Convolutional Neural Network for Thailand's Eastern Economic Corridor (EEC) land cover classification using overlapping process on satellite images

被引:17
作者
Chermprayong, P. [1 ]
Hongkarnjanakul, N. [2 ]
Rouquette, D. [3 ]
Schwob, C. [4 ]
Mezeix, L. [1 ]
机构
[1] Burapha Univ, Fac Engn, 169 Long Hard Bangsaen Rd, Chon Buri 20131, Thailand
[2] Geoinformat & Space Technol Dev Agcy, Space Krenovat Pk SKP 88 Moo 9, Chon Buri 20230, Thailand
[3] Airbus Def & Space, 31 Rue Cosmonautes, F-31400 Toulouse, France
[4] Airbus Singapore, 12 Seletar Aerosp Link, Singapore 797553, Singapore
关键词
Convolutional neural network; Artificial intelligence; Land cover; Image processing; Overlapped images; Satellite image; LOP BURI PROVINCE; HIGH-RESOLUTION; SEMANTIC SEGMENTATION; AERIAL IMAGES; SCENARIOS; IMPACT;
D O I
10.1016/j.rsase.2021.100543
中图分类号
X [环境科学、安全科学];
学科分类号
08 ; 0830 ;
摘要
Land cover is a powerful tool in urban management as a source of information to support authorities' decision making. In this paper, land cover of Eastern Economic Corridor cities in Thailand is performed using the aggregation of results from two Convolutional Neural Network models, with both using the same architecture. The first model is for the detection of water and the second is for the classification of land type consisting of 3 classes: city, forest and crop. In Firstly, the size of the 4 class existing dataset is increased resulting in an accuracy of 98% and 93% for the binary and three class CNN model respectively. To improve the land cover on satellite images an overlapping process is introduced in order to reduce the classification area from 0.25 km(2) to 0.004 km(2), using the same image resolution of 8 m per pixel. The use of the overlapping allows to propose a largely better land cover where the contour of the detected class is well produced. Moreover, this methods shows its better ability to detect smaller surface size and especially for the water, crops and forest class. However, it is showed that the overlapping method does not improved the accuracy of the prediction that is mainly related to the dataset size. Finally, the robustness of the proposed method to quickly perform a global land cover with limited computer power is demonstrated.
引用
收藏
页数:14
相关论文
共 81 条
[1]   Large-Scale Hydrological Modelling of the Upper Parana River Basin [J].
Abou Rafee, Sameh A. ;
Uvo, Cintia B. ;
Martins, Jorge A. ;
Domingues, Leonardo M. ;
Rudke, Anderson P. ;
Fujita, Thais ;
Freitas, Edmilson D. .
WATER, 2019, 11 (05)
[2]   Remote sensing of 10 years changes in the vegetation cover of the northwestern coastal land of Red Sea, Saudi Arabia [J].
Alharthi, Awad ;
El-Sheikh, Mohamed A. ;
Elhag, Mohamed ;
Alatar, Abdulrahman A. ;
Abbadi, Ghanim A. ;
Abdel-Salam, Eslam M. ;
Arif, Ibrahim A. ;
Baeshen, Ariej A. ;
Eid, Ebrahem M. .
SAUDI JOURNAL OF BIOLOGICAL SCIENCES, 2020, 27 (11) :3169-3179
[3]  
Anil K., 2020, APPL ACOUST, V167
[4]   Fully automatic multi-temporal land cover classification using Sentinel-2 image data [J].
Baamonde, Sergio ;
Cabana, Martino ;
Sillero, Neftali ;
Penedo, Manuel G. ;
Naveira, Horacio ;
Novo, Jorge .
KNOWLEDGE-BASED AND INTELLIGENT INFORMATION & ENGINEERING SYSTEMS (KES 2019), 2019, 159 :650-657
[5]   Assessing Consistency of Five Global Land Cover Data Sets in China [J].
Bai, Yan ;
Feng, Min ;
Jiang, Hao ;
Wang, Juanle ;
Zhu, Yunqiang ;
Liu, Yingzhen .
REMOTE SENSING, 2014, 6 (09) :8739-8759
[6]  
BEJOY A, 2020, BIOCYBERN BIOMED ENG, V40, P1436
[7]  
BRAHIM A, 2020, BIOSYST ENG, V198, P31
[8]   Uncertainties in land use data [J].
Castilla, G. ;
Hay, G. J. .
HYDROLOGY AND EARTH SYSTEM SCIENCES, 2007, 11 (06) :1857-1868
[9]   Semantic Segmentation of Aerial Images With Shuffling Convolutional Neural Networks [J].
Chen, Kaiqiang ;
Fu, Kun ;
Yan, Menglong ;
Gao, Xin ;
Sun, Xian ;
Wei, Xin .
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, 2018, 15 (02) :173-177
[10]  
Chenxiao Z., 2020, INT J APPL EARTH OBS, V88