Improving the accuracy of image-based forest fire recognition and spatial positioning

Forest fires are frequent natural disasters. It is necessary to explore advanced means to monitor, recognize and locate forest fires so as to establish a scientific system for the early detection, real-time positioning and quick fighting of forest fires. This paper mainly expounds methods and algorithms for improving accuracy and removing uncertainty in image-based forest fire recognition and spatial positioning. Firstly, we discuss a method of forest fire recognition in visible-light imagery. There are four aspects to improve accuracy and remove uncertainty in fire recognition: (1) eliminating factors of interference such as road and sky with high brightness, red leaves, other colored objects and objects that are lit up at night, (2) excluding imaging for specific periods and azimuth angles for which interference phenomena repeatedly occur, (3) improving the thresholding method for determining the flame border in image processing by adjusting the threshold to the season, weather and region, and (4) integrating the visible-light image method with infrared image technology. Secondly, we examine infrared-image-based methods and approaches of improving the accuracy of forest fire recognition by combining the spectrum threshold with an object feature value such as the normalized difference vegetation index and excluding factors of disturbance such as interference signals, extreme weather and high-temperature animals. Thirdly, a method of visible analysis to enhance the accuracy of forest fire positioning is examined and realized; the method includes decreasing the visual angle, selecting central points, selecting the largest spots, and judging the selection of fire spots according to the central distance. Case studies are examined and the results are found to be satisfactory