Remote Sensing Vegetation Classification Method Based on Vegetation Index and Convolution Neural Network

Mingzhu Xu; Hao Xu; Peng Kong; Yanlan Wu

doi:10.3788/LOP202259.2428005

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 24 >Page 2428005 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 24, 2428005 (2022)

Remote Sensing Vegetation Classification Method Based on Vegetation Index and Convolution Neural Network

Mingzhu Xu¹, Hao Xu², Peng Kong², and Yanlan Wu^1,3,4,*

Author Affiliations

¹School of Resources and Environmental Engineering, Anhui University, Hefei 230601, Anhui, China

²Institute of Spacecraft System Engineering, Beijing 100094, China

³Information Materials and Intelligent Sensing Laboratory of Anhui Province, Hefei 230601, Anhui, China

⁴Anhui Engineering Research Center for Geographical Information Intelligent Technology, Hefei 230601, Anhui, China

show less

DOI: 10.3788/LOP202259.2428005 Cite this Article Set citation alerts

Mingzhu Xu, Hao Xu, Peng Kong, Yanlan Wu. Remote Sensing Vegetation Classification Method Based on Vegetation Index and Convolution Neural Network[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2428005 Copy Citation Text

show less

Fig. 1. Flow chart of network structure

Download full size

Fig. 2. Dense blocks

Download full size

Fig. 3. Atrous spatial pyramid pooling (ASPP) block

Download full size

Fig. 4. Multiresolution feature fusion modes. (a) Mode 1); (b) mode 2); (c) mode 3)

Download full size

Fig. 5. Example of vegetation samples from remote sensing images

Download full size

Fig. 6. City classification results before and after HRDN joined NDVI

Download full size

Fig. 7. Rural classification results before and after HRDN joined NDVI

Download full size

Fig. 8. Result map of vegetation classification in urban areas by HRDN, Deeplab-V3+, BiseNet, and DCCN

Download full size

Fig. 9. Result map of vegetation classification in rural areas by HRDN, Deeplab-V3+, BiseNet, and DCCN

Download full size

Fig. 10. Result map of vegetation classification and extraction by proposed method

Download full size

Use	Sample	Model testing	Universal validation 1	Universal validation 2
Sensor	GF2-PMS2	GF2-PMS2	GF6-PMS	GF2-PMS2
Region	Hefei	Hefei	Hefei	Beijing
Spatial resolution /m	1	1	2	1
Image acquisition date	2015-08-03	2016-08-27 2015-08-03	2018-10-04	2016-08-27
Midline coordinates	E117.3，N31.7	E116.8，N40.4 E117.3，N31.7	E117.4，N32.1	E116.8，N40.4

Table 1. Image parameters

Region	Image	Cultivated land		Grassland		Forest		Aquatic vegetation		Mean F1 score	Mean IOU	OA
Region	Image	F1	IOU	F1	IOU	F1	IOU	F1	IOU	Mean F1 score	Mean IOU	OA
City	Image 1	14.53	9.01	18.22	10.02	63.33	46.33	/	/	32.03	21.79	87.02
	Image 2	15.67	8.50	13.59	7.29	67.49	50.93	67.28	50.70	41.01	29.36	89.79
	Image 3	/	/	12.44	6.63	90.27	82.27	/	/	51.36	44.45	94.88
	Image 4	/	/	14.94	8.07	83.56	71.77	/	/	49.25	39.92	93.90
	All images	15.10	8.76	14.80	8.00	76.16	62.83	67.28	50.70	43.41	33.88	91.40
Rural	Image 5	89.05	80.26	87.80	78.26	79.20	65.56	78.07	64.02	83.53	72.03	78.49
	Image 6	86.19	75.74	81.89	69.33	77.21	62.87	77.32	63.03	80.65	67.74	77.90
	Image 7	86.71	76.54	84.44	73.07	80.99	68.05	80.68	67.61	83.21	71.32	82.60
	Image 8	72.08	56.35	70.97	55.00	61.11	44.00	60.68	43.11	66.21	49.62	61.16
	All images	83.51	72.22	81.28	68.92	74.63	60.12	74.19	59.44	78.40	65.18	75.04
（a）
Region	Image	Cultivated land		Grassland		Forest		Aquatic vegetation		Mean F1 score	Mean IOU	OA
Region	Image	F1	IOU	F1	IOU	F1	IOU	F1	IOU	Mean F1 score	Mean IOU	OA
City	Image 1	53.23	36.27	57.49	40.34	82.45	70.14	/	/	64..39	48.92	93.48
	Image 2	56.02	38.91	62.31	45.25	84.5	96.56	84.48	73.12	71.83	63.46	94.79
	Image 3	/	/	77.61	63.42	98.34	96.73	/	/	87.98	80.08	99.25
	Image 4	/	/	75.32	60.41	97.87	95.83	/	/	86.60	78.12	99.38
	All images	54.63	37.59	68.18	52.36	90.79	89.82	84.48	73.12	82.14	67.65	96.73
Rural	Image 5	90.60	82.82	88.58	79.50	85.49	74.66	84.82	73.64	87.37	77.66	87.20
	Image 6	94.33	89.27	92.50	86.05	91.71	84.69	91.67	84.62	92.55	86.16	93.08
	Image 7	94.34	89.29	93.14	87.16	92.52	86.08	92.43	85.93	93.11	87.11	94.41
	Image 8	88.29	79.04	86.44	76.12	89.03	80.23	88.59	79.52	88.09	78.73	90.70
	All images	91.89	85.11	90.17	82.21	89.69	81.42	89.38	80.93	90.28	82.41	91.35
（b）

Table 2. Comparison of classification accuracy of HRDN before and after adding NDVI. (a), (b) are classification accuracies of fused NDVI model and unfused NDVI model, respectively

Image 1	Cultivated land	Grassland	Forest	Aquatic vegetation	UA /%
Cultivated land	47.35	2.22	1.21	/	60.95
Grassland	8.52	56.64	4.45	/	60.33
Forest	33.24	31.31	85.02	/	90.84
Aquatic vegetation	0	0	0.01	/	0
PA /%	47.35	56.64	85.02	/
Kappa：84.50%
（a）
Image 2	Cultivated land	Grassland	Forest	Aquatic vegetation	UA /%
Cultivated land	58.62	6.73	0.14	0	53.63
Grassland	6.47	60.63	2.87	27.66	66.91
Forest	31.86	28.72	93.15	1.1	84.63
Aquatic vegetation	0	0.02	0.02	67.26	58.37
PA /%	58.62	60.63	93.15	67.26
Kappa：85.85%
（b）
Image 3	Cultivated land	Grassland	Forest	Aquatic vegetation	UA /%
Cultivated land	/	0	0	/	0
Grassland	/	81.38	0.42	/	79.97
Forest	/	18.43	99.56	/	98.68
Aquatic vegetation	/	0.12	0.20	/	0
PA /%	/	81.38	99.35	/
Kappa：98.37%
（c）
Image 4	Cultivated land	Grassland	Forest	Aquatic vegetation	UA /%
Cultivated land	/	0.37	0.14	/	0
Grassland	/	85.08	1.09	/	70.40
Forest	/	14.16	97.89	/	99.47
Aquatic vegetation	/	0.05	0.01	/	0
PA /%	/	85.08	97.89	/
Kappa：98.29%
（d）

Table 3. Confusion matrices of images in urban areas. (a)-(d) are confusion matrices of four images in urban area respectively

Image 5		Cultivated land		Grassland		Forest		Aquatic vegetation		UA /%
Cultivated land		87.51		2.00		8.14		27.34		93.92
Grassland		2.03		25.05		2.18		0.71		11.80
Forest		7.85		60.66		80.87		22.70		77.96
Aquatic vegetation		0.27		4.65		0.54		42.39		66.73
PA /%		87.51		25.05		80.87		42.39
Kappa：80.61%
（a）
Image 6		Cultivated land		Grassland		Forest		Aquatic vegetation		UA /%
Cultivated land		94.71		16.74		5.90		1.09		93.96
Grassland		0.67		37.79		0.53		0.23		53.04
Forest		4.47		36.79		92.50		7.27		88.87
Aquatic vegetation		0.06		0.24		0.42		89.87		92.32
PA /%		94.71		37.79		92.5		89.87
Kappa：90.05%
（b）
Image 7	Cultivated land		Grassland		Forest		Aquatic vegetation		UA /%
Cultivated land	93.51		7.65		5.32		0.96		95.19
Grassland	0.20		70.67		1.55		0.08		52.67
Forest	5.41		18.45		91.74		9.61		91.76
Aquatic vegetation	0.24		0.09		0.16		86.54		85.99
PA /%	93.51		70.67		91.74		86.54
Kappa：91.78%
（c）
Image 8	Cultivated land		Grassland		Forest		Aquatic vegetation		UA /%
Cultivated land	88.8		40.70		6.01		13.83		87.79
Grassland	0.14		26.37		0.24		0.26		70.91
Forest	10.19		29.37		93.22		15.85		89.07
Aquatic vegetation	0.68		0.71		0.26		69.53		83.87
PA /%	88.8		26.37		93.22		69.53
Kappa：86.23%
（d）

Table 4. Confusion matrices of images in rural areas. (a)-(d) are confusion matrices of four images in rural areas respectively

Region	Image	Cultivated land		Grassland		Forest		Aquatic vegetation		Mean F1 score	Mean IOU	OA
Region	Image	F1	IOU	F1	IOU	F1	IOU	F1	IOU	Mean F1 score	Mean IOU	OA
City	Image 1	39.60	24.69	36.60	22.40	79.31	65.72	/	/	51.84	37.60	92.48
	Image 2	55.19	38.11	38.50	23.84	80.10	66.80	80.06	66.74	63.46	48.87	93.32
	Image 3	/	/	30.68	18.12	92.60	86.21	/	/	61.64	52.17	95.77
	Image 4	/	/	35.37	21.48	91.17	83.77	/	/	63.27	52.63	96.23
	All images	47.40	31.40	35.29	21.46	85.99	75.80	80.06	66.74	60.05	46.21	94.45
Rural	Image 5	90.00	81.81	89.19	80.49	85.50	74.67	84.72	73.49	87.35	78.99	85.96
	Image 6	91.47	84.29	90.07	81.93	88.35	79.13	88.40	79.21	89.57	81.14	88.57
	Image 7	87.98	78.54	86.66	76.46	85.10	74.07	85.00	73.92	86.19	75.75	87.53
	Image 8	77.59	63.39	76.95	62.46	79.12	65.46	78.57	64.70	70.06	64.00	78.92
	All images	86.76	77.01	85.71	75.34	84.52	73.33	84.17	72.83	83.29	74.97	85.25
（a）
Region	Image	Cultivated land		Grassland		Forest		Aquatic vegetation		Mean F1 score	Mean IOU	OA
Region	Image	F1	IOU	F1	IOU	F1	IOU	F1	IOU	Mean F1 score	Mean IOU	OA
City	Image 1	0	0	35.99	21.94	67.16	50.56	/	/	51.57	36.25	88.42
	Image 2	0	0	35.52	21.59	68.77	52.41	68.72	52.35	57.67	42.12	90.36
	Image 3	/	/	67.95	51.45	92.84	86.64	/	/	80.40	69.05	95.64
	Image 4	/	/	51.00	34.23	89.86	81.59	/	/	70.43	57.91	95.91
	All images	0	0	47.62	32.30	79.66	67.80	68.72	52.35	65.337	50.82	92.58
Rural	Image 5	90.43	82.53	89.68	81.30	85.86	75.22	84.84	73.66	87.70	78.17	86.48
	Image 6	91.13	83.71	89.70	81.33	86.79	76.67	86.65	76.44	88.57	79.54	87.43
	Image 7	89.73	81.37	88.01	78.59	86.30	75.91	85.89	75.28	87.48	77.79	88.13
	Image 8	80.86	67.86	80.05	66.73	81.63	68.97	80.95	67.99	80.87	67.89	82.27
	All images	88.04	78.87	86.86	76.99	85.15	74.19	84.58	73.34	86.16	75.85	86.08
（b）
Region	Image	Cultivated land		Grassland		Forest		Aquatic vegetation		Mean F1 score	Mean IOU	OA
Region	Image	F1	IOU	F1	IOU	F1	IOU	F1	IOU	Mean F1 score	Mean IOU	OA
City	Image 1	16.53	9.01	18.22	10.02	63.32	46.33	/	/	32.69	21.79	87.02
	Image 2	15.67	8.50	13.59	7.29	67.49	50.93	67.28	50.70	41.01	29.36	89.79
	Image 3	/	/	12.44	6.63	90.27	82.27	/	/	51.36	44.45	94.88
	Image 4	/	/	14.94	8.07	83.56	71.77	/	/	49.25	39.92	93.90
	All images	16.51	8.76	14.80	8.00	76.16	62.83	67.28	50.70	43.58	33.88	91.40
Rural	Image 5	89.05	80.26	87.80	78.26	79.20	65.56	78.07	64.02	83.53	72.03	78.49
	Image 6	86.19	75.74	81.89	69.33	77.21	62.87	77.32	63.03	80.65	67.74	77.90
	Image 7	86.71	76.54	84.44	70.07	80.99	68.05	80.68	67.61	83.21	70.57	82.60
	Image 8	72.08	56.35	70.97	55.00	61.11	44.00	60.25	43.11	66.10	49.62	61.16
	All images	83.51	72.22	81.28	68.17	74.63	60.12	74.08	59.44	78.37	70.11	75.04
（c）

Table 5. Classification results of three methods. (a), (b), (c) are classification and extraction results of Deeplab-V3+, BiseNet, and DCCN models, respectively

Region	Cultivated land		Grassland		Forest		Aquatic vegetation		Mean F1 score	Mean IOU	OA	Kappa
Region	F1	IOU	F1	IOU	F1	IOU	F1	IOU	Mean F1 score	Mean IOU	OA	Kappa
GF-6 Hefei image 1	89.04	80.24	85.99	75.42	84.60	73.30	84.17	72.66	86.53	75.41	88.42	82.97
GF-6 Hefei image 2	59.79	42.64	62.56	45.51	84.22	72.74	84.17	72.67	72.69	58.39	93.55	84.14
GF-2 Beijing image 1	89.08	80.31	86.63	76.41	88.89	80.00	88.64	79.60	88.31	79.08	88.95	82.78
GF-2 Beijing image 2	71.40	55.52	64.89	48.03	85.82	75.17	85.71	75.00	76.95	63.43	92.41	83.32

Table 6. Extraction accuracy of vegetation classification by proposed method

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享