Research Of Forest Fuel Type Classification Based On Remote Sensing

In this article we studied and discussed the classification of forest fuel using theRS(Remote Sensing) method. We classified forest fuel types mainly based on thedifference of living type and dominant species. Considering the vegetation condition ofTahe Forest Bureau, we compared the spectrum characteristic curve of main needleleaved species, including Larix gmelini, Pinus sylvestns, Picea asperata; and main broadleaved species, including Betula platyphylla, Populus davidiana, and Salix; also mixedforest, hard use land, swamp, water and residential area. These include forest fuel typesand 2 other types of water and residential area. Considering the existing classificationmethod, and also effects of forest age on burning characteristics, we determined the finalforest fuel type as cultivated Lanx gmelini forest, natural Lanx gmelini forest, Pinussylvestris forest, broad leaved forest, mixed forest, hard use land, and swamp. We alsodescribed each type.In this article, we chose the 2002 ETM+ remote sensing image as raw material, alsowe used relief map, forest region compartment map and so on to determine the exact fueltype, and classified the forest fuel. The methods we used include ISODATA method,Maximum likelihood classification of supervised classification, and also BP neutralnetwork method. Using the ISODATA method we classified the image into 9 differenttypes including 7forest fuel types and 2 other types; using the MLC and BP method wefurther divided natural Larix gmelini forest into middle-aged Lanx gmelini forest andmature Lanx gmelini forest.Using the ISODATA method we initially classified the image into 40 sub types, andcombined them into 9 classes. The total accuracy is 64.33ï¼…, and the accuracy of Pinussylvestris forest and broad leaved forest is very low. Using the MLC method, wecompared the OIF value of different band combination, and found that the 4,5,3 bandcombination is most suitable for taking the MLC classification. And further divided naturalLarix gmelini forest into middle-aged Lanx gmelini forest and mature Larix gmelini forest.Total accuracy is 79ï¼…. When combine the two types into natural Lanx gmelini forest, wegot total accuracy as 82.00ï¼…, increased by 16.67ï¼…. Based on these above, we used BPneutral network as another method to classify image of 4,5,3 band as RGB, whenclassified into 10 types, the total accuracy is 84.67ï¼…, classify as natural Larix gmeliniforest, the accuracy is 87.67ï¼…. comparing with the 4,5,3 MLC method, the total accuracyincreased by 5.67ï¼….We compared the producer accuracy, user accuracy and total accuracy of different fuel types. Results show that as the sequence of ISODATA, MLC, and BP neutral network,the three kinds of accuracy generally presented increasing trends, only one or two typeshave a fluctuated trend. Total accuracy compare show tat the BP neutral network methodhas the highest total accuracy.