Crop classification of modern agricultural park based on time-series Sentinel-2 images

Dongyan Zhang; Zhen Dai; Xingang Xu; Guijun Yang; Yang Meng; Haikuan Feng; Qi Hong; Fei Jiang

doi:10.3788/IRLA20200318

Journals >Infrared and Laser Engineering >Volume 50 >Issue 5 >Page 20200318 > Article

Infrared and Laser Engineering
Vol. 50, Issue 5, 20200318 (2021)

Crop classification of modern agricultural park based on time-series Sentinel-2 images

Dongyan Zhang¹, Zhen Dai^1、2, Xingang Xu^2、*, Guijun Yang², Yang Meng², Haikuan Feng², Qi Hong¹, and Fei Jiang^1、3

Author Affiliations

¹National Engineering Research Center for Agro-Ecological Big Data Analysis & Application, Anhui University, Hefei 230601, China

²Beijing Agricultural Information Technology Research Center, Beijing 100097, China

³School of Information Engineering, Suzhou University, Suzhou 234000, China

show less

DOI: 10.3788/IRLA20200318 Cite this Article

Dongyan Zhang, Zhen Dai, Xingang Xu, Guijun Yang, Yang Meng, Haikuan Feng, Qi Hong, Fei Jiang. Crop classification of modern agricultural park based on time-series Sentinel-2 images[J]. Infrared and Laser Engineering, 2021, 50(5): 20200318 Copy Citation Text

show less

Fig. 1. Geographical location and sample distributions of the study area

Download full size | View in the Article

Fig. 2. Temporal changes of NDVI (a), RVI (b), EVI (c) and Ref (NIR) (d) for main crops

Download full size | View in the Article

Fig. 3. Flow chart of crop classification extraction

Download full size | View in the Article

Fig. 4. Crop classification results using Decision Tree (a), Random Forest (b), Support Vector Machine (c), Maximum Likelihood (d)

Download full size | View in the Article

Type	May			June			July			August			September			October
Type	E	M	L	E	M	L	E	M	L	E	M	L	E	M	L	E	M
Note: E means the early 10 days of a month,M is the middle 10 days，and L represents the lately 10 days.
Rice			Sowing			Tillering			Heading			Filling			Maturity
Soybean	Sowing				Seeding			Flowering		Podding			Filling			Maturity
Stevia			Transplanting					Branching				Flowering			Maturity
Corn	Sowing		Seeding				Jointing			Tasseling		Filling				Maturity
Dry rice		Sowing				Tillering				Heading		Filling				Maturity

Table 1. Growth period of five crops in the study area

View in the Article

Data time	Sensor	Quality
2019-05-22	Sentinel-2A	Best
2019-06-11	Sentinel-2A	Good
2019-06-21	Sentinel-2A	Best
2019-07-01	Sentinel-2A	Good
2019-08-15	Sentinel-2B	Best
2019-08-30	Sentinel-2A	Best
2019-09-14	Sentinel-2B	Best
2019-09-24	Sentinel-2B	Best
2019-10-04	Sentinel-2B	Best

Table 2. Data lists of Sentinel-2 images

View in the Article

Sentinel-2 bands	Wavelength/μm	Reflection/m
Band1-Coastal aerosol	0.443	60
Band2-Blue	0.490	10
Band3-Green	0.560	10
Band4-Red	0.665	10
Band5-Vegetation red edge	0.705	20
Band6-Vegetation red edge	0.740	20
Band7-Vegetation red edge	0.783	20
Band8-NIR	0.842	10
Band8A-Vegetation red edge	0.865	20
Band9-Water vapour	0.945	60
Band10-SWIR-Cirrus	1.375	60
Band11-SWIR1	1.610	20
Band12-SWIR2	2.190	20

Table 3. Spectral bands of the Sentinel-2 sensors (S2A & S2B)

View in the Article

Indicator	Description	Source
Notes: In the formula, $ {\mathrm{\rho }}_{\mathrm{N}\mathrm{I}\mathrm{R}} $ is the near-infrared band reflectivity, $ {\mathrm{\rho }}_{\mathrm{R}\mathrm{E}\mathrm{D}} $ is the red band reflectivity, $ {\mathrm{\rho }}_{\mathrm{B}\mathrm{L}\mathrm{U}\mathrm{E}} $ is the blue band reflectivity and L is the soil adjustment coefficient of 1.
Normalized Difference Vegetation Index(NDVI)	${\rm NDVI} = \dfrac{ { {\rho _{\rm NIR} } - {\rho _{\rm RED} } } }{ { {\rho _{\rm NIR} } + {\rho _{\rm RED} } } }$	Ref.[15]
Ratio Vegetation Index(RVI)	${\rm RVI} = \dfrac{ { {\rho _{\rm NIR} } } }{ { {\rho _{\rm RED} } } }$	Ref. [16]
Enhanced Vegetation Index(EVI)	${\rm EVI} = 2.5×\dfrac{ { {\rho _{\rm NIR} } - {\rho _{\rm RED} } } }{ { {\rho _{\rm NIR} } + 6.0×{\rho _{\rm RED} } - 7.5×{\rho _{\rm BLUE} } + L} }$	Ref. [17-18]
Near Infrared Ray(Ref(NIR))	The reflection of Band-8 in Tab.3	Ref. [19]

Table 4. Classification indicators used in the study

View in the Article

Type	Calculation formula
Notes: where k represents the number of rows and columns of the confusion matrix, X_ii represents the value on the diagonal of the confusion matrix that is the number of pixels correctly classified,N represents the total number of pixels verified, X_i represents the i row of the confusion matrix. The sum of elements, X_j represents the sum of elements in the j column of the confusion matrix.
Mapping accuracy	$\mathrm{P}\mathrm{A}=\dfrac{ {X}_{ii} }{ {X}_{j} }×100\%$
User accuracy	$\mathrm{U}\mathrm{A}=\dfrac{ {X}_{ii} }{ {X}_{i} }×100\%$
Overall accuracy	$\mathrm{O}\mathrm{A}=\displaystyle\sum _{i=1}^{k}\dfrac{ {X}_{ii} }{N}×100\%$
Kappa coefficient	${{K} } = \dfrac{ {N\displaystyle\sum\nolimits_{i = 1}^k { {X_{ii} } } - \sum\nolimits_{i = 1}^k { {X_i}{X_j} } } }{ { {N^2} - \displaystyle\sum\nolimits_{i = 1}^k { {X_i}{X_j} } } }$

Table 5. Formulas of accuracy evaluation

View in the Article

Type	Soybean	Rice	Stevia	Corn	Dry rice	Total	User accuracy
Soybean	3288	4	0	0	320	3612	91.03%
Rice	0	5680	16	33	176	5905	96.19%
Stevia	0	0	2039	0	59	2098	97.19%
Corn	0	1	0	8451	221	8673	97.44%
Dry rice	8	9	147	1055	4820	6039	79.81%
Mapping accuracy	99.76%	99.75%	92.60%	88.57%	86.13%

Table 6. Confusion matrix result of Decision Tree

View in the Article

Type	Soybean	Rice	Stevia	Corn	Dry rice	Total	User accuracy
Soybean	3294	0	0	0	393	3687	89.34%
Rice	0	5692	19	183	216	6110	93.16%
Stevia	0	0	1993	0	8	2001	99.60%
Corn	0	0	2	9301	41	9344	99.54%
Dry rice	4	2	142	102	4938	5188	95.18%
Mapping accuracy	99.88%	99.96%	92.44%	97.03%	88.24%

Table 7. Confusion matrix result of Random Forest

View in the Article

Classification method	Overall accuracy	Kappa coefficient
Maximum Likelihood	86.5%	0.823
Support Vector Machine	91.6%	0.890
Decision Tree	92.2%	0.897
Random Forest	95.8%	0.944

Table 8. Overall accuracy estimation and Kappa coefficient of classification based on each method

Download Citation

Save the article for my favorites

Paper Information