To ensure sustainable fisheries, many complex on-vessel activities are periodically monitored to provide data to assist the assessment of stock status and ensure fishery regulations are being met. Such monitoring is often performed manually which is an exhaustive and expensive process. Consequently, several forms of Electronic Monitoring (EM) have emerged recently and include the use of electronic monitoring using on-board video cameras and Automatic Identification System (AIS). Unfortunately, insufficient cameras, ineffective camera position or obstructions, may lead to objects or behaviours of interest not being observed. In addition, more subtle, anomalous behaviours characteristic of behaviours of interest may still be captured. With the increasing success of deep learning methods, this article identifies the scope and challenges of using state-of-the-art deep learning approaches to anomaly detection in fisheries, and in particular to automatically detect abnormal behaviours from on-board video cameras and AIS data in line with current fishing practices and regulations. This study will take us one step closer towards automatic anomaly detection frameworks that can potentially replace existing manual monitoring methods.

为了确保可持续渔业，定期监测许多复杂的船上活动，以提供数据来协助评估种群状况并确保遵守渔业法规。这种监控通常是手动执行的，这是一个详尽且昂贵的过程。因此，最近出现了多种形式的电子监控（EM），包括使用车载摄像机和自动识别系统（AIS）进行电子监控。不幸的是，相机不足、相机位置无效或障碍物可能导致无法观察到感兴趣的物体或行为。此外，仍然可以捕获感兴趣行为的更微妙、异常的行为特征。随着深度学习方法的日益成功，本文确定了使用最先进的深度学习方法进行渔业异常检测的范围和挑战，特别是自动检测机载摄像机和 AIS 的异常行为数据符合当前的捕捞实践和法规。这项研究将使我们向自动异常检测框架更近一步，该框架有可能取代现有的手动监控方法。