Monitoring and Extracting Relevant Parameters to Detect Forest Fire Using Satellite Imagery

A forest is a complex ecosystem which is predominantly composed of trees, shrubs and is usually a closed canopy. Forests are warehouses of a large variety of life forms such as plants, mammals, birds, insects and reptiles etc. Forests have major importance which provide an environment for many plants and animals thus protects and sustains the diversity of nature and absorbs the suspended particles of air which helps in reducing pollution. But due to some human and natural effects fire may occurs in forests.

Forest fires have disastrous social, economic and environmental impacts. Forest fire fighting involves extensive human resources and it is a very dangerous activity, in which there are many casualties every year. Many different technologies and techniques have been applied in forest surveillance and forest-fire detection. In existing methods, detection of forest fires are done based on smoke velocity, where there is a possibility of misleading in detection because fog or snow may look like forest smoke and even usage of sensors may lead to high false alarm rate. Hence, satellite or digital images are said to be one of the efficient source in the detection of forest fire. Satellite monitoring of forest fire has been possible in recent decades. Earth is observed by a growing number of satellites every day but only less percentage of these observations are appropriate to detect the forest fire. In this paper, the satellite images of temperature are extracted using image clustering by which abnormal or high temperature region compare to surrounding region will be extracted. The abnormal temperature is analyzed based on temperature scale. Image extraction process will be done based on k-means clustering algorithm. The feature extracted cluster image is selected to analyze it further by applying wavelet transform. The Haar wavelet transform is chosen here as it is an efficient technique, where decomposition is applied to the image in rows and columns by transforming from data space to wavelet space in frequency domain. Further, haar wavelet transform can automatically invert a RGB image into gray scale image and denoise the image, and present it in one dimension.

In order to detect the forest fire we have to estimate certain range of values from the feature extracted image. We compute a mean values for original, gray scale and clustered images of temperature. Out of all these mean values, clustered image mean values will be considered to detect fire. After analyzing the mean value dataset an appropriate value is chosen using median of means which is estimated at 10.14. So, whenever clustered image mean value exceeds the chosen value of 10.14 may be considered as a fire in forest otherwise no fire. Based on the data received from the forest department, this proposed model yields an average accuracy of 89.50% in detecting the forest fire.