Ship similarity comparison method
Through the use of deep learning models in the similarity of fishing boats for division and regularization, and combined with high -level image processing technology such as brightness adjustment, noise, distorted correction and sharpness, the LBP feature vector extraction and comparison are used to solve the traditional algorithm in the fishing boat in the fishing boat The problem of poor effects in similarity judgments has achieved more accurate evaluation of the similarity of fishing boats, and provides more effective technical support for marine fishery monitoring.
Patent Information
- Application Number
- CN202411252873.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2025-01-28
AI Technical Summary
The existing technology has two main problems in the judgment of similarity of fishing boats: First, the impact of sea surface background leads to inaccurate similarity calculation results; second, traditional algorithms are difficult to capture subtle differences in the appearance and structure of fishing boats, making it difficult to accurately distinguish different fishing boats. .
The use of a ship similarity contrast method, including the segmentation of the fishing boats, the regularization of the division of the picture, the comprehensive processing of image quality, and the feature value of the image. Specific steps include using deep learning models (such as YOLO framework) for fishing boat segmentation, regularization of image, application of automatic brightness and contrast adjustment, image dehumidification, lens distortion school, color balance and saturation adjustment, image sharpness and other technical processing images Finally, use the LBP feature vector extraction and comparison to evaluate similarity.
Through these technical means, we can more accurately judge the similarity of fishing boats, reduce the impact of the background of the sea surface, capture the subtle differences between the appearance and structure of the fishing boat, so as to achieve a more accurate comparison of similarity on fishing boats, and provide marine fishery monitoring and management. More effective technical support.
Smart Images
Figure CN119380047A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer image recognition in the marine field, and in particular to a ship similarity comparison method. Background Art
[0002] In the context of current technological development, image similarity calculation has become highly accurate and widely used in various fields, such as using algorithms such as Euclidean distance, cosine distance, and Hamming distance to evaluate image similarity. However, these general similarity calculation methods often fail to achieve the expected results when applied to specific fields, especially marine fishery monitoring. In particular, we found several significant problems in the identification of fishing boat similarity.
[0003] First of all, as the subject being compared, fishing boats are often integrated with the vast sea surface, and their main parts do not appear particularly prominent in the entire picture. When using traditional similarity algorithms, due to the large area of the sea surface, the algorithm is often strongly affected by the sea surface background, resulting in that even if there are large differences in the fishing boats themselves, the similarity calculation results of the entire picture are almost the same.
[0004] Secondly, many fishing boats do not differ much in appearance and structure. The differences between them are often reflected in the personal arrangements and decorations of the shipowners. These subtle differences may not be obvious in the pictures. Therefore, it is difficult for traditional similarity algorithms to capture these subtle differences from the data, making it impossible to accurately distinguish different fishing boats.
[0005] To address these issues, we need to further research and develop specific similarity calculation methods applicable to the field of marine fisheries to more accurately identify the similarities of fishing vessels and provide more effective technical support for marine fisheries monitoring and management. Summary of the invention
[0006] In order to solve the problems existing in the prior art, the present invention provides a ship similarity comparison method to solve the problem that fishing boats have high similarity and are difficult to distinguish.
[0007] To solve the above problems, the technical solution of the present invention is as follows: A ship similarity comparison method comprises the following steps:
[0008] S1, segmentation of fishing boats in the image;
[0009] S2, regularization of the segmented images, setting fixed pixels, and normalizing all images according to fixed pixels;
[0010] S3, comprehensive processing of image quality;
[0011] S4. Extraction and calculation of image feature values.
[0012] Furthermore, the S1 includes building a deep learning model, using the YOLO framework, separating the fishing boats in the picture, with 250-280 training rounds, a learning rate of 0.0001-0.0003, a decay point of 210-260, a decay coefficient of 0.05-0.2, an initial value of the learning rate of 0, a linear growth step of 3000-5000, and an optimizer using a momentum algorithm and L2 regularization, with a regularization coefficient of 0.0005.
[0013] Furthermore, a dynamic maximization scheme may be adopted in S2 to batch calculate the largest picture in the pictures in S1, set the largest picture as a fixed pixel, and regularize other pictures according to the fixed pixel.
[0014] Further, the S3 includes:
[0015] Step (1) uses automatic brightness and contrast adjustment technology to improve the overall appearance of the fishing boat image; by analyzing the brightness and contrast distribution in the image, the parameters are automatically adjusted to optimize the visual effect. For areas that are too dark or too bright, the adjustment technology will perform local enhancement to ensure that the details of the fishing boat are clearly visible in the image.
[0016] Further, the S3 includes:
[0017] Step (2) uses image denoising technology to eliminate noise in the image; the noise includes sensor noise, transmission noise, etc. The image clarity and quality are improved by using image denoising technology.
[0018] Further, the S3 includes:
[0019] Step (3) uses lens distortion correction technology; by analyzing the distortion pattern produced by the lens, the image is automatically corrected.
[0020] Further, the S3 includes:
[0021] Step (4) applies color balance and saturation adjustment techniques.
[0022] Further, the S3 includes:
[0023] Step (5) uses image sharpening technology to enhance the edge and details of the fishing boat; by enhancing the high-frequency information in the image, the edge of the fishing boat is clearer and the details are richer.
[0024] Furthermore, in S4, the LBP feature vector is used to extract the feature value of the ship, and the comparison between the feature values is used as the similarity judgment standard of the fishing boats, and finally normalization is performed to obtain the similarity value.
[0025] Compared with the prior art, the present invention has the following beneficial effects:
[0026] Step (1) in S3 automatically adjusts parameters to optimize visual effects by analyzing the brightness and contrast distribution in the image. For areas that are too dark or too bright, the automatic brightness and contrast adjustment technology will perform local enhancement to ensure that the details of the fishing boat are clearly visible in the image.
[0027] Step (3) in S3 uses lens distortion correction technology. After correction, the fishing boat in the image will present a more realistic shape and size, avoiding distortion caused by lens distortion;
[0028] Step (4) in S3 applies color balance and saturation adjustment technology to ensure that the color of the fishing boat in the image is realistic and bright, and closer to the fishing boat in the real world;
[0029] Step (5) in S3 uses image sharpening technology to enhance the edge and details of the fishing boat, and by enhancing the high-frequency information in the image, the edge of the fishing boat is clearer and the details are richer;
[0030] It can more accurately identify the similarity of fishing vessels, provide more effective technical support for marine fishery monitoring and management, and achieve more accurate comparison of the similarity of fishing vessels in fishing port monitoring images. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0032] like Figure 1 As shown, a ship similarity comparison method comprises the following steps:
[0033] S1. Segmentation of fishing boats in the picture: Build a deep learning model and use the YOLO framework to segment the fishing boats in the picture. The number of training rounds is 270, the learning rate is 0.000125, the decay points are 216 and 243, and the decay coefficient is 0.1. The initial value of the learning rate is 0, and the linear growth step is 4000. The optimizer uses the momentum algorithm and L2 regularization, and the regularization coefficient is 0.0005. Considering the differences in video accuracy and scenes, the optimal training parameters may need to be adjusted.
[0034] S2. Regularization of segmented images: Since the separated fishing boats are irregular shapes, the images need to be regularized. In practice, this application adopts a maximization scheme to uniformly normalize the images into images of size 4096*1800. The specific normalization size should be determined according to the separated fishing boats in the actual scene. A dynamic maximization scheme can be used to batch calculate the largest image in each image, and regularize other images according to the pixels of this image.
[0035] S3. In the comprehensive processing of images, a series of advanced image processing technologies are used to restore the appearance and details of fishing boats in the real world as much as possible. These technologies not only target problems such as uneven lighting and lens distortion, but also take into account other factors that may affect image quality, such as noise and lack of contrast.
[0036] (1) First, the overall appearance of the fishing boat image is improved using automatic brightness and contrast adjustment technology. This method automatically adjusts parameters to optimize the visual effect by analyzing the brightness and contrast distribution in the image. For areas that are too dark or too bright, the algorithm will perform local enhancement to ensure that the details of the fishing boat are clearly visible in the image.
[0037] (2) Next, we use advanced image denoising techniques to remove the noise in the image. Noise can be caused by many factors, such as sensor noise, transmission noise, etc. By applying denoising algorithms, we can effectively reduce these noises and improve the clarity and quality of the image.
[0038] (3) To address the lens distortion problem, lens distortion correction technology is used. This technology automatically corrects the image by analyzing the distortion pattern produced by the lens. After correction, the fishing boats in the image will show a more realistic shape and size, avoiding the distortion caused by lens distortion.
[0039] (4) In addition, in order to further enhance the visual effect of the image, color balance and saturation adjustment techniques need to be applied. These techniques can ensure that the colors of the fishing boats in the image are realistic and bright, and closer to the fishing boats in the real world.
[0040] (5) Finally, image sharpening techniques are used to enhance the edges and details of the fishing boat; by enhancing the high-frequency information in the image, this can make the edges of the fishing boat clearer and the details richer. This is crucial for the subsequent feature extraction task.
[0041] S4. Image feature value extraction and calculation: After comprehensive comparison of multiple similarity algorithms, this application finally determined the LBP feature vector to extract the feature value of the ship, and used the comparison between the feature values as the similarity judgment standard of the fishing boats, and finally normalized it to obtain the similarity value.
[0042] The above specific implementation methods are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to examples, a person skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.
Claims
1. A ship similarity comparison method, characterized in that: The following steps are involved: S1, segmentation of fishing boats in the image; S2, regularization of the segmented images, setting fixed pixels, and normalizing all images according to fixed pixels; S3, comprehensive processing of image quality; S4. Extraction and calculation of image feature values.
2. The comparison method according to claim 1, characterized in that: The S1 includes building a deep learning model, using the YOLO framework, separating the fishing boats in the picture, with 250-280 training rounds, a learning rate of 0.0001-0.0003, a decay point of 210-260, a decay coefficient of 0.05-0.2, an initial value of the learning rate of 0, a linear growth step of 3000-5000, and an optimizer using a momentum algorithm and L2 regularization, with a regularization coefficient of 0.0005.
3. The comparison method according to claim 1, characterized in that: In S2, a dynamic maximization solution may be adopted to batch calculate the largest picture in the pictures in S1, set the largest picture as a fixed pixel, and regularize other pictures according to the fixed pixels.
4. The comparison method according to claim 1, characterized in that: The S3 includes: Step (1) uses automatic brightness and contrast adjustment technology to improve the overall appearance of the fishing boat image; by analyzing the brightness and contrast distribution in the image, the parameters are automatically adjusted to optimize the visual effect. For areas that are too dark or too bright, the automatic brightness and contrast adjustment technology will perform local enhancement to ensure that the details of the fishing boat are clearly visible in the image.
5. The comparison method according to claim 4, characterized in that: The S3 includes: Step (2) uses image denoising technology to eliminate noise in the image; the noise includes sensor noise, transmission noise, etc. The image clarity and quality are improved by using image denoising technology.
6. The comparison method according to claim 5, characterized in that: The S3 includes: Step (3) uses lens distortion correction technology; by analyzing the distortion pattern produced by the lens, the image is automatically corrected.
7. The comparison method according to claim 6, characterized in that: The S3 includes: Step (4) applies color balance and saturation adjustment techniques.
8. The comparison method according to claim 7, characterized in that: The S3 includes: Step (5) uses image sharpening technology to enhance the edge and details of the fishing boat; by enhancing the high-frequency information in the image, the edge of the fishing boat is clearer and the details are richer.
9. The comparison method according to claim 1, characterized in that: In the S4, the LBP feature vector is used to extract the feature value of the ship, and the comparison between the feature values is used as the similarity judgment standard of the fishing boats, and finally normalization is performed to obtain the similarity value.