Flexibility and high capabilities of classification methods have made it one of the main methods for extracting information from satellite images, which will be used for generating spatial information. Image classification is a decision-making process in which we are recognizing a class to assign the desired pixel with maximum confidence to it. Maximum likelihood method due to the simplicity of the algorithm, low computational cost, and high reliability is one of the most applicable classification methods. Due to the spectral overlap, it is possible that the probability calculated for several classes become close to each other, or the amount of the highest probability becomes very small in this method. In such cases, it is not possible to assign a class to a pixel with confidence. Thus, to have more accuracy, a threshold is used as a controller. Using threshold in classification leads to omitting some pixels at labeling and increases the accuracy and reliability in labeling other pixels.

          In this thesis, in order to classify all of the pixels in the image and increase the accuracy, two new methods in satellite images classification are suggested. The DTML supervised classification method which is pixel-based will be able to classify images through some stages. At first, each pixel is located at the highest level of a decision tree. DTML method classifies each pixel in the small number of classes with the formation of hypothesis based on pixel transfer from each level to lower one and the accuracy of this hypothesis. In this method, decision making for transferring each pixel to the lower level is based on the Maximum Likelihood algorithm. The iterative process continues until one of the two conditions is satisfied: A state having a single class is reached or no new state is reached after the iteration. In this case, the pixel is classified to a mixed class according to its level. The DTFL is a sub-pixel supervised classification method based on the tree method and the Fuzzy algorithm. At the beginning of the image classification with DTFL method, each pixel is located at the highest level of a decision tree. DTFL method classifies each pixel using an iterative process with the formation of a hypothesis based on pixel transfer from each level to lower one and the accuracy of this hypothesis. This method is based on making decisions using Fuzzy logic. The iterative process continues until one of the two conditions is satisfied: A state having a single class is reached or no new state is reached after the iteration. In this case, the pixel is classified to a mixed class of default classes based on its level.

          In order to the accuracy assessment, a simulated image and the GeoEye-1 satellite image of Azadi Stadium were classified using DTML and DTFL methods, and the results were compared with the Maximum Likelihood classifier. DTFL classification method comprised of eighth Fuzzy membership functions of triangular shape, trapezoidal shape, ð shape, bell shape, Gaussian, differential S shape, multiplicative S shape, and a Fuzzy membership function based on Bayes’ Law and twelve decision making law of minimum, maximum, Multiplicative, Collective, Collective second power, mathematical mean, geometric mean, harmonic mean, minimum/ maximum, mathematical mean* minimum/ maximum, geometric mean* minimum/ maximum, harmonic mean* minimum/ maximum. In the classification of the simulated image, DTML method reached the overall accuracy of 95.51% and the Kappa Coefficient of 94.05%, which DTFL method in its highest accuracy, using Gaussian membership function and geometric mean decision-making law reached the overall accuracy of 96.13% and the Kappa Coefficient of 94.84%. Clearly, these methods reached better results compared with the overall accuracy of 94.94% and the Kappa Coefficient of 93.25% resulted from the maximum likelihood method. Defining a threshold in the maximum likelihood method to achieve the same accuracy of the DTML and DTFL methods, respectively, 0.84% and 1.86% of the pixels are taking the non-classified label. In classifying of the image of the study area, DTML method reached the overall accuracy of 91.27% and the Kappa Coefficient of 91.16%, which the DTFL method in its best accuracy, using triangular membership function and harmonic mean decision-making law reached the overall accuracy of 96.14% and the Kappa Coefficient of 96.06%. These methods reached better accuracy compared with the overall accuracy of 89.91% and the Kappa Coefficient of 89.77% resulted from the maximum likelihood method. Defining a threshold in the maximum likelihood method to achieve the same accuracy of the DTML and DTFL methods, respectively, 2.76% and 8.73% of the pixels are taking the non-classified label.

          According to the classification results of the simulated image and satellite image, Satellite image classification using the tree method and the Fuzzy algorithm (proposed DTFL method) leads to higher overall accuracy and the Kappa Coefficient and the users trust to the results of this approach would be more.