Application Research On Fire Detection Based On Hybrid Intelligent Algorithm

Forest fire is a kind of worldwide natural disasters, but domestic and foreignscholars have never ceased to the study of forest fire automatic detection technology.Forest fire detection based on video image technology is a kind of effective fire detectiontechnology which overcomes the shortcomings of the traditional fire detectingtechnology based on the sensor which is difficult to adapt to the complex environmentand has lower accuracy.It uses video image processing technology、 mathematical morphology method、pattern recognition and artificial intelligence technology in this paper. It puts forward aset of effective forest fire flame detection method after researching the colorcharacteristics of the forest fire flame shape texture characteristics and dynamiccharacteristics. There are five research achievements in this paper:In the first, it puts forward an improved image segmentation algorithm based oncolor models. The algorithm improves by introducing a new constraint condition andmathematical morphology image processing technology to the original algorithmaccording to the original RGB-HSI color model under the condition of the lighterbackground objects which is unable to accurately segment the prospect goal. Theexperiments show that the improved image segmentation algorithm can accuratelysegment the foreground target area and effectively eliminate the bright backgroundinterference, with very strong practicability.In the second, it puts forward the effective feature extraction algorithm respectivelyaccording to color features of forest fire flame、 shape features、 texture features anddynamic features. It identify for suspected flame area in video images by using variety offeatures.In the third, it puts forwards an improved hybrid particle swarm algorithm byimproving the inertia weight and introducing artificial immune algorithm、geneticalgorithm according to the standard PSO algorithm which falls into local optimum easilywith the lower precision. And then, it improved hybrid particle swarm algorithm tooptimize the initial value of BP neural network according to the original BP neuralnetwork algorithm which is sensitive to the initial weights and bias value, but cannot getaccurate initial conditions. It makes the BP neural network operations on a good initialvalue, which effectively enhances the stability of the algorithm to improve theconvergence speed and convergence precision of the algorithm. In the fourth, it extracts characteristic values for different fusion target area to makea decision on whether the area is the fire flame by using BP neural network based onhybrid particle swarm algorithm optimization from the prospects for video image. Theexperiments show that hybrid particle swarm algorithm to optimize the BP neuralnetwork is better than original BP neural network convergence rate stable performancewith higher precision and faster convergence.In this paper, the improved image segmentation algorithm can effectively segmentthe foreground objects in video image by modifying the parameters of constraints inorder to apply more use of the environment, so it has strong practicability. And it putsforward a kind of BP neural network based on hybrid particle swarm optimizationalgorithm which successfully applied to the recognition of forest fire flame. Theexperiments show that using hybrid particle swarm algorithm to optimize the BP neuralnetwork integrated extracted from video image has the higher accuracy which canidentify many of the features of fire flame, compared with the original BP algorithm. It isa kind of effective method of forest fire detection.