The Study Of High-temperature Targets Identification Method Improvement In Shortwave Infrared Remote Sensing

Identification and property retrieval for high temperature targets on the surfacesuch as forest fires, grassland fires, coal seam spontaneous combustions, heap coking,oil well torch, volcanic eruptions have great significance to environmental monitoring,disaster warning and so on. Pixel reflectance for high temperature targets (visualreflectance) in short-waves infrared bands (SWIR,1.3-3.0μm) are obviously higherthan surface features of normal temperature, which are lower in the photographyinfrared bands (0.76-1.3μm).So SWIR bands can be regarded as advantage bands ofidentification for high temperature targets. Now identification relative methods aboutSWIR are NDFI (normalized differential fire index), mahalanobis distance multipletruncation, mahalanobis distance multiple discriminant analysis, factor analysis and soon.As NDFI method suffers the interference of water and colorbonds, hightemperature targets can’t be identified effectively. In order to solve the problem, weconduct improvement to NDFI method on the basis of study about spectrum of typicalnormal temperature surface features and high temperature targets, so we can removethe interference of water. Then Fisher two types discrimination is adopted to classifycolorbonds and high temperature targets precisely. The main research achievementsare shown below:1. On the basis of study for typical normal temperature surface features and hightemperature targets, Landsat8OLI data spectrum database is constructed. Thedatabase comes from29scenes remote sensing image in different regions. Thedatabase includes normal temperature surface features (water, vegetable, cultivatedland, settlement place, road) and high temperature targets (forest fires, heap coking,oil well torch, volcanic eruptions, burning the grass on waste land, metal smelting factory).It includes14types and44species spectrum, and it provides abundanttraining samples. In the study, the reflectance values of7thband (SWIR band) for OLIhigh temperature targets are higher the normal temperature surface features whichexclude colorbonds, and they are lower than normal temperature surface features inthe5thband (photography infrared bands). In the result of identification for NDFI,high temperature targets, part of water and colorbonds have similar NDFI values, sohigh temperature targets can’t be identified effectively.2. Because the water has lower reflectivity in5thband (photography infraredbands), and the colorbonds have higher reflectivity in7thband (SWIR), lead to theresult of high-temperature targets identification mixed with corresponding largerNDFI value. Aiming at this problem, according to the different spectral characteristicof water，colorbonds and high-temperature target in visible bands, combined2thband(blue) and4thband (red) into the NDFI algorithm to improve the NDFI method, sothat it can remove all the water abnormal interference and part of colorbondsdisturbance.3. According to the confusing problem of colorbonds, using fisher two typesdiscrimination to distinguish objects. On the bases of OLI data bands selecting,combined with the high temperature targets and colorbonds multi bands informationto establish discrimination function for discrimination. The results show that thetheoretical precision is96.92%and the actual accuracy is94.61%using seven bands.Selecting higher contribution1thband (deep blue),2thband (blue),4thband (red) and5thband (photography infrared), the theoretical precision is97.69%, and the actualaccuracy is96.23%. In7thband (SWIR), the high temperature targets and colorbondshave higher reflectivity, so that the SWIR band has smaller contribution. Consideringthe effect of temporal object spectrum, the best discrimination variable can be selectedby combining multi temporal remote sensing data. In the results of discrimination in7bands, The difference of NDFI (improved) in two phase has the greatest contributionto two types discrimination result, it was50.27%, the theoretical precision is98.97%,and the actual accuracy is95.87%. Therefore, combined with the multi temporal remote sensing data is conductive to high temperature targets identification, NDFI(improved) difference can be used as the important basis for high temperature targetsidentification.