SATELLITE IMAGE ENHANCEMENT USING GANS: A DEEP LEARNINGBASED SUPER-RESOLUTION FRAMEWORK

The growing demand for accurate geospatial analysis in sectors such as climate monitoring, precision agriculture, urbaninfrastructure planning, and disaster response has intensified the need for high-resolution satellite imagery. However, practicallimitations such as sensor hardware constraints, atmospheric interference, and limited bandwidth during data transmission oftenresult in low-resolution or visually degraded images. These artifacts can hinder decision-making processes and reduce theeffectiveness of downstream tasks like object detection, segmentation, and classification.In response to these challenges, this study proposes a deep learning-based image enhancement framework using a ConvolutionalNeural Network (CNN)-driven Generative Adversarial Network (GAN). The system is composed of a generator that learns totransform blurry, low-resolution satellite images into sharper, more detailed representations, and a discriminator that encouragesthe generator to produce outputs indistinguishable from real high-resolution imagery. Unlike conventional super-resolutiontechniques based on interpolation or overly complex modelslike ESRGAN, the proposed architecture leverages a streamlined CNNdesign optimized for computational efficiency and practical usability.The training process utilizes paired datasets of low- and high-resolution satellite images, with preprocessing steps that normalizeimage input and unify spatial dimensions. The model was trained using binary crossentropy loss and evaluated based onperceptual quality and visual accuracy. A dedicated inference pipeline allows real-time enhancement of user-provided images viacommand-line input, saving the improved result to an output directory for easy access. Experimental results confirm the model’scapacity to recover fine details, sharpen structures, and improve image interpretability, even in noisy or blurred input conditions.This research highlights the potential of lightweight CNN-GAN systems as accessible, deployable solutions for enhancingsatellite imagery in low-resource environments, field operations, or educational platforms. The approach bridges the gap betweenacademic advancement and practical applicability in Earth observation domains.