Search

KR-20260066492-A - AI SELF-LEARNING SYSTEM AND METHOD FOR SHIP SITUATIONAL AWARENESS

KR20260066492AKR 20260066492 AKR20260066492 AKR 20260066492AKR-20260066492-A

Abstract

The present invention relates to an AI autonomous learning system and method for ship situation awareness, comprising: a sensing unit for sensing data to perceive the surrounding conditions of a ship; a ship situation awareness device for perceiving the surrounding conditions of a ship by utilizing sensor data sensed from heterogeneous sensors provided in the sensing unit; a data labeling device for marking an object to be recognized based on the sensor data and performing labeling; and an automatic learning device for automatically performing learning by analyzing the situation awareness result data perceived by the ship situation awareness device and the labeling result data labeled by the data labeling device, and providing a situation awareness model based on the learning result to the ship situation awareness device. The ship situation awareness device is configured to generate the situation awareness result data in real time by utilizing sensor data from each sensor provided in the sensing unit based on the situation awareness model, thereby updating the situation awareness model through autonomous learning based on actual sensor data collected during the operation of the ship, so that the situation awareness model can be optimized by matching it to the domain of the actual operating environment without human effort or intervention.

Inventors

  • 이충근

Assignees

  • 한화오션 주식회사

Dates

Publication Date
20260512
Application Date
20241104

Claims (10)

  1. A sensing unit that senses data to recognize the surrounding conditions of a ship, A ship situation awareness device that recognizes the surrounding conditions of a ship by utilizing sensor data sensed from a heterogeneous sensor provided in the above-mentioned sensing unit, A data labeling device that performs labeling by marking an object to be recognized based on the above sensor data, and It includes an automatic learning device that analyzes situation recognition result data recognized by the ship situation recognition device and labeling result data labeled by the data labeling device to automatically perform learning, and provides a situation recognition model based on the learning result to the ship situation recognition device. The above-described ship situation awareness device is an AI autonomous learning system for ship situation awareness characterized by generating the above-described situation awareness result data in real time by utilizing sensor data from each sensor provided in the sensing unit based on the above-described situation awareness model.
  2. In paragraph 1, The above data labeling device is characterized by performing non-real-time labeling on sensor data collected from the sensing unit using current data sensed up to the point of labeling and future data sensed after the point of labeling. This describes an AI autonomous learning system for ship situation awareness.
  3. In paragraph 2, The above data labeling device is an AI autonomous learning system for ship situation awareness characterized by automatically performing data labeling tasks using an unsupervised learning algorithm independent of ship situation awareness.
  4. In any one of paragraphs 1 through 3, The above automatic learning device includes a data analysis unit that analyzes the situation recognition result data and the labeling result data to determine whether the data requires learning, and It includes a self-learning unit that automatically learns data determined by the data analysis unit as data requiring learning and updates the situation recognition model, and An AI autonomous learning system for ship situation recognition, characterized in that the above data analysis unit determines that the above labeled result data is data requiring learning when a difference occurs between the above data as a result of comparing the above situation recognition result data and the above labeled result data.
  5. In paragraph 4, The above-described self-learning unit stores the labeling result data determined to be data requiring learning in a database, automatically learns and updates the situation recognition model when the amount of data requiring learning reaches a preset amount, and provides the updated situation recognition model to the ship situation recognition device, characterized by an AI autonomous learning system for ship situation recognition.
  6. (a) A step of sensing data to recognize the surrounding conditions of the vessel using a sensing unit, (b) A step of recognizing the surrounding situation of a ship based on artificial intelligence regarding sensor data sensed from heterogeneous sensors provided in the sensing unit of the ship situation recognition device, (c) A step of performing labeling by marking an object to be recognized based on the sensor data in a data labeling device and (d) a step of automatically performing learning by analyzing the situation recognition result data recognized in step (b) and the labeling result data labeled in step (c) in an automatic learning device, and providing a situation recognition model based on the learning result to the ship situation recognition device. AI autonomous learning method for ship situation recognition, characterized in that in step (b) above, the ship situation recognition device generates the situation recognition result data in real time by utilizing sensor data from each sensor provided in the sensing unit based on the situation recognition model.
  7. In paragraph 6, AI autonomous learning method for ship situation recognition, characterized in that, in step (c) above, the data labeling device performs non-real-time labeling by utilizing current data sensed up to the point of labeling the sensor data collected from the sensing unit and future data sensed after the point of labeling.
  8. In Paragraph 7, AI autonomous learning method for ship situation awareness, characterized in that in step (c) above, the data labeling device automatically performs data labeling tasks by utilizing a non-supervised learning algorithm independent of ship situation awareness.
  9. In any one of paragraphs 6 through 8, The above step (d) includes (d1) a step of analyzing the situation recognition result data and the labeling result data using a data analysis unit provided in the automatic learning device to determine whether the data requires learning, and (d2) includes a step of automatically learning the data determined to be data requiring learning in step (d1) using a self-classing unit to update the situation recognition model, and AI autonomous learning method for ship situation recognition, characterized in that in step (d1) above, the data analysis unit compares the situation recognition result data and the labeling result data, and if a difference occurs between the data, the labeling result data is determined to be data requiring learning.
  10. In Paragraph 9, AI autonomous learning method for ship situation recognition, characterized in that in step (d2) above, the self-learning unit stores the labeling result data determined to be data requiring learning in a database, automatically learns and updates the situation recognition model when the amount of data requiring learning reaches a preset amount, and provides the updated situation recognition model to the ship situation recognition device.

Description

AI Self-Learning System and Method for Ship Situational Awareness The present invention relates to an autonomous vessel, and more specifically, to an AI autonomous learning system and method for recognizing the vessel situation through artificial intelligence autonomous learning on data collected during operation in an autonomous vessel. Generally, autonomous vessels are equipped with cameras and various sensors in addition to the basic navigation equipment found on conventional ships. During operation, the onboard cameras, sensors, and navigation equipment collect video and various data regarding the vessel's surroundings. Based on the collected data, situational awareness information is generated, and autonomous navigation is performed based on this information. In addition, while autonomous vessels are fundamentally operated on the premise of autonomous navigation, manual operation by crew members on board or remote control via a land-based remote control center is required in certain situations, such as during docking or undocking, or when a collision risk is determined. Here, for manual operation or remote control of a vessel, a person must directly verify situational awareness information, including video of the vessel's surroundings and various sensor data, to control the vessel's operation. As such, technology for perceiving surrounding conditions is essential for the autonomous and safe operation of ships. Context-aware technology utilizes sensors capable of detecting external information, such as radar (Radio Detection and Ranging, RADAR), lidar (Light Detection and Ranging or Laser Imaging, Detection and Ranging, LIDAR/LiDAR), cameras, and the Automatic Identification System (AIS). For example, the following patent documents 1 and 2 disclose a situation awareness technology applied to an autonomous vessel according to the prior art. Meanwhile, with the recent development of Artificial Intelligence (AI) technology, situational awareness technologies utilizing AI models are being developed. The aforementioned situational awareness refers to identifying the presence and location of surrounding vessels, objects, land, etc. Therefore, a learning-based AI method was proposed using situational awareness technology that utilizes heterogeneous sensors. In other words, conventional situational awareness technology utilized supervised learning to identify targets, where a person informs the user of the target's existence in advance and the user learns and generalizes the relevant information. This supervised learning method requires a 'data set' stored from previously collected data, including human cognitive results. As such, since the aforementioned situational awareness is a technology that perceives surrounding conditions based on criteria set by humans, situational awareness utilizing AI models is gaining attention. On the other hand, AI-based situational awareness techniques have an inherent problem in that they require training an AI model. In other words, the AI model can make appropriate judgments only if the domain domain of the data provided for training matches the domain domain of the data collected from ships in actual operation. However, the domain generalized from previously collected data does not match the domain of the actual operating environment. This domain mismatch can cause performance degradation in the operating environment of the learned situational awareness solution. Consequently, conventional methods involve collecting and learning from various data, with the expectation that the learned domain will match the actual operational domain. This approach requires a large amount of data and necessitates data collected from an environment similar to the operating environment. As a solution to this, domain matching is sometimes achieved by performing learning based on actual operational data. In this case, additional human effort is required, such as collecting operational data and incorporating the perceptions of administrators or developers. In order to reduce the aforementioned domain difference, a method of retraining using data collected at the time of operation has been proposed; however, the aforementioned retraining method had the problem of being difficult to implement periodic AI model training because it required a large workforce, such as managers or developers, to intervene and train after collecting operational data. In other words, conventional ship situation awareness technology performs learning using existing data rather than the environment installed on the ship; therefore, a problem existed where performance degradation occurred in the situation awareness solution due to a domain mismatch with the actual environment. To address these issues, when performing learning using data collected during the operation process, there was a problem of reduced efficiency because human intervention was required to label and learn the collected data, which necessitated a large work