ICGAN-CRSI: Integrating Conditional Generative Adversarial Networks for Efficient Classification of Remote Sensing Satellite Images

Remote sensing imagery is extensively utilized across various industries, including monitoring the environment, planning for cities, agriculture, and managing disasters. Image classification is crucial in advancing these applications, yet the existing literature on large-scale, application-dependent datasets and models remains limited. Satellite image classification faces challenges such as data quality, volume, and distribution, which complicate analysis. Additional complexities arise from handling variations in scale and resolution. Moreover, the rapid growth in remote sensing datasets, driven by increased satellite missions and sensor advancements, necessitates new methodologies. This study introduces high-capacity convolutional neural network (CNN) architectures with large-scale datasets, focusing on the AID (Aerial Image Dataset) for satellite image classification. We propose several CNN architectures with scaling methods to utilize spatial information for categorizing images into thirty distinct classes. To enhance classification performance, a novel model, Integrating Conditional Generative Adversarial Networks for Efficient Classification of Remote Sensing Satellite Images (ICGAN-CRSI), is presented, integrates CGAN Conditional Generative Adversarial Networks, enabling the creation of diverse realistic samples. This powerful integration with CGANs enriches the training set, helps to reduce overfitting, and allows the model to capture a wider array of patterns. The image enhancement phase leverages BM3D, an image denoising technique, to reduce noise and boost spatial resolution. Various architectures were tested, including MobileNetV2, VGG19, ResNet50, and Xception. The addition of dense and dropout layers preserves gradient details during classification, leading to improved accuracy. Findings show that the Xception model delivers the highest performance, attaining an accuracy rate of 95.3%.