Computer classification of four major components of surface fuel in Northeast China by image: the first step towards describing spatial heterogeneity of surface fuels by images

A hierarchical scheme for describing spatial heterogeneity of forest fuels by images is proposed here which involves efforts at least in three tiers: the first one is description of spatial heterogeneity of fuels, mainly surface fuels, by image at experimental scale, the second is that at managerial scale, and the third is to establish a linkage between the descriptions at the above two scales. The description of spatial heterogeneity of forest fuels at the experimental scale is the basis of the whole methodology. Its major task is describing spatial distribution and loads of different fuel compositions by images. The perquisite for describing the spatial variation of compositions by images is that the components are consistently separable by computer images. If so, the following questions should be answered preceding the realization of the scheme: how about the accuracy? How consistent is the method? What is the effect of sample size on classifierʹs consistency and what is the minimum sample size required by a consistent classifier? Currently, litter work has been done in the image description of spatial heterogeneity of forest fuels at the experimental scale. As the first step towards the goal, an experiment was conducted on automatic computer identification of four major components of a kind of surface fuel commonly seen in Northeast China by their images in order to explore the answers to the above questions. The results indicate that the four components can be separated with a mean correctness as high as 96.63% by their computer images through a consistent linear classier using nine carefully chosen image features. Sample size exerts little effect on classifierʹs consistency, thus 30–50 images for each component will be sufficient to build a good classifier for simplestructured images and the size recommended by the literature [Cai, Pattern recognition, Press of Northwest Electronics and Communication College, Xi’an, 1986] for more complicated images. These conclusions indicate the preliminary feasibility of the scheme proposed here for fuel heterogeneity description. Further tests of the above conclusions on different and more complicated fuel components are still needed. And it is also important to explore methods for segmenting fuel images by these components. When this has been accomplished, the goal of description of spatial heterogeneity of fuel compositions by images will be achieved.