Ship water gauge identification method, system and equipment based on deep learning

Abstract

The invention discloses a ship water gauge identification method, system and equipment based on deep learning. The method comprises the following steps: acquiring a real-time image of a ship water gauge; preprocessing the real-time image of the ship water gauge; inputting the preprocessed real-time image of the ship water gauge into a pre-trained deep learning network model, and identifying the water line and the water gauge scale of the ship water gauge by using the deep learning network model to obtain the real-time draught state of the ship; and storing the identified water line and the water gauge scale of the ship water gauge in a database. According to the application, the deep learning network model is adopted, and the ship water gauge scale and the waterline can be accurately identified in a complex environment; through an image preprocessing technology, identification of the ship water gauge can adapt to various light conditions and complex sea conditions, stable identification in severe weather is guaranteed, and environmental adaptability is improved. In addition, historical data are stored through a cloud database, real-time query and data backtracking are supported, and traceability and accuracy are improved.

Description

Technical Field

[0001] The present application relates to the field of ship transportation technology, and in particular to a ship water level identification method, system and device. Background Art

[0002] The ship's draft gauge is a device used to display the draft of a ship, and is an important basis for monitoring ship load and navigation safety. In modern shipping and port management, the reading of the ship's draft gauge directly affects the efficiency of port loading and unloading operations, navigation safety, cargo weight measurement, port charges and other key links. Therefore, accurate and real-time acquisition of ship's draft gauge data is crucial to ensure efficient port management, shipping safety and accuracy of trade settlement.

[0003] Existing ship water gauge identification mostly relies on manual reading and laser radar ranging systems. Manual identification usually involves the crew manually reading the scale value of the water gauge, which is simple and intuitive. However, manual identification relies on manual reading, which is not only inefficient, but also susceptible to human subjective errors. At the same time, it is greatly affected by the environment. External factors such as light changes and water wave disturbances will bring uncertainty to manual identification. In addition, under severe weather conditions, manual operation has certain safety risks.

[0004] Some ports have adopted automation technologies such as LiDAR systems. LiDAR systems use laser beam ranging to measure the height difference between the water gauge and the water surface. However, the detection accuracy of LiDAR systems is also easily affected by external environmental interference and is greatly affected by the environment. In addition, equipment such as LiDAR is expensive, with high equipment costs and complex maintenance, making it difficult to use widely.

[0005] Whether it is manual recognition or laser-based automatic recognition, the recognition accuracy will be significantly reduced in bad weather and when the light changes greatly, and the environmental adaptability is poor. In addition, the existing systems often lack effective real-time data storage and management functions, lack data retention mechanisms, and cannot trace back and verify historical data. Summary of the invention

[0006] The present application provides a ship draft gauge recognition method, system and device based on deep learning, aiming to solve the technical problem of poor environmental adaptability of existing ship draft gauge recognition methods.

[0007] In a first aspect, a ship water gauge recognition method based on deep learning comprises:

[0008] Get real-time images of the ship's water gauge;

[0009] Preprocess the real-time image of the ship's water gauge;

[0010] The pre-processed real-time image of the ship draft gauge is input into the pre-trained deep learning network model, and the deep learning network model is used to identify the water level line and the draft gauge scale of the ship draft gauge to obtain the real-time draft status of the ship;

[0011] The identified water level line and the water level scale of the ship are stored in the database.

[0012] In the above scheme, optionally, preprocessing the real-time image of the ship water gauge includes:

[0013] Remove noise from real-time images of ship water gauges;

[0014] Perform illumination compensation on real-time images except for ship water gauge;

[0015] Perform water surface fluctuation processing on real-time images except for ship water gauge;

[0016] Perform edge enhancement on real-time images except for ship water gauges.

[0017] In the above scheme, optionally, the deep learning network model is a convolutional neural network model combined with a YOLO model or a Mask R-CNN model.

[0018] In the above solution, optionally, the training process of the deep learning network model includes:

[0019] Collect image sample data of ship water gauge;

[0020] Perform data annotation for each image in the image sample data;

[0021] Preprocess each image in the image sample data;

[0022] Dividing the image sample data into a training set, a validation set, and a test set, and using the training set to train a deep learning network model;

[0023] Evaluate the performance of deep learning network models and adjust the hyperparameters of deep learning network models based on the evaluation results.

[0024] In the above scheme, further optionally, the image sample data of the ship water gauge includes images under different lighting conditions, sea surface fluctuations, and weather changes; during the training process, the weights are optimized by the optimization algorithm to minimize the loss function, and the recognition accuracy of the water gauge scale and water level line is gradually improved; when evaluating the performance of the deep learning network model, the accuracy, recall rate or F1 score is used to evaluate the performance of the model in the water gauge scale and water level line recognition task.

[0025] In the above solution, optionally, the method further includes:

[0026] Determine whether the ship is abnormal based on its real-time draft status;

[0027] If the ship is judged to be abnormal, an early warning message will be output.

[0028] In the above scheme, further optionally, when it is judged that the ship is overloaded or unloaded unevenly based on the draft state of the ship, it is determined that the ship is abnormal.

[0029] In the second aspect, a ship water gauge recognition system based on deep learning includes:

[0030] An image acquisition module is used to acquire a real-time image of the ship's water gauge;

[0031] An image preprocessing module is used to preprocess the real-time image of the ship's water gauge;

[0032] The draft gauge recognition module is used to input the pre-processed real-time image of the ship draft gauge into the pre-trained deep learning network model, and use the deep learning network model to recognize the water level line and the draft gauge scale of the ship draft gauge to obtain the real-time draft status of the ship;

[0033] The data storage module is used to store the identified water level line and the water gauge scale of the ship water gauge in a database.

[0034] In a third aspect, a ship water gauge recognition device based on deep learning includes:

[0035] An image acquisition device is installed at the dock to capture real-time images of the ship's water level;

[0036] A computer device on which the ship water level recognition system based on deep learning described in the second aspect is arranged.

[0037] In a fourth aspect, a computer program product comprises a computer program / instruction, wherein the computer program / instruction implements the steps of the method described in the first aspect when executed by a processor.

[0038] Compared with the prior art, this application has at least the following beneficial effects:

[0039] The embodiment of the present application provides a method for identifying a ship's draft gauge using a deep learning network model, by acquiring a real-time image of the ship's draft gauge, preprocessing the real-time image of the ship's draft gauge, inputting the preprocessed real-time image of the ship's draft gauge into a pre-trained deep learning network model, using the deep learning network model to identify the water level line and the draft gauge scale of the ship's draft gauge, obtaining the real-time draft status of the ship, and storing the identified water level line and the draft gauge scale of the ship's draft gauge in a database, so that the scale and water level line on the draft gauge can be identified using the deep learning network model, the draft depth (draft status) of the ship can be output in real time, and the data can be stored in a cloud database for subsequent query and analysis; the present application adopts a deep learning network model, which can accurately identify the ship's draft gauge scale and waterline in a complex environment; before the real-time image of the ship's draft gauge is input into the deep learning network model, the image preprocessing technology is used to enable the recognition of the ship's draft gauge to adapt to various light conditions and complex sea conditions, ensure stable recognition in bad weather, and improve environmental adaptability; in addition, historical data is saved in a cloud database, real-time query and data backtracking are supported, and traceability and accuracy are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of a process flow of a ship water gauge recognition method based on deep learning provided in one embodiment of the present application;

[0041] Figure 2 Another schematic diagram of a ship water gauge recognition method based on deep learning provided in one embodiment of the present application;

[0042] Figure 3 A block diagram of the module architecture of a ship water gauge identification system based on deep learning is provided for one embodiment of the present application. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0044] In the description of the present application: unless otherwise specified, expressions such as “include”, “comprising”, “having”, etc. also mean “not limited to” (certain units, components, materials, steps, etc.).

[0045] In one embodiment, Figure 1 As shown, a ship water gauge recognition method based on deep learning is provided, comprising the following steps:

[0046] S1, obtain the real-time image of the ship's water gauge.

[0047] Specifically, high-definition cameras and sensors installed at the dock can be used to capture real-time images of the ship's water level. Image acquisition requires high resolution and wide dynamic range to ensure that the details of the water level can be accurately captured under various lighting conditions.

[0048] Step S1 is to perform image acquisition: obtain the ship's water level image through a high-definition camera, and transmit the data to the processing system.

[0049] S2, preprocessing the real-time image of the ship water gauge.

[0050] After the image is acquired, it is first necessary to process the image to improve the image quality and make the scale and water level line on the water ruler clearer.

[0051] Step S2 is to perform image preprocessing: using techniques such as denoising, illumination compensation and edge enhancement to ensure that the water gauge scale and water level line are clearly visible in the image.

[0052] Specifically, the preprocessing of the real-time image of the ship water gauge includes:

[0053] (1) Remove noise from the real-time image of the ship's water level gauge to reduce unnecessary influence.

[0054] The noise in the image may come from environmental interference, camera sensor limitations, improper shooting angles, etc. Noise will affect the clarity of the water gauge scale and water level line. Common noise removal methods are:

[0055] Gaussian Blur: used to smooth images and reduce noise;

[0056] Median Filtering: Removes salt and pepper noise by taking the median of neighboring pixels and retains edge information;

[0057] Bilateral Filtering: It not only smoothes the image, but also preserves the edge information of the image.

[0058] (2) Perform illumination compensation on real-time images except for the ship water gauge to improve the loss or blur of details in the image caused by insufficient or excessive light.

[0059] Due to changes in light in the shooting environment, the brightness of the image may be uneven, resulting in overexposure or darkening of some areas, affecting the recognition effect. In order to ensure that the image can be clearly recognized under various lighting conditions, the commonly used lighting compensation methods are:

[0060] Histogram Equalization: By adjusting the contrast and brightness of the image, the brightness distribution of the image is made more uniform and the details of the image are enhanced;

[0061] Adaptive Histogram Equalization (CLAHE): Similar to traditional histogram equalization, but using local area equalization, it can adaptively adjust the illumination in different local areas and is more suitable for complex lighting conditions;

[0062] Brightness and Contrast Adjustment: Adjust the brightness and contrast of an image based on its global or local brightness to ensure that details in the image stand out more clearly.

[0063] (3) Processing the water surface fluctuations of the real-time images except the ship water gauge;

[0064] Water surface fluctuations or ship shaking may cause ripple effects in the water gauge image, affecting recognition. To solve this problem, commonly used processing methods include:

[0065] Water wave removal algorithm: Remove the interference of water surface fluctuations through image processing techniques such as wavelet transform or deep learning-based denoising methods.

[0066] (4) The real-time image of the ship’s water gauge is enhanced by edge enhancement to enhance the edges of the water gauge scale and water level line, making it easier for the deep learning algorithm to identify.

[0067] Image sharpening can increase the contrast of the image, especially the scale of the water gauge and the edge of the water level line, making these key parts more obvious. Commonly used sharpening methods include:

[0068] Laplacian Operator: A common edge detection method that can enhance details and edges in images;

[0069] Sobel operator: used to detect horizontal and vertical edges of the image and help highlight the scale lines of the water ruler;

[0070] Unsharp Masking (USM): Enhance the edges of the image and the scales and water level lines on the water ruler through high-pass filtering.

[0071] S3, input the preprocessed real-time image of the ship draft gauge into the pre-trained deep learning network model, use the deep learning network model to identify the water level line and the draft gauge scale of the ship draft gauge, and obtain the real-time draft status of the ship.

[0072] Among them, the deep learning network model uses a convolutional neural network model combined with the YOLO model or the Mask R-CNN model. YOLO is a real-time target detection algorithm that divides the entire image into multiple grids and predicts the object category and bounding box in each grid. The advantage of YOLO lies in its high efficiency. It can complete image recognition in a very short time and is suitable for real-time systems. Mask R-CNN is a model for object detection and instance segmentation. In addition to providing category labels and bounding boxes, it can also generate accurate pixel-level segmentation masks for each target, which is very effective for fine feature recognition such as water gauge scale lines.

[0073] Step S3 is to perform water gauge recognition: the scale lines and waterline on the water gauge are automatically recognized through the deep learning model to calculate the draft of the ship.

[0074] Furthermore, the training process of the deep learning network model includes:

[0075] (1) Collect image sample data of ship water gauge.

[0076] Collect a large amount of sample data with ship water level images. These images can be from real port environments or generated through simulation. The image sample data should include different lighting conditions, sea surface fluctuations, weather changes, etc. to ensure that the model has strong environmental adaptability.

[0077] (2) Perform data labeling on each image in the image sample data.

[0078] In these images, the scale lines and water level lines of the water gauge are annotated so that the deep learning model can learn how to recognize these features. For each image, experts or annotators need to accurately mark the positions of the scale lines and water level lines of the water gauge, usually by box selection, key point annotation or pixel-level annotation methods. The annotated data will be used as training "labels" to guide model learning.

[0079] (3) Preprocess each image in the image sample data.

[0080] In the data preparation stage, the original images are processed to remove noise, enhance lighting, and repair the influence of factors such as water surface fluctuations.

[0081] (4) The image sample data is divided into a training set, a validation set, and a test set. The training set is used to train the deep learning network model, the validation set is used to adjust the parameters, and the test set is used to evaluate the final performance of the model.

[0082] During the training process, the weights are optimized through the optimization algorithm to minimize the loss function, and the recognition accuracy of the water gauge scale and water level line is gradually improved.

[0083] The deep learning model is trained using labeled image data. During the training process, the model can optimize its weights through the back-propagation algorithm, gradually improving the recognition accuracy of the water gauge scale and water level line. In order to prevent overfitting, cross-validation and data enhancement (such as random rotation, scaling, cropping, etc.) can be used.

[0084] Using appropriate loss functions during training, such as cross entropy loss (for classification problems) and mean squared error (MSE, for regression problems), or combining multi-task losses for object detection tasks (such as location loss and classification loss), can help the model accurately locate the scale lines and water level lines of the water gauge during training.

[0085] Commonly used optimization algorithms include Adam, SGD, etc., which minimize the loss function by gradually updating the weights.

[0086] (5) Evaluate the performance of the deep learning network model and adjust the hyperparameters of the deep learning network model based on the evaluation results.

[0087] When evaluating the performance of the deep learning network model, precision, recall or F1 score is used to evaluate the model's performance in the water gauge scale and water level line recognition tasks.

[0088] During the training process, indicators such as precision, recall, and F1 score are used to evaluate the performance of the model in the water gauge scale and water level line detection tasks. If an object detection model is used, indicators such as AP (Average Precision) can also be used to evaluate the detection effect of the model.

[0089] According to the evaluation results, adjust the model's hyperparameters (such as learning rate, batch size, number of network layers, etc.) and further optimize the model. If it is found that the model is not effective under certain specific conditions (such as low light, turbulent sea surface, etc.), it may be necessary to solve it by adding corresponding training data.

[0090] In other words, during the entire training process, the deep learning model learns the features of the water gauge scale and water level line through a large amount of labeled data, and achieves high-precision recognition through the improved YOLO or Mask R-CNN model. Through data preprocessing, model training and tuning, stability and efficiency can be ensured in complex environments, thereby providing reliable technical support for the automatic recognition of ship water gauges. Specifically, the water gauge scale and water level line can be detected and recognized based on the improved YOLOV8 model.

[0091] S4, storing the identified water level line and the water level scale of the ship water level in a database.

[0092] The identified data is automatically stored in the database for subsequent tracing and analysis.

[0093] Step S4 is data storage: the identified data is automatically stored in a database for real-time monitoring and historical data tracing.

[0094] Furthermore, the method further comprises:

[0095] Determine whether the ship is abnormal based on its real-time draft status;

[0096] If the ship is judged to be abnormal, an early warning message will be output.

[0097] Among them, when it is judged that the ship is overloaded or unloaded unevenly based on the draft status of the ship, it is determined that there is an abnormality in the ship.

[0098] In other words, monitor the ship’s draft in real time, detect abnormal conditions (such as overloading), and provide early warning.

[0099] This step is anomaly detection and early warning: the system continuously monitors the ship's draft status and automatically issues an early warning if overloading or other abnormal conditions occur.

[0100] Another schematic diagram of the method provided in the embodiment of the present application can be seen in Figure 2 .

[0101] The embodiment of the present application provides a ship draft gauge recognition method based on computer vision, which uses a deep learning algorithm to solve the problems in the current technology. The ship draft gauge recognition method based on deep learning provided by the embodiment of the present application can automatically capture the ship draft gauge image, identify the scale and water level on the draft gauge, output the ship's draft depth (draft status) in real time, and store the data in a cloud database for subsequent query and analysis.

[0102] Through image enhancement, noise suppression and other processing, the water gauge image can be clarified, making the water level and scale in the image more obvious, thereby reducing the impact of environmental changes (such as uneven lighting, waves or water surface reflections, etc.) on image quality. In this way, the deep learning model can learn and recognize more efficiently on relatively stable images.

[0103] By learning a large number of water gauge images, the deep learning model can adaptively identify water level lines and scales in different environments. Compared with manual or radar recognition methods, the deep learning model can automatically extract features from images and adapt to various complex backgrounds and conditions, showing stronger environmental adaptability.

[0104] Traditional methods usually perform unstably in different environments and are easily limited by specific lighting, weather, etc. However, deep learning models, especially those trained with a large amount of data, can automatically learn a variety of features and have strong generalization capabilities, so they can cope with different environmental changes (such as day and night, sunny and rainy days, etc.).

[0105] The model can recognize water gauge images in a variety of environments, and can make relatively accurate judgments even in bad weather or special circumstances (such as light reflection, rain interference, etc.). For example, through training under a large number of different conditions, the deep learning model can "learn" how to distinguish water levels under wave disturbances and how to identify water gauge scales under different lighting conditions. Through technologies such as transfer learning, the trained deep learning model can be transferred and used in different environments without having to start training from scratch each time, which greatly improves the efficiency and practicality of the model.

[0106] Traditional water gauge reading methods are usually manual and rely on manual reading, with low requirements for real-time and automation. This solution achieves automation and real-time by inputting the real-time image of the ship's water gauge into the deep learning network model: the deep learning model is efficient in image processing, can quickly process real-time image data, and instantly output the recognition results of water level lines and scales, thereby meeting the needs of real-time monitoring during the ship's navigation.

[0107] The identified water level and water gauge data are directly stored in the database, which not only improves the automation of data collection, but also facilitates subsequent data analysis, historical record query and tracing of ship draft status. Compared with the manual recording of traditional methods, the deep learning-based method can achieve efficient data accumulation and management, which is convenient for ports and shipping companies to conduct subsequent analysis and decision-making.

[0108] In addition, with continuous use, the deep learning model can be continuously trained and optimized through new data, so that the accuracy and environmental adaptability of the system are continuously improved. Compared with traditional recognition methods, the deep learning-based system can automatically adjust its recognition strategy and algorithm parameters according to new environmental data, thereby continuously improving the accuracy of recognition and adapting to changing environmental conditions.

[0109] The environmental adaptability of the ship water gauge recognition system can be significantly improved through the deep learning-based water gauge recognition method. This method combines image preprocessing, the powerful feature learning and automatic recognition capabilities of the deep learning model, and real-time data storage and management functions, fundamentally solving the problems of poor adaptability and low accuracy of traditional methods in different environments and conditions. In this way, the real-time monitoring of ship water gauges can be ensured to be more accurate and reliable, and human errors can be reduced, thereby improving overall shipping safety and port operation efficiency.

[0110] In summary, the ship water gauge recognition method based on deep learning provided in the embodiment of the present application has at least the following effects:

[0111] (1) High-precision recognition: Using advanced deep learning models such as improved YOLOV8, it can accurately identify ship water gauge scales and waterlines in complex environments.

[0112] (2) Strong environmental adaptability: Through image preprocessing and enhancement technology, it can adapt to various lighting conditions and complex sea conditions, ensuring stable recognition in severe weather.

[0113] (3) Automation and real-time monitoring: It realizes the automatic identification of ship water gauges and real-time monitoring of data, greatly reducing the need for manual operation.

[0114] (4) Data retention and management: Historical data is stored in the cloud database, supporting real-time query and data backtracking, improving traceability and accuracy.

[0115] (5) Intelligent early warning: It can automatically detect anomalies, such as overloading or uneven unloading, and issue early warnings in a timely manner to ensure the safety and efficiency of port operations.

[0116] In one embodiment, Figure 3 As shown, a ship water gauge recognition system based on deep learning is provided, which includes the following program modules:

[0117] An image acquisition module 301 is used to acquire a real-time image of a ship's water gauge;

[0118] The image preprocessing module 302 is used to preprocess the real-time image of the ship's water gauge; perform denoising, illumination compensation and edge enhancement on the collected image to ensure clear presentation of the water gauge scale and water level line;

[0119] The draft gauge recognition module 303 is used to input the pre-processed real-time image of the ship's draft gauge into a pre-trained deep learning network model, and use the deep learning network model to recognize the water level line and the draft gauge scale of the ship's draft gauge to obtain the real-time draft status of the ship; based on the deep learning algorithm, the draft gauge scale and the water level line are automatically recognized to calculate the draft depth of the ship;

[0120] The data storage module 304 is used to store the identified water level and the water gauge scale of the ship in a database; store the identified water gauge data, and support historical data backtracking.

[0121] Furthermore, the water gauge recognition module uses a convolutional neural network (CNN) combined with a YOLO or Mask R-CNN model to recognize the water gauge scale and waterline.

[0122] Furthermore, the image preprocessing module includes noise removal, brightness enhancement and water surface fluctuation processing to cope with image acquisition under complex environmental conditions.

[0123] Furthermore, the system also includes an intelligent monitoring and early warning module, which is used to determine whether the ship is abnormal according to the real-time draft status of the ship; if the ship is determined to be abnormal, an early warning message is output. That is, the intelligent monitoring and early warning module monitors the draft depth of the ship in real time and automatically issues an early warning when an abnormal situation is detected.

[0124] The specific implementation content of each module can also be found in the above definition of a ship water level recognition method based on deep learning, which will not be repeated here.

[0125] As above, the ship water gauge recognition system based on deep learning provided in the embodiment of the present application has at least the following effects:

[0126] (1) High-precision recognition: Using advanced deep learning models such as improved YOLOV8, it can accurately identify ship water gauge scales and waterlines in complex environments.

[0127] (2) Strong environmental adaptability: Through image preprocessing and enhancement technology, the system can adapt to various lighting conditions and complex sea conditions, ensuring stable recognition in severe weather.

[0128] (3) Automation and real-time monitoring: The system realizes the automatic identification of ship water gauges and real-time monitoring of data, greatly reducing the need for manual operation.

[0129] (4) Data retention and management: Historical data is stored in the cloud database, supporting real-time query and data backtracking, which improves the traceability and accuracy of the system.

[0130] (5) Intelligent early warning: The system can automatically detect anomalies, such as overloading or uneven unloading, and issue early warnings in a timely manner to ensure the safety and efficiency of port operations.

[0131] In one embodiment, a ship water gauge recognition device based on deep learning is provided, which includes:

[0132] An image acquisition device is installed at the dock to capture real-time images of the ship's water level;

[0133] A computer device is provided with a ship water gauge identification system based on deep learning provided in the above embodiment.

[0134] In other words, the ship water gauge identification equipment includes the following main parts:

[0135] Image acquisition device: High-definition cameras and sensors installed at the dock are used to capture real-time images of the ship's water gauge.

[0136] Image preprocessing module: This module is used to deal with problems such as noise and light changes in the image to ensure that the acquired water level image is clear and visible.

[0137] Deep learning recognition algorithm: Detection and recognition of water gauge scales and water level lines based on the improved YOLOV8 model.

[0138] Data storage and management module: The identified data is automatically stored in the database for subsequent tracing and analysis.

[0139] Intelligent monitoring and early warning system: monitors the ship's draft in real time, detects abnormal conditions (such as overloading), and provides early warnings.

[0140] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which involves all or part of the processes in the above-mentioned embodiment method.

[0141] In one embodiment, a computer program product is also provided, including a computer program / instruction, which implements all or part of the process in the above-mentioned embodiment method when executed by a processor.

[0142] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A ship water gauge recognition method based on deep learning, characterized in that: include: Get real-time images of the ship's water gauge; Preprocess the real-time image of the ship's water gauge; The pre-processed real-time image of the ship draft gauge is input into the pre-trained deep learning network model, and the deep learning network model is used to identify the water level line and the draft gauge scale of the ship draft gauge to obtain the real-time draft status of the ship; The identified water level line and the water level scale of the ship are stored in the database.

2. The ship water gauge recognition method based on deep learning according to claim 1 is characterized in that: The preprocessing of the real-time image of the ship water gauge includes: Remove noise from real-time images of ship water gauges; Perform illumination compensation on real-time images except for ship water gauge; Perform water surface fluctuation processing on real-time images except for ship water gauge; Perform edge enhancement on real-time images except for ship water gauges.

3. The ship water gauge recognition method based on deep learning according to claim 1 is characterized in that: The deep learning network model is a convolutional neural network model combined with a YOLO model or a Mask R-CNN model.

4. The ship water gauge recognition method based on deep learning according to claim 1 is characterized in that: The training process of the deep learning network model includes: Collect image sample data of ship water gauge; Perform data annotation for each image in the image sample data; Preprocess each image in the image sample data; Dividing the image sample data into a training set, a validation set, and a test set, and using the training set to train a deep learning network model; Evaluate the performance of deep learning network models and adjust the hyperparameters of deep learning network models based on the evaluation results.

5. The ship water gauge recognition method based on deep learning according to claim 4 is characterized in that: The image sample data of ship water gauges contains images under different lighting conditions, sea surface fluctuations, and weather changes. During the training process, the weights are optimized through the optimization algorithm to minimize the loss function, and the recognition accuracy of water gauge scales and water level lines is gradually improved. When evaluating the performance of the deep learning network model, the accuracy, recall rate or F1 score is used to evaluate the performance of the model in the water gauge scale and water level line recognition task.

6. The ship water gauge recognition method based on deep learning according to claim 1 is characterized in that: The method further comprises: Determine whether the ship is abnormal based on its real-time draft status; If the ship is judged to be abnormal, an early warning message will be output.

7. The ship water gauge recognition method based on deep learning according to claim 6 is characterized in that: When it is judged that the ship is overloaded or unloaded unevenly based on the draft state of the ship, it is determined that the ship is abnormal.

8. A ship water gauge recognition system based on deep learning, characterized in that: include: An image acquisition module is used to acquire a real-time image of the ship's water gauge; An image preprocessing module is used to preprocess the real-time image of the ship's water gauge; The draft gauge recognition module is used to input the pre-processed real-time image of the ship draft gauge into the pre-trained deep learning network model, and use the deep learning network model to recognize the water level line and the draft gauge scale of the ship draft gauge to obtain the real-time draft status of the ship; The data storage module is used to store the identified water level line and the water gauge scale of the ship water gauge in a database.

9. A ship water gauge recognition device based on deep learning, characterized in that: include: An image acquisition device is installed at the dock to capture real-time images of the ship's water level; A computer device on which the ship water gauge identification system based on deep learning as described in claim 8 is arranged.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

Images