Forest Litter Smoldering Fire Detection Based On Infrared Thermal Imager

Forest litter mainly includes the branches and leaves,the trunk of restrains of the forest undergrowth and herbaceous plants,some moss and lichen etc organic material.When the weather is extremely dry,or the litter's moisture content is too low,or there is thunder fire or external fire,it may lead to the humus smoldering.After the fire was put out,the litter is prone to burning,it leads to the spread of the smoldering fire.Because there is no flame of the smoldering fire and it is in the early stages of the fire,it is easy to be ignored.In this paper,we selected mongolica forest litter as sample,and conducted mongolica litter smoldering detection experiments by using infrared thermal imager,and offered data based of mongolica forest smoldering fire detection.(1)In the aspect of theoretical analysis,combined with the basic components of the infrared thermal imager and the physical principle of thermal imaging,this paper expounded the basic features of infrared ray and the basic law of infrared radiation,introduced some basic methods of the infrared image processing in detail,including median filtering method,the wavelet transform denoising method,the segmentation method based on edge detection and threshold segmentation method.These theories laid solid theoretical basis of the paper main research work.(2)In the aspect of experimental study,under the laboratory conditions,make the use of the two functions which are temperature test and imaging of infrared thermal imager to conduct the mongolica litter smoldering fire detection experiments.At the same time,we regarded the temperature,the shape parameter change rate and area change rate as the criterions to the smoldering fire.(1)Heated the litter and used the thermal infrared imager to measure the litter temperature in different forest wind speed,different detection distance and different angle.At the same time,we established a multiple quadratic nonlinear fitting equation about the relative error of infrared thermal imaging temperature measurement and detection distance and angle under the conditions of different wind speed.The equations offered data to the mongolica forest smoldering fire detection.(2)Adopted the wavelet threshold denoising method to remove most of the clutter and noise.Secondly adopted the morphological open operation to smooth the images' boundary,completed infrared images' preprocessing.Finally used the maximum between-cluster variance method to segment the infrared images and extracted the shape parameter change rate and area change rate of the target.(3)Trained the BP neural network and the input vectors were temperature,the shape parameter change rate and area change rate,the output was the probability for smoldering fire.Training results showed that the error transmitted and iterated reversely 12 times,then it was convergent.Secondly,testing the BP neural network which was trained by 15 groups of samples,the results showed that the smoldering fire recognition rate was 96.6%.