Research On Forest Fire Detection Based On Infrared Images

In the past, smoke sensors and temperature sensors are the main method used to monitor forest fires, with single criteria and low recognition accuracy. In this paper, the infrared image processing and pattern recognition techniques are used to design an infrared image fires detecting system. The acquired environment information is used to make further decisions simultaneously. Forest fire detection system consists of infrared image preprocessing, image segmentation, feature extraction and pattern recognition. Among them, feature extraction and pattern recognition are the focus of the study.In image processing, the inter- frame difference method is used to determine whether there are potential fire masks, and median filtering method is used to reduce image noise. After pretreatment, the classical algorithm is used for infrared image segmentation so that getting the fire suspected area. Region segmentation algorithm suitable for grayscale infrared image is proposed, which obtained better segmentation by algorithm simulation. After obtained binary segmentation image, morphological methods are used for further processing.In feature extraction, feature geometry, texture features, histogram features and dynamic features of potential fire mask are extracted. At the same time, feature extraction algorithm is applied on the experimental sample. O ne-dimensional, two-dimensional and three-dimensional features have been observed and analyzed. In order to further improve the accuracy rate of fire detection, smoke and temperature monitoring system based on wireless sensor networks is used to collect environmental characteristics of transmission corridor as an auxiliary feature.In fire pattern recognition, SVM and Ada Boost are used to model the infrared image fire recognition. Then different normalization methods, kernel functions and cross-validation method are used to select the optimal parameters of SVM. Finally, the two algorithms are merged using a linear weighted method for getting higher fire recognition accuracy rate.Finally, the reliability of image processing algorithms, the effectiveness of the fire area features and the accuracy of pattern recognition algorithm are proved to be effective by simulation. Infrared image fire detection system and smoke recognition system based on wireless sensor networks are developed. System was installed near a transmission corridor of Miyi, and work well.