Forest Fire Discrimination Research Based On FY4 Remote Sensing

Satellite monitoring is always the key of forest fire monitoring system.For improving the forest fire discrimination accuracy and real-time monitoring ability,further reducing the damage rate of forest fire and increasing the forest fire extinguishing rate,a satellite with high time resolution and high accuracy of forest fire monitoring is urgently required.At present,the geostationary meteorological satellites have the highest time resolution in forest fire monitoring,and FY4 is the one with highest time-effective in China.Using the characteristics of high time resolution and sensitive to heat source information of mid infrared channel,FY4 remote sensing data can quickly identify medium and small size of fire points.And it also can detect the real-time development of forest fire,providing first-hand information for forest fire fighting.Here are the main research jobs of this thesis:(1)By analyzing the strengths and weakness of the existing forest fire satellite monitoring algorithm,doing research on comprehensive algorithm of forest fire satellite discrimination based on the independent information time and space,this thesis proposed the method of of time and space to distinguishing forest fires,which can improve the accuracy of extracting forest fire information.(2)Carry out characteristic spectral analysis of the 14 bands of FY4 remote sensing data,and then detect the most suitable band or combination of bands for each feature through the detection of the ground band,and finally use the optimal band selection method and support vector machine(SVM)Analyze and get the most suitable band and band combination for forest fire monitoring.(3)Doing research on the method of distinguishing forest fire points combining time and space by analyzing the changes of bands brightness temperature and pixel values of two phases of FY4 remote sensing data images before and after the fire.(4)Use the NDVI vegetation index,cloud,water detection,and fixed heat source information to exclude abnormal fire points and false fire points.Combined with the fire radiation biomass FRP burning fire point to estimate the fire point intensity,the fire area is estimated by the maximum inter-class variance,and finally real-time monitoring of forest fires is realized.Here are the main research results:(1)The most suitable band combination for forest fire identification is(B7,B8,B12).Its SVM ground objects classification accuracy was 99.21%,and the Kappa coefficient was 0.855.And it was the optimal band combination with the highest classification accuracy of forest fires identification.This band combination met the result of optimal band combination screening.(2)The temperature of the ground objects can only produce a change of 1K within 10-15 minutes normally.When the temperature rises over 5K within 10-15 minutes,and the difference value between the 7th and 12th bands is 20K(10K at night),it can be regarded as a fire.By contrary,the forest fire can be judged out when the difference value less than 10K(20K in the daytime).Hence the difference value of temperature between the 7th band and 12th band can be the criterion of forest fires.(2)Discriminating forest fires through FY4 remote sensing data can increase the efficiency of forest fire monitoring to 15 minutes per time.Using the algorithm of FY4 space-time combination to distinguish fire points can effectively solve the problems of medium and small fire points monitoring,false alarm of high temperature abnormal points and low temperature fire points monitoring,but the monitoring effects of tiny temperature fire points are ineffective due to the lack of spatial resolution.(3)In the single moment verification,when the time differs by 4 minutes,the local accuracy can be more than 75%.In the continuous fire monitoring,the development of the forest fire can be clearly reflected through the temperature variation diagram of pixel.Through the monitor of forest fire intensity and forest fire area and combing with high spatial resolution image,the trend of fire can be estimated,which provides theoretical basis for forest fire command and fighting.