Visual monitoring of High-Sea fishing activities using deep learning-based image processing

Abstract

Due to the increasing shortage of fish in the seas, Vessel Monitoring Systems (VMS) play a very important role in fishing activity monitoring, control and surveillance. In this context, the detection of fishing activities in prohibited zones is a critical task. Although position, speed and other information are provided by the VMS system, detecting fishing activities using only this type of information may be rather ineffective. The purpose of this paper is to propose an intelligent video-surveillance system that is able to automatically detect if (and when) fishing activities occur by automatically visually monitoring the vessel activities. After detection, the Control Center would be informed of relevant events and receive visual data through a low-bandwidth, low-cost satellite link to check if illegal activities are effectively being performed. To achieve this goal, the proposed visual monitoring solution adopts a deep-learning approach twice: First, to detect the fishing activities at the vessel, a well-known convolutional neural network designed for image classification is exploited. Second, to enhance at the Control Center the heavily compressed decoded images, transmitted with a low-bandwidth satellite channel, a convolutional neural network suitable for image restoration is also used. The proposed monitoring system follows a new approach for the problem addressed based on image data and can significantly improve the VMS even detection and validation efficiency by exploiting advanced image processing tools, thus helping to ensure the availability and sustainability of fish stocks for generations to come.