Identification and Delineation of Forest Fire Using Various Spatial Data Mining Techniques

Abstract

The ascent in the volume of spatial data owes its expansion to the technical newlineprogression in technologies that aid in spatial data acquisition, mass storage and newlinenetwork interconnection. Thus, the requisite for automated detection of spatial newlineknowledge from voluminous spatial data arises. In recent times, spatial data mining newlineplays a crucial role in the real world applications especially in the natural resources newlineassessment and management. Among the wide range of applications, mitigation of newlinenatural disaster has become imperative especially so in the case of forest fire. Due to newlinethe importance of fire in influencing the global environment, the demands of newlineautomatic early fire delimitation systems has received a significant amount of newlineattention and in this context, there have been numerous studies using conventional as newlinewell as satellite data. But however so, use of color space models implemented with newlinemining algorithms as carried out in the present study is novice and has helped to newlinedetect and delineate forest fire regions from an image. newlineIn the present study, forest fire regions from the images are extracted using various newlinespatial data mining techniques. To understand the intricacies of color in identification newlineof such fire pixels, RGB images showing forest fire are transformed into different newlinecolor space modes viz., YCbCr, CIE XYZ and CIELAB. After such transformation newlineanisotropic diffusion segmentation is carried out as part of the pre-processing phase, newlinewhich is later subjected to other two phases of the mining techniques - training phase newlineand testing phase. The study examines with its two crucial phases training and newlinetesting of forest fire initially using a single image and later multiple images using newlinesix images as a set of image. Spatial data mining techniques such as fuzzy, Radial newlineBasis Function Neural Network (RBFNN), Support Vector Machine (SVM) and newlineAdaptive Neuro Fuzzy Inference System (ANFIS) are applied on the images and newlinerespective efficiency in region growing of fire pixels are s