توضیحات
ABSTRACT
Instead of assigning every map pixel to a single class, fuzzy classification includes information on the class assigned to each pixel but also the certainty of this class and the alternative possible classes based on fuzzy set theory. The advantages of fuzzy classification for vegetation mapping are well recognized, but the accuracy and uncertainty of fuzzy maps cannot be directly quantified with indices developed for hard- boundary categorizations. The rich information in such a map is impossible to convey with a single map product or accuracy figure. Here we introduce a suite of evaluation indices and visualization products for fuzzy maps generated with ensemble classifiers. We also propose a way of evaluating classwise prediction certainty with “dominance profiles” visualizing the number of pixels in bins according to the probability of the dominant class, also showing the probability of all the other classes. Together, these data products allow a quantitative understanding of the rich information in a fuzzy raster map both for individual classes and in terms of variability in space, and also establish the connection between spatially explicit class certainty and traditional accuracy metrics. These map products are directly comparable to widely used hard boundary evaluation procedures, support active learning-based iterative classification and can be applied for operational use.
INTRODUCTION
The prevailing approach in remote sensing is that each output map pixel belongs to one and only one class, but the shortcomings of such classification have been identified early on (Foody, 1992; Mairota et al., 2015). Crisp classificationincludes a strong reduction of the information in the sensor data by binarizing gradients and omitting information on alternatives to the selected class. This often compromises the applicability of remote sensing derived vegetation maps (Townsend, 2000), but alternative approaches remain rare. Fuzzy mapping (also known as soft classification) assigns a probability of membership for each class to each pixel. It includes information on the sub-dominant classes, can handle smooth transitions and uncertain identification, and is therefore particularly well suited for vegetation mapping. Ensemble classifiers such as random forests or neural networks are becoming increasingly popular, but although these inherently output fuzzy data, they are mostly still used for creating hard-boundary maps. Some reasons for this may be that many users ask for clear and unambiguous results even if this is not justified by the objects they are mapping. Quantitatively conveying the information in a fuzzy map is considered difficult towards non-specialists. Also, compatibility with standard data formats and especially vectorization remains problematic since several alternative approaches exist. But most important of all, fuzzy maps are regularly criticized because their accuracy is not straightforward to quantify. Many possible metrics of fuzzy classification accuracy exist, but most are difficult to compare with crisp classification maps and no standards have been accepted.
Year: 2016
Publisher : WG II/4
By : A. Zlinszky , A.Kania
File Information: English Language/ 8 Page / size: 1,592 KB
Download: click
سال : 2016
ناشر : WG II/4
کاری از : A. Zlinszky , A.Kania
اطلاعات فایل : زبان انگلیسی / 8 صفحه / حجم : KB 1,592
لینک دانلود : روی همین لینک کلیک کنید
نقد و بررسیها
هنوز بررسیای ثبت نشده است.