Research On Forest Fire Image Segmentation Algorithm Based On Partial Shape

Three disasters which damage the forest are fire, plant disease, and pests. Among them the damage of forest fire is the most serious. Not only does it damage the valuable forest resources and lead to the deaths of a large number of plants and animals, but also causes the transformation of local microclimate and leads to the destruction in nature's balance. Around the world, many countries pay much attention to researching on monitoring and detecting the forest fire.Objects in nature have fractal characteristics. Nowadays, more and more scholars study image fire detection algorithm based on simply fractal theory. Not only does it inherit the advantages of non-contact monitoring from the image fire detection, but also uses the fractal characteristics of fire and smoke to segment fire image. It becomes the latest hot spot in the field of fire detection at home and abroad.The thesis uses the local fractal dimension to describe the smoke and flame of forest fire. Three segmentation algorithms of forest fire are proposed.The main research is as follows:Firstly, a smoke segmentation algorithm based on calculating the local fractal dimension of forest fire image by four pyramid model is proposed. This segmentation algorithm improves the traditional triangle prism model which is used to calculate the local fractal dimension of the image into four pyramid model. The four pyramid model is used to calculate local fractal dimension of forest fire image for segmenting the smoke in forest fire image.Secondly, a flame segmentation algorithm of forest fire image based YCbCr and local fractal dimension by four prism model is proposed. This algorithm improves the model which is used to calculate the local fractal dimension from four pyramid model to four prism model and converts RGB color image of forest fire into YCbCr space. The advantage of this algorithm is to ensure the segmentation accuracy and to reduce the amount of calculation. At the same time, it reduces the influence of background factors on the segmentation results.Thirdly, a flame segmentation algorithm of forest fire image based on RGB and local fractal dimension by four prism model is put forward. First of all, this algorithm extracts the three-channel components of forest fire image. Then, it uses the four prism model to calculate the local fractal dimension of each component image. Finally, it adds the results of component images to obtain the final segmentation results.At last, this thesis makes some simulation experiments of three segmentation algorithms. Then it compares and analyzes the advantages and disadvantages of them and sets up the foundation for the future research on new segmentation methods of forest fire image.