Comparative Study of Deep Learning Models for UAV Visual Image Fire Detection : An Analysis of Pre-Trained Architectures in Keras

With the development of Unmanned aerial vehicles vision technology, UAV images play an important role in national ecological environment protection, mineral resources exploration, Marine environment monitoring, land use investigation, water resources development and other fields. The UAV image has the characteristics of high clarity, large scale, and can fly at ultra-low altitude and high visual distance, which helps people to complete some dangerous work. However, the current UAV image recognition technology can only identify larger targets, and the recognition ability of small items needs to be improved. In this paper, a variety of Keras-based UAV image recognition systems are built, analyzed and compared. AlirezaShamsoshoara fire images are used for training tests, and the structure of several deep learning models is revealed, which provides help for the development of deep-level UAV image recognition systems.