Maritime radar oil film segmentation method and system based on improved salient feature and som

Abstract

The application provides a maritime radar oil film segmentation method and system based on improved saliency features and SOM, and relates to the technical field of oil film image segmentation.The technical points of the application include: preprocessing the obtained original maritime radar image;extracting the saliency features of the preprocessed image to obtain an image of an area of interest;using an improved self-organizing mapping network to segment the oil film in the image of the area of interest to obtain a segmented image.By introducing the saliency features, the texture roughness and boundary irregularity of the oil film under different scales are effectively described, and the ability to distinguish the oil film from the seawater background under complex sea conditions is improved; meanwhile, in combination with the improved self-organizing mapping network, the problems that the traditional clustering method is prone to falling into local optimization and the classification boundary is fuzzy when processing high-dimensional and nonlinear oil film features are effectively solved, and the precision and stability of the oil film region segmentation are further improved.The application improves the intelligent level of oil film monitoring under complex environments.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil film image segmentation, and particularly relates to a maritime radar oil film segmentation method and system based on improved saliency features and SOM. BACKGROUND

[0002] The offshore oil film image separation technology is one of the core technologies for solving the monitoring and treatment of marine oil leakage pollution, and the core goal is to accurately extract the oil film region from the complex marine background (such as seawater, waves, foam, and light changes) to provide data support for pollution range assessment, leakage source tracing, and subsequent cleaning operations.

[0003] In the existing image segmentation method, the feature extraction of the oil film image cannot effectively depict the texture roughness and boundary irregularity of the oil film at different scales; at the same time, the existing self-organizing map network (Self-Organizing Map, SOM) is prone to fall into local optimum and the classification boundary is fuzzy when clustering based on high-dimensional and nonlinear oil film features; thereby causing the oil film segmentation method based on the self-organizing map network to have certain defects in precision and stability. SUMMARY

[0004] In view of the above problems, the present application provides a maritime radar oil film segmentation method and system based on improved saliency features and SOM.

[0005] According to an aspect of the present application, a maritime radar oil film segmentation method based on improved saliency features and SOM is provided, which comprises:

[0006] Pretreating the obtained maritime radar original image;

[0007] Extracting saliency features of the pretreated image to obtain an image of interest;

[0008] Using an improved self-organizing map network to segment the oil film of the image of interest to obtain a segmented image.

[0009] Further, the extraction of saliency features of the pretreated image to obtain an image of interest comprises: filling the pretreated image, setting a sliding window, and calculating the saliency features of the filled image in each sliding window; using a dynamic threshold T to perform binaryzation processing on the saliency image F composed of the saliency features of the multiple sliding windows: assigning the pixel value less than the dynamic threshold T in the saliency image F as 1, and assigning other pixel values as 0; post-processing the binaryzation image to obtain the image of interest.

[0010] Further, the calculation of the saliency features of the image in each sliding window comprises: for the center position (i, j) of each sliding window, calculating the threshold value through the Gaussian weighted mean value Based on the threshold of each sliding window Statistical sliding window within which the following conditions are met number of foreground pixels , Indicates the image after filling In position The pixel values ​​are given, where x and y are the coordinate offsets of each pixel relative to the window center (i,j); the perimeter E of the sliding window is calculated; and the number of foreground pixels within the sliding window is calculated. The saliency features of the image within each sliding window are obtained by comparing the ratio of the logarithm of the sliding window to the logarithm of the perimeter E of the sliding window, and then normalized.

[0011] Furthermore, the threshold is calculated using a Gaussian weighted average. The calculation formula is:

[0012] ;

[0013] in, The fill size is equal to the sliding window radius; C is a constant offset. Indicates the Gaussian kernel weights;

[0014] The formula for calculating the perimeter E of the sliding window is: ;in This indicates the side length of the sliding window.

[0015] Furthermore, the formula for calculating the dynamic threshold T is as follows:

[0016] ;

[0017] in, The global mean of the saliency image F; The global pixel standard deviation of the saliency image F; The polarization coefficient of the saliency image F; and These are the standard deviation adjustment coefficient and the polarization attenuation coefficient, respectively.

[0018] Furthermore, the post-processing of the binarized image to obtain the region of interest image includes: performing region filtering on the binarized image to remove noise points with a connected region area of ​​less than 50 pixels, inverting the background, and marking the connected regions to generate a binary region of interest mask; removing the upper half of the image without waves and removing independent small target regions to obtain the region of interest image.

[0019] Furthermore, the improvements in the improved self-organizing mapping network include:

[0020] The weight vector of each neuron j The update formula is:

[0021] ;

[0022] Where t represents the current training iteration number of the self-organizing map network; Indicates the learning rate. ; x represents the pixel feature vector extracted from the region of interest image; The improved Gaussian neighborhood function is represented as follows:

[0023] ;

[0024] in, The position of the j-th neuron in the one-dimensional linear structure; The best matching unit for x; Indicates distance from Manhattan; Represents the neighborhood radius.

[0025] According to another aspect of the present invention, a marine radar oil film segmentation system based on improved saliency features and SOM is proposed, the system comprising:

[0026] The preprocessing module is configured to preprocess the acquired raw marine radar images in pairs;

[0027] The region of interest extraction module is configured to extract salient features from the preprocessed image to obtain the region of interest image;

[0028] The oil film segmentation module is configured to use an improved self-organizing map network to segment the oil film in the region of interest image and obtain a segmented image.

[0029] According to another aspect of the present invention, a terminal device is provided, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the marine radar oil film segmentation method based on improved saliency features and SOM as described above.

[0030] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the marine radar oil film segmentation method based on improved saliency features and SOM as described above.

[0031] The beneficial technical effects of this invention are:

[0032] This invention proposes a marine radar oil film segmentation method and system based on improved saliency features and Self-Organizing Map (SOM). By introducing saliency features, the method effectively characterizes the texture roughness and boundary irregularities of oil films at different scales, improving the ability to distinguish oil films from the seawater background under complex sea conditions and enhancing the robustness and accuracy of detection. Simultaneously, by combining an improved self-organizing map network and introducing a dynamic neighborhood function and adaptive learning rate mechanism, the method effectively solves the problems of traditional clustering methods easily getting trapped in local optima and blurred classification boundaries when dealing with high-dimensional, nonlinear oil film features, further improving the accuracy and stability of oil film region segmentation. This invention provides a new technical path for the efficient identification and accurate extraction of marine oil films, improves the intelligent level of oil film monitoring in complex environments, and can strongly support scientific decision-making in marine environmental emergency response and pollution control. Attached Figure Description

[0033] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0034] Figure 1 This is a flowchart of the marine radar oil film segmentation method based on improved saliency features and SOM as described in the embodiments of the present invention;

[0035] Figure 2 This is an example of a preprocessed image in an embodiment of the present invention;

[0036] Figure 3 This is an example of a normalized saliency image in an embodiment of the present invention;

[0037] Figure 4 This is an example of a region of interest image in an embodiment of the present invention;

[0038] Figure 5 This is an example of a segmentation result image in an embodiment of the present invention;

[0039] Figure 6 This is an example of an image fused with the segmentation result image and the preprocessed image in an embodiment of the present invention;

[0040] Figure 7 This is an example of an image after the fused image has been converted back to a polar coordinate system in an embodiment of the present invention;

[0041] Figure 8 This is a schematic diagram of the marine radar oil film segmentation system based on improved saliency features and SOM as described in an embodiment of the present invention. Detailed Implementation

[0042] The principles and spirits of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only for better understanding of the present application by those skilled in the art, and do not limit the scope of the present application in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0043] Those skilled in the art know that the embodiments of the present application can be implemented as a system, a device, an apparatus, a method or a computer program product. Therefore, the present disclosure can be embodied in the form of entire hardware, entire software (including firmware, resident software, microcode, etc.), or a combination of hardware and software. In this context, it should be understood that any number of elements in the drawings are used for illustration only, not for limitation, and any naming is only for distinction, not for any limiting meaning.

[0044] The present application aims at the problem that offshore oil film images are difficult to extract accurately under complex sea conditions (such as illumination change, wave interference, etc.), and proposes an intelligent segmentation method combining improved saliency features and self-organizing mapping network (SOM). The traditional SOM is prone to local optimization and fuzzy classification boundary when dealing with high-dimensional and nonlinear oil film features. The present application introduces a dynamic neighborhood function and an adaptive learning rate mechanism, combines a Gaussian neighborhood and an improved distance function, and enhances the convergence and clustering stability of the SOM network. At the same time, the saliency features effectively describe the roughness of the oil film texture and the irregularity of the boundary, and improve the ability to distinguish oil film and seawater under complex background. Thus, the accurate positioning and stable segmentation of the oil film region are realized, which provides key technical support for offshore oil spill monitoring and management.

[0045] The embodiment of the present application proposes a maritime radar oil film segmentation method based on improved saliency features and SOM, as shown in Figure 1 The method comprises the following steps:

[0046] S1, preprocessing the acquired maritime radar original image;

[0047] S2, extracting saliency features of the preprocessed image to obtain an image of interest;

[0048] S3, using an improved self-organizing mapping network to segment the oil film of the image of interest to obtain a segmented image.

[0049] The method starts from S1. In S1, the acquired maritime radar original image is preprocessed.

[0050] According to an embodiment of the present invention, in the image preprocessing stage, the original marine radar image is first transformed into a coordinate system, that is, the polar coordinates of the image are transformed into a Cartesian coordinate system. Then, noise reduction for co-frequency interference and speckle noise suppression are performed, followed by local contrast enhancement. The preprocessed image is shown below. Figure 2 As shown.

[0051] Then in S2, the salient features of the preprocessed image are extracted to obtain the region of interest image.

[0052] According to an embodiment of the present invention, the method includes: S21, performing image filling on the preprocessed image and setting a sliding window, and calculating the salient features of the filled image within each sliding window; S22, performing binarization processing on the salient image F composed of the salient features of multiple sliding windows using a dynamic threshold T; S23, performing post-processing on the binarized image to obtain an interest region image.

[0053] S21 specifically includes: First, performing image filling on the preprocessed image using symmetrical filling; and setting a sliding window, in this embodiment, the window size is 15×15. For each window center position (i,j), a threshold is calculated using a Gaussian weighted mean. :

[0054] ;

[0055] in, This represents the fill size, which is equal to the window radius. w represents the side length of the sliding window. Symmetrical filling is used to ensure that the sliding window can still cover the complete neighborhood when it is at the edge of the image and to avoid edge effects by mirroring the boundary values. Indicates the image after filling In position The pixel value; C is a constant offset, whose core function is to adjust the overall offset of the threshold to adapt to the overall grayscale difference between the oil film and the background; G(x,y) represents the Gaussian kernel weight, defined as:

[0056] ;

[0057] Where x and y are the coordinate offsets of each pixel relative to the window center (i,j), and define the position of each pixel within the kernel to calculate its spatial distance from the center point; The standard deviation is the Gaussian distribution, which controls the rate of weight decay. The larger the value, the smoother the weight distribution; conversely, the smaller the value, the more concentrated the weights are in the center. Make the sum of all kernel weights equal to 1, that is This ensures that the threshold calculation is a weighted average rather than an absolute value summation and avoids bias due to kernel size or other factors. The changes lead to inconsistencies in the threshold scale.

[0058] Then, the number of pixels in the sliding window satisfying is counted :

[0059] ;

[0060] wherein is an indicator function that returns 1 when the condition is true, and 0 otherwise; i.e. when the pixel value is less than the local threshold.

[0061] Then, the perimeter E of the sliding window is calculated; in the saliency computation, the perimeter E usually refers to the boundary length of the target region in the binary image. For the binary region in the sliding window, its perimeter is calculated by the actual boundary pixels through the edge detection algorithm. In the present embodiment, the sliding window side length is used to calculate the perimeter E: ; i.e. approximately represents the "boundary span" in the single direction.

[0062] Then, the saliency feature of the image in the sliding window is calculated:

[0063] ;

[0064] The saliency feature of the local region is estimated by calculating the ratio of the logarithm of the number of foreground pixels in the sliding window to the logarithm of the perimeter E of the sliding window.

[0065] Then, the saliency feature is linearly normalized: ; wherein D represents a matrix composed of the saliency feature values of all the sliding windows, is the minimum value in the matrix D; is the maximum value in the matrix D. The normalized saliency image F composed of the saliency features of multiple sliding windows is shown as Figure 3 .

[0066] S22 specifically comprises: obtaining the threshold T using the dynamic Gaussian threshold calculation method based on the obtained normalized image, and dynamically adjusting the weight of the standard deviation through the exponential term so that the threshold can be self-adaptive to the statistical characteristics of the image content; wherein the calculation formula of the threshold T is:

[0067] ;

[0068] ​wherein, is the global mean of the saliency image F; is the global pixel standard deviation of the saliency image F; is the polarization coefficient of the saliency image F; and are the standard deviation adjustment coefficient and the polarization attenuation coefficient respectively, and in the embodiment of the present application, the values are set to 2.0 and 0.2 respectively.

[0069] Subsequently, the normalized saliency image F is binarized using a dynamic threshold T:

[0070] ;

[0071] wherein is the binary segmentation result, 1 representing the foreground (high-dimensional area) and 0 representing the background.

[0072] S23 specifically comprises: performing region filtering on the binarized image to remove noise points with a connected region area less than 50 pixels, background inversion, and labeling the connected regions to generate a binary interest area mask; and finally performing post-processing to remove the area without sea waves in the upper half of the image and remove independent small target areas to obtain a final interest area image, as shown in Figure 4 .

[0073] Then in S3, an improved self-organizing mapping network (SOM) is used to segment the interest area image to obtain a segmentation image.

[0074] According to the embodiment of the present application, the SOM is a neural network of unsupervised learning, which realizes the nonlinear mapping of high-dimensional data to low-dimensional space through competitive learning. The SOM structure is a neuron network of 1 row x 10 columns, each neuron has a weight vector , wherein D is the feature dimension. In the training process, the Euclidean distance is calculated between each input sample x and all neuron weights : ; the neuron with the minimum distance (Best Matching Unit, BMU) is selected as the winner, wherein the BMU is the best matching unit, i.e. the neuron to which the sample is most matched; and then the weight is updated, i.e. the weights of the neurons in the neighborhood are updated according to the neighborhood function :

[0075] ;

[0076] wherein j represents the index of the neuron in the output layer grid; t represents the current training iteration number of the SOM neural network; is the learning rate; x represents the input sample, i.e. the pixel feature vector extracted from the interest area image; is a Gaussian neighborhood function, whose formula is:

[0077] ;

[0078] where, represents the best matching unit position; represents the i-th neuron position; is a neighborhood radius, which gradually decays with training iterations, used to control the neighborhood range.

[0079] Each pixel is mapped to the nearest neuron (BMU) by SOM to generate a cluster label. The label matrix is reshaped to the original image size to form a segmentation result.

[0080] Compared with the traditional SOM, the embodiments of the present application have the following improvements:

[0081] 1) Optimize the SOM network structure and training parameters:

[0082] The initial value of the learning rate is set to 0.05, and the decay rate parameter is enabled, and the decay rate is 95% for each iteration, so as to avoid oscillation in the later training, that is, the calculation formula of the learning rate is: ; This mechanism keeps the learning rate at a larger value in the early training (accelerates convergence), and then gradually decays by a fixed ratio (95%), and the calculation formula of the learning rate is integrated into the weight update formula.

[0083] The above dynamic decay strategy takes into account the needs of fast clustering in the early training and fine adjustment in the later period, effectively avoids oscillation, and improves the convergence stability and generalization ability.

[0084] 2) Improve the neighborhood function is :

[0085] The present application adopts a one-dimensional linear topology structure, arranges the neurons as a linear structure mode of 1 K, and changes the neighborhood function to Manhattan distance. That is, the calculation formula of the neighborhood function value is:

[0086] ;

[0087] where, is the position (scalar) of the j-th neuron in the one-dimensional linear structure; is the position of the winning neuron (BMU), that is, the best matching unit of x; is the neighborhood radius; represents the Manhattan distance.

[0088] The structure adaptability, i.e., one-dimensional topology, is more in line with the image scanning sequence (such as line-by-line processing), and can reduce irrelevant neuron interference.

[0089] The improved self-organizing mapping network is used for oil film segmentation of the region of interest image, and a segmentation result image is as shown in Figure 5 .

[0090] Further, the oil film segmentation image is post-processed to obtain a final oil film segmentation image. According to the embodiment of the present application, the segmentation result image is fused into the pre-processed image, as shown in Figure 6 ; and finally, the result is converted back to the polar coordinate system, as shown in Figure 7 .

[0091] The embodiment of the present application also proposes a maritime radar oil film segmentation system based on improved saliency features and SOM, as shown in Figure 8 , which comprises:

[0092] The pre-processing module 410 is configured to pre-process the acquired original maritime radar image;

[0093] The region of interest extraction module 420 is configured to extract saliency features of the pre-processed image to obtain a region of interest image;

[0094] The oil film segmentation module 430 is configured to use the improved self-organizing mapping network to perform oil film segmentation on the region of interest image to obtain a segmentation image.

[0095] Each module is connected through wired and / or wireless means to realize data transmission between modules.

[0096] The function of the maritime radar oil film segmentation system based on improved saliency features and SOM according to the embodiment of the present application can be explained by the aforementioned maritime radar oil film segmentation method based on improved saliency features and SOM, so the unexplained part of the system embodiment can be referred to the method embodiment above, and will not be repeated here.

[0097] Another embodiment of the present application proposes a terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the maritime radar oil film segmentation method based on improved saliency features and SOM according to the aforementioned embodiment when executing the computer program.

[0098] Since the terminal device introduced in the embodiment is the electronic device used in implementing the method for maritime radar oil film segmentation based on improved saliency features and SOM in the embodiment of the application, based on the method for maritime radar oil film segmentation based on improved saliency features and SOM in the embodiment of the application, those skilled in the art can understand the specific implementation mode of the electronic device of the embodiment and various changes thereof, so how the electronic device realizes the method in the embodiment of the application will not be introduced in detail here.

[0099] Another embodiment of the application provides a computer readable storage medium comprising a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to perform the method for maritime radar oil film segmentation based on improved saliency features and SOM described in the above embodiment when the computer program runs.

[0100] It should be noted that although several units, modules or sub-modules are mentioned in the above detailed description, such division is merely exemplary and not mandatory. In fact, according to the embodiments of the application, the features and functions of two or more modules described above can be embodied in one module. Conversely, the features and functions of one module described above can be further divided into modules for embodiment.

[0101] In addition, although the operations of the method of the application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all of the shown operations must be performed to achieve the desired result. Additionally or alternatively, certain steps can be omitted, a plurality of steps can be combined into one step, and / or one step can be divided into a plurality of steps.

[0102] Although the spirit and principles of the application have been described with reference to several specific embodiments, it should be understood that the application is not limited to the disclosed specific embodiments, and the division of aspects does not mean that the features in these aspects cannot be combined to benefit. Such division is only for the convenience of expression. The application is intended to cover various modifications and equivalent arrangements included in the spirit and scope of the appended claims.

Claims

1. A method for oil slick segmentation in marine radar based on improved saliency features and SOM, characterized in that, The method comprises the following steps: preprocessing the obtained marine radar raw image; extracting the saliency features of the preprocessed image to obtain an image of a region of interest; the method comprises the following steps: The preprocessed image is filled with image, and a sliding window is set, and the saliency feature of the filled image in each sliding window is calculated, including: for the center position (i, j) of each sliding window, the threshold value is calculated by Gaussian weighted mean ; according to the threshold value of each sliding window , the number of foreground pixels in the sliding window satisfying is counted , , wherein f represents the pixel value of the filled image at position , x and y are the coordinate offsets of each pixel relative to the window center (i, j); the perimeter E of the sliding window is calculated; the saliency feature of the image in each sliding window is obtained by calculating the ratio of the logarithm of the number of foreground pixels in the sliding window to the logarithm of the perimeter E of the sliding window, and the saliency feature is normalized. using a dynamic threshold T to perform binaryzation processing on a saliency image F composed of saliency features of multiple sliding windows: assigning a pixel value less than the dynamic threshold T in the saliency image F as 1, and assigning other pixel values as 0; performing post-processing on the binaryzation image to obtain an image of a region of interest; using an improved self-organizing mapping network to perform oil film segmentation on the image of the region of interest to obtain a segmented image; the improvement in the improved self-organizing mapping network comprises: weight vector of each neuron q The update formula is: ； Wherein, t represents the current training iteration number of the self-organizing mapping network; denotes the learning rate, ; z represents the pixel feature vector extracted from the region of interest image; denotes the improved Gaussian neighborhood function: ； wherein, position of the qth neuron in the one-dimensional linear structure; best matching unit for z; denotes the Manhattan distance; denotes the neighborhood radius.

2. The method for shipborne radar oil slick segmentation based on improved saliency feature and SOM according to claim 1, characterized in that, The threshold value is calculated by a Gaussian weighted mean The calculation formula is: ； wherein, represents a fill size, which is equal to the sliding window radius; C is a constant offset; represents a Gaussian kernel weight; The calculation formula of the circumference E of the sliding window is: ; wherein represents the side length of the sliding window.

3. The method for ship radar oil film segmentation based on improved saliency feature and SOM according to claim 1, characterized in that, the calculation formula of the dynamic threshold T is: ； wherein, denotes the global mean of the saliency image F; denotes the global pixel standard deviation of the saliency image F; denotes the polarization coefficient of the saliency image F; and denote the standard deviation adjustment coefficient and the polarization decay coefficient, respectively.

4. The method for ship radar oil film segmentation based on improved saliency feature and SOM according to claim 1, characterized in that, the post-processing of the binaryzation image to obtain an image of a region of interest comprises: performing region filtering on the binaryzation image to remove noise points with a connected region area less than 50 pixels, background inversion, and labeling the connected region to generate a binary region of interest mask; removing regions without sea waves in the upper half of the binary region of interest mask and removing independent small target regions to obtain an image of a region of interest.

5. The marine radar oil film segmentation system based on improved saliency feature and SOM, characterized in that, The system is implemented based on the method for marine radar oil film segmentation based on improved saliency features and SOM according to any one of claims 1-4; the system comprises: a preprocessing module configured to preprocess the obtained marine radar raw image; a region of interest extraction module configured to extract the saliency features of the preprocessed image to obtain an image of a region of interest; an oil film segmentation module configured to use an improved self-organizing mapping network to perform oil film segmentation on the image of the region of interest to obtain a segmented image.

6. A terminal device, characterized by comprising: The computer program is stored in the memory and configured to be executed by the processor, and the processor implements the method for marine radar oil film segmentation based on improved saliency features and SOM according to any one of claims 1-4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the method for marine radar oil film segmentation based on improved saliency features and SOM according to any one of claims 1-4 when the computer program is running.

Images