Deep learning-based Object Detection in Underwater Communications System

• Published in: 2023 Second International Conference on Electrical, Electronics, Information and Communication Technologies (ICEEICT) pp. 1 - 6
• Main Authors: Sangari, M. Siva, Thangaraj, K., Vanitha, U., Srikanth, N., Sathyamoorthy, J., Renu, K.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 05.04.2023
• Subjects: Convolution; Correlation; Correlation filter tracking algorithm; Filtering algorithms; Object detection; Region based Convolutional Neural Networks; Robot sensing systems; Target tracking; Underwater communication; Underwater communications
• DOI: 10.1109/ICEEICT56924.2023.10157072

Being at the nexus of robotics and ocean engineering, underwater robots have been a developing research area. They can be used for deep sea infrastructure inspections, oceanographic mapping, and environmental monitoring. Autonomous navigation skills are essential for doing these activities successfully, especially given the poor communication conditions in underwater locations. Autonomous navigation technologies, such as path planning and tracking, have been one of the fascinating but difficult issues in the field of study due to the extremely dynamic and three-dimensional settings. Due to their short detection ranges and poor visibility, cameras have not received much attention as an underwater sensor. However, using visual data from cameras is still a popular technique for underwater sensing, and it works particularly well for close-range detections. In this study, the enhancement of underwater vision is achieved by combining the max-RGB and shades of grey methods. Then, to solve the problem of poorly illuminated underwater images, a technique known as RCNN (Region-based Convolutional Neural Network) is proposed. This procedure tells the mapping relationship how to create the illumination map. Following image processing, an RCNN strategy for underwater detection and classification is recommended. Two improved strategies are then used to change the RCNN structure in accordance with the properties of underwater vision. In order to deal with the challenges of object tracking and detection in underwater communication, a correlation filter tracking algorithm (CFTA) method was created. The properties of the invariant moment and area were looked at after the object's region had been extracted using a threshold segment and morphological technique. The findings show that the suggested method is effective for underwater target tracking based on RCNN-CFTA in the aquatic environment. Simulated evaluation of these methods' performance demonstrates the potency of the suggested strategies.