Wild Forest fire is one of the most unavoidable and vital natural threat for all creatures. Forest fire creates more dangerous hazard to environmental systems, infrastructure and ecology . Forest fire detection is one of the most crucial steps to conserve the entire ecosystem. This brings necessity to recognize forest fires as soon as possible to prevent the losses and to reduce the degradation. This paper proposed a novel forest fire detection algorithm which accurately detects the fire region from forest fire still images by using Discrete Wavelet Transform (DWT) and Nonsubsampled Contourlet Transform (NSCT)(DWT-NSCT-FFD). The proposed system consists of two phases, training and testing. Initially all the images in the database are converted into LAB color space and intensity normalization is applied. To classify the fire pixels Extreme Learning Machine (ELM) classifier is used. Several features of image like Static, dynamic textures, color, and arithmetic features are extracted then these features are cascaded and given as input to the ELM classifier and corresponding mask as target in the training phase. In the testing phase, the trained ELM will find the flame portion in an input image based on the trained data. Completed Robust Local Binary Pattern (CRLBP) is applied to extract static texture. Then GLCM (Grey Level Co-occurrence Matrix) and Gabor transform are implemented to obtain the feature vectors. Dynamic texture features are extracted by using Discrete Wavelet Transform and Nonsubsampled Contourlet Transform. The experimental results indicate the validity and efficiency of the proposed algorithmagainst existing methods. It is shown that the proposed method accurately detects and identifies fire when compared to the conventional techniques.
Volume 11 | 01-Special Issue
Pages: 772-785