PhD Oral Exam - Chi Yuan, Mechanical Engineering
When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.
Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.
Due to their rapid response capability and maneuverability, extended operational range, and improved personnel safety, unmanned aerial vehicles (UAVs) with vision-based systems have great potentials for forest fire surveillance and detection. Over the last decade, it has shown an increasingly strong demand for UAV-based forest fire detection systems, as they can avoid many drawbacks of other forest fire detection systems based on satellites, manned aerial vehicles, and ground equipments. Despite this, the existing UAV-based forest fire detection systems still possess numerous practical issues for their use in operational conditions. In particular, the successful forest fire detection remains difficult, given highly complicated and non-structured environments of forest, smoke blocking the fire, motion of cameras mounted on UAVs, and analogues of flame characteristics. These adverse effects can seriously cause either false alarms or alarm failures.
In order to successfully execute missions and meet their corresponding performance criteria and overcome these ever-increasing challenges, investigations on how to reduce false alarm rates, increase the probability of successful detection, and enhance adaptive capabilities to various circumstances are strongly demanded to improve the reliability and accuracy of forest fire detection system. According to the above-mentioned requirements, this thesis concentrates on the development of reliable and accurate forest fire detection algorithms which are applicable to UAVs. These algorithms provide a number of contributions, which include: (1) a two-layered forest fire detection method is designed considering both color and motion features of fire; it is expected to greatly improve the forest fire detection performance, while significantly reduce the motion of background caused by the movement of UAV; (2) a forest fire detection scheme is devised combining both visual and infrared images for increasing the accuracy and reliability of forest fire alarms; and (3) a learning-based fire detection approach is developed for distinguishing smoke (which is widely considered as an early signal of fire) from other analogues and achieving early stage fire detection.