With the continuous development of information technology and the popularization of the internet and artificial intelligence (AI), fishing-vessel and fishing-port management is undergoing broad digital transformation. Although many management units actively adopt various digital means to manage and control fishing vessels and ports, problems remain — such as low recognition rates of vessel registration numbers, lagging risk identification and high false-alarm rates — largely due to the inherent characteristics of traditional AI algorithms.
To solve these problems, GW-OCEAN cooperated with Huawei to launch "GW · Smart Port," a multimodal large model for smart fishing ports powered by Ascend’s formidable computing power. The system deeply combines traditional AI algorithms with large-model technology to achieve a new level of digitalization in fishing-vessel and fishing-port management.
Targeting pain points such as low vessel-number recognition rates, untimely risk identification and high false-alarm rates, "GW · Smart Port" adopts new-generation AI technologies based on large models to improve the fishing-vessel and fishing-port management system: using deep learning to learn complex patterns and regularities from large volumes of data for more accurate vessel-number and risk recognition; using transfer learning and reinforcement learning to migrate models already mature in other fields into fishing-vessel and fishing-port management; combining multiple data types such as imagery, video, sound, radar and AIS, and using multimodal fusion technology for more comprehensive risk identification and event monitoring; and adopting methods such as self-supervised learning to learn richer representations from unlabeled data, improving the system’s adaptability to new situations.
In terms of product functions, "GW · Smart Port" can break through the impact of factors such as weather, angle and occlusion in complex fishing-port and fishing-vessel scenes, performing precise image segmentation and automatic target annotation; through AI image-segmentation technology it accurately recognizes multiple complex targets within an area — including fishing-vessel type and hull number — and fuses AIS information to record fishing-vessel tracks; through image-recognition technology it effectively identifies safety-related behaviors such as vessel capsizing, collision behavior, reverse navigation and man-overboard, and issues early-warnings and forecasts for potential or known risky behaviors.
GW-OCEAN completed the migration of its R&D platform to the heterogeneous computing architecture of Huawei’s Qingdao Ascend AI Ecosystem Innovation Center, using Huawei Atlas 910 servers and the MindSpore deep-learning framework to incubate the large model, and completed the product development of "GW · Smart Port" in a short time. Compared with other fishing-port management products, "GW · Smart Port" highly integrates information such as video surveillance, radar, AIS and electronic charts, reduces large-model dataset requirements by 90% and data-annotation workload by 50%, supports zero-shot learning and multimodal learning, substantially improves recognition accuracy, and is compatible with existing systems. In the future, the technologies of "GW · Smart Port" will be widely applied in scenarios such as fishing-vessel and fishing-port management, fishery law-enforcement, fishery harvesting, maritime emergency management, marine-environment protection and integrated command-and-dispatch.
