Intelligent Underwater Object Detection and Image Restoration for Autonomous Underwater Vehicles

Unmanned Underwater Vehicles (UUVs) have been reliable and economical technological solutions to perform undersea monitoring tasks in comparison to manned vehicles. However, in many situations, UUV is unable to fulfill complex undersea research tasks since target objects appear distorted due to light absorption and scattering. Besides, ocean surveying undergoes severe power requirements compared to terrestrial systems because of battery-driven low-storage vehicles like Unmanned Underwater Vehicles (UUVs). Therefore, limited power supply, motion resistance of water medium, and distorted target object appearance can delay the mission and reduce the efficiency of UUV in their underwater operations. Considering the resource-constrained undersea monitoring setup, we propose an intelligent two-stage framework for expeditious monitoring of underwater scenes. First, an effective deep neural network is employed for underwater object/region of interest (ROI) detection. Then the detected ROI is restored using an efficient restoration method, thereby improving the visual quality of the degraded images and aiding the navigating and monitoring tasks of UUVs. Our method has been objectively and subjectively assessed using 9 evaluation metrics and our key results reveal mAP of 94.35% and an Underwater Color Image Quality Evaluation (UCIQE) score of 3.09, surpassing state-of-the-art methods for object detection. Furthermore, the execution time of 0.550 secs is required for object detection and dehazing, making this proposal suitable for UUVs to perform automatic undersea object detection and dehazing within operational running requirements.