AUTO-IDENTIFYING FOREST FIRE-POINTS IN METEOROLOGICAL SATELLITE IMAGES BASED ON NEURAL NETWORK

Identification of forest fire-points in meteorological satellite images is the basis of monitoring forest fire using meteorological satellite data. The traditional approach of identification is visual interpretation. In the visual interpretation, better precision is obtained in practice, but experienced interpreter is needed and it is difficult to settle for auto-identification with computer, so that there is always unfavorable in the description of information distribution and the improvement of efficiency. In order to find out the forest fire in time and eradicate it quickly, it is very significant to make the fire-points auto-identified without reducing the precision. The artificial neural network theory (short for NN) developing rapidly in recent years provides a new means for solving this problem. NN has the basal characters of human brains, such as learning, recollecting and generalizing. The peculiarity of NN is massive parallel computing, distributive memory of information, nonlinear dynamics of consecutive time, global behaviour, great fault-tolerance and robust, self-organization, self-learning and real time processing. BP (Back Propagation Learning Algorithm) model is a NN used widely. The method using BP model to automatically identify fire-points in meteorological satellite images are discussed and the test in the range of Hubei Province is presented. Besides the NOAA/AVHRR data from 1994 to 1999, the earth's surface distribution is used. 43 meteorological satellite images with fire-points are selected, and 31 of them are regarded as trained collection, the others as tested collection. The result of the test shows that a fire-point is correlative with 7 character factors, which are the radiation-values of the 5 channels of NOAA/AVHRR, the earth's surface distribution and the difference of the contiguous temperature. The disciplined NN has recollected the characters of fire-points and non-points and has ability to identify points in images. Comparing NN with visual interpretation, the conclusion is drawn that NN can auto-identify fire-points in meteorological satellite images with the almost same precision.