Fuzzy Slope Position Segmentation Based On Random Forest

Not only a wealth of resources are contained in the nature, but some series of complex phenomenon also taking place constantly. For these phenomena including climate change, bio-ecological, social and cultural, economic development and military installation, having a huge impact on people’s lives. Generally speaking, due to the diverse topography, there are significant differences in these resources and phenomena. Therefore, only through an accurate, reliable and systematical way to obtain these related data, can we get a good insight into the nature. So as to the scientific, reasonable, accurate and site-specific development planning for economic development, exploitation and utilization as well as other civil engineering can be draw up. Especially for these complex areas, modeling is untoward on account of the sophisticated surface condition, which to a large extent increases lots of difficulties to many related researches severely. Therefore, classification of terrain is of great practical significance.This paper focuses on how to fuzzy divide complex landform type more quickly an d efficiently. Firstly, the background, research status and problems of current slope posit ion segmentation as well as random forest algorithm were analysis explicitly. Meanwhil e, some relevant theories on them were also detailed presented as well. Conducted in-de pth research and analysis of the existing algorithm for fuzzy landform classification, we found that most of the fuzzy slope position segmentation algorithms had serious proble ms of so vast data need to be processed, the complex calculation process and high timecost, or may ignore the different influence that various terrain have on classification duri ng the model designed. Then in order to solve these problems, an overall solution was p roposed. Subsequently, a fuzzy slope position segmentation algorithm, with a window mechanism, based on random forest, was presented in this paper.Fuzzy slope segmentation algorithm based on random forest is mainly focused on t wo points. For the first, during the process of classifying, different terrain factor would bring landform different effects. For the second, slope category is a relative concept, wit h the distance between locations in space increasing, however the influence degree of la ndform they have with each other would inversely decline, even when the distance reac hed far enough the influence could be neglected. On accordance with the assumptions a bove, the typical slope point sets were extracted combined with expert knowledge, as tra ining and validation sets. Then the relationship between terrain factors and landforms w as analyzed by means of random forest model, and by calculating the importance that di fferent terrain factors have on kinds of landforms as the classification weighting factors.In addition, a number of varying windows was set, for each of the window, the distance between unclassified location and typical location was further calculated. For those typi cal examples which is in the window, firstly obtaining their corresponding fuzzy terrain elements, then taking advantage of the weighted factor principle to calculate their compr ehensive similarity. In that having taken the effects of different terrain factors on multi-c lass recognition into account, therefore not only can this algorithm improve the classific ation accuracy but also reduce the running time severely.Finally, an area of yongxinzhen in jiangjin, Chongqing was taken as the study area,experimenting with fuzzy slope position segmentation algorithm based on random fores t, fuzzy classification algorithms with minimal operator and fuzzy classification algorith ms based on mean value were conducted on it. And then, the ultimate consequences wer e compared in terms of kappa value and accuracy. The result turned out that both in kap pa value and accuracy, fuzzy slope segmentation algorithm based on random forest is su perior to others. What’s more, it can be known by comparing the results obtained using various windows, when the increase of window size reaches a certain level, the effect of the classification would tend to a stable state, however, the time-cost increasing all the t ime. Thus, the hypothesis which was proposed at the beginning of the paper was further validated. Lastly, the optimal classification scheme was obtained when taking the classif ication effects and time-cost into consideration, and was applied in a larger area to verif y its validity. In addition, the graph of the slope position partitioning result was drawn i n ArcGIS.