Hierarchical Multilabel Ship Classification in Remote Sensing Images Using Label Relation Graphs

In this paper, a ship classification framework based on deep learning is proposed. We focus on the influence of the incident angle factor for the classification results on deep learning-based methods. A representative SAR ship dataset containing three types of ship and the coverage of incidence angle is approximately from 20° to 60° is created. We evaluated the training-test performance of four deep learning models on the dataset. Taking cargo ship as the example, the experimental results show that when using data with different range of incident angles for training, the classification performance on test set with different range of incident angles varies greatly. The first analysis of the incident angle factor in SAR ship classification using deep learning methods allowed researchers to select appropriate data when using the deep learning method to classify ships in SAR images, and may suggest satellite parameters based on the classification results.

Ship classification has long been a challenging task in synthetic aperture radar (SAR) applications. With the improvement of SAR resolution, it has gradually become possible based on high-resolution SAR images. This letter presents a joint feature representation method for characterizing the superstructure scattering features of merchant ships, and then a hierarchical scheme for classifying merchant ships in high-resolution SAR images based on integrated macro and detailed superstructure features is proposed. Comprehensive experiment on three types of high-resolution SAR samples demonstrates that the proposed features and method have the ability to improve the performance of merchant ship classification in SAR images.

Due to technological advances, a massive amount of data is produced daily, presenting challenges for application areas where data needs to be labelled by a domain specialist or by expensive procedures, in order to be useful for supervised machine learning purposes. In order to select which data points will provide more information when labelled, one can make use of active learning methods. Active learning (AL) is a subfield of machine learning which addresses methods to build models with fewer, but more representative instances. Even though AL has been vastly studied, it has not been thoroughly investigated in hierarchical multi-label classification, a learning task where multiple class labels can be assigned to an instance and these labels are hierarchically structured. In this work, we provide a public framework containing baseline and state-of-the-art algorithms suitable for this task. Additionally, we also propose a new algorithm, namely Hierarchical Query-By-Committee (H-QBC), which is validated on datasets from different domains. Our results show that H-QBC is capable of providing superior predictive performance results compared to its competitors, while being computationally efficient and parameter free.

Many ships have AIS devices that can provide information such as the types of ships, which can help the maritime authorities manage the marine traffic in a better way. However, when some ships do not have AIS devices installed or turn off these devices, it is desirable to identify the types of ships by the trajectory features provided by radar. In order to achieve this goal, first, the trajectories are generated based on the obtained AIS points and radar points, and then the radar trajectories are associated with the AIS trajectories to obtain the labels of the radar trajectories. Next, three types of features of radar trajectories are extracted. Due to the small amount of experimental data and the problem of class imbalance, this paper proposes a heterogeneous ensemble learning method based on EasyEnsemble and SMOTE when training the ship classification model. The experimental results show that the proposed method is superior to homogeneous ensemble learning and heterogeneous ensemble learning without SMOTE methods. Moreover, the method can identify almost all the minority class samples and has certain application value.

Estimating the lightship mass is an important factor in the early stage of ship design given its impact on shipbuilding costs. The mass of the hull and equipment, as well as its distribution, affects the hull strength, deadweight, and stability of the ship. Furthermore, a good estimate of the lightship mass in the preliminary phase is very important for defining the construction costs as accurately as possible. As the time available to the designer in the conceptual or preliminary phase of ship design is often limited, the methods used to determine the lightship mass must be reliable, fast, and effective. The paper presents and analyzes methods for estimating the lightship mass that are based on empirical data of built ships. Several empirical methods were combined and calibrated to establish the least-squares method which combination gives the value of the lightship mass closest to the one given in the collected database for three types of merchant ships: tankers, bulk carriers, and container ships. The paper presents the results of the calculations of the masses of hull steel, outfitting, machinery, and superstructure. It can be concluded that the presented model for all three types of merchant ships gives satisfactory results.

Zhang, H. and Zhao, X., 2015. Quantitative analysis of organizational behavior of container shipping in the upper and middle reaches of the Yangtze River based on Hub-and–Spoke network.By investigating the current situation of container shipping organization at the middle and upper reaches of the Yangtze River, it was discovered that current transport organizations do not follow the basic law of “large ship with long line, small with short line” but conform to a government subsidies policy. This paper constructs a integer programming optimization model of the container Hub-and-Spoke transport in the middle and upper reaches of Yangtze River and indicates that the container Hub-and–Spoke transport organization mode in the middle and upper reaches of Yangtze River with the axis of the section of Wuhan Newport-Shanghai Yangshan port area is optimized, compared with the existing transport organization mode, which can lead to 5% and 8.4% drop in generalized total transport cost and fuel consumption respectively given the existing ship types as well as 15.3% and 20.1% decreases respectively given the optimal ship types. As the government subsidy policy misled the behavior of the ship market managers, the negative effects caused by the behavior of government and ship managers have also been quantified and analyzed, and a policy aimed at the development of the optimization of the container transport organization in the middle and upper reaches of Yangtze River have been put forward.

Layer-by-layer training is an alternative end-to-end backpropagation approach for training deep convolutional neural networks. Layer-by-layer training on specific architectures can yield very competitive results. In the ImageNet database (www.imagenet. org), layer-by-layer trained networks can perform comparable to many modern end-to-end trained networks. This article compares the performance gap between the two training procedures over a wide range of network architectures and further analyzes the potential limitations of layer-by-layer training. The results show that layer-by-layer learning quickly saturates after a certain critical level due to overfitting of early levels in neural networks. Several approaches that have been used to solve this problem are discussed, and a methodology for improving layer-by-layer learning in various neural network architectures is discussed. Fundamentally, this research highlights the need to open up the black box that represents modern deep neural networks and explore the layer-by-layer interactions between intermediate hidden layers within deep networks through the lens of layer-by-layer learning.

The study of the behaviour of ships in the real sea has become a real challenge in recent years due to climate change, especially in the European sea basin. One of the main categories of types of ships that cross these waters are transport ships. The development of the economy and trade have a major influence on this type of ships, which must be studied and improved in order to be fully exploited. Investigating the ship’s behavior is not a new topic, but the changing environment of the transport ship’s specifications requires continuous study. An alternative to model testing is the use of numerical simulation through which performance can be fully evaluated using only computers, without the need to produce a scale model for testing. Due to improved computer performance, but also the disadvantages of towing testing, numerical simulations have become increasingly popular for determining ship characteristics. The shorter time in which results can be obtained, as well as the ease of controlling and changing the input parameters, contributes to the preference to use numerical simulations, to the detriment of basin tests. In this paper, the behavior of a Post-Panamax benchmark container ship in regular waves was studied. Numerical simulations were performed using SHIPFLOW MOTIONS, a solver that treats potential nonlinear flows for regular waves and provides accurate results by the free surface potential flow panel method. For our case, the simulations were performed for a range of six speeds corresponding to Froude numbers ranged between 0.174 and 0.218, for five nondimensional wavelengths (λ/Lpp) and for four wave heights. Because body discretization is one of the main factors for obtaining accurate results, an additional study was conducted in calm waters on body refinement and the influence that the mesh has on results, for a range of ten grids with different densities.

Inland river ships (IRSs) use diesel with a lower sulfur content and emit relatively low emissions, making it challenging to monitor their emissions. Sniffer monitoring equipment was installed from August 2020 to June 2022 at the Gezhou Dam of the Yangtze River and monitored emissions from 8,238 IRSs passing through the lock. We partnered with the maritime department to select 100 ships passing through the lock to extract fuel oil samples for direct fuel sulfur content (FSC) detection, which determined the actual FSC of the passing ships. The monitoring data from these 100 ships indicated that the relative error of the SO2 emission factors (EFs) and FSC results is significant at the 10-parts-per-million level. The monitoring data from the remaining 8,138 ships showed that the EFs of NO, NO2, PM2.5, and PM10 were 24.02 ± 16.92 g kg−1, 10.30 ± 18.08 g kg−1, 0.72 ± 0.60 g kg−1, and 0.92 ± 0.70 g kg−1, respectively. The NOx EFs of container ships are higher than those of other ship types, while the PM EFs for different ship types do not significantly differ. Based on these EFs, we calculated the average emission rates for different types of ships passing through locks, which is a real-time measurement method for estimating ship emissions. In addition, a comparison of ship EF measurements over the past 20 years revealed that EF values for SO2, NOx, and PM exhibited a downward trend, with the calculated results of the current study determined to be the lowest numerical level.

Domain-invariant representation learning using an unsupervised domain adversarial adaptation deep neural network

Parametric geometric modeling plays an important role in ship’s hull form optimization by use of computational fluid dynamic (CFD) analysis. However, it is difficult to create satisfactory parametric modeling for some curved shapes, such as a ship’s bulbous bow. In this study, the cubic Bezier curve and curve-plane intersection methods are applied to generate the parametric design of a bulbous bow in a solid modeling procedure by taking into account the input of 4 (four) design parameters. For this, a suitable application program interface script within the ANSYS Design Modeler was developed. An application to the ship resistance minimization by use of CFD was made to show that the proposed method could be implemented properly. In this respect, the parametric design of the bulbous bow of a container ship (KRISO Container Ship type) was chosen to be modified. First, it was shown that the computational results generated by CFD were close to the experimental data for the original ship hull form. The developed optimization method was subsequently applied to find the optimal bulbous bow. Finally, the dependence of the optimum bulbous bow on a ship’s speed (some variations of Fn values) was investigated and the results were compared to each other.

To improve collision risk prediction in high-traffic coastal waters and support real-time decision-making in maritime navigation, this study proposes a regional collision risk prediction system integrating the Computed Distance at Collision (CDC) method with an Adaptive Neuro-Fuzzy Inference System (ANFIS). Unlike Distance at Closest Point of Approach (DCPA), which depends on the position of Global Positioning System (GPS) antennas, Computed Distance at Collision (CDC) directly reflects the actual hull shape and potential collision point. This enables a more realistic assessment of collision risk by accounting for the hull geometry and boundary conditions specific to different ship types. The system was designed and validated using ship motion simulations involving bulk and container ships across varying speeds and crossing angles. The CDC method was used to define collision, almost-collision, and near-collision situations based on geometric and hydrodynamic criteria. Subsequently, the FIS–CDC model was constructed using the ANFIS by learning patterns in collision time and distance under each condition. A total of four input variables—ship speed, crossing angle, remaining time, and remaining distance—were used to infer the collision risk index (CRI), allowing for a more nuanced and vessel-specific assessment than traditional CPA-based indicators. Simulation results show that the time to collision decreases with higher speeds and increases with wider crossing angles. The bulk carrier exhibited a wider collision-prone angle range and a greater sensitivity to speed changes than the container ship, highlighting differences in maneuverability and risk response. The proposed system demonstrated real-time applicability and accurate risk differentiation across scenarios. This research contributes to enhancing situational awareness and proactive risk mitigation in Maritime Autonomous Surface Ship (MASS) and Vessel Traffic System (VTS) environments. Future work will focus on real-time CDC optimization and extending the model to accommodate diverse ship types and encounter geometries.

Container ships are an important means of water transportation, especially for inter-island shipping. One type of modern container ship is the Kriso Ship (KCS), which was designed in 1997. This ship uses a driving device called a propulsor, which can be either mechanical or non-mechanical. One commonly used mechanical propulsion device is the propeller. The MAU propeller, developed in Japan, is one of the types used on container ships. When selecting a propeller, one must consider thrust, torque, and efficiency, usually measured through the KT-KQ-J diagram in open water testing. Each propeller has different characteristics, so it requires special research. Modifying the propeller skew angle can affect thrust, torque and efficiency. This research shows the optimal skew angle for the AU5-80 propeller is 35° skew, as it produces the best efficiency. The results of this test can later be used as recommendations to students and shipyards about propeller selection.

Direct observation using satellites and long range video surveillance is not possible for ship classification in adverse weather and during night. Radar and more specifically radar imaging offers a solution for the above adverse conditions. Ship Classification using radar is of utmost important to defense of any country to manage vast naval resources and to tell the friend from foe. Automatic ship classification based on radar images has been very successful in determining the ship class as well as other details to reliably recognize a ship type using machine vision. Inverse Synthetic Aperture Radar (ISAR) Imaging which relies on a stationary radar and a moving object with preferably superstructure will result in an image that is somewhat unique to a particular ship class. There have been many attempts to classify these ISAR images automatically with varying degree of success. The results we present here using Moment Invariants (Hu Moments) are indeed superior to many other feature-based classification approaches as they have strong invariant properties.

The transport of liquefied natural gas (LNG) has significant impact on traffic capacity of waterways, especially the approach channels shared by LNG carriers and other types of ships (general cargo ships, container ships, etc.). Few studies take the behavioral characteristics of LNG carriers and their impacts into consideration. In this paper, we propose a framework for capacity analysis of shared approach channels based on the spatial–temporal consumption method. It consists of three modules: (1) the tide module predicts the tidal height and tidal time for identifying the time windows for LNG carriers; (2) the spatial–temporal consumption module is introduced to calculate the capacity of approach channels; (3) the LNG carrier navigation module is for analyzing the characteristics of LNG carriers and the impact on the capacity of approach channels. A spatial–temporal indexed chart is designed to visualize the utilization of the spatial–temporal resources. A case study on the approach channel of Yueqing Bay near the east coast of China is conducted to verify the effectiveness of the framework. The utilization rates of the approach channel and the impact of LNG carriers are presented using our method. The results of the case study indicate that the proposed traffic capacity analyzing framework can provide support for making traffic management strategies.