Mapping rice paddy distribution using remote sensing by coupling deep learning with phenological characteristics

Zhu，A. Xing1,2,3,4,5; Zhao，Fang He2,4; Pan，Hao Bo1; Liu，Jun Zhi1

2021-04-01

10.3390/rs13071360

Remote Sensing   影响因子和分区

2072-4292

Two main approaches are used in mapping rice paddy distribution from remote sensing images: phenological methods or machine learning methods. The phenological methods can map rice paddy distribution in a simple way but with limited accuracy. Machine learning, particularly deep learning, methods that learn the spectral signatures can achieve higher accuracy yet require a large number of field samples. This paper proposed a pheno-deep method to couple the simplicity of the phenological methods and the learning ability of the deep learning methods for mapping rice paddy at high accuracy without the need of field samples. The phenological method was first used to initially delineate the rice paddy for the purpose of creating training samples. These samples were then used to train the deep learning model. The trained deep learning model was applied to map the spatial distribution of rice paddy. The effectiveness of the pheno-deep method was evaluated in Jin’an District, Lu’an City, Anhui Province, China. Results show that the pheno-deep method achieved a high performance with the overall accuracy, the precision, the recall, and AUC (area under curve) being 88.8%, 87.2%, 91.1%, and 94.4%, respectively. The pheno-deep method achieved a much better performance than the phenological alone method and can overcome the noises in the training samples from the phenological method. The overall accuracy of the pheno-deep method is only 2.4% lower than that of the deep learning alone method trained with field samples and this difference is not statistically significant. In addition, the pheno-deep method requires no field sampling, which would be a noteworthy advantage for situations when large training samples are difficult to obtain. This study shows that by combining knowledge-based methods with data-driven methods, it is possible to achieve high mapping accuracy of geographic variables using remote sensing even with little field sampling efforts.

Deep learning Machine learning Phenological characteristics Remote sensing Rice paddy distribution Spatial prediction

SCI ; EI

英语

其他

WOS:000638791300001

20211610217237

Knowledge based systems ; Learning systems ; Mapping ; Remote sensing ; Sampling

Surveying:405.3 ; Expert Systems:723.4.1

2-s2.0-85104069180

Scopus

1.Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application,School of Geography,Nanjing Normal University,Nanjing,210023,China
2.State Key Laboratory of Resources and Environmental Information System,Institute of Geographic Sciences and Natural Resources Research,Chinese Academy of Sciences,Beijing,100101,China
3.Department of Geography,University of Wisconsin-Madison,Madison,53706,United States
4.College of Resources and Environment,University of Chinese Academy of Sciences,Beijing,100049,China
5.Center for Social Sciences,Southern University of Science and Technology,Shenzhen,518055,China

Zhu，A. Xing,Zhao，Fang He,Pan，Hao Bo,et al. Mapping rice paddy distribution using remote sensing by coupling deep learning with phenological characteristics[J]. Remote Sensing,2021,13(7).

Zhu，A. Xing,Zhao，Fang He,Pan，Hao Bo,&Liu，Jun Zhi.(2021).Mapping rice paddy distribution using remote sensing by coupling deep learning with phenological characteristics.Remote Sensing,13(7).

Zhu，A. Xing,et al."Mapping rice paddy distribution using remote sensing by coupling deep learning with phenological characteristics".Remote Sensing 13.7(2021).