Dehazing of Remote Sensing Images using Deep Learning

Image Dehazing in remote sensing is essential to enhance clarity and improve precision in feature and data analysis, which is vital for tasks such as land monitoring, environmental research, and object detection. This study introduces an innovative deep learning methodology for dehazing images, employing a hybrid architecture that merges convolutional neural networks (CNNs) with Transformer models. The suggested model incorporates a pre-trained ResNet50V2 to extract features alongside strategically positioned Transformer blocks within the network, aimed at capturing global context and long-range relationships in images. By utilizing multi-head self-attention mechanisms, the Transformer blocks significantly boost the model’s capability to recover intricate details in hazy images, surpassing traditional CNN-based dehazing methods. Our dehazing approach utilizes ResNet50V2 to develop robust hierarchical representations from hazy images, which are then refined by Transformer blocks. The implementation of Conv2D transpose layers guarantees precise upsampling and reconstruction of high-quality clear images. This integration of CNN and Transformer layers enables the model to better maintain structural information and restore clarity while preventing overfitting through multi-scale feature fusion.The model is assessed using a dataset comprising pairs of hazy and clear images, with performance gauged by metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM). The results highlight exceptional dehazing performance, reflected in high PSNR and SSIM values, demonstrating notable enhancements in image quality. This study emphasizes the efficacy of Transformer models in tasks related to image restoration and presents a novel architecture that could be utilized in various image enhancement applications.