Classification of forest fire area from firms hotspot data using deep learning technique | |
| Author | Sapkota, Shishir |
| Call Number | AIT Thesis no.RS-20-09 |
| Subject(s) | Forests and forestry--Classification Forest fires--Data processing Deep learning (Machine learning) Remote-sensing images--Interactive multimedia |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Remote Sensing and Geographic Information System |
| Publisher | Asian Institute of Technology |
| Abstract | Forest fires are the most common disasters occurring in forests worldwide. It affects a large number of wildlife, plants, and the ecosystem as a whole. Every year the Northern part of Thailand is highly affected by forest fires. So, it is necessary for firefighting authority to find the exact location of the forest as soon as possible to take action promptly. The MODIS sensor provides daily hotspot data (MCD 14DL) which is a point data that represents active fire in the ground. But every MODIS hotspot does not represent the bum area. The burned area is given by MODIS monthly burned area data product (MCD64A 1), which is not much of use for prompt response system. Since the temporal gap between MODIS hotspot (MCD 14DL) and burned area data (MCD64AI) is one month, it is necessary to find the exact forest fire area within a few days of bum. In this study, we developed a deep learning classification model that classifies a hotspot in the forest as bum or unburn. The burn hotspots in the forest area are considered as forest fire area. Different parameters such as brightness temperature, CONFIDENCE, STC (Spatio Temporal Clusters), Month, Latitude Region, Geographical Region, and last but not the least the aerosol parameters (Aerosol Optical Depth) for day 1 and day 2 of hotspot occurrence is considered for model development. The aerosol parameters are changed due to the smoke coming out of forest fire. This deep learning model is useful for forest fire authorities because it identifies the actual forest fire area with 2 days of temporal gap. |
| 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) | Miyazaki, Hiroyuki;Virdis, Salvatore G.P.; |
| Scholarship Donor(s) | Asian Institute of Technology Fellowship; |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2020 |