|
|
|
Abstract
Assessment of Mixed Pixels Decomposition Accuracy
Ali Ghafouri
KN Toosi University of Technology,
Iran
Email: ghafouri.ali@gmail.com
Traditional image classification assigns each pixel to a single class by assuming all pixels within the image are pure. The Mixed Pixels Decomposition attempts to separate mixtures of reflectance defining mixed cells. Specifically, this technique attempts to discern and quantify the components contributing to each sampled cell. The method assumes that the reflectance of each cell is a linear combination of the components contributing, although a non-linear contribution model is possible (Roberts, 1998).
ASTER satellite data were used to study spectral features and classifications of land cover especially agricultural fields of northeastern Markazi province, Iran. Ground investigation data allowed the evaluation of quantitative precision and accuracy of spatial distribution.For evaluation of subpixel classification accuracy, few methods and measurements such as entropy and cross entropy have been proposed, which have some limitations; Cross entropy needs to a fuzzy ground truth dataset, the matter that is not available simply. For this purpose, a correctness parameter has been introduced (á) for subpixel accuracy assessment. This correctness parameter expresses the matching rate of the results of subpixel classification especially in fraction maps with the ground truth data. The proposed method for the accuracy assessment of the subpixel classifiers makes it possible to inspect the classes individually. Additionally similar to traditional methods, each class can be investigated individually in respect of the corresponding commission and omission errors.
|
|
|