Evaluate the impact of Majority filter kernel size on increasing the accuracy of remotely sensed images

Classification of remotely sensed images is one of the most important operations in image processing, and various methods have been proposed to increase the accuracy of the data classification. The aim of this study is to evaluate the impact of the kernel size of the Majority filter on improving the accuracy of the final classification of images. For this reason, a window of Landsat 8, OLI image from Ardabil County was selected. After applying the required preprocessing, the multispectral and panchromatic bands of this sensor were combined with the FFT-IHS method. The image was classified using the maximum likelihood algorithm with 8 classes. The overall accuracy and Kappa coefficient of this classification were 88.3 and 0.87, respectively. The size of the 3 * 3, 5 * 5, 7 * 7, 9 * 9, 11 * 11 and 13 * 13 kernels were selected to perform the post-processing operations to increase the accuracy of classification. The results showed that by increasing the size of the kernel to 5 * 5 and 7 * 7, the overall accuracy and kappa coefficients increased to 91.6, 91.8 and 0.91 for both kernels, respectively. Applying a larger kernel size reduces the overall accuracy and Kappa coefficient as in it's down to 77.2 and 0.74 in the 13 * 13 kernel, respectively. Therefore, the size of the 5 * 5 and 7 * 7 kernels will be introduced as the optimal kernel depending on the nature of the images that used.