Hybrid deep learning based vanished object detection and tracking in underwater image processing

Underwater images often suffer from poor clarity and color distortion due to light absorption and dispersion. This paper proposes a Hybrid Deep Learning-based Vanished Object Detection method to enhance underwater image quality and improve object detection accuracy. In the preprocessing stage, Shade-of-Grey and Max-RGB techniques are combined to correct lighting distortions. A hybrid deep learning model, integrating CNN and RNN, is used for image enhancement and object classification based on underwater visual traits. To further improve object tracking and recognition, a Kalman Filter (KF) is incorporated into the deep learning structure. The method is implemented using Python and evaluated on both synthetic and real underwater datasets. Performance is measured using precision, recall, accuracy, F1-score, runtime, and RMSE. Results show a significant improvement in color accuracy and visual quality, with the proposed model achieving up to 99.1% accuracy at a learning rate of 80, outperforming existing techniques.