Early Classification and Detection of Melon Graft Healing State Based on Hyperspectral Imaging

Fig. 2. Overall research plan

Fig. 3. Segmentation diagram
(a): Segmentation of region of interest and background region; (b): Spectra of region of interest and background region; (c): Mask

Fig. 4. CARS-SPA algorithm
(a): Feature band extracted by CARS algorithm;(b): Feature band extracted by CARS-SPA algorithm

Fig. 5. The choice of grafting difference information
(a): Average spectra of grafted survival and non-viable seedlings on the 7th day; (b): Average absorption spectra of grafted survival and non-viable seedlings on the 7th day; (c): Location of the extracted feature wavelengths using DIS-CARS-SPA algorithm

Table 1. Two classification results of grafted survival seedlings and non-viable seedlings on the same day based on SVM constructed by different pretreatments

Table 1. Two classification results of grafted survival seedlings and non-viable seedlings on the same day based on SVM constructed by different pretreatments

天数	P/%
天数	无	SG	FD	SD	MSC	SNV	SG-SD	SNV-SD	SNV-SG-SD
1	99.11	99.11	99.33	99.33	99.33	99.33	99.33	99.05	99.33
2	98.82	98.82	99.76	99.76	99.29	99.76	99.56	100.00	100.00
3	96.73	97.01	98.63	99.23	98.09	99.18	99.25	99.18	99.87
4	98.11	98.11	97.64	99.05	98.02	98.85	98.86	99.02	99.58
5	97.34	97.58	99.03	99.15	98.55	99.03	99.08	99.03	100.00
6	96.54	96.04	98.36	98.85	98.15	98.36	97.98	99.38	99.69
7	97.85	97.62	99.64	99.81	99.04	99.25	99.86	99.76	100.00

Table 2. Two classification results of survival/non-viable seedlings grafted on different days using SVM constructed based on SNV-SG-SD pretreatment
View table
View in Article
Table 2. Two classification results of survival/non-viable seedlings grafted on different days using SVM constructed based on SNV-SG-SD pretreatment
类型 P/%
A B C D A-B A-C B-C A-D
嫁接成活苗 90.17 97.68 99.15 - 99.33 99.65 99.23 -
嫁接非成活苗 97.03 - - 99.76 - - - 99.95

Table 3. Fourteen classification results of grafted survival/non-viable seedlings by SVM based on feature extraction after SNV-SG-SD preprocessing

Table 3. Fourteen classification results of grafted survival/non-viable seedlings by SVM based on feature extraction after SNV-SG-SD preprocessing

输入类型	预处理	特征提取	分类模型	特征波段数	P/%
1~7 d甜瓜嫁接苗	SNV-SG-SD	CARS-SPA	SVM(RBF)	32	96.26
1~7 d甜瓜嫁接苗	SNV-SG-SD	DIS-CARS-SPA	SVM(RBF)	26	96.85
1~7 d甜瓜嫁接苗	SNV-SG-SD	-	SVM(RBF)	150	93.48

Table 4. Parameter selection of SVM model based on different optimization algorithms

Table 4. Parameter selection of SVM model based on different optimization algorithms

输入类型	预处理	特征提取	分类模型	SVM参数	P/%
1~7 d甜瓜嫁接苗	SNV-SG-SD	DIS-CARS-SPA	GS-SVM(RBF)	c=1 000, g=1 000	96.85
1~7 d甜瓜嫁接苗	SNV-SG-SD	DIS-CARS-SPA	PSO-SVM(RBF)	c=1 000, g=10	87.89
1~7 d甜瓜嫁接苗	SNV-SG-SD	DIS-CARS-SPA	QPSO-SVM(RBF)	c=1 000, g=100	92.62
1~7 d甜瓜嫁接苗	SNV-SG-SD	DIS-CARS-SPA	GA-SVM(RBF)	c=100, g=10	75.32
1~7 d甜瓜嫁接苗	SNV-SG-SD	DIS-CARS-SPA	QGA-SVM(RBF)	c=100, g=100	85.27

Table 5. Parameter selection of SVM model based on different kernel functions