@phdthesis{oai:kyutech.repo.nii.ac.jp:00007799,
 author = {Muhammad Hasif bin Azami},
 month = {2022-11-24},
 note = {1: Introduction||2: Research Background and Literature Reviews||3: Research Methodology||4: Results||5: Discussion||6: Conclusion and Recommendation, An increasing number of wildfire cases every year has caused fear around the world. Scientists and researchers agreed that this catastrophe occurred due to climate change. Dry and windy conditions had worsened the situation in the affected area. Properties and life losses have created serious concerns for the authority to find a solution for preparing and fighting the fire promptly. Since the late ‘70s, leveraging satellite technology has brought helpful insight to monitor, detect, and assess wildfire events. NOAA AVHRR is one of the oldest Earth Observation （EO） satellites with the main objective of detecting and mapping forest fires. The MODIS fire product regularly upgrades the sensor technology and launches the satellites into space. However, with the advancement of current technologies, a miniaturized satellite called CubeSat creates a novel mission design by reducing the satellite development time, increasing the launching batch in a constellation method, and enhancing the detection result wildfire. The prime limitations of CubeSat are the size, weight, and power （SWaP）, which lead to the optimization design of the payload and the communication subsystem. The big image data acquired by the CubeSat creates a bottleneck effect between the satellite and the ground station due to the low downlink data rate. Deep learning （DL） techniques are improving in the computer vision area. Image classification, detection, and segmentation are used in neural network architecture designed by artificial intelligence researchers. In this study, the convolution neural network （CNN） algorithm was chosen for the pre-processing onboard CubeSat for wildfire detection as well as for the graphical user interface （GUI） used on the ground post-processing. The first and crucial step was to develop a custom dataset for wildfire images by leveraging satellite imagery. Defining the specifications of the CubeSat payload to which the CNN was implemented could support selecting the accurate resolution and bands for acquiring the satellite images. The KITSUNE satellite is a 6-unit CubeSat platform implementing the CNN onboard for wildfire image classification. It serves as the secondary mission to support the main mission of a 5-m class EO. The on-ground testing revealed that the CNN could classify wildfire occurrences on the satellite system using the MiniVGGNet network with an overall accuracy of 98 % and an F1-score of 97% success rate in 137 seconds. Other models were also compared, such as ResNet and MiniGoogLeNet implemented on the GUI with 97% and 96% F1-score, respectively. Overall, this research showed the feasibility of CubeSat of executing CNN onboard in orbit, particularly for wildfire detection., 九州工業大学博士学位論文 学位記番号：工博甲第556号 学位授与年月日：令和4年9月26日, 令和4年度},
 school = {九州工業大学},
 title = {Pre- and post-processing of space imaging for wildfire detection using convolution neural network approach},
 year = {}
}