WO2021017577A1 - 一种基于集成学习的船舶类型仿冒监测方法 - Google Patents
一种基于集成学习的船舶类型仿冒监测方法 Download PDFInfo
- Publication number
- WO2021017577A1 WO2021017577A1 PCT/CN2020/090547 CN2020090547W WO2021017577A1 WO 2021017577 A1 WO2021017577 A1 WO 2021017577A1 CN 2020090547 W CN2020090547 W CN 2020090547W WO 2021017577 A1 WO2021017577 A1 WO 2021017577A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- ship
- type
- data
- historical
- feature
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Definitions
- the invention relates to a ship type monitoring method, in particular to a ship type counterfeiting monitoring method based on integrated learning
- the present invention provides a ship type counterfeiting monitoring method based on integrated learning.
- the methods include innovative methods such as feature selection based on AIS historical data, historical data preprocessing and feature generation, and evaluation function setting.
- the historical trajectory messages used in the present invention are all AIS trajectory messages complying with the NEMA0183 protocol.
- Each message includes ship name, MMSI number, ship type, course, speed, heading, longitude, latitude, and time. Information such as stamp, intelligence source, batch number, jurisdiction code, responsibility area code, sea and air identification, etc.
- the time stamp information records the time of the ship at each location, and the MMSI number is the unique ID of the ship in the AIS system.
- the historical data selection, pre-processing and feature generation methods after many experiments, found that the longitude, latitude, speed, heading, ship heading, and time stamp in the ship’s AIS message are used to describe the ship’s navigation characteristics. It has the best effect to realize ship type judgment. Historical AIS data must go through processes such as outlier elimination and type adjustment to prevent outliers from affecting the model monitoring results. In the experiment, it is found that a single track message is used as a feature to train the classification model, and the error is larger. A better method is to splice the important data items of a ship's continuous multiple track messages into one feature for use Model training. Therefore, a method for splicing and generating sliding window features is provided in the present invention to generate features that are ultimately used for model training.
- the evaluation function setting method since type counterfeiting does not commonly occur in various types of ships, the probability of its occurrence in fishing vessel types is much greater than that of cargo ships, passenger ships and other types. Therefore, it is necessary to customize the evaluation function during model training to intervene in the model training process, so that the finally generated model is more sensitive to the monitoring of fishing boats and other types of frequent counterfeiting phenomena.
- a method for monitoring counterfeiting of ship types based on integrated learning including the following steps:
- Step 1 Obtain the ship's historical track message data used for model training, clean the ship's historical track message data, and adjust the data type;
- Step 2 Select feature data items, perform format transformation, and normalize the transformed features
- Step 3 Select a classifier, set an evaluation function for model training, and obtain a classification model
- Step 4. Perform real-time judgment, monitoring and warning on the ship target type according to the classification model.
- the step 1 includes:
- Step 1-1 clean historical data: scan all historical ship track message data used for model training, and clean historical data according to the following rules: delete historical ship track message data whose speed, course and heading are less than 0, Ship historical track message data with latitude and longitude on land, and ship historical track message data with course and heading greater than 360 degrees;
- Step 1-2 perform historical data deduplication: determine the track points with the same time, position, and heading as duplicate points, and delete the duplicate points in the ship's historical track message data to remove them;
- Steps 1-3 adjust the data type: set the corresponding regular expression to match the ship name of the AIS message for some of the ship types with characteristics named, and match the data of other types of ship historical track messages to this type
- the ship type of the ship historical track message data of the ship name naming feature is modified to this type. If the name of a fishing boat generally contains "YU", "YANG ZHI" and other related characters and ends with a number from 4 to 6, you can set the regular expression pattern as follows:
- Its representative meaning is a ship name that contains characters such as YU, YU CHUAN, YANG ZHI, YU YANG, YU BU, BU LAO and ends with at least 4 digits.
- This type of ship name is unique to fishing boats. If there is a message conforming to the regular expression in the AIS message data of the cargo ship, passenger ship, etc., the ship type data item of the message is modified to a fishing boat.
- the step 2 includes:
- Step 2-1 After many experiments, it is found that the longitude, latitude, speed, course, ship heading, and time stamp in the ship's AIS message can describe the ship's navigation characteristics well, and the effect of judging ship type is the best . Therefore, the MMSI, longitude, latitude, speed, heading, ship heading, and time stamp in the ship’s historical track message data are selected as the characteristic data items to be stored separately, and the ship’s historical track message data is stored according to MMSI (Maritime Mobile Communication Service Identifier).
- MMSI Maritime Mobile Communication Service Identifier
- MMSI Code, Maritime Mobile Service Identify
- timestamp is the secondary key, that is, the items with the same MMSI are sorted from smallest to largest according to the time stamp.
- Step 2-2 use sliding window for feature stitching: set the sliding window size n and sliding step length m, and use the sliding window method to combine the longitude, latitude, speed, and heading in the same MMSI continuous ship historical track message data , Ship heading and timestamp are spliced into a feature and stored.
- the feature dimension is 6n.
- the time difference between the historical trajectory message data of two adjacent ships in a feature does not exceed 900 seconds. If it exceeds, the sliding window will move forward and re Features in the splicing window;
- the feature label is the code of the ship type of the ship’s AIS message (for example, passenger ships, cargo ships, fishing ships, oil tankers, and tugboats can be set to code 0, 1, 2, 3, 4);
- Step 2-3 transform the timestamp: Since most ship sailing rules are periodic, take the remainder of the timestamp and the number of seconds in a day, and add the time difference with time zone 0 to transform it into the number of seconds of the day ,
- the specific transformation formula for my country’s sea area in the East Eight District is as follows:
- timestamp represents the timestamp
- time represents the timestamp after transformation
- Steps 2-4 normalize the new features: calculate the mean ⁇ and variance ⁇ of each dimension feature in all sample spaces, use the normalization formula to transform each dimension feature, and save the ⁇ and As a normalized model, the transformation formula is:
- x represents a new feature
- x' represents a normalized feature
- all normalized features form a training sample.
- the step 3 includes:
- Step 3-1 use Classification and Regression Tree (CART) as the base classifier for ensemble learning; use ensemble learning combined with serial structure, that is, each layer has only one CART, and the classification error of the previous layer is used as the next A layer of CART input (integrated learning classification algorithms such as GBDT, XGBoost, etc., which meet the above structural characteristics can be used to implement the method of the present invention);
- CART Classification and Regression Tree
- Step 3-2 use the error rate error, the mean square error MSE, and the area under the receiver operating characteristic curve roc_auc as the evaluation function, and modify the evaluation function of the integrated learning according to the actual needs;
- Step 3-3 Use the integrated learning algorithm described in steps 3-1 and 3-2 to learn and train the training samples obtained in steps 2-4, generate a classification model, and save it.
- Step 3-2 the perturbation modification of the evaluation function of the integrated learning according to actual needs includes: when it is necessary to focus on monitoring fishing boats disguised as other ships, only the error rate error of the fishing boat part is calculated as the objective function:
- pred yu_other represents the number of fishing boats predicted to be other ships
- train yu represents the true number of fishing boat samples in the training sample.
- step 3-2 the disturbance modification of the evaluation function of integrated learning according to actual needs includes: when it is necessary to focus on monitoring fishing boats disguised as other ships, adding a weight coefficient to the fishing boat::
- weight is a real number greater than 1, which indicates the weight of the fishing boat error calculation
- pred other_yu indicates the number of fishing boats predicted by other boats
- train indicates the total number of sample data.
- the step 4 includes:
- Step 4-1 record the ship's real-time track message, the number of records must be greater than the sliding window size n, where the message value should comply with the rules for cleaning historical data in step 1-1, otherwise, re-record the ship's real-time track message;
- Step 4-2 generate real-time type monitoring features: when a new message is received, the latest n continuous ship real-time trajectory messages are processed by the method in step 2 to obtain the normalized characteristics;
- Step 4-3 abnormality monitoring and reporting: input the normalized features into the classification model, use the classification model to determine the type of the ship, and record the abnormality if it is inconsistent with the type in the ship's real-time track message; set the threshold for the number of abnormalities, When the number of consecutive abnormalities exceeds the threshold, a suspected counterfeiting alarm is reported, and if the subsequent monitoring determines that it is normal, the alarm is reported.
- the present invention solves the problem of counterfeiting monitoring of ship types.
- traditional maritime supervision if staff want to discover counterfeit ship types, they can only estimate based on experience and use the position, speed, heading and other information in the ship’s AIS message.
- This method is not only extremely inefficient, but also often accurate. not tall.
- the present invention first clarifies the characteristic information required for type judgment and monitoring and its generation method; then provides the composition structure and related settings of a suitable machine learning classification algorithm; and finally provides a specific process method for real-time monitoring.
- the type of counterfeit monitoring method provided by the present invention has a faster monitoring speed and a higher monitoring accuracy in actual use, and can simultaneously monitor the entire sea area in real time, compared to the traditional method using manual experience The efficiency has been greatly improved. Using the method of the invention can solve the problems of poor efficiency and low accuracy of traditional ship type counterfeiting monitoring.
- Figure 1 is the overall flow chart of model training and real-time monitoring
- Figure 2 is a flow chart of data cleaning and feature generation
- Figure 3 is a schematic diagram of a sliding window feature generation method
- Figure 4 is a flow chart of judging and monitoring the type of a message.
- a method for monitoring counterfeiting of ship types based on integrated learning includes the following steps:
- first set the cleaning rules including but not limited to the location should be in the area of responsibility and cannot be on land, the speed cannot be negative, the heading and heading cannot be negative and cannot be greater than 360 degrees, the historical data
- the track points whose data items such as position, speed and heading are in compliance with the outlier point cleaning rules are removed;
- the regular expression pattern can be set as follows:
- Its representative meaning is a ship name that contains characters such as YU, YU CHUAN, YANG ZHI, YU YANG, YU BU, BU LAO and ends with at least 4 digits.
- This type of ship name is unique to fishing boats. If there is a message conforming to the regular expression in the AIS message data of the cargo ship, passenger ship, etc., the ship type data item of the message is modified to a fishing boat.
- the code of the ship type of the ship’s AIS message such as passenger ship, cargo ship, fishing vessel, oil tanker, and tugboat can be set to code 0, 1, 2, 3, 4;
- time stamp is taken from the number of seconds in a day, and the time difference with time zone 0 is added to convert it to the number of seconds of the day.
- time difference is added to convert it to the number of seconds of the day.
- the method of the present invention uses CART as the base classifier for ensemble learning; ensemble learning using serial iterative structure combination, that is, each layer has only one CART, and the classification error of the previous layer is used as the input of the CART of the next layer; Integrated learning classification algorithms such as GBDT, XGBoost, etc. can be used to implement the method of the present invention;
- the evaluation function of ensemble learning can be disturbed and modified according to actual needs to increase the weight of the corresponding type and accelerate the training iteration process; if you want to focus on monitoring and pretending to be When fishing boats of other ships, only the error of the fishing boat part can be calculated as the evaluation function, such as:
- pred yu_other represents the number of fishing boats predicted as other ships
- train yu represents the true number of fishing boat samples in the data.
- add a weight coefficient to the fishing boat part such as:
- weight is a real number greater than 1, which indicates the weight of the fishing boat error calculation
- pred other_yu indicates the number of fishing boats predicted by other boats
- train indicates the total number of sample data.
- the longitude, latitude, speed, heading, heading, and time stamp in the last n continuous real-time messages are spliced into a feature, and the time is combined using the method in (23)
- the stamp item is transformed into the number of seconds of the day; the saved normalization model is used to normalize the feature;
- the classification model is used to determine the type of ship, and if it is inconsistent with the message type, abnormalities will be recorded; set the threshold for the number of abnormalities, generally an integer between 10-30. The smaller the threshold, the higher the sensitivity of the system. When the number of abnormalities exceeds the threshold, a suspected counterfeit alarm will be reported, and if the subsequent monitoring determines that it is normal, the alarm will be reported.
- the present invention comprehensively utilizes big data and artificial intelligence technology to study and propose feasible solutions from a technical perspective, and gives specific implementation steps.
- This invention can successfully detect ships with counterfeit AIS message types, and provide powerful technical support for the Ministry of Maritime Affairs and Fisheries to help them further reduce the probability of water traffic accidents. It is believed that it is used in my country's maritime and fishery departments, especially in the Bohai Sea Rim. , Zhoushan, Beibu Gulf and other regions with rich fishery resources have broad market prospects.
- the present invention provides a method for monitoring counterfeit ship types based on integrated learning. There are many methods and ways to implement this technical solution. The above are only preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art In other words, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention. All the components that are not clear in this embodiment can be implemented using existing technology.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
船舶类型 | 测试总数量 | 预测错误数量 | 预测错误率 |
客船 | 150000 | 360 | 0.24% |
货船 | 200000 | 2240 | 1.12% |
渔船 | 200000 | 2700 | 1.35% |
油轮 | 150000 | 930 | 0.62% |
拖船 | 100000 | 2640 | 2.64% |
Claims (7)
- 一种基于集成学习的船舶类型仿冒监测方法,其特征在于,包括以下步骤:步骤1,获取用于模型训练的船舶历史航迹报文数据,对船舶历史航迹报文数据进行清洗,并调整数据类型;步骤2,选择特征数据项,并进行格式变换,对变换生成后的特征进行归一化处理;步骤3,选择分类器,设置评估函数进行模型训练,得到分类模型;步骤4,根据分类模型实时对船舶目标类型进行判断监测与告警。
- 根据权利要求1所述的方法,其特征在于,所述步骤1包括:步骤1-1,清洗历史数据:扫描全部用于模型训练的船舶历史航迹报文数据,根据如下规则清洗历史数据:删除速度、航向和船艏向小于0的船舶历史航迹报文数据、经纬度在陆地位置的船舶历史航迹报文数据,以及航向和船艏向大于360度的船舶历史航迹报文数据;步骤1-2,进行历史数据去重:将时间、位置、航向均相同的航迹点判定为重复点,删除船舶历史航迹报文数据中的重复点进行去除;步骤1-3,进行数据类型调整:对部分命名有特征的船舶类型,设置对应的正则表达式对AIS报文的船名进行匹配,将其他类型的船舶历史航迹报文数据中符合该类型船名命名特征的船舶历史航迹报文数据的船舶类型修改为该类型。
- 根据权利要求2所述的方法,其特征在于,所述步骤2包括:步骤2-1,选择特征数据项:选择船舶历史航迹报文数据中的MMSI、经度、纬度、速度、航向、船艏向、时间戳作为特征数据项单独存储,将船舶历史航迹报文数据根据MMSI和时间戳从小到大排序,其中MMSI为排序主键,时间戳为副键,即先按照MMSI从小到大排序,MMSI相同的项按照时间戳从小到大排序;步骤2-2,使用滑动窗口进行特征拼接:设置滑动窗口大小n和滑动步长m,使用滑动窗口的方法将同一个MMSI的连续两个以上的船舶历史航迹报文数据中的经度、纬度、速度、航向、船艏向、时间戳拼接成一条特征并存储,特征维度为6n,一条特征中相邻两条船舶历史航迹报文数据之间时间差不超过900秒,如果超过则滑动窗口前进一步,重新拼接窗口内特征;特征标签为该船舶AIS报文的船舶类型的代号;步骤2-3,对时间戳进行变换:将时间戳与一天的秒数取余,并加上与0时区时差,将其变换为当日的秒数,对于处于东八区的我国海域来说具体变换公式如下:time=timestamp%86400+28800,其中,timestamp表示时间戳,time表示变换后的时间戳;步骤2-4,对新的特征进行归一化处理:计算每一维特征在全部样本空间中的均值μ和方差σ,使用归一化公式对每一维特征进行变换,并保存下μ和σ作为归一化模型,变换公式为:x’=(x-μ)/σ,其中,x表示新的特征,x’表示归一化后的特征,所有归一化后的特征组成训练样本。
- 根据权利要求3所述的方法,其特征在于,所述步骤3包括:步骤3-1,使用分类回归树CART作为集成学习的基分类器;使用串行结构组合的集成学习,即每一层只有一个CART,上一层的分类误差作为下一层CART的输入;步骤3-2,根据实际需求对集成学习的评估函数进行扰动修改;步骤3-3,使用符合步骤3-1和3-2描述的集成学习算法对步骤2-4得到的训练样本进行 学习训练,生成分类模型并进行保存。
- 根据权利要求4所述的方法,其特征在于,步骤3-2中,所述根据实际需求对集成学习的评估函数进行扰动修改,包括:当需要着重监测伪装成其他船舶的渔船时,只计算渔船部分的错误率error作为目标函数:error=pred yu_other/train yu,其中pred yu_other表示将渔船预测成其他船舶的数量,train yu表示训练样本中渔船样本的真实数量。
- 根据权利要求4所述的方法,其特征在于,步骤3-2中,所述根据实际需求对集成学习的评估函数进行扰动修改,包括:当需要着重监测伪装成其他船舶的渔船时,对渔船增加权重系数:error=(pred yu_other*weight+pred other_yu)/train,其中weight为一个大于1的实数,表示将渔船的误差计算权重;pred other_yu表示将其他船预测成渔船的数量,train表示训练样本总数量。
- 根据权利要求6所述的方法,其特征在于,所述步骤4包括:步骤4-1,记录船舶实时航迹报文,记录数量需大于滑动窗口大小n,其中报文数值应符合步骤1-1中清洗历史数据的规则,否则重新记录船舶实时航迹报文;步骤4-2,生成实时类型监测特征:收到一条新报文时,将最近n条连续船舶实时航迹报文采用步骤2的方法进行处理,得到归一化后的特征;步骤4-3,异常监测与报告:将归一化后的特征输入分类模型,使用分类模型判断船舶的类型,如果与船舶实时航迹报文中的类型不一致则记录异常;设置异常数量阈值,当连续异常数量超过阈值时则报告疑似仿冒告警,如之后监测判断正常则报告消警。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
ZA2021/04574A ZA202104574B (en) | 2019-07-29 | 2021-06-30 | Ship-type-spoofing detection method employing ensemble learning |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910687682.0A CN110633353B (zh) | 2019-07-29 | 2019-07-29 | 一种基于集成学习的船舶类型仿冒监测方法 |
CN201910687682.0 | 2019-07-29 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021017577A1 true WO2021017577A1 (zh) | 2021-02-04 |
Family
ID=68969581
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/090547 WO2021017577A1 (zh) | 2019-07-29 | 2020-05-15 | 一种基于集成学习的船舶类型仿冒监测方法 |
Country Status (3)
Country | Link |
---|---|
CN (1) | CN110633353B (zh) |
WO (1) | WO2021017577A1 (zh) |
ZA (1) | ZA202104574B (zh) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113283653A (zh) * | 2021-05-27 | 2021-08-20 | 大连海事大学 | 一种基于机器学习和ais数据的船舶轨迹预测方法 |
CN113870620A (zh) * | 2021-10-19 | 2021-12-31 | 遨海科技有限公司 | 一种多ais设备同时开启的船舶识别方法 |
CN114492571A (zh) * | 2021-12-21 | 2022-05-13 | 西北工业大学 | 一种基于相似度距离的船舶轨迹分类方法 |
CN114510961A (zh) * | 2022-01-03 | 2022-05-17 | 中国电子科技集团公司第二十研究所 | 基于递归神经网络和北斗定位的船舶行为智能监控算法 |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110633353B (zh) * | 2019-07-29 | 2020-05-19 | 南京莱斯网信技术研究院有限公司 | 一种基于集成学习的船舶类型仿冒监测方法 |
CN111177140B (zh) * | 2020-01-02 | 2023-07-28 | 云南昆船电子设备有限公司 | 一种卷烟制丝线生产过程数据清洗系统和方法 |
CN112833882A (zh) * | 2020-12-30 | 2021-05-25 | 成都方位导向科技开发有限公司 | 一种动态加权航线自动推荐方法 |
CN112861968A (zh) * | 2021-02-07 | 2021-05-28 | 上海普适导航科技股份有限公司 | 一种海洋大数据处理系统 |
CN116506513B (zh) * | 2023-06-26 | 2023-09-26 | 广州中海电信有限公司 | 一种根据船舶航行状态实时调整船舶数据传输的系统 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9940520B2 (en) * | 2015-05-01 | 2018-04-10 | Applied Research LLC. | Automatic target recognition system with online machine learning capability |
CN108921219A (zh) * | 2018-07-03 | 2018-11-30 | 中国人民解放军国防科技大学 | 基于目标航迹的机型识别方法 |
CN109214107A (zh) * | 2018-09-26 | 2019-01-15 | 大连海事大学 | 一种船舶航行行为在线预测方法 |
CN109508634A (zh) * | 2018-09-30 | 2019-03-22 | 上海鹰觉科技有限公司 | 基于迁移学习的船舶类型识别方法及系统 |
CN110018453A (zh) * | 2019-03-28 | 2019-07-16 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | 基于飞机航迹特征的智能机型识别方法 |
CN110633353A (zh) * | 2019-07-29 | 2019-12-31 | 南京莱斯网信技术研究院有限公司 | 一种基于集成学习的船舶类型仿冒监测方法 |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2007133085A1 (en) * | 2006-05-15 | 2007-11-22 | Telefonaktiebolaget Lm Ericsson (Publ) | A method and system for automatic classification of objects |
US11210939B2 (en) * | 2016-12-02 | 2021-12-28 | Verizon Connect Development Limited | System and method for determining a vehicle classification from GPS tracks |
CN107145903A (zh) * | 2017-04-28 | 2017-09-08 | 武汉理工大学 | 一种基于卷积神经网络图片特征提取的船舶类型识别方法 |
CN107506444B (zh) * | 2017-08-25 | 2020-09-11 | 中国人民解放军海军航空大学 | 中断航迹接续关联机器学习系统 |
CN108664933B (zh) * | 2018-05-11 | 2021-12-28 | 中国科学院空天信息创新研究院 | 一种用于sar图像船舶分类的卷积神经网络的训练方法及其分类方法、船舶分类模型 |
CN108961468B (zh) * | 2018-06-27 | 2020-12-08 | 广东海洋大学 | 一种基于集成学习的船舶动力系统故障诊断方法 |
CN109712071B (zh) * | 2018-12-14 | 2022-11-29 | 电子科技大学 | 基于航迹约束的无人机图像拼接与定位方法 |
CN109800796A (zh) * | 2018-12-29 | 2019-05-24 | 上海交通大学 | 基于迁移学习的船舶目标识别方法 |
CN109960692B (zh) * | 2019-03-12 | 2021-03-05 | 中国电子科技集团公司第二十八研究所 | 船舶航向模型的数据可视化方法、设备及计算机存储介质 |
-
2019
- 2019-07-29 CN CN201910687682.0A patent/CN110633353B/zh active Active
-
2020
- 2020-05-15 WO PCT/CN2020/090547 patent/WO2021017577A1/zh active Application Filing
-
2021
- 2021-06-30 ZA ZA2021/04574A patent/ZA202104574B/en unknown
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9940520B2 (en) * | 2015-05-01 | 2018-04-10 | Applied Research LLC. | Automatic target recognition system with online machine learning capability |
CN108921219A (zh) * | 2018-07-03 | 2018-11-30 | 中国人民解放军国防科技大学 | 基于目标航迹的机型识别方法 |
CN109214107A (zh) * | 2018-09-26 | 2019-01-15 | 大连海事大学 | 一种船舶航行行为在线预测方法 |
CN109508634A (zh) * | 2018-09-30 | 2019-03-22 | 上海鹰觉科技有限公司 | 基于迁移学习的船舶类型识别方法及系统 |
CN110018453A (zh) * | 2019-03-28 | 2019-07-16 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | 基于飞机航迹特征的智能机型识别方法 |
CN110633353A (zh) * | 2019-07-29 | 2019-12-31 | 南京莱斯网信技术研究院有限公司 | 一种基于集成学习的船舶类型仿冒监测方法 |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113283653A (zh) * | 2021-05-27 | 2021-08-20 | 大连海事大学 | 一种基于机器学习和ais数据的船舶轨迹预测方法 |
CN113283653B (zh) * | 2021-05-27 | 2024-03-26 | 大连海事大学 | 一种基于机器学习和ais数据的船舶轨迹预测方法 |
CN113870620A (zh) * | 2021-10-19 | 2021-12-31 | 遨海科技有限公司 | 一种多ais设备同时开启的船舶识别方法 |
CN113870620B (zh) * | 2021-10-19 | 2023-07-21 | 遨海科技有限公司 | 一种多ais设备同时开启的船舶识别方法 |
CN114492571A (zh) * | 2021-12-21 | 2022-05-13 | 西北工业大学 | 一种基于相似度距离的船舶轨迹分类方法 |
CN114492571B (zh) * | 2021-12-21 | 2024-03-01 | 西北工业大学 | 一种基于相似度距离的船舶轨迹分类方法 |
CN114510961A (zh) * | 2022-01-03 | 2022-05-17 | 中国电子科技集团公司第二十研究所 | 基于递归神经网络和北斗定位的船舶行为智能监控算法 |
Also Published As
Publication number | Publication date |
---|---|
CN110633353B (zh) | 2020-05-19 |
ZA202104574B (en) | 2021-08-25 |
CN110633353A (zh) | 2019-12-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021017577A1 (zh) | 一种基于集成学习的船舶类型仿冒监测方法 | |
Zhang et al. | A big data analytics method for the evaluation of ship-ship collision risk reflecting hydrometeorological conditions | |
CN109190636B (zh) | 一种遥感图像舰船目标信息提取方法 | |
WO2022068417A1 (zh) | 一种基于ais数据的船舶徘徊检测方法 | |
CN113553682B (zh) | 一种数据驱动的多层次船舶航路网络构建方法 | |
Zissis et al. | A distributed spatial method for modeling maritime routes | |
Rawson et al. | A machine learning approach for monitoring ship safety in extreme weather events | |
Zhao et al. | A novel similarity measure for clustering vessel trajectories based on dynamic time warping | |
Wang et al. | Use of AIS data for performance evaluation of ship traffic with speed control | |
Feng et al. | Collision risk assessment for ships’ routeing waters: An information entropy approach with Automatic Identification System (AIS) data | |
CN115294804B (zh) | 一种基于船舶状态监测的海缆安全预警方法及系统 | |
Tang et al. | Detection of abnormal vessel behaviour based on probabilistic directed graph model | |
CN110633892A (zh) | 一种ais数据提取延绳钓捕捞状态的方法 | |
Sevgili et al. | A data-driven Bayesian Network model for oil spill occurrence prediction using tankship accidents | |
CN116257565A (zh) | 船舶异常行为检测方法 | |
Spiliopoulos et al. | Knowledge extraction from maritime spatiotemporal data: An evaluation of clustering algorithms on Big Data | |
Yang et al. | Evaluation of port emergency logistics systems based on grey analytic hierarchy process | |
Kandel et al. | A data-driven risk assessment of Arctic maritime incidents: Using machine learning to predict incident types and identify risk factors | |
Zhang et al. | How liner shipping heals schedule disruption: A data-driven framework to uncover the strategic behavior of port-skipping | |
CN115239110A (zh) | 一种基于改进topsis方法的通航风险评价方法 | |
Zhou et al. | Macroscopic collision risk model based on near miss | |
Radhakrishnan et al. | Machine learning based automated process for predicting the anomaly in AIS data | |
Wang et al. | Analysis of navigation characteristics of inland watercraft based on DBSCAN clustering algorithm | |
Cui et al. | Research on the development of ship target detection based on deep learning technology | |
Duan et al. | AIS-based operational phase identification using Progressive Ablation Feature Selection with machine learning for improving ship emission estimates |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20848646 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20848646 Country of ref document: EP Kind code of ref document: A1 |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20848646 Country of ref document: EP Kind code of ref document: A1 |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 14/11/2022) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20848646 Country of ref document: EP Kind code of ref document: A1 |