Fire Detection Based On Flame Feature

With the development of economy and the adjustment of national policy, more and more attention has been paid to the industry for ship construction. At the same time, fire disasters that may occur in ships are also receiving more and more attention from researchers. Since ship fires often generate great harm and cause high casualty rates, the prevention of ship fires has become a topic of high importance. With the rapid development of digital image processing and pattern recognition technologies, image based fire detector technologies have demonstrated their unique and obvious advantages in the early detection of potential fire disasters and thus have become the focus of research in recent years.In this thesis, the characteristics and the causes of the fire are described in detail. A survey of the current status of the research in intelligent fire alarm systems in ships at both home and abroad is also presented. In addition, an intelligent fire alarming system that utilizes image processing techniques is developed based on the existing work. The system contains the following three parts.In the first part, in order to improve the quality of the images captured by the monitor and improve the accuracy of fire detection system, each image is processed by a series of preprocessing steps. Based on the preprocessing, the advantages and disadvantages of a few different color spaces are analyzed. Based on the analysis, the RGB-HSI color model is used to segment an image and obtain the fire and background areas in the image.In the second part, based on the target areas obtained with the RGB-HSI color model,the segmentation results of each image are described and analyzed quantitatively. A variety of quantitative descriptions of a flame area are developed. Specifically, we describe the static characteristics of a flame area from three aspects, including the color, shape and texture features. The third order color moment for color features of a flame area, the circular degree that describes the shape characteristics of the area, and the energy, entropy,correlation and inertia moments related to the texture features of the area are extracted and used as the input of the classifier in the third part.In the third part, the features obtained in the second part form an eight-dimensional input space, and a fire identification system can be constructed based on a support vector machine. Part of the data is used as the training sample to train the support vector machine,and the other part is used as the prediction sample to test the recognition accuracy of thesystem. The simulation results show that the support vector machine is able to achieve excellent recognition rates on both fire flames and other interference sources after being trained with the training data. It can thus be used in the development of intelligent fire alarm systems in ships.