Research On Forest Fire Smoke Recognition Based On Video Images

Smoke is the most significant visible phenomenon in the early stages of forest fires, the visible light video monitoring technology for its moderate cost and real time features has become an important forest fire monitoring technology now, so video-based forest fire smoke recognition has practical significance.In this paper, a systems framework of forest fire smoke recognition methods that integrates dynamic testing and static testing is proposed. In this framework several methods such as moving target detection, image feature extraction, pattern classification and recognition are studied. A forest fire smoke identification system based on this study has been applied in Jiufeng National Forest Park.Major results of this study are as follows:(1)A forest fire moving target detection algorithm. According to the moving feature of smoke, it is based on the improved background estimation and combines color criterion. The background updating model is improved by adding the original background image and the optimal values of the main parameters are identified. This algorithm has excellent smoke capture and preliminary filtration capacity of disruptors such as flying birds, swaying branches or moving cars.(2)A forest fire smoke image feature extraction method based on the pulse coupled neural network (PCNN). The adjustment method of adaptive PCNN model is improved, and the recommended parameters’ values of PCNN model for forest fire smoke detection are proposed.(3)A forest fire smoke characteristics data expression method for the PCNN output characteristic data that is Johnson time signatures combined with entropy sequence. With a comparative study to GLCM method, PCNN method significantly improves the recognition rate of forest fire smoke and reduces the false positive rate.(4)A comparison of classification performance of BP neural network, Self-organizing neural networks, Probabilistic neural networks and Support vector machine for the forest fire smoke characteristics data extracted by PCNN. Determined the optimization parameters of the four classifier and analyzed the experiment results from the recognition rate, operating performance and the impact of normalized operation, the experimental results show that, the feature data extracted by PCNN has excellent validity and reliability And the results also show Support vector machine obtained the highest recognition rate, at the same time, it has lower false negative rate and false alarm rate, the classification performance is also good. Support vector machine is more suitable for the classification of the forest fire smoke feature data extracted by PCNN.(5)An automatic forest fire smoke recognition system integrated wireless networks, video monitoring and other techniques. This system implements the results of this study and now has been set in Jiufeng National Forest Park in Beijing. It successfully captured a true forest fire and through all the simulated forest fires tests.