Open AccessExtraction of Crop Planting Structure in County Based on Multi-temporal Images of Sentinel-2
Author(s) -
Zhengqian Li,
Feng Xiong
Publication year - 2021
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/632/5/052031
Subject(s) - cohen's kappa , principal component analysis , pattern recognition (psychology) , feature (linguistics) , dimension (graph theory) , feature extraction , computer science , texture (cosmology) , artificial intelligence , vegetation (pathology) , set (abstract data type) , autocorrelation , mathematics , remote sensing , data mining , statistics , geography , image (mathematics) , medicine , linguistics , philosophy , pathology , programming language , pure mathematics
Accurate land information using was the foundation of the modern agriculture. The Sential-2 images were used as data sources in this paper. Firstly, spectral curve analysis of the features in the study area was performed and non-vegetated areas were excluded by using decision tree method. Since the same texture feature in each band had a certain autocorrelation, the principal component analysis method was used to reduce the dimension. In the end, 49 features including original band, vegetated band, texture feature could be achieved, then 3 models with different input which were classified by random forest method could be set up. The result showed that the model which used original band, vegetation index and decreased dimension texture feature had the highest accuracy. The total classification accuracy and Kappa index was 91.03% and 0.81771 respectively. Therefore, some conclusions were figured out: (1) the classification method based on multiple features was helpful to improve the accuracy; (2) the time series characteristics of vegetation index could help improve the classification accuracy of non-main crops; (3) decreasing the dimension of texture feature could improve the efficiency and cartographic quality of the classification.