Satellite Image Classification using Deep Learning

Abstract :

In recent years, Deep Learning (DL) applied to Remote Sensing (RS) imagery has demonstrated remarkable capabilities in image classification by autonomously selecting optimal features for specific tasks. Selecting an appropriate DL architecture, a subset of Machine Learning (ML), involves multiple training layers. Among these, the Convolutional Neural Network (CNN) has been widely adopted to address complex problems such as image classification and object recognition through a sequence of feed-forward layers. CNNs process images directly, enabling effective extraction and representation of distinctive features. A typical CNN comprises convolutional, pooling, and fully connected layers. In this study, a CNN-based model was trained on a Kaggle dataset using a high-performance Graphics Processing Unit (GPU). The proposed architecture exhibited superior efficiency in classifying satellite images into defined categories. Experimental results demonstrate that the model achieved a classification accuracy of 94.59%, highlighting its potential as a robust tool for satellite image classification in RS applications.