Underwater Object Detection Using Transfer Learning and Deep Neural Network

Object detection underwater is one of the most important tasks in various applications: marine biology, environ- mental monitoring, and underwater exploration. In this paper, we discuss a novel approach to underwater object detection based on transfer learning and deep neural networks. We make use of pre- trained models in large datasets to resolve problems of reduced visibility, distorted colors, and varying lighting conditions. We perform a systematic examination of various transfer learning methods and their relation to the accuracy of detection. We then perform experiments with the proposed model on a dataset prepared under water images and show significant improvements in detection capability over the standard methods. Our findings are commensurate with the deep learning's potential of overcoming the constraints of underwater object detection, that finds a promising application in advanced applications of underwater robotics and conservation in the near future.