In season resilient land cover and crop feature extraction and classification | |
| Author | Zaw Thu Htet |
| Call Number | AIT Thesis no.RS-25-01 |
| Subject(s) | Crops--Classification Land cover Neural networks (Computer science) |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Remote Sensing and Geographic Information Systems |
| Publisher | Asian Institute of Technology |
| Abstract | Climate change has significantly exacerbated flooding events, disrupting agricultural practices and food security, particularly in Thailand's Thachin River basin. These challenges necessitate advanced monitoring and analysis of land use and crop patterns to support adaptive decision-making. This study explores the potential of Convolutional Neural Network (CNN)-based models, including 1D, 2D, and 3D CNNs, for land use and land cover classification across three distinct seasons (cool, hot, and wet). Seasonal classification is hypothesized to provide robust insights into dynamic agricultural practices affected by climate-induced disruptions. Data inputs were preprocessed and aligned with the respective CNN architectures to evaluate model performance. Incremental analyses of input bands and timesteps were studied and revealed that 1DCNN is most effective for the wet season, while 2DCNN offers versatility across seasons when spatial features are well-represented. Notably, 3DCNN demonstrated consistent and robust performance across all seasons, making it the preferred approach for comprehensive classification and mapping tasks in the Thachin River basin. Additionally, 3DCNN-based models showed promising potential as feature extractors for transfer learning, achieving comparable accuracy when integrated with traditional machine learning techniques. Despite their promise, CNN-based methodologies face challenges, including high computational demands, domain-specific expertise requirements, and issues with data reliability. Misalignments between recorded planting dates and Thailand’s phenology calendar underscore the need for precise and consistent data collection to enhance model accuracy. Moreover, limitations in the Day of Year (DoY) calculation modules within modified 3D CNN architectures—particularly with shortened timesteps—raise concerns about overfitting. This study highlights the need for methodological advancements and improved data reliability to develop resilient, adaptive solutions for climate-impacted agricultural systems. |
| Year | 2025 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Remote Sensing and Geographic Information Systems (RS) |
| Chairperson(s) | Sarawut Ninsawat; |
| Examination Committee(s) | Chaklam Silpasuwanchai;Mozumder, Chitrini; |
| Scholarship Donor(s) | Greater Mekong Scholarship (RTG); |
| Degree | Thesis (M.Sc.) - Asian Institute of Technology, 2025 |