Underwater object detection plays a crucial role in identifying and analyzing objects beneath water surfaces. This project develops a comprehensive system integrating image dehazing using the Generic Model-Agnostic Convolutional Neural Network (GMAN) to remove haze-like degradation caused by light scattering and absorption, thereby restoring clarity, contrast, and color fidelity in underwater images. Subsequently, the YOLOv11 algorithm performs efficient detection of multiple underwater objects from dehazed images and live video input. The system features a user-friendly interface with modules including Home, Register, Login, Image Upload (for dehazing and detection), and Logout. The dataset for training and testing is the SUIM dataset from Roboflow, containing annotated underwater images. The project employs Python as the back-end language and Streamlit for building an interactive framework. By preprocessing images with GMAN dehazing, the system enhances detection accuracy under challenging underwater conditions, such as varying lighting and visibility levels. Results, including dehazed images and detected objects, are displayed in a structured manner. This approach supports underwater monitoring, exploration, and research activities by providing clearer visuals and accurate object identification. Overall, the project demonstrates the effective integration of deep learning algorithms for dehazing and object detection in underwater environments.

Keywords: Underwater object detection, image dehazing, GMAN (Generic Model-Agnostic Convolutional Neural Network), YOLOv11, SUIM dataset, light scattering and absorption, visibility enhancement, color restoration, Streamlit interface, live video detection, deep learning, marine exploration, Python backend.