A Novel Integration Technique for Satellite Image Classification

In recent times, satellite remote sensing technology has revolutionized the analysis of climatic changes and image classification. However, the primary issue with satellite images arises due to variations in their features such as textures, lightning, and resolution, leading to inaccurate classification. This study proposes a combination of Support Vector Machines (SVM), Extended Local Binary Patterns (ELBP), and Convolutional Neural Networks (CNNs) for the effective classification of satellite images. In the pre-processing stage, the input image is analyzed by a median filter, which removes noise by replacing abnormal pixel values with the median values that are appropriate for the following stages. Then, CNN performs a comparison of the extracted features obtained from the ELBP method. Ultimately, the SVMs are utilized to classify the test images into their respective classes. The proposed ELBP-CNN-SVM model achieves an accuracy of 98.42% and outperforms the existing SDS-Network approach when tested on the Caltech dataset.