Unsupervised Classification of Remote Sensing Images via Generative Adversarial Networks and Transfer Learning

Abstract

With the continuous advancement of remote sensing technology, the application of remote sensing images in fields such as environmental monitoring and urban planning has been significantly expanded. Accurate classification of remote sensing images is essential for effective image analysis and interpretation. However, traditional supervised classification methods rely heavily on large volumes of labeled data, which are often costly and difficult to obtain in practical scenarios. To address this challenge, unsupervised remote sensing image classification has attracted increasing research interest. Recently, the introduction of Generative Adversarial Networks (GANs) and transfer learning has provided new strategies and technical pathways for unsupervised classification tasks. GANs enhance feature representation by generating images that closely resemble the original data, while transfer learning enables existing knowledge to be leveraged for improved classification performance in target tasks. Although notable progress has been achieved, existing unsupervised classification methods still face considerable challenges. Traditional unsupervised learning approaches often exhibit low classification accuracy under complex environmental conditions, particularly in feature extraction and noise resistance. While deep learning-based methods have improved classification performance to some extent, their effectiveness remains limited by factors such as training data volume and network architecture design. Therefore, enhancing the classification accuracy and robustness of remote sensing images by combining the strengths of GANs and transfer learning remains a critical research problem. In this study, an unsupervised remote sensing image classification method based on GANs and transfer learning was proposed. Initially, remote sensing images were augmented using GANs to generate richer feature representations, thereby improving the effectiveness of subsequent classification. Subsequently, an unsupervised classification method that incorporates transfer learning was introduced, enabling the utilization of existing model knowledge to further enhance classification accuracy. Experimental results demonstrate that the proposed method achieved superior classification accuracy and robustness in remote sensing image classification tasks, offering a promising new direction for the development of unsupervised remote sensing image classification techniques.