Sugarcane farm classification using object-based image analysis and machine learning approach | |
| Author | Warot Watahong |
| Call Number | AIT Thesis no.RS-20-08 |
| Subject(s) | Sugarcane Remote-sensing images Image analysis--Data processing Machine learning--Classification |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Remote Sensing and Geographic Information System |
| Publisher | Asian Institute of Technology |
| Abstract | This research aims to study and present about an alternative technique to generate the sugarcane farm plot from the satellite image with high resolution by using OBIA approach. Additionally, this research also performed sugarcane plantation period classification with Sentinel-2 data with multi-temporal based on machine learning technique integrated with selected vegetation indices. Since multi-temporal data can be provided continuous time series data, so it can use for classifying the period with the phenology data. This alternative classification technique that applied in this study may be provide a satisfying sugarcane farm plot and sugarcane classification map since OBIA and machine learning are powerful technique for classification. Furthermore, satellite image which has high resolution integrated with OBIA method will provide a good potential for delineating the sugarcane farm. The result of research can provide a useful information that can use for inspecting the sugarcane farm. This research, to make a sugarcane farm plot by using single date of Gaofen-2 image based on OBIA. It provided 64% of correction percentage when compared with reference farm data. For plantation period classification, this research using multi-temporal data of Sentinel-2 satellite image (October 2017 to September 2018) to classify the period for the segment result that we did not know the exact plant date. We applied machine learning model: RF classifier and SVM classifier to the multi-temporal of SA VI and NDVI dataset. We found that RF classifier model that used NDVI data with plant sugarcane farm provided 62% of overall accuracy (actual period) and 66% for ±1 period when compared with single date data, the highest OA from single date data is only 27% which is lower than using multi-temporal data. Furthermore, for the model evaluation using recall value and Fl score. RF classifier provided a performance better than SVM classifier. The highest is RF classifier model that used NDVI data with plant sugarcane farm, it provided 0.62 for recall value and 0.73 for Fl score. Conclusively, the government or sugarcane company can use and apply the result of this research to develop and manage their sugarcane farm area data and also plan their strategies to manage their profits from sugarcane farm. It also applies to large scale area with different crop type in Thailand. |
| Year | 2020 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Remote Sensing (RS) |
| Chairperson(s) | Sarawut Ninsawat; |
| Examination Committee(s) | Tripathi, Nitin Kumar;Mozumder, Chitrini; |
| Scholarship Donor(s) | Royal Thai Government Fellowship; |
| Degree | Thesis (M.Sc.) - Asian Institute of Technology, 2020 |