A remote measurement method of ship water gauge based on drone

Abstract

The present invention discloses a method for remotely measuring a ship's draft gauge based on an unmanned aerial vehicle (UAV). The method uses a draft gauge measurement system primarily composed of a web-based control terminal, a terminal, an algorithm terminal, and a database to detect draft gauge draft values. The drone first collects raw video containing the water surface area and draft gauge, then filters the raw video at the terminal to obtain valid images. The algorithm terminal then analyzes the valid images using an image recognition method to obtain the draft gauge draft value. The web-based control terminal manages users, data, and permissions, issues task instructions to the terminal, and receives calculation results from the algorithm terminal. This method utilizes the high efficiency and flexibility of UAVs, combines deep learning technologies such as target detection, character recognition, and target segmentation, and performs optimization and improvements. This method can address deficiencies in the safety and convenience of draft gauge measurement operations and produce more accurate measurement results. Experiments have shown that the method has high operating speed, high detection accuracy, and flexible detection areas.

Description

Technical Field

[0001] The invention relates to a remote measurement method of a ship draft gauge based on an unmanned aerial vehicle (UAV), which is a ship draft gauge scale detection technology. Background Art

[0002] Vessel draft gauge weighing is the most widely used measurement method in the bulk cargo shipping industry both domestically and internationally. Reading the draft values ​​on a vessel's six-sided draft gauge is a crucial factor in its accuracy. Due to the lack of specialized draft gauge weighing instruments, the current method of reading draft values ​​is primarily based on visual observation. This method is subject to subjective factors, the inability to record data during observation, and the limited number of observation points. This results in inaccurate and imprecise readings. Furthermore, the inconvenient observation environment increases safety risks for personnel, resulting in heavy workload and high labor costs.

[0003] In addition to traditional visual inspection methods, there are currently many solutions to try to solve this problem, such as installing dual pressure sensors to detect the draft depth of the ship's side, emitting ultrasonic waves through ultrasonic sensors and calculating the waterline position information based on the time required for the ultrasonic waves to reflect from the water surface back to the deck, and directly using laser ranging sensors to directly obtain the water surface position. These methods are all based on physical sensors, and have problems such as difficult installation and construction, and accuracy that is easily affected by the environment.

[0004] Currently, there are some methods for automatically reading draft gauges based on image and video algorithms. These image recognition methods primarily include modules for waterline positioning, character recognition, and numerical value estimation. Traditional image recognition methods first detect the waterline through edge processing, then use image binarization to locate the scale digits. Finally, they use template matching and other methods for recognition, and finally estimate the draft gauge scale value through fitting. This method has poor generalization and robustness, is limited to certain draft gauge scenarios, and is susceptible to factors such as hull rust and the waterline. Consequently, the accuracy of both waterline detection and digit recognition cannot be guaranteed.

[0005] In recent years, related methods based on deep learning have also been proposed. This type of method uses deep learning network training to realize scale character recognition and positioning, and uses image segmentation network to locate waterline position information. The two are combined to estimate the waterline using methods such as least squares method. The problem with this type of method is that the misrecognition rate of character positioning and recognition for the entire image is high, and the bounding box of character positioning is not fit enough, which will cause large reading errors. In addition, this type of method cannot solve the problem of water gauge tilt and rotation angle in the image. Moreover, due to the non-planar rotation and tilt angle of the hull itself, the value of the scale characters and their spacing are nonlinear, which easily increases the reading error. Summary of the Invention

[0006] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a method for remote measurement of ship draft gauges based on drones to improve the accuracy of port draft gauge readings and improve the efficiency of this work.

[0007] Technical solution: To achieve the above purpose, the technical solution adopted by the present invention is:

[0008] A method for remote measurement of ship draft gauges based on drones is proposed. The draft gauge measurement system mainly composed of a web master control terminal, a terminal, an algorithm terminal and a database is used to detect the draft gauge draft value. First, the drone collects the original video containing the water surface area and the draft gauge. Then, the original video is filtered at the terminal to obtain valid images. Then, the algorithm terminal analyzes the valid images through an image recognition method to obtain the draft gauge draft value. The web master control terminal manages users, data and permissions, issues task instructions to the terminal and receives the calculation results of the algorithm terminal.

[0009] The water gauge measurement system adopted by the present invention generally uses a WEB master terminal for administrators, and a mobile terminal (such as a mobile phone or a laptop) as a terminal for on-site operators. For a specific water gauge draft numerical measurement, the process is roughly as follows: first, the administrator creates a task on the WEB master terminal and sends it to the on-site operator; second, the on-site operator views his or her own work task on the mobile terminal, clicks on the corresponding function, mobilizes the drone to collect video of the ship's water gauge, and uploads it to the server; then, the algorithm performs artificial intelligence-based recognition on the received original video, obtains the water gauge draft numerical result, and synchronizes the result to the WEB master terminal and the mobile terminal; the on-site operator can make on-site corrections to the feedback results, and the corrected results are sent to the WEB master terminal for reference; finally, the WEB master terminal writes the original video, pictures, water gauge draft numerical measurement results and correction results and other data into the database to complete the task.

[0010] Preferably, the WEB master terminal, terminal, algorithm terminal and database are completely decoupled and are all deployed in the form of a restful interface to facilitate subsequent functional expansion and upgrading.

[0011] Preferably, the original video is screened to obtain a valid picture, which meets the resolution requirements (the picture cannot be too blurry and the character area cannot be too small) and contains a water surface area and a water ruler at the same time, wherein the water surface runs across the entire valid picture and the area below the water surface occupies 1 / 3±α of the area of ​​the entire valid picture, the water ruler stands on the vertical center line of the entire valid picture, the range of the water ruler's lateral deviation from the vertical center line is limited to 10% of the width of the entire valid picture, the angle of the water ruler's deviation from the vertical center line is limited to ±30°, and at least one letter M appears in the water ruler, and the length of the water ruler revealed in the entire picture is 2 to 2.5M (that is, in the entire picture, only a small section of the water ruler in the range of 2 to 2.5M can be seen, such as 5 to 7M or 1 to 3.5M, which is revealed in the entire picture); α is a set value; the posture of the drone during collection is set based on the requirements for valid pictures.

[0012] In the prior art, water level measurement mainly relies on manual acquisition of original videos. There is no unified standard for the video shooting method, angle, clarity, etc., and the collected original videos cannot be effectively and standardizedly structured. An important design point of the present invention is that drones are used in the acquisition of original videos, so that more stringent shooting requirements can be put forward, which is convenient for later data processing and can also improve the accuracy of detection. Combined with the later requirements for effective pictures, the shooting requirements of drones can be standardized. It is more appropriate to limit the shooting requirements of drones to the requirements for effective pictures or slightly relax them. The original videos shot in accordance with these requirements will greatly improve the accuracy of the algorithm's automatic recognition and allow on-site personnel to clearly see the true reading of the water level.

[0013] Preferably, the algorithm includes a water gauge area detection module, a water gauge character positioning module, a character recognition module, a water gauge fitting module, a water surface detection module, a water gauge reading module and a water gauge reading verification module;

[0014] The water gauge area detection module extracts the water gauge area from the entire valid image;

[0015] The water ruler character positioning module locates the coordinates and boundaries of each character in the water ruler area;

[0016] The character recognition module recognizes characters in the water gauge area;

[0017] The water ruler fitting module combines the coordinates of each character and the recognition result to reconstruct the complete water ruler;

[0018] The water surface detection module detects the boundary between the water surface and the hull in the entire effective image and obtains the water surface intersection point;

[0019] The water gauge reading module reads the water gauge value by combining the reconstructed complete water gauge with the water surface intersection point;

[0020] The water gauge reading verification module combines all the read water gauge values, eliminates the error value through the fitting function, and obtains the optimized water gauge value.

[0021] Preferably, the water gauge area detection module uses the YOLO algorithm to perform target detection.

[0022] Preferably, after extracting the water gauge area, the water gauge area detection module performs image processing including edge detection, binarization, and dilation and corrosion on the water gauge area to obtain the outer rectangle of the character area, and calculates the rotation and translation matrix for tilt correction of the water gauge area based on the tilt angle and rotation center of the outer rectangle of the character area; performs tilt correction on the water gauge area according to the rotation and translation matrix and provides it to the water gauge character positioning module and the character recognition module; performs tilt correction on the entire effective image according to the rotation and translation matrix and provides it to the water surface detection module.

[0023] Preferably, the water ruler character positioning module adopts a text segmentation method based on image segmentation, that is, first classification is performed at the pixel level to determine the text target to which each pixel point belongs, and then all text targets are integrated to obtain a probability map of the text segmentation area, and finally the enclosing curve of the text segmentation area is obtained through post-processing.

[0024] The accuracy of current artificial intelligence algorithms cannot reach complete accuracy, so their recognition results are often unreliable, and the directly generated water gauge results have large errors. In addition, in dock and port environments, the water surface is not completely static, but rather exhibits irregular fluctuations. Therefore, it is necessary to reconstruct the identified water gauge and perform standardized fitting on the horizontal plane to improve the accuracy of the water gauge reading. Preferably, when performing water gauge reconstruction, the water gauge fitting module can integrate certain a priori standards into the water gauge reconstruction algorithm to make the reconstructed water gauge more standardized and accurate; these a priori standards include: 1. There should be only one "8632" between two "M" characters; 2. "8632" should be arranged in a vertical column with little difference in the distance between characters; 3. There should be one or two characters before each "M" character; 4. If there is a two-digit number before the character "M", the first digit should not be greater than 3; 5. All characters should be relatively clustered horizontally and relatively discrete vertically.

[0025] Preferably, the water surface detection module uses a two-stage image segmentation method to determine the boundary between the water surface and the hull. During the first stage of image segmentation, the entire valid image is segmented at the water surface position. During the second stage of image segmentation, a local image (e.g., a 512×512 local image) is taken upward from the water surface obtained by the first segmentation, or a local image (e.g., a 512×512 local image) is taken downward from the nearest complete character on the water surface obtained by the first segmentation. The local image is segmented again at the water surface position, and the result of the second segmentation is used as the boundary between the water surface and the hull. The two image segmentation methods can be the same or different, for example, both use the envelope algorithm mentioned in this article.

[0026] Considering that the waterline (water surface) is formed by water waves, the size of the water fluctuation affects the final value. The value obtained at the peak is higher than the true value, and the value obtained at the trough is lower than the true value. Therefore, it is necessary to process the water wave curve to obtain a more accurate waterline. Preferably, the water surface detection module uses an envelope algorithm to calculate the water surface position, specifically:

[0027] Lower envelope:

[0028] Upper envelope:

[0029] Water surface position:

[0030] Where: y(t) represents the calculated value of the water surface position at time t, y down (t) represents the corrected water surface position at time t on the lower envelope, y up (t) represents the water surface position at time t on the corrected upper envelope line; y0(t) represents the actual position of the water surface at time t, down(t) represents the fitting value of the water surface position at time t obtained by the lower envelope fitting function, and up(t) represents the fitting value of the water surface position at time t obtained by the upper envelope fitting function.

[0031] A segment of original video can be used to read out several groups of waterline results. Due to errors or errors in recognition or calculation, some of them have large errors from the true values, while some have small errors. Usually, the values ​​with large deviations account for a small proportion of the whole, and common errors are similar, so correct or incorrect data are usually concentrated around a few values. Preferably, the water surface detection module uses the K-means clustering method (also known as the k-means algorithm) to perform unsupervised learning on the values ​​of all water surface positions obtained, first find the cluster center of the values, and then delete all the values ​​of the clusters with fewer values, and repeat this cycle until the value function of the K-means clustering method is met, retaining the values ​​around the final cluster center, and using the values ​​of the water surface positions finally retained as the water surface intersection points for reading the water gauge values.

[0032] Using the algorithm of the present invention to read the draft gauge value requires high hardware resources, is time-consuming, and has complex logical relationships; therefore, it is possible to consider utilizing server performance and using a multi-channel parallel method to calculate the six-sided video of the ship, reducing the system time consumption by 6 times. Preferably, all algorithms in the draft gauge measurement system are encapsulated using a standard interface to achieve the effect of enabling one sentence of code, which is convenient for later maintenance and upgrades; for the six-sided draft gauge values ​​to be measured for the same bow ship, six groups of drones are used to simultaneously work to collect the original video of the six sides of the ship. The draft gauge measurement system uses a multi-channel parallel method to simultaneously calculate the six-sided original video and synchronously obtain the draft gauge values ​​of the six sides.

[0033] Beneficial effects: The drone-based ship draft gauge remote measurement method provided by the present invention utilizes the advantages of drones' high efficiency and flexibility, combines deep learning target detection, character recognition, target segmentation and other technologies, and optimizes and improves them. It can make up for the deficiencies in the safety and convenience of draft gauge measurement operations, and obtain more accurate measurement results. On-site workers can complete measurement tasks safely and conveniently through handheld terminals. Experiments have shown that the present invention has fast operation speed, high detection accuracy, strong timeliness, flexible detection area, reliable draft gauge collection and measurement results, and can also reduce the work pressure of on-site workers, reduce operating costs, improve overall operating efficiency, improve operating safety, and avoid errors caused by personal subjective factors, thereby avoiding the occurrence of three-party disputes. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is an architectural diagram of the water gauge measurement system of the present invention;

[0035] Figure 2 A flow chart showing how the algorithm of the present invention reads the water gauge value;

[0036] Figure 3 This is an example of the inland port identification result in the embodiment;

[0037] Figure 4 This is an example of coastal port identification results in the embodiment. DETAILED DESCRIPTION

[0038] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] A drone-based remote measurement method for ship draft gauges uses a draft gauge measurement system consisting primarily of a web-based control terminal, a handheld terminal, an algorithm, and a database to detect draft gauge draft values. The drone first collects raw video footage of the water surface area and draft gauge. The handheld terminal then filters the raw video to obtain valid images. The algorithm then analyzes the valid images using image recognition methods to obtain the draft gauge draft values. The web-based control terminal is primarily responsible for data management, result query, user management, permission management, task management, and system scheduling; the handheld terminal is primarily responsible for data collection, data processing, data transmission, and enabling identification; the algorithm is primarily responsible for data analysis, intelligent identification, and result calculation of the collected raw video; and the database is primarily responsible for structured storage of all data. The web-based control terminal, terminal, algorithm, and database are completely decoupled and deployed using RESTful interfaces to facilitate subsequent functional expansion and upgrades.

[0040] In this example, the water gauge measurement system adopts a private cloud-deployed WEB master terminal + operator handheld terminal solution, and its architecture is as follows: Figure 1 As shown. The WEB main control end serves as the overall control center of the entire water level measurement system, issuing tasks to the handheld terminal. After the operator controls the drone to complete the shooting of the original video, the original video is imported into the handheld terminal through wireless transmission or manual import, and the handheld terminal completes the screening of valid images. The filtered valid images are sent to the algorithm end for AI core calculation, and the calculation results are sent to the existing inspection platform to generate the final inspection report. AI core calculations and data storage and management are all performed on the cloud server. The main system parameters of this system are shown in Table 1:

[0041] Table 1 Main parameters of the system

[0042]

[0043] In this example, an inland port and a coastal port were selected to conduct remote measurements of ship water gauges, which were then compared with manual readings to verify the feasibility of this system.

[0044] The drone in this example sets shooting requirements based on the requirements for valid images. Specifically, the video frame meets the resolution requirements (the image cannot be too blurry, and the character area cannot be too small) and contains both the water surface area and the water ruler. The water surface runs across the entire video frame and the area below the water surface occupies 1 / 3±α of the video frame area. The water ruler stands on the vertical centerline of the video frame. The lateral deviation of the water ruler from the vertical centerline is limited to 10% of the video frame width. The angle of the water ruler from the vertical centerline is limited to ±30°. At least one letter M appears in the water ruler, and the length of the water ruler visible in the entire image is 2 to 2.5 meters.

[0045] The algorithm end of this example mainly includes a draft gauge area detection module, a draft gauge character positioning module, a character recognition module, a draft gauge fitting module, a water surface detection module, a draft gauge reading module and a draft gauge reading verification module; the draft gauge area detection module extracts the draft gauge area from the entire valid image; the draft gauge character positioning module locates the coordinates and boundaries of each character in the draft gauge area; the character recognition module recognizes the characters in the draft gauge area; the draft gauge fitting module reconstructs a complete draft gauge based on the coordinates and recognition results of each character; the water surface detection module detects the boundary line between the water surface and the hull in the entire valid image and obtains the water surface intersection point; the draft gauge reading module reads the draft gauge value based on the fitted complete draft gauge and the water surface intersection point; the draft gauge reading verification module eliminates the error value through the fitting function based on all the read draft gauge values ​​to obtain the optimized draft gauge value.

[0046] like Figure 3 The following are the results of water gauge identification for inland port ships. The main features are: the water surface is relatively calm, the wind and waves are small, and the impact of scale identification fluctuation range is small; there are fewer types of ships and the characteristics are relatively fixed; there are many non-standard ships, and it is difficult to solve all situations.

[0047] like Figure 4 The following are the results of water gauge recognition for ships in coastal ports. The main features are: the water surface is relatively undulating, the wind and waves are strong, and the fluctuation range of scale recognition has a greater impact; there are many types of ships, and more data needs to be collected for learning and training; there are fewer non-standard ships, and most ships can be successfully identified.

[0048] When the on-site operators complete the work according to the system process, the collected original video, water gauge recognition result video, water gauge representative frame image, water gauge result set, and task-related information (such as port / terminal information, operator information, drone information, etc.) are all structured and stored in the database for subsequent traceability and retrieval.

[0049] The water gauge area detection module in this example uses an object detection algorithm to determine the location of the water gauge in the input image. This is a very core task. During the project research, we evaluated various object detection algorithms and found that the YOLO algorithm was the most accurate under the same speed criteria and the fastest under the same accuracy criteria. Therefore, the YOLO algorithm was selected as the object detection algorithm in this example.

[0050] The purpose of the water ruler character positioning module in this example is to use a character detection algorithm to locate the position of characters in the detected water ruler area. The character detection algorithm in this case adopts a text segmentation method based on image segmentation, that is, first classifying at the pixel level to determine the text target to which each pixel belongs, and then combining all text targets to obtain a probability map of the text segmentation area, and finally obtain the enclosing curve of the text segmentation area through post-processing.

[0051] The character recognition module in this example uses a text detection algorithm to locate text regions in an image and then identify the text within them. In simple terms, character detection addresses the question of where characters are in an image, while character recognition addresses the question of what they are.

[0052] The water gauge fitting module in this example reconstructs the recognized water gauge to improve its accuracy. During the reconstruction process, the recognized characters are corrected using a priori standards, improving the accuracy of the reconstructed water gauge.

[0053] The surface detection module in this example uses segmentation to determine the position of the water surface and the hull, thereby identifying the waterline. Segmentation methods include semantic segmentation and instance segmentation. Semantic segmentation extends foreground-background separation, separating image components with distinct semantic meanings. Instance segmentation extends the detection task, describing the outline of the target (more detailed than the detection bounding box).

[0054] Unsupervised learning is performed on the values ​​of all water surface positions determined by the water surface detection module. First, the cluster center of the values ​​is found, and then all values ​​of the cluster with fewer values ​​are deleted. This cycle is repeated until the value function of the K-means clustering method is satisfied. The values ​​around the final cluster center are retained, and the values ​​of the water surface positions that are finally retained are used as the water surface intersection points for reading the water gauge values.

[0055] The water gauge reading module in this example combines the reconstructed complete water gauge with the water surface intersection point to read the water gauge value.

[0056] The water gauge reading verification module in this example combines all the read water gauge values, eliminates the error value through the fitting function, and obtains the optimized water gauge value.

[0057] In general, on the algorithm side, this example optimizes the accuracy of reading water gauge values ​​using existing methods by expanding the data set, adding a water gauge fitting module based on prior conditions, using a local water surface segmentation method, and filtering the water surface position values ​​using the K-means clustering method.

[0058] Table 2 Comparative test results of the existing method and the method of the present invention

[0059]

[0060] As can be seen from Table 2, the detection accuracy and efficiency of the method of the present invention are significantly improved. The original videos captured are clear and standardized, and the waterline detection, character selection and recognition accuracy are all high. The error of the final reading result of the present invention is also significantly reduced.

[0061] In addition, the system device of the present invention can support two configurations:

[0062] 1. Manual version: drone + tablet + server, automatic recognition, requires human control;

[0063] 2. Automatic version: drone + drone nest, autonomous flight, automatic identification, to achieve true unmanned operation.

[0064] At the same time, the system of the present invention can also be promoted to major domestic docks and ports; it can also be expanded to be applied in scenarios such as water level measurement, water depth measurement, oil level measurement, object volume measurement, and can also be promoted to multiple fields such as water conservancy, energy, production and manufacturing, and teaching. It can save a lot of manpower and material resources, and has huge market prospects in the future.

[0065] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any form, and any technical solutions obtained by equivalent replacement or equivalent transformation fall within the scope of protection of the present invention.

Claims

1. A method for remote measurement of ship water gauge based on drone, characterized by: A draft gauge measurement system, primarily consisting of a web-based control terminal, a terminal, an algorithm, and a database, is used to detect draft gauge values. First, a drone is used to capture raw video of the water surface and draft gauge. The terminal then filters the raw video to obtain valid images. The algorithm then analyzes these images using image recognition methods to determine the draft gauge value. The web-based control terminal manages users, data, and permissions, issues task instructions to the terminal, and receives calculation results from the algorithm. The algorithm includes a water gauge area detection module, a water gauge character positioning module, a character recognition module, a water gauge fitting module, a water surface detection module, a water gauge reading module and a water gauge reading verification module; The water gauge area detection module extracts the water gauge area from the entire valid image; The water ruler character positioning module locates the coordinates and boundaries of each character in the water ruler area; The character recognition module recognizes characters in the water gauge area; The water ruler fitting module combines the coordinates of each character and the recognition result to reconstruct the complete water ruler; The water surface detection module detects the boundary between the water surface and the hull in the entire valid image to obtain the water surface intersection point; the water surface detection module uses a two-stage image segmentation method to determine the boundary between the water surface and the hull. During the first stage of image segmentation, the entire valid image is segmented at the water surface position. During the second stage of image segmentation, a local sample is taken upward from the water surface obtained by the first segmentation, or a local sample is taken downward from the nearest complete character on the water surface obtained by the first segmentation, and the local sample is subjected to a second segmentation of the water surface position. The result of the second segmentation is used as the boundary between the water surface and the hull. The water gauge reading module reads the water gauge value by combining the reconstructed complete water gauge with the water surface intersection point; The water gauge reading verification module combines all the read water gauge values, eliminates the error value through the fitting function, and obtains the optimized water gauge value.

2. The method for remote measurement of ship draft gauge based on drone according to claim 1, characterized in that: The original video is screened to obtain valid images. The valid images meet the resolution requirements and contain both the water surface area and the water ruler. The water surface runs across the entire valid image and the area below the water surface occupies 1 / 3±α of the entire valid image area. The water ruler stands on the vertical centerline of the entire valid image. The range of the water ruler's lateral deviation from the vertical centerline is limited to 10% of the width of the entire valid image. The angle of the water ruler's deviation from the vertical centerline is limited to ±30°. At least one letter M appears in the water ruler, and the length of the water ruler revealed in the entire image is 2 to 2.5 meters. α is a set value. The posture of the drone during collection is set based on the requirements for valid images.

3. The method for remote measurement of ship water gauge based on drone according to claim 1, characterized in that: The water surface detection module uses an envelope algorithm to calculate the water surface position, specifically: Lower envelope: Upper envelope: Water surface position: Where: y(t) represents the calculated value of the water surface position at time t, y down (t) represents the corrected water surface position at time t on the lower envelope, y up (t) represents the water surface position at time t on the corrected upper envelope line; y0(t) represents the actual position of the water surface at time t, down(t) represents the fitting value of the water surface position at time t obtained by the lower envelope fitting function, and up(t) represents the fitting value of the water surface position at time t obtained by the upper envelope fitting function.

4. The method for remote measurement of ship draft gauge based on an unmanned aerial vehicle according to claim 1, characterized in that: The water surface detection module uses the K-means clustering method to perform unsupervised learning on the values ​​of all water surface positions obtained, and uses the values ​​of the water surface positions finally retained as the water surface intersection points for reading the water gauge values.

5. The method for remote measurement of ship draft gauge based on drone according to claim 1, characterized in that: After extracting the water gauge area, the water gauge area detection module performs image processing including edge detection, binarization, and dilation and corrosion on the water gauge area to obtain the outer rectangle of the character area, and calculates the rotation and translation matrix for tilt correction of the water gauge area based on the tilt angle and rotation center of the outer rectangle of the character area; the water gauge area is tilt-corrected according to the rotation and translation matrix and then provided to the water gauge character positioning module and the character recognition module; the entire effective image is tilt-corrected according to the rotation and translation matrix and then provided to the water surface detection module.

6. The method for remote measurement of ship draft gauge based on drone according to claim 1, characterized in that: The water gauge area detection module uses the YOLO algorithm to perform target detection.

7. The method for remote measurement of ship draft gauge based on drone according to claim 1, characterized in that: The water ruler character positioning module adopts a text segmentation method based on image segmentation, that is, it first performs classification at the pixel level to determine the text target to which each pixel belongs, then integrates all text targets to obtain a probability map of the text segmentation area, and finally obtains the enclosing curve of the text segmentation area through post-processing.

8. The method for remote measurement of ship draft gauge based on drone according to claim 1, characterized in that: The water ruler fitting module integrates the priori standard of the water ruler into the water ruler reconstruction algorithm, corrects the coordinates and recognition results of each character, and reconstructs a complete water ruler.

Images