Triplet network with autoencoder for remote sensing image scene recognition and classification

Remote sensing image scene categorization is a crucial process that involves assigning semantic labels to images based on their content,which has wide-ranging applications in a number of fields. The inherent potential of deep neural networks to learn attractive features has substantially sped up progress in the field, resulting in considerable breakthroughs in remote sensing and scene classification powered by deep learning. Nevertheless, the existing literature lacks a comprehensive evaluation of research findings in deep learning for remote sensing image scene categorization, despite the notable advancements in this domain. While CNN has shown remarkable efficacy in extracting image features, ranging there are persistent challenges that hinder its optimal performance in the domain of categorization of images from data collected from remote areas. The intrinsic attributes linked to remote sensing image data, such as intraclass variability and inter-class similarities, result in a notable decrease in the classification accuracy of the model. Moreover, the discernible impact of noise and interference further exacerbates limitations in CNN performance within this specific context. In response to these challenges, this study introduces an innovative classification methodology that seamlessly integrates a triplet network and autoencoder. Furthermore, the paper systematically conducts a rigorous comparative assessment, evaluating the performance of diverse image classification models across various remote sensing datasets. This holistic approach aims to contribute to the evolving landscape of remote sensing image scene classification by addressing existing gaps and advancing our understanding of effective model architectures.