Enhancing Remote Sensing Image Classification: A Strategic Integration of Deep Learning Technique and Transfer Learning Approach

Image classification within the realm of remote sensing technology plays a vital role across various sectors, including military operations, geographical mapping, surveillance activities, and environmental monitoring. Procurement of sharp and accurate satellite imagery is presently uncomplicated. This investigation is geared toward implementing image classification through a sophisticated deep learning algorithm and assessing its prognostic capabilities. This study employs a comprehensive dataset comprising 5,631 satellite images, encompassing representations of cloud-covered regions, desert landscapes, verdant territories, and aquatic zones. The evolution and widespread adoption of deep neural networks, along with their methodologies, have experienced substantial growth. This research harnesses one of the most pivotal deep learning frameworks, Convolutional Neural Networks (CNNs). The efficacy of the CNN model deployed here is noteworthy, specifically employing the VGG-16 CNN architecture. The model exhibits significant performance improvement and achieves excellent validation accuracy. The evaluation findings demonstrate the model’s strong ability to categorize satellite imagery accurately, which strengthens its generalization capacity.

I. Introduction

Remote sensing is an interdisciplinary field that enhances our understanding of the Earth’s surface by capturing information without direct physical contact. Remote sensing is useful in many areas, such as agriculture, environmental surveillance, planning cities, and disaster management, when combined with Geographic Information Systems (GIS) and analytics. In satellite image analysis or classification, the major tasks include image classification, image understanding, and object detection. Land surface classification, which provides information on forests, urban areas, agricultural fields, and water areas, is one of the main applications of satellite image classification. From the traditional approach of manual interpretation of aerial photographs, the field underwent a transformative shift with the onset of satellite technology. The introduction of multispectral and hyper spectral sensors facilitated the capture of information beyond the visible spectrum, greatly enhancing our ability to characterize the Earth’s surface.