10): 1030004. Sheng Gao, Qiaohua Wang, Dandan Fu, Qingxu Li. Nondestructive Detection of Sugar Content and Firmness of Red Globe Grape by Hyperspectral Imaging[J]. Acta Optica Sinica, 2019, 39(10): 1030004.

红提糖度和硬度的高光谱成像无损检测下载： 1012次

图 1. 三种放置模式下红提果粒的高光谱图像。(a)横放;(b)果柄侧朝下;(c)果柄侧朝上

Fig. 1. Hyperspectral images in three placement orientations. (a) Horizontal; (b) fruit stalk-side down; (c) fruit stalk-side up

图 2. 高光谱图像中背景与红提区域的反射率

Fig. 2. Reflectivity of background and red globe grape area in hyperspectral images

图 3. 红提果粒的高光谱图像处理。(a) 726.6 nm处高光谱的图像;(b)掩模板图像;(c)掩模后红提区域的图像

Fig. 3. Hyperspectral image processing of red globe grapes. (a) Hyperspectral image at 726.6 nm; (b) mask template image; (c) masked image of red globe grape area

图 4. 红提样本的原始光谱

Fig. 4. Originalspectra of red globe grape samples

图 5. 红提糖度的GA特征波长选取图。(a)GA筛选图;(b)RMSECV变化图

Fig. 5. GA characteristic wavelength extraction of sugar content of red globe grape. (a) GA-screened image; (b) change of RMSECV

图 6. 红提糖度的SPA特征波长选取图。(a)RMSE变化图;(b)SPA选取的变量

Fig. 6. SPA characteristic wavelength extraction of sugar content of red globe grape. (a) Change of RMSE; (b) selected variables of SPA

图 7. 红提糖度的CARS特征波长选取图。(a)采样变量数;(b) RMSECV;(c)回归系数路径

Fig. 7. CARS characteristic wavelength extraction of sugar content of red globe grape. (a) Number of sampled variables; (b) RMSECV; (c) paths of regression coefficients

图 8. 红提糖度的UVE特征波长选取图

Fig. 8. UVE characteristic wavelength extraction of sugar content of red globe grape

图 9. 基于GA-RF的红提糖度的最优模型

Fig. 9. Optimal model for sugar content of red globe grape based on GA-RF

图 10. 基于MA-SPA-RF的红提硬度的最优模型

Fig. 10. Optimal model for firmness of red globe grape based on MA-SPA-RF

表 1采用不同预处理方法得到的全波长PLSR预测模型

Table1. Full-band PLSR prediction model using different preprocessing methods

表 2利用SPXY算法划分样本集的数据统计

Table2. Datastatistics of partitioning sample sets by SPXY algorithm

表 3不同放置位置下得到的全波长PLSR预测模型

Table3. Full-band PLSR prediction model with different placementorientations

表 4基于红提特征波长建立的糖度和硬度预测模型的结果

Table4. Results of prediction model for sugar content and firmness based on characteristic wavelengths of red globe grape

表 5糖度和硬度最优模型的特征波点列表

Table5. Characteristic wave points of optimal model for sugar content and firmness

关于本站 Cookie 的使用提示

全站搜索

红提糖度和硬度的高光谱成像无损检测 下载： 1012次

图 1. 三种放置模式下红提果粒的高光谱图像。(a)横放;(b)果柄侧朝下;(c)果柄侧朝上

Fig. 1. Hyperspectral images in three placement orientations. (a) Horizontal; (b) fruit stalk-side down; (c) fruit stalk-side up

图 2. 高光谱图像中背景与红提区域的反射率

Fig. 2. Reflectivity of background and red globe grape area in hyperspectral images

图 3. 红提果粒的高光谱图像处理。(a) 726.6 nm处高光谱的图像;(b)掩模板图像;(c)掩模后红提区域的图像

Fig. 3. Hyperspectral image processing of red globe grapes. (a) Hyperspectral image at 726.6 nm; (b) mask template image; (c) masked image of red globe grape area

图 4. 红提样本的原始光谱

Fig. 4. Originalspectra of red globe grape samples

图 5. 红提糖度的GA特征波长选取图。(a)GA筛选图;(b)RMSECV变化图

Fig. 5. GA characteristic wavelength extraction of sugar content of red globe grape. (a) GA-screened image; (b) change of RMSECV

图 6. 红提糖度的SPA特征波长选取图。(a)RMSE变化图;(b)SPA选取的变量

Fig. 6. SPA characteristic wavelength extraction of sugar content of red globe grape. (a) Change of RMSE; (b) selected variables of SPA

图 7. 红提糖度的CARS特征波长选取图。(a)采样变量数;(b) RMSECV;(c)回归系数路径

Fig. 7. CARS characteristic wavelength extraction of sugar content of red globe grape. (a) Number of sampled variables; (b) RMSECV; (c) paths of regression coefficients

图 8. 红提糖度的UVE特征波长选取图

Fig. 8. UVE characteristic wavelength extraction of sugar content of red globe grape

图 9. 基于GA-RF的红提糖度的最优模型

Fig. 9. Optimal model for sugar content of red globe grape based on GA-RF

图 10. 基于MA-SPA-RF的红提硬度的最优模型

Fig. 10. Optimal model for firmness of red globe grape based on MA-SPA-RF

表 1采用不同预处理方法得到的全波长PLSR预测模型

Table1. Full-band PLSR prediction model using different preprocessing methods

表 2利用SPXY算法划分样本集的数据统计

Table2. Datastatistics of partitioning sample sets by SPXY algorithm

表 3不同放置位置下得到的全波长PLSR预测模型

Table3. Full-band PLSR prediction model with different placementorientations

表 4基于红提特征波长建立的糖度和硬度预测模型的结果

Table4. Results of prediction model for sugar content and firmness based on characteristic wavelengths of red globe grape

表 5糖度和硬度最优模型的特征波点列表

Table5. Characteristic wave points of optimal model for sugar content and firmness

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

红提糖度和硬度的高光谱成像无损检测下载： 1012次