Underwater image processing is crucial for various applications, but the quality of underwater images is often degraded due to factors such as light attenuation and dispersion in the aquatic environment. This study proposes a two-stage technique to address these issues, involving image enhancement using LightDehazeNet and classification using Zero-Shot Learning (ZSL). The enhancement stage employs LightDehazeNet to remove haze, adjust contrast, and sharpen the images using methods like Contrast Limited Adaptive Histogram Equalization (CLAHE), gamma correction and sharpening. These enhancements address the challenges of light distortion and loss of detail, enabling better visualization and analysis of underwater scenes. In the classification stage, a pre-trained CLIP model is used with ZSL to classify images through text-image correspondence, eliminating the need for task-specific training data. This approach overcomes the limitation of scarce annotated underwater datasets and offers high adaptability across different classification tasks. The proposed work’s performance is evaluated using metrics such as PSNR and SSIM for image enhancement and accuracy for classification. The results demonstrate the work’s effectiveness in enhancing underwater images and classifying them accurately. The proposed work achieves a PSNR of 27.39 ± 3.37 dB and an SSIM of 0.81 ± 0.08, comparable to state-of-the-art methods. The classification accuracy of 86% using ZSL highlights the ability to distinguish between various image classes without extensive labeled data. This research benefits the advancement of underwater imaging applications in fields such as marine biology, navigation, underwater exploration, and environmental monitoring.