Based On Forest Fires In Multi-scale Fractal Image Recognition Algorithm

The forest fire usually breaks out abruptly and randomly, which always brings huge loss in a very short time. In this occasion, the forecast of occurrence and the development of the forest fire is becoming an important problem to be solved urgently.In this thesis, the multifractal parameters are used to describe the smoke and the flame. First, extract eligible points by using difference among multifractal spectrum of the smoke, forest and cloud. Then, distinguish smoke by the characteristics of distribution of pixel. In this way, smork and flames will be segmented from complex background area. In the Matlab simulation, the multifractal spectrum f(a) is successf-ully used to identify the smoke which is caused by the forest fire from the whole image.In this thesis, Contourlet transform is used to image recognition in forest fires. A new segmentation algorithm for forest fire image based on multifractal and Contourlet transform is proposed. The algorithm firstly decomposes the image into several subbands with multi-scale, location and multi-direction by Contourlet transform. Then, multifractal analysis is performed for low frequency sub-band. The multifractal spectrum is used as the characteristic parameters to filter pixels which mark different scenery. At last, according to the distribution of these pixels, fire and smoke segmentation is realized. Experimental results show that the algorithm can segment flame and smoke in effect, and improve the efficiency of segmentation for forest fire image. This algorithm has good accuracy and high real time.