We present a survey on marine object detection based deep neural network approaches, which are state-of-the-art approaches for the development of autonomous ship navigation, maritime surveillance, shipping management, and other intelligent transportation system applications in future. The fundamental task surveillance navigation is to construct reachable visual perception that requires high efficiency accuracy detection. Therefore, high-performance learning-based algorithms high-quality...

The safety of ships during nighttime navigation has always been a major concern. With the widespread application technologies such as intelligent recognition, detection, and unmanned ship at night, maritime light pollution significantly affected effectiveness these safety. Therefore, effectively eliminating become an urgent challenge that needs to be addressed. This paper presents model based on spatial frequency blocks (SFBs) solve problem in sea images. includes ResNet-50, encoder,...

Increasing vessel traffic in narrow, winding inland waterways has heightened the risk of accidents, driving need for improved surveillance and management. This study addresses challenge real-time processing synchronization voluminous video AIS data effective waterway We developed a method utilizing smart buoys equipped with sensors edge computing devices, enabling dynamic spatiotemporal fusion. The integration advanced computer vision techniques target detection allows analysis provides...

The development of China’s digital waterways has led to the extensive deployment cameras along inland waterways. However, limited processing and utilization resources hinder ability provide waterway services. To address this issue, paper introduces a novel perception approach based on an intelligent navigation marker system. By integrating multiple sensors into markers, fusion camera video data automatic identification system (AIS) is achieved. proposed method enhanced one-stage object...

The autonomous decision-making model for ship navigation requires extensive interaction and trial-and-error in real, complex environments to ensure optimal performance efficiency across various scenarios. However, existing approaches still encounter significant challenges addressing the temporal features of state space tackling dynamic collision avoidance tasks, primarily due factors such as environmental uncertainty, high dimensionality space, limited decision robustness. This paper...

Abstract Vessel Traffic Service (VTS) significantly improves the navigation efficiency of ports. This paper proposes a model called Joint Extraction Triples from VHF Speech (JER‐VHF) to ensure VTS. Numerous texts are extracted Very High Frequency (VHF) speech communication contents and these organized into dataset named VHFDT. The proposed model's transforming task transforms voice this triple representation. VHFDT has large number overlapping triples. Therefore, combined with three...

Abstract In order to address issues that arise in complex scenarios, an edge-enhanced YOLOv5 algorithm is proposed. The fusion of edge features with original feature maps serves enhance the accuracy detection process. employs Sobel and Canny operators for extraction, integrating these low-level through addition concatenation. Testing urban desert settings demonstrated enhanced models exhibited superior a reduction false missed compared YOLOv5. This methodology has potential target...

Recent years, both the domestic and international frontiers made headway in realm of intelligent navigation for ships, so vision-based ship monitoring systems smart ships had a great development prospect. However, night-time environment at sea, images usually have low contrast, visibility generous noise splash. Deep learning algorithms are affected by these problems, which not only limit detecting recognizing capability but also bring unstable elements aspect safety. To solve this study...

Underwater target classification is an important research topic in the field of underwater acoustic signal processing, and marine mammal vocalizations great significance for conservation them. In this paper, calls humpback whales, short-finned pilot whales sperm are taken as objects plotted spectrograms. Aiming to address issue small targets spectrograms, we proposed improved model based on YOLOv5 integrate Vision Transformer BiFormer module enhance detection accuracy each sample. The test...

The passive location methods for underwater target is particularly critical to both military and civilian domains, with Time Difference of Arrival (TDOA) gaining widespread utilization across diverse applications. In this paper, the method based on TDOA distributed multi-agent systems (UTL-DMS) proposed apply in shallow water environment. solved by constructing a set pseudo-linear equations, requiring three observers system. simulation results show that average error 1.48 m. are constructed....

In order to find a model-based algorithm achieve more rapid and efficient real-time route planning for inland water, this paper studies variety of algorithms. By studying the (Deep Q_Network, DQN) in deep reinforcement learning algorithm, rasterizes data water use it as state space neural network, obtains action output network according space. calculate current position agent taken by raster, next is obtained. Connect all positions travels, we can get planned water. comparing experimental...

Visual object detection is one of the most important aspects autonomous ship's perception. In order to make various deep learning-based target models have a unified performance evaluation standard, we provide an image dataset in ship navigation scenarios and its corresponding benchmarks, optimize classification strategy, use real scene train milestone models, which effectively proves that SOTA model has uneven specific scenarios. It difficult realize industrial deployment visual perception...