A dynamic self-adaptive super-resolution lightweight ship number recognition method, device and medium

Abstract

The application discloses a kind of dynamic self-adapting super light-weight ship number identification method, equipment and medium, belong to dynamic image identification technical field, for solving in existing ship number identification, it is difficult to realize the dynamic tracking and identification of moving ship, ship number identification accuracy is not high, real-time processing efficiency is relatively low technical problem.Method includes: the moving pixel difference calculation between adjacent frames to the ship video data that gimbals camera is shot;The initial hull image is carried out about the denoising enhancement processing under hull contour, and the preprocessed hull image is obtained;The preprocessed hull image is carried out multi-scale frequency domain division processing based on wavelet transform, and the multi-scale component feature is obtained;Through inverse wavelet transform, the high-resolution hull image is obtained by the spatial domain reconstruction processing of the medium-high frequency component feature in multi-scale component feature;Text recognition processing is carried out to high-resolution hull image, and the ship name number information corresponding to the ship is obtained.

Description

Technical Field

[0001] This application relates to the field of dynamic image recognition technology, and in particular to a dynamic adaptive super-resolution lightweight ship number recognition method, device and medium. Background Technology

[0002] Currently, port operations require tracking and managing vessels entering and leaving the port, especially during fishing bans. Strictly controlling the departure of fishing vessels from their posts is a crucial means of controlling illegal fishing during these periods. On the other hand, verifying the consistency between vessel names and scheduling information through vessel name identification assists in optimizing berth allocation and loading / unloading operations, improving vessel management and throughput. Furthermore, combining vessel navigation trajectories with port tidal data can predict optimal berthing times and automatically generate scheduling suggestions, promoting fully automated port operations. All these applications require the effective identification of vessel names.

[0003] Current ship name recognition technology mainly consists of two stages: the first stage is ship target detection, and the second stage is ship number localization and recognition. For ship target detection, Fast R-CNN or YOLO series target detection technologies are mainly used. After obtaining the ship's hull target through target detection, OCR technology is then used to recognize the ship number text. However, this current method still faces certain technical bottlenecks when processing ship number recognition:

[0004] (1) Dynamic tracking and identification of moving vessels is not currently achieved. At present, the identification of vessel numbers generally assumes that the vessel hull is already in the picture and the vessel number is clearly visible in the picture. However, in real-world scenarios, the situation is much more complex. On the one hand, the vessels are moving and their speeds are inconsistent, which means that the vessel hull may not be in the frame of the target camera. On the other hand, the size of the vessel hull and the clarity of the vessel number are different, which leads to a sharp drop in the accuracy of vessel number identification.

[0005] (2) The clarity of the ship number varies under different weather conditions and at different distances. The ship's hull is affected by weather conditions, which may result in a lower clarity of the ship number. For example, in foggy or rainy weather, this also leads to a certain decrease in the accuracy of ship number identification.

[0006] (3) Insufficient real-time processing performance of ship number recognition. In practical application scenarios, ship number recognition algorithms need to be able to process the data transmitted from the camera in real time. However, both steps of the ship number recognition algorithm consume a certain amount of computing time, resulting in low real-time processing efficiency. Summary of the Invention

[0007] This application provides a dynamic adaptive super-resolution lightweight ship number identification method, device, and medium to solve the following technical problems: In existing ship number identification, it is difficult to achieve dynamic tracking and identification of moving ships, the accuracy of ship number identification is not high, and the real-time processing efficiency is relatively low.

[0008] The embodiments of this application adopt the following technical solutions:

[0009] On one hand, this application provides a dynamic adaptive super-resolution lightweight ship number recognition method, including: calculating the difference in moving pixels between adjacent frames of ship video data captured by a gimbal camera to determine the ship's movement vector and the ship's hull pixel ratio; performing motion tracking control on the gimbal camera according to the movement vector to obtain tracking and positioning information; and performing focusing processing on the ship based on the tracking and positioning information under the hull pixel ratio to obtain an initial ship image; performing denoising and enhancement processing on the initial ship image under the hull contour to obtain a preprocessed ship image; performing multi-scale frequency domain partitioning processing on the preprocessed ship image based on wavelet transform to obtain multi-scale component features; performing spatial domain reconstruction processing on the mid-high frequency component features in the multi-scale component features through inverse wavelet transform to obtain a high-resolution ship image; and performing text recognition processing on the high-resolution ship image to obtain the ship name information corresponding to the ship.

[0010] This application embodiment utilizes a gimbal camera to dynamically adapt to the recognition needs of vessels under different motion states, improving the flexibility and adaptability of the recognition system. By employing lightweight super-resolution technology, computational complexity and resource consumption are reduced, making the method suitable for resource-constrained environments such as mobile devices or edge computing devices. It can also accurately determine the vessel's movement vector, thereby achieving precise motion tracking control of the gimbal camera. Furthermore, it ensures the rapid acquisition of clear initial images of the vessel even during movement, which is crucial for text recognition. Multi-scale frequency domain partitioning based on wavelet transform can extract multi-scale component features, which helps to better capture image details, thereby improving the accuracy of text recognition. Moreover, reconstructing mid-to-high frequency components through inverse wavelet transform provides better image quality for text recognition. Finally, through text recognition processing, the vessel's name and number information can be accurately extracted, which has significant application value in fields such as vessel monitoring and traffic management.

[0011] In one feasible implementation, the motion pixel difference between adjacent frames of the ship video data captured by the gimbal camera is calculated to determine the ship's motion vector and the ship's hull pixel ratio. Specifically, this includes: using a lightweight Yolov10-small model in the ship target detection system associated with the gimbal camera to perform target detection processing on the ship within the gimbal camera's field of view, identifying and obtaining target detection information; controlling the gimbal camera to acquire and process the ship's video stream based on the target detection information, obtaining the ship video data; calculating the pixel difference between the center points of the ship's hull outline in any adjacent frames of the ship video data to obtain the motion pixel difference; and calculating the pixel changes related to the ship's motion trajectory based on the motion pixel difference to obtain the motion vector; wherein the motion vector includes: the center point pixel's movement speed and the center point pixel's movement direction; extracting and processing the first video frame containing the complete ship hull outline pixels from the ship video data to obtain a reference image; and comparing the target pixel value corresponding to the complete ship hull outline pixels with the overall reference pixel size of the reference image to obtain the ship hull pixel ratio.

[0012] In one feasible implementation, motion tracking control is performed on the gimbal camera based on the movement vector to obtain tracking and positioning information. Specifically, this includes: after recognizing the movement vector, calculating the coordinates of the ship target in the ship video stream using a preset YOLOv5 model to obtain the coordinates of the center point of the ship's bounding box; performing target tracking processing based on the Lucas-Kanade optical flow method on the coordinates of the center point of the ship's bounding box based on the movement vector, and generating a ship movement vector using a Cartesian coordinate system as the reference coordinate system; mapping the ship movement vector to motion control under the gimbal motor to obtain gimbal motor control quantities; wherein, the gimbal motor control quantities include: horizontal rotation angle and pitch speed proportional factors; and dynamically adjusting the parameters of the gimbal motor control quantities using a preset PID controller to generate tracking and positioning information for tracking the coordinates of the center point of the ship's bounding box.

[0013] In one feasible implementation, based on the tracking and positioning information, focusing processing is performed on the vessel according to the pixel ratio of the vessel hull to obtain an initial vessel image. Specifically, this includes: using a gimbal camera controlled by the tracking and positioning information to perform initial focusing control on the vessel, ensuring that the overall pixels of the vessel hull are located in the central region of the image, and generating a preliminary focusing strategy; wherein the initial focusing control includes: optical zoom control, digital zoom control, and focus-holding control; based on the initial focusing strategy and the vessel's movement vector, adaptive adjustment of the focusing area of ​​the overall pixels of the vessel hull is performed to obtain a dynamic adjustment strategy for the focusing window; after executing the preliminary focusing strategy and the dynamic adjustment strategy for the focusing window, adaptive focusing iteration is performed on the focusing quality of the vessel using a preset Tenengrad gradient function, outputting the initial vessel image; wherein the adaptive focusing iteration includes: adaptive step size adjustment and reverse fine-tuning adaptive adjustment.

[0014] In one feasible implementation, the initial hull image undergoes denoising and enhancement processing related to the hull contour to obtain a preprocessed hull image. Specifically, this includes: extracting the hull contour from the initial hull image using a preset Canny threshold and depth edge detection technology, obtaining initial hull contour features; denoising the initial hull contour features using multi-fusion region processing, obtaining denoised hull contour features; wherein the multi-fusion region includes: a core region, a background region, and a contour region; dynamically adjusting the denoised hull contour features based on the overall pixel size characteristics of the hull, obtaining sharpened hull contour features; and performing texture restoration and color correction processing on the sharpened hull contour features to output the preprocessed hull image.

[0015] In one feasible implementation, the preprocessed hull image is subjected to multi-scale frequency domain partitioning based on wavelet transform to obtain multi-scale component features. Specifically, this includes: performing a transformation calculation from the spatial domain to the frequency domain on the preprocessed hull image using discrete wavelet transform technology to obtain frequency domain data; and performing multi-scale decomposition on the frequency domain data using multi-scale wavelet transform to generate the multi-scale component features at different frequency domain scales.

[0016] In one feasible implementation, the mid-to-high frequency component features in the multi-scale component features are reconstructed in the spatial domain using inverse wavelet transform to obtain a high-resolution ship hull image. Specifically, this includes: removing the low-frequency component features from the multi-scale component features to obtain the mid-to-high frequency component features; performing optimization control on the mid-to-high frequency component features based on a nonlocal low-rank regularization algorithm under image self-similarity; and performing spatial domain image reconstruction processing on the optimized mid-to-high frequency component features using the inverse wavelet transform to obtain the high-resolution ship hull image.

[0017] In one feasible implementation, text recognition processing is performed on the high-resolution ship hull image to obtain the ship's name information. Specifically, this includes: inputting the high-resolution ship hull image into an OCR algorithm; and using the OCR algorithm to recognize the characters on the ship hull in the high-resolution ship hull image to obtain the ship's name information. The characters include at least: Chinese characters, numbers, and English letters.

[0018] Secondly, embodiments of this application also provide a dynamically adaptive super-resolution lightweight ship number identification device, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, enabling the at least one processor to execute a dynamically adaptive super-resolution lightweight ship number identification method as described in any of the above embodiments.

[0019] Thirdly, embodiments of this application also provide a non-volatile computer storage medium, wherein the storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each program including instructions, wherein when the instructions are executed by a terminal, the terminal executes a dynamic adaptive super-resolution lightweight ship number recognition method described in any of the above embodiments.

[0020] This application provides a dynamically adaptive super-resolution lightweight ship number identification method, device, and medium. Compared with the prior art, the embodiments of this application have the following beneficial technical effects:

[0021] 1. Dynamic Adaptive Capability: It can dynamically adapt to the identification needs of ships in different motion states, improving the flexibility and adaptability of the identification system.

[0022] 2. Lightweight processing: By using lightweight super-resolution technology, the computational complexity and resource consumption are reduced, making the method suitable for resource-constrained environments, such as mobile devices or edge computing devices.

[0023] 3. High-precision positioning: By calculating the difference in moving pixels between adjacent frames, the movement vector of the ship can be accurately determined, thereby achieving precise motion tracking and control of the gimbal camera.

[0024] 4. Fast focusing: Focusing is performed according to the pixel ratio of the hull, which ensures that a clear initial image of the hull can be obtained quickly even when the hull is moving, which is crucial for text recognition.

[0025] 5. Image preprocessing: Denoising and enhancement processing improves the quality of the preprocessed ship hull image, which is helpful for the subsequent text recognition process and reduces recognition errors.

[0026] 6. Multi-scale feature extraction: Multi-scale frequency domain partitioning based on wavelet transform can extract multi-scale component features, which helps to better capture image details and thus improve the accuracy of text recognition.

[0027] 7. High-resolution reconstruction: High-frequency components were reconstructed through inverse wavelet transform, resulting in a high-resolution image of the ship's hull, which provides better image quality for text recognition.

[0028] 8. Accuracy of text recognition: Through text recognition processing, the ship's name and number information can be accurately extracted, which has important application value in fields such as ship monitoring and traffic management. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0030] Figure 1 A flowchart of a dynamic adaptive super-resolution lightweight ship number recognition method provided in an embodiment of this application;

[0031] Figure 2 A schematic diagram of ship name identification provided in an embodiment of this application;

[0032] Figure 3 This is a schematic diagram of the structure of a dynamically adaptive super-resolution lightweight ship number recognition device provided in an embodiment of this application. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0034] It should be noted that this application solves the above-mentioned technical problems by introducing dynamic adaptive hull tracking technology, image super-resolution technology, and lightweight technology based on model quantization:

[0035] (1) Dynamic Adaptive Tracking Technology for Gimbals: By introducing a pixel-level motion speed detection algorithm for ships within the camera, the pixel-level movement distance of the ship is calculated. Simultaneously, based on the rotational speed of the gimbal, dynamic tracking of the ship is achieved, thus positioning the ship at the center of the camera's image. Furthermore, the focus of the gimbal is adjusted according to the pixel size of the ship target, ensuring that the ship's area within the camera's image is no less than 2 / 3. This results in obtaining a higher quality image of the target.

[0036] (2) Super-resolution technology for ship hull photos based on wavelet transform: By introducing wavelet transform technology, high-frequency components of multiple scales are extracted, edge details are enhanced, and edge features of the ship hull are further extracted, thereby solving the problem of blurred ship hull images and fonts caused by weather.

[0037] (3) Lightweight ship number recognition technology: In the target detection stage, the ship hull photos taken in the port are first labeled. Then, based on transfer learning, the target detection algorithm is optimized. Finally, model quantization technology is used to convert the 32-bit floating-point parameters of the underlying structure of the model into 8-bit integers, thereby improving the inference efficiency of the model.

[0038] This application provides a dynamically adaptive super-resolution lightweight ship number recognition method, such as... Figure 1 As shown, the dynamic adaptive super-resolution lightweight ship number recognition method specifically includes steps S101-S106:

[0039] S101. Calculate the difference in moving pixels between adjacent frames of the video data of the ship captured by the PTZ camera to determine the ship's movement vector and the ship's hull pixel ratio.

[0040] Specifically, the first step is to use the lightweight Yolov10-small model in the ship target detection system associated with the gimbal camera to perform target detection processing on the ship within the field of view of the gimbal camera, and identify and obtain target detection information.

[0041] Furthermore, based on the target detection information, the pan-tilt camera is controlled to collect and process the video stream of the vessel, thereby obtaining vessel video data.

[0042] Furthermore, the pixel difference between the center points of the ship's outline in any adjacent frames of the ship video data is calculated to obtain the moving pixel difference. Then, the pixel changes related to the ship's motion trajectory are calculated from the moving pixel difference to obtain the motion vector. The motion vector includes the moving speed and direction of the center point pixel.

[0043] Furthermore, the first video frame containing the complete hull outline pixels from the ship video data is extracted and processed to obtain a reference image.

[0044] Furthermore, the target pixel value corresponding to the complete hull outline pixels must be compared with the reference pixel value of the whole image corresponding to the reference image to obtain the hull pixel ratio.

[0045] As a feasible implementation method, based on the detection of ship speed, direction, and size using a gimbal, when a ship enters the detection range of the gimbal camera, the ship may be too far away to clearly see its number. This application first performs ship target detection on the video image. Based on the difference in moving pixels of the detected ship center between two frames, it calculates the moving speed and relative direction of the ship's pixel center point, and simultaneously calculates the ratio between the detected ship size and the pixel size of the entire image.

[0046] In one embodiment, Figure 2 This application provides a schematic diagram of ship name identification, such as... Figure 2 As shown, the ship target detection uses a lightweight Yolov10-small model. Based on the Yolov10-small model, post-training quantization (PTQ) is performed. The trained Yolov10-small model is directly quantized using an asymmetric algorithm, mapping the FP32 data to uint8 data to obtain a lightweight model, thus improving inference efficiency. The formula used is as follows: In other words, a lightweight Yolov10-small model associated with the gimbal camera is first used for ship target detection. Within the field of view of the gimbal camera, the ship is detected in real time to obtain its position and size information. Then, based on the target detection information, the gimbal camera is controlled to track the ship and acquire continuous video frame data.

[0047] In one embodiment, any two adjacent frames from the video data are selected. The hull outline in each frame is identified, and the coordinates of the outline's center point are calculated. The pixel coordinate difference between the center points of the hull outline in adjacent frames is calculated to obtain the moving pixel difference. Then, based on the pixel difference, the moving speed and direction of the center point pixel are calculated to determine the movement vector. The first frame containing the complete hull outline is extracted from the video data as a reference image. The reference image is then preprocessed to ensure its quality is suitable for subsequent calculations. The pixel values ​​of the complete hull outline in the reference image are measured. Finally, the hull outline pixel values ​​are compared with the total pixel values ​​of the reference image to obtain the hull pixel ratio.

[0048] S102. Based on the movement vector, perform motion tracking control on the gimbal camera to obtain tracking and positioning information. Then, based on the tracking and positioning information, perform focusing processing on the vessel according to the stated hull pixel ratio to obtain an initial hull image.

[0049] Specifically, after identifying the motion vector, the coordinates of the ship target in the ship video stream are calculated using a preset YOLOv5 model to obtain the coordinates of the center point of the ship's bounding box.

[0050] Furthermore, based on the movement vector, target tracking processing based on the Lucas-Kanade optical flow method is required to analyze the coordinates of the center point of the hull bounding box, and the hull movement vector is generated using the Cartesian coordinate system as the reference coordinate system.

[0051] Furthermore, the hull movement vector is mapped to motion control parameters related to the gimbal motors to obtain the gimbal motor control variables. These gimbal motor control variables include: horizontal rotation angle and pitch speed scaling factors.

[0052] Furthermore, by using a preset PID controller, the control parameters of the gimbal motor are dynamically adjusted to generate tracking and positioning information for tracking the coordinates of the center point of the ship's bounding box.

[0053] Furthermore, by using a PTZ camera controlled by tracking positioning information, initial focusing control is performed on the vessel to ensure that all pixels of the vessel are located in the center area of ​​the image, and an initial focusing strategy is generated. This initial focusing control includes optical zoom control, digital zoom control, and focus-holding control.

[0054] Furthermore, based on the initial focusing strategy and the ship's movement vector, the focusing area of ​​the overall ship pixels is adaptively adjusted to obtain a dynamic adjustment strategy for the focusing window.

[0055] Furthermore, after implementing the initial focusing strategy and the dynamic adjustment strategy of the focusing window, the focusing quality of the ship is adaptively iterated using a preset Tenengrad gradient function to output an initial image of the ship. This adaptive focusing iteration includes adaptive step size adjustment and reverse fine-tuning adaptive adjustment.

[0056] As a possible implementation method, such as Figure 2 As shown, the gimbal moves the camera image according to the pixel movement speed and relative direction of the ship, and according to the rotation speed of the gimbal, fixing the ship in the center of the image. At the same time, the gimbal focuses according to the relative size of the ship, so as to ensure that the ship occupies at least 2 / 3 of the entire image, thereby obtaining an image of the ship in the center of the image and occupying the main area of ​​the image, that is, the initial ship image.

[0057] In one embodiment, a motion tracking control system can be used to track and locate the initial hull image. The motion tracking control system may consist of the following modules:

[0058] 1. Video capture module: responsible for capturing video streams from ships.

[0059] 2. Target Detection Module: Detects ship targets based on the YOLOv5 model and calculates the coordinates of the center point of the ship's bounding box.

[0060] 3. Target tracking module: The Lucas-Kanade optical flow method is used to track the ship's hull and generate the ship's movement vector.

[0061] 4. Gimbal Control Module: Based on the ship's movement vector, the module controls the movement of the gimbal motor to achieve camera tracking.

[0062] 5. Focusing control module: Based on tracking and positioning information, it focuses on the ship and generates an initial image of the hull.

[0063] In other words, the motion tracking control system enables real-time tracking and positioning of vessels, improving the ability to identify and track them. Furthermore, motion control of the gimbal camera based on motion vectors enhances tracking accuracy. Focusing processing also yields high-quality initial images of the hull, providing a solid foundation for subsequent processing.

[0064] S103. Perform noise reduction and enhancement processing on the initial hull image based on the hull contour to obtain a preprocessed hull image.

[0065] Specifically, the initial hull contour features are obtained by first extracting the hull outline from the initial hull image using a preset Canny threshold and depth edge detection technology.

[0066] Furthermore, the initial hull contour features are subjected to denoising processing under multi-fusion regions to obtain denoised hull contour features. The multi-fusion regions include: core region, background region, and contour region.

[0067] Furthermore, based on the overall pixel size characteristics of the hull, it is also necessary to dynamically adjust the anisotropic edges of the denoised hull contour features to obtain sharpened hull contour features.

[0068] Furthermore, texture restoration and color correction are performed on the sharpened hull outline features to output a pre-processed hull image.

[0069] In one embodiment, such as Figure 2 As shown, the code approach for the edge fusion stage based on Canny thresholding and deep edge detection technology is as follows:

[0070] def edge_fusion(img):

[0071] #Dual-threshold Canny detection (low threshold 50, high threshold 150)

[0072] canny_edges=cv2.Canny(img,50,150)

[0073] #HED Deep Edge Detection (using a pre-trained model)

[0074] hed_edges=hed_model.predict(img)

[0075] #Weighted fusion (weight coefficient α = 0.7)

[0076] fused_edges=cv2.addWeighted(canny_edges,0.7,hed_edges,0.3,0)

[0077] return fused_edges

[0078] Then, multi-region denoising is performed: Core region: non-local mean denoising (search window 15x15). Background region: Gaussian blur (σ=2.0) combined with morphological opening operation. Contour region: adaptive bilateral filtering (spatial σ=10, intensity σ=15). Then, anisotropic sharpening is performed, which can be done using the following code:

[0079] pixel_size=calculatePixelSize(contour);

[0080] [Gx,Gy]=imgradientxy(contour);

[0081] adjusted_contour=anisotropicDiffusion(contour,...

[0082] 'ConductionMethod','exponential',...

[0083] 'MaxIterations', 10,...

[0084] 'Lambda', 0.05 / pixel_size);

[0085] Next, texture restoration is performed, which involves reconstructing the texture using an improved TV-L1 model with local contrast-limited histogram equalization (CLAHE). Then, color correction is performed, which requires chromatic aberration compensation based on the CIE Lab color space δE2000, and finally, white balance correction is completed.

[0086] S104. Perform multi-scale frequency domain partitioning processing on the preprocessed hull image based on wavelet transform to obtain multi-scale component features.

[0087] Specifically, the preprocessed ship hull image is first transformed from the spatial domain to the frequency domain using discrete wavelet transform technology to obtain frequency domain data.

[0088] Furthermore, the frequency domain data is decomposed into multiple scales using multi-scale wavelet transform to generate multi-scale component features at different frequency domain scales.

[0089] In one embodiment, the discrete wavelet transform is first initialized, followed by multi-scale decomposition, including: First-level decomposition: generating the LL1 low-frequency approximation component (retaining 85% energy); extracting the HL1 horizontal high-frequency, LH1 vertical high-frequency, and HH1 diagonal high-frequency components; finally, using the Mallat algorithm for fast downsampling. Second-level decomposition: performing a second decomposition on LL1 to obtain LL2-HH2; dynamically adjusting the scale factor α = 1.414. Third-level decomposition: performing the final decomposition on LL2; introducing a scale guard band (15% of the Nyquist frequency). Finally, based on the multi-scale component feature tree, multi-scale decomposition is performed on the horizontal structural features, vertical structural features, and surface defect features in the frequency domain data, outputting multi-scale component features at different frequency domain scales.

[0090] S105. By using inverse wavelet transform, the mid-to-high frequency component features in the multi-scale component features are reconstructed in the spatial domain to obtain a high-resolution ship hull image.

[0091] Specifically, the low-frequency component features in the multi-scale component features are first removed to obtain the mid-to-high frequency component features.

[0092] Furthermore, based on the nonlocal low-rank regularization algorithm, optimization control is performed on the mid-to-high frequency component features under the condition of image self-similarity.

[0093] Furthermore, by using inverse wavelet transform, the mid-to-high frequency component features after optimization control are processed for spatial domain image reconstruction to obtain a high-resolution ship hull image.

[0094] In one embodiment, the lowest frequency component is discarded first, and then, based on a nonlocal low-rank regularization algorithm, optimization control related to image self-similarity is performed on the mid-to-high frequency component features. The possible code implementation is as follows:

[0095] def NLR_denoise(coeffs):

[0096] # Construct similar block groups (7×7 search window, 3×3 block size)

[0097] patch_groups=build_NL_groups(coeffs)

[0098] #Minimize the weighted nuclear norm (λ=0.1)

[0099] optimized = []

[0100] for group in patch_groups:

[0101] U,S,V = np.linalg.svd(group)

[0102] #Singular value contraction (τ = 0.7 * max(S))

[0103] S_hat=np.maximum(S-0.7*np.max(S),0)

[0104] optimized.append(U@np.diag(S_hat)@V)

[0105] return reconstruct(optimized)

[0106] Then, nonlocal similarity is combined with low-rank constraints to effectively suppress artifacts while ensuring reconstruction accuracy. Specifically, image self-similarity is used to optimize high-frequency detail reconstruction and suppress noise interference to improve super-resolution reconstruction results. Finally, the remaining mid-to-high frequency components are reconstructed into a new spatial domain using inverse wavelet transform. Based on the reconstructed new spatial domain, a high-resolution ship hull image is output.

[0107] S106. Perform text recognition processing on the high-resolution ship image to obtain the ship's name and number information.

[0108] Specifically, a high-resolution image of the ship's hull is input into an OCR algorithm. The OCR algorithm then recognizes the characters on the hull in the high-resolution image to obtain the ship's name and serial number. These characters include at least letters, numbers, and English letters.

[0109] In addition, embodiments of this application also provide a dynamically adaptive super-resolution lightweight ship number recognition device, such as... Figure 3 As shown, the dynamically adaptive super-resolution lightweight ship number recognition device 300 specifically includes:

[0110] At least one processor 301. And a memory 302 communicatively connected to the at least one processor 301. The memory 302 stores instructions executable by the at least one processor 301, enabling the at least one processor 301 to execute:

[0111] The motion pixel difference between adjacent frames of the video data of the ship captured by the gimbal camera is calculated to determine the ship's motion vector and the ship's hull pixel ratio.

[0112] Based on the motion vector, motion tracking control is performed on the gimbal camera to obtain tracking and positioning information; and based on the tracking and positioning information, focusing processing is performed on the vessel according to the pixel ratio of the vessel hull to obtain an initial vessel image.

[0113] The initial hull image is subjected to denoising and enhancement processing based on the hull contour to obtain a preprocessed hull image.

[0114] The preprocessed hull image is subjected to multi-scale frequency domain partitioning based on wavelet transform to obtain multi-scale component features.

[0115] By using inverse wavelet transform, the mid-to-high frequency component features in the multi-scale component features are reconstructed in the spatial domain to obtain a high-resolution ship hull image.

[0116] Text recognition processing is performed on high-resolution ship images to obtain the ship's name and number information.

[0117] This application embodiment utilizes a gimbal camera to dynamically adapt to the recognition needs of vessels under different motion states, improving the flexibility and adaptability of the recognition system. By employing lightweight super-resolution technology, computational complexity and resource consumption are reduced, making the method suitable for resource-constrained environments such as mobile devices or edge computing devices. It can also accurately determine the vessel's movement vector, thereby achieving precise motion tracking control of the gimbal camera. Furthermore, it ensures the rapid acquisition of clear initial images of the vessel even during movement, which is crucial for text recognition. Multi-scale frequency domain partitioning based on wavelet transform can extract multi-scale component features, which helps to better capture image details, thereby improving the accuracy of text recognition. Moreover, reconstructing mid-to-high frequency components through inverse wavelet transform provides better image quality for text recognition. Finally, through text recognition processing, the vessel's name and number information can be accurately extracted, which has significant application value in fields such as vessel monitoring and traffic management.

[0118] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0119] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0120] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0121] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0122] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0123] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0124] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0125] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0126] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0127] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0128] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of this specification.

Claims

1. A dynamic adaptive super-resolution lightweight ship number recognition method, characterized in that, The method includes: The motion pixel difference between adjacent frames of the video data of the ship captured by the gimbal camera is calculated to determine the ship's motion vector and the pixel ratio of the ship's hull. Specifically, this includes: The lightweight Yolov10-small model in the ship target detection system associated with the gimbal camera is used to perform target detection processing on the ship within the field of view of the gimbal camera, and target detection information is identified and obtained. Based on the target detection information, the pan-tilt camera is controlled to collect and process the video stream of the vessel to obtain the vessel video data; The pixel difference between the center points of the ship's outline in any adjacent frames of the ship's video data is calculated to obtain the moving pixel difference; and the pixel change related to the ship's motion trajectory is calculated on the moving pixel difference to obtain the moving vector; wherein, the moving vector includes: the moving speed of the center point pixel and the moving direction of the center point pixel; The first video frame containing the complete hull outline pixels in the ship video data is extracted and processed to obtain a reference image; The target pixel value corresponding to the complete hull outline pixels is compared with the whole image reference pixel value corresponding to the reference image to obtain the hull pixel ratio. Based on the movement vector, motion tracking control is performed on the gimbal camera to obtain tracking and positioning information; and based on the tracking and positioning information, focusing processing is performed on the vessel according to the hull pixel ratio to obtain an initial hull image. The initial hull image is subjected to noise reduction and enhancement processing based on the hull contour to obtain a preprocessed hull image. The preprocessed hull image is subjected to multi-scale frequency domain partitioning based on wavelet transform to obtain multi-scale component features. By using inverse wavelet transform, the mid-to-high frequency component features in the multi-scale component features are reconstructed in the spatial domain to obtain a high-resolution ship hull image. The high-resolution ship image is processed by text recognition to obtain the ship's name and number information.

2. The dynamic self-adaptive super-resolution light-weight ship identification method according to claim 1, characterized in that, Based on the movement vector, motion tracking control is performed on the gimbal camera to obtain tracking and positioning information, specifically including: After the motion vector is identified, the coordinates of the ship target in the ship video stream are calculated using a preset YOLOv5 model to obtain the coordinates of the center point of the ship bounding box. Based on the movement vector, target tracking processing based on the Lucas-Kanade optical flow method is performed on the coordinates of the center point of the hull bounding box, and the hull movement vector is generated with the Cartesian coordinate system as the reference coordinate system. The hull movement vector is mapped to motion control under the gimbal motor to obtain the gimbal motor control quantity; wherein, the gimbal motor control quantity includes: horizontal rotation angle and pitch speed proportionality factor; The control parameters of the gimbal motor are dynamically adjusted using a preset PID controller to generate tracking and positioning information for tracking the coordinates of the center point of the hull's bounding frame.

3. The dynamic self-adaptive super-resolution light-weight ship identification method according to claim 1, characterized in that, Based on the tracking and positioning information, the vessel undergoes focusing processing according to the hull pixel ratio to obtain an initial hull image, specifically including: controlling the pan-tilt-zoom camera under the tracking positioning information to perform initial focus control on the ship to make the whole ship pixels located in the center area of the image and generate a preliminary focus strategy, wherein the initial focus control comprises optical zoom control, digital zoom control and focus keeping control; performing adaptive adjustment of the focus area of the whole ship pixels according to the preliminary focus strategy and based on the movement vector of the ship to obtain a focus window dynamic adjustment strategy; after executing the preliminary focus strategy and the focus window dynamic adjustment strategy, performing adaptive focus iteration on the focus quality of the ship by using a preset Tenengrad gradient function to output the initial ship body image, wherein the adaptive focus iteration comprises step adaptive adjustment and reverse fine adjustment adaptive adjustment.

4. The dynamic self-adaptive super-resolution light-weight ship identification method according to claim 1, characterized in that, performing denoising enhancement processing on the initial ship body image with respect to the ship body contour to obtain a preprocessed ship body image, specifically comprising: extracting the ship body contour of the initial ship body image under the fused edge by using a preset Canny threshold and a depth edge detection technology to obtain initial ship body contour features; performing denoising processing on the initial ship body contour features in multiple fusion regions to obtain denoised ship body contour features, wherein the multiple fusion regions include a core region, a background region and a contour region; performing dynamic adjustment of the denoised ship body contour features with respect to anisotropic edges based on the size characteristics of the whole ship pixels to obtain sharpened ship body contour features; performing texture recovery and color correction processing on the sharpened ship body contour features to output the preprocessed ship body image.

5. The dynamic self-adaptive super-resolution light-weight ship identification method according to claim 1, characterized in that, performing multi-scale frequency domain division processing on the preprocessed ship body image based on wavelet transform to obtain multi-scale component features, specifically comprising: performing conversion calculation on the preprocessed ship body image from the spatial domain to the frequency domain by using discrete wavelet transform technology to obtain frequency domain data; performing multi-scale decomposition on the frequency domain data by using multi-scale wavelet transform to generate the multi-scale component features in different frequency domain scales.

6. The dynamic self-adaptive super-resolution light-weight ship identification method according to claim 1, characterized in that, performing spatial domain reconstruction processing on the medium-high frequency component features in the multi-scale component features by using inverse wavelet transform to obtain a high-resolution ship body image, specifically comprising: performing elimination processing on the low-frequency component features in the multi-scale component features to obtain the medium-high frequency component features; performing optimization control on the medium-high frequency component features with respect to image self-similarity based on a non-local low-rank regularization algorithm; performing image reconstruction processing on the medium-high frequency component features after optimization control by using the inverse wavelet transform to obtain the high-resolution ship body image.

7. The dynamic self-adaptive super-resolution light-weight ship identification method according to claim 1, characterized in that, performing character recognition processing on the high-resolution ship body image to obtain the ship name number information corresponding to the ship, specifically comprising: inputting the high-resolution ship body image into an OCR algorithm; performing recognition processing on the characters of the ship body in the high-resolution ship body image by using the OCR algorithm to obtain the ship name number information, wherein the characters at least include characters, numbers and English.

8. A dynamic adaptive super-resolution light-weight vessel identification device, characterized by, The device comprises: at least one processor; and, A memory connected in communication with the at least one processor; wherein The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of claim 1-7.

9. A non-transitory computer storage medium, comprising: The storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, and each program includes instructions, and the instructions enable the terminal to perform the method of claim 1-7 when the terminal executes the instructions.

Images