An infrared sea surface ship target detection method and system with resistance to fish scale wave interference

Abstract

The present invention discloses a method and system for infrared sea surface ship target detection that is resistant to fish-scale wave interference. The method relates to the technical field of infrared sea surface ship target detection, and includes acquiring a target infrared sea surface image; constructing a gradient-domain-based fish-scale wave suppression algorithm based on gradient-domain processing, gradient field restoration, guided filtering, and an atmospheric scattering model; constructing an infrared ship target detection model resistant to fish-scale wave interference based on the gradient-domain-based sea surface fish-scale wave suppression algorithm and detection model; and inputting the target infrared sea surface image into the infrared ship target detection model resistant to fish-scale wave interference to obtain a ship target recognition result. The present invention suppresses sea surface fish-scale light clutter, improving the contrast of the target area, ship target recognition accuracy in complex environments, and anti-interference capability.

Description

Technical Field

[0001] The present invention relates to the technical field of infrared sea surface ship target detection, in particular to an infrared sea surface ship target detection method and system resistant to fish scale wave interference. Background Art

[0002] Aiming at the demand for intelligent identification of ship targets in complex sea environments, the target detection model based on deep learning has obvious advantages in the field of natural images. However, when it is applied to infrared sea surface images, the signal-to-noise ratio of infrared images caused by complex environmental factors is relatively low, and the target may be submerged by the noise, especially under the interference of clutter such as fish scale waves, presenting flickering texture features, which will lead to misidentification and missed identification.

[0003] Therefore, there is an urgent need for an infrared sea surface ship target detection method and system that is resistant to fish scale wave interference to solve the noise problem of the target under the interference of clutter such as fish scale waves. Summary of the Invention

[0004] To solve the above problems, this application proposes an infrared sea surface ship target detection method and system that is resistant to fish scale wave interference, which suppresses fish scale light clutter on the sea surface, improves the contrast of the target area, and the accuracy of ship target recognition and anti-interference ability in complex environments.

[0005] The present invention provides an infrared sea surface ship target detection method that is resistant to fish scale wave interference, comprising the following steps:

[0006] S1. Acquire target infrared sea surface image;

[0007] S2. Based on gradient domain processing, gradient field restoration, guided filter processing and atmospheric scattering model, a fish-scale wave suppression algorithm based on gradient domain is constructed;

[0008] S3. Based on the sea surface fish scale wave suppression algorithm and detection model based on the gradient domain, an infrared ship target detection model that is resistant to fish scale wave interference is constructed;

[0009] S4. Input the target infrared sea surface image into the infrared ship target detection model that is resistant to fish scale wave interference to obtain a ship target recognition result; the ship target recognition result includes: target category, position and predicted probability.

[0010] Preferably, in S2, a fish-scale wave suppression algorithm based on gradient domain is constructed according to gradient domain processing, gradient field restoration, guided filter processing and atmospheric scattering model, specifically including:

[0011] S201, obtaining historical infrared sea surface images and historical ship target recognition results;

[0012] S202, performing gradient domain processing and gradient field restoration on the historical infrared sea surface image in sequence according to the Scharr operator, gradient domain transformation equation, cumulative distribution function, and Poisson equation to obtain a gradient field restored image;

[0013] S203, using a guided filtering equation to perform guided filtering processing on the gradient field restoration image to obtain a guided filtering processed image;

[0014] S204: Using an atmospheric scattering model, the contrast of the guided filter processed image is enhanced to obtain a historical ship target recognition result.

[0015] Preferably, in S202, according to the Scharr operator, the gradient domain transformation equation, the cumulative distribution function and the Poisson equation, the historical infrared sea surface image is sequentially subjected to gradient domain processing and gradient field restoration to obtain a gradient field restored image, specifically including:

[0016] The Scharr operator is used to calculate the gradient of the historical infrared sea surface image to obtain the gradient amplitude of the original gradient field;

[0017] Based on the gradient amplitude of the original gradient field, the minimum gradient threshold is set to determine the target area;

[0018] The gradient domain transformation equation is used to compress the dynamic range of the target area and obtain the transformation gradient amplitude;

[0019] After mapping the transformed gradient amplitude through the cumulative distribution function, the transformed gradient amplitude is used to construct a new gradient field;

[0020] Using Poisson's equation, the gradient domain of the new gradient field is converted into a spatial domain image to obtain the gradient field restoration image.

[0021] Preferably, the expression of the gradient amplitude of the original gradient field is:

[0022] ;

[0023] in, Represents the gradient magnitude of the image, that is, the gradient vector The size of is the horizontal direction, y is the longitudinal direction, is the gradient component in the horizontal direction (x direction), G y is the gradient component in the vertical direction (y direction).

[0024] Preferably, the gradient domain transformation equation is:

[0025] ;

[0026] in, is the gradient amplitude The new gradient amplitude after gradient domain transformation, is the maximum gradient amplitude, is the minimum gradient threshold, is the maximum gradient threshold, is a logarithmic function.

[0027] Preferably, in S203, guided filtering equations are used to perform guided filtering processing on the gradient field restoration image to obtain a guided filtering processed image, which specifically includes:

[0028] Determine a guide image and an input image; the guide image is a historical infrared sea surface image; the input image is a gradient field restoration image;

[0029] Based on the guided image, the input image is filtered using the guided filtering equation to obtain a guided filtering processed image.

[0030] Preferably, the guided filtering equation is:

[0031] ;

[0032] in, is the pixel position, Pixel A local window centered on is any pixel in the local window, The pixel values ​​of the image are processed for guided filtering. is the pixel value of the guidance image, the coefficient and are the coefficients of the linear function when the window center is at pixel k.

[0033] Preferably, in S204, an atmospheric scattering model is used to enhance the contrast of the guided filter processed image to obtain a historical ship target recognition result, which specifically includes:

[0034] Based on historical infrared sea surface images, guided filter processed images, atmospheric light intensity and transmittance, an atmospheric scattering equation is constructed to obtain an atmospheric scattering model.

[0035] The guided filter processed image is input into the atmospheric scattering model to obtain the historical ship target recognition result.

[0036] Preferably, the atmospheric scattering equation is:

[0037] ;

[0038] in, is the pixel position, is a contrast-enhanced infrared sea surface image. is a historical infrared sea surface image. To process the image for guided filtering, is the atmospheric light intensity, is the transmittance, is the ambient light item.

[0039] Preferably, in S3, an infrared ship target detection model resistant to fish-scale wave interference is constructed according to a sea surface fish-scale wave suppression algorithm and detection model based on a gradient domain, specifically including:

[0040] Construct training and test sets based on historical infrared sea surface images and historical ship target recognition results;

[0041] The gradient domain-based sea surface fish scale wave suppression algorithm and the detection model are connected to obtain an untrained infrared ship target detection model that is resistant to fish scale wave interference.

[0042] The training set is input into an untrained infrared ship target detection model resistant to fish scale wave interference, and a stochastic gradient descent optimizer is used to train the untrained infrared ship target detection model resistant to fish scale wave interference to obtain a trained infrared ship target detection model resistant to fish scale wave interference;

[0043] Inputting the test set into the trained infrared ship target detection model resistant to fish scale wave interference, adjusting the trained infrared ship target detection model resistant to fish scale wave interference, and obtaining a trained infrared ship target detection model resistant to fish scale wave interference;

[0044] The trained infrared ship target detection model resistant to fish scale wave interference is determined as the infrared ship target detection model resistant to fish scale wave interference.

[0045] The present invention provides an infrared sea surface ship target detection system that is resistant to fish scale wave interference, comprising:

[0046] Target acquisition module, used to acquire target infrared sea surface images;

[0047] An algorithm building module is used to construct a gradient domain-based fish-scale wave suppression algorithm based on gradient domain processing, gradient field restoration, guided filter processing, and atmospheric scattering model;

[0048] A model building module is used to build an infrared ship target detection model that is resistant to fish-scale wave interference based on a sea surface fish-scale wave suppression algorithm based on a gradient domain;

[0049] The target recognition module is used to input the target infrared sea surface image into the infrared ship target detection model that is resistant to fish scale wave interference to obtain the ship target recognition result; the ship target recognition result includes: target category, position and predicted probability.

[0050] In summary, the infrared sea surface ship target detection method and system with resistance to fish scale wave interference of the present invention has the following advantages over traditional technologies:

[0051] (1) The sea surface fish scale wave suppression algorithm based on the gradient domain is used to construct an infrared ship target detection model that is resistant to fish scale wave interference. This model suppresses the sea surface fish scale light clutter, improves the contrast of the target area, and improves the ship target recognition accuracy and anti-interference ability in complex environments.

[0052] (2) The fish-scale wave suppression algorithm based on the gradient domain proposed in this invention, due to the introduction of gradient domain processing restoration, guided filtering and atmospheric scattering model, can not only effectively improve the target prediction confidence in scenes with different fish-scale wave interference levels, but also significantly improve the visual effect.

[0053] The technical method of the present invention is further described in detail below through the accompanying drawings and examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The grayscale image, grayscale histogram, gradient image and gradient amplitude histogram of the fish scale wave infrared sea surface image of the present invention are as follows; wherein, Figure 1 (a) is a grayscale image of the infrared sea surface image containing fish scale waves; Figure 1 (b) is the grayscale histogram of the infrared sea surface image containing fish scale waves; Figure 1 (c) is the gradient map of the infrared sea surface image containing fish scale waves; Figure 1 (d) is the gradient amplitude histogram of the infrared sea surface image containing fish scale waves;

[0055] Figure 2 The grayscale image, grayscale histogram, gradient image and gradient amplitude histogram of the fish scale-free infrared sea surface image of the present invention are as follows; wherein, Figure 2 (a) is the grayscale image of the infrared sea surface without fish scale wave; Figure 2 (b) is the grayscale histogram of the infrared sea surface image without fish scale wave; Figure 2 (c) is the gradient map of the infrared sea surface image without fish scale wave; Figure 2 (d) is the gradient amplitude histogram of the infrared sea surface image without fish scale wave;

[0056] Figure 3 This is a flow chart of an infrared sea surface ship target detection method that is resistant to fish scale wave interference according to the present invention;

[0057] Figure 4 This is a diagram of the architecture of the infrared sea surface ship target detection model that is resistant to fish scale wave interference according to the present invention;

[0058] Figure 5 This is a processing effect diagram of the fish-scale wave suppression algorithm based on the gradient domain in the present invention; Figure 5(a) is a schematic diagram of the original infrared image of scene 1; Figure 5 (b) is a schematic diagram of the effect of gradient domain processing and restoration of the original infrared image of scene 1; Figure 5 (c) is a schematic diagram showing the effect of gradient domain processing, restoration, and guided filtering on the original infrared image of scene 1. Figure 5 (d) is a schematic diagram showing the effect of the original infrared image of scene 1 after gradient domain processing and restoration, guided filtering processing, and atmospheric scattering model processing; Figure 5 (e) is a schematic diagram of the original infrared image of scene 2; Figure 5 (f) is a schematic diagram of the effect of gradient domain processing and restoration of the original infrared image of scene 2; Figure 5 (g) is a schematic diagram of the effect of the original infrared image of scene 2 being processed by gradient domain, restoration and guided filtering in sequence; Figure 5 (h) is a schematic diagram of the effect of the original infrared image of scene 2 after gradient domain processing and restoration, guided filtering processing and atmospheric scattering model; Figure 5 (i) is a schematic diagram of the original infrared image of scene three; Figure 5 (j) is a schematic diagram of the effect of gradient domain processing and restoration of the original infrared image of scene 3; Figure 5 (k) is a schematic diagram of the effect of the original infrared image of scene 3 being processed by gradient domain, restoration and guided filtering in sequence; Figure 5 (l) is a schematic diagram of the effect of the original infrared image of scene 3 after gradient domain processing and restoration, guided filtering processing and atmospheric scattering model;

[0059] Figure 6 This is a comparison chart of the detection effect of the typical fish-scale wave suppression algorithm of the present invention on the fish-scale wave test set; Figure 6 (a) is the detection effect diagram of the YOLOv11n model in scene 1; Figure 6 (b) is the detection effect diagram of scene 1 sharp mask method combined with YOLOv11n model; Figure 6 (c) is the detection effect diagram of scene 1 unsharp masking method combined with YOLOv11n model; Figure 6 (d) shows the detection effect of scenario 1 based on the gradient histogram enhanced equalization algorithm combined with the YOLOv11n model; Figure 6 (e) is the detection effect diagram of scene 1 using the gradient domain-based fish scale wave suppression algorithm combined with the YOLOv11n model; Figure 6 (f) is the detection effect diagram of the YOLOv11n model in scene 2; Figure 6 (g) is the detection effect diagram of scene 2 unsharp mask method combined with YOLOv11n model; Figure 6(h) is the detection effect diagram of scene 2 unsharp masking method combined with YOLOv11n model; Figure 6 (i) shows the detection effect of scenario 2 based on the gradient histogram enhanced equalization algorithm combined with the YOLOv11n model; Figure 6 (j) in the figure shows the detection effect of the gradient domain-based fish scale wave suppression algorithm combined with the YOLOv11n model in scene 2; Figure 6 (k) is the detection effect diagram of the YOLOv11n model in scene three; Figure 6 (l) in the figure is the detection effect diagram of the scene triple-sharpening mask method combined with the YOLOv11n model; Figure 6 (m) in the figure is the detection effect diagram of the scene triple unsharp masking method combined with the YOLOv11n model; Figure 6 (n) in the figure is the detection effect diagram of scenario three based on the gradient histogram enhanced equalization algorithm combined with the YOLOv11n model; Figure 6 (o) in the figure shows the detection effect of the gradient domain-based fish scale wave suppression algorithm combined with the YOLOv11n model in scene three; Figure 6 (p) in the figure is the detection effect diagram of the YOLOv11n model in scene 4; Figure 6 (q) in the figure is the detection effect diagram of scene 1 sharp mask method combined with YOLOv11n model; Figure 6 (r) in the figure is the detection effect diagram of scene 1 unsharp masking method combined with YOLOv11n model; Figure 6 (s) in the figure is the detection effect diagram of scenario 1 based on the gradient histogram enhanced equalization algorithm combined with the YOLOv11n model; Figure 6 (t) in the figure is the detection effect diagram of the gradient domain-based fish scale wave suppression algorithm combined with the YOLOv11n model in scene 1;

[0060] Figure 7 The figure is a schematic diagram of a module of an infrared sea surface ship target detection system that is resistant to fish scale wave interference according to the present invention. DETAILED DESCRIPTION

[0061] The technical method of the present invention is further described below through the accompanying drawings and embodiments. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and values ​​described in these embodiments do not limit the scope of this application.

[0062] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0063] Technologies, systems, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0064] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0065] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0066] Before the infrared sea surface ship target detection is performed on the target infrared sea surface image, the grayscale histogram and gradient histogram are obtained, such as Figure 2 and Figure 3 As shown in the figure, the distribution of infrared sea surface images is analyzed from two angles.

[0067] From the rising or falling trend of the histogram, Figure 1 (b) Figure 1 (d) Figure 2 (b) and Figure 2 As shown in (d), the distribution trend of the grayscale image of the infrared sea surface image is different under different lighting conditions and interference conditions, such as Figure 1 (a) Figure 1 (c) Figure 2 (a) and Figure 2 As shown in (d), there is no obvious pattern and it is difficult to distinguish the target from the background. In the infrared sea surface image with fish scale waves, as shown in Figure 1 (b), the grayscale value of the sea wave background ranges from 100 to 140, and the average grayscale value of the target area is calculated to be 115. In the infrared sea surface image without fish scale waves, as shown in (d), there is no obvious pattern and it is difficult to distinguish the target from the background. Figure 2 As shown in (b), the grayscale value of the wave background ranges from 30 to 100, with a more even distribution. The average grayscale value of the target is 102. This shows that in infrared sea surface images with fish-scale wave interference, the target is submerged in the noise due to the high brightness of the fish-scale wave portion, making it difficult to distinguish between the target and the background. This indicates that using only grayscale features is difficult to distinguish between the target and the background, and may result in both the wave interference and the ship target being enhanced.

[0068] like Figure 1 (b) and Figure 2 As shown in (b) in the figure, it can be seen that although there are several peaks in the grayscale histogram, and the entire gradient histogram is obviously biased towards small gradient value distribution, Figure 1 (d) and Figure 2 As shown in (d) in the figure, the gradient amplitude histogram with and without fish scale wave interference has only one peak. Figure 1 As shown in (d), the wave gradient values ​​are mainly concentrated in the range of 1-5, and the target average gradient value is 22. In the infrared sea surface image without fish scale waves, Figure 2 As shown in (d), the wave gradient values ​​are primarily concentrated between 4 and 6, while the target has an average gradient value of 24. The gradient histogram shows a distinct gradient boundary between the target and the background. Therefore, the algorithm designed in this paper can use a gradient threshold to distinguish the ship target from sea clutter, thereby removing the influence of the waves. Furthermore, although the pixel area covered by the ship target is small, the surrounding gradient values ​​are large. In such images, the target area is more prominent in the gradient value, making the use of gradient domain processing a good idea.

[0069] Therefore, the present invention provides an infrared sea surface ship target detection method that is resistant to fish scale wave interference, such as Figure 3 Shown, including:

[0070] Step S1: Acquire a target infrared sea surface image.

[0071] Step S2: Based on gradient domain processing, gradient field restoration, guided filtering, and an atmospheric scattering model, a gradient domain-based fish-scale wave suppression algorithm is constructed. The present invention utilizes gradient domain processing, gradient field restoration, guided filtering, and an atmospheric scattering model to construct a gradient domain-based fish-scale wave suppression algorithm to improve contrast.

[0072] Step S2 can be replaced by the following steps S201 to S204, as follows:

[0073] Step S201: Acquire historical infrared sea surface images and historical ship target recognition results.

[0074] Step S202: Gradient domain processing and gradient field restoration are performed on the historical infrared sea surface image in sequence according to the Scharr operator, gradient domain transformation equation, cumulative distribution function, and Poisson equation to obtain a gradient field restored image. The present invention uses gradient domain processing to filter out fish scale wave clutter noise and uses gradient mapping to improve the contrast of ship targets.

[0075] Step S202 specifically includes:

[0076] Step S2021: Based on the gradient amplitude of the original gradient field, a minimum gradient threshold is set to determine the target area, and then a gradient domain transformation equation is used to perform dynamic range compression on the target area to obtain a transformed gradient amplitude.

[0077] Step S2021 specifically involves first performing gradient calculation, then setting the gradient threshold, and finally performing gradient domain processing.

[0078] Gradient calculation: Use the Scharr operator to calculate the lateral direction and vertical direction y The horizontal gradient value of the gradient and longitudinal gradient values , which will have better rotational symmetry than the Sobel operator. is the original gradient field of the historical infrared sea surface image, where g is the gradient threshold. The expression of the gradient amplitude of the original gradient field is:

[0079] ;

[0080] in, Represents the gradient magnitude of the image, that is, the gradient vector The size of is the horizontal direction, y is the longitudinal direction, is the gradient component in the horizontal direction (x direction), G y is the gradient component in the vertical direction (y direction).

[0081] Gradient threshold setting: Based on the analysis of infrared sea surface image targets and clutter characteristics, it can be seen that there are boundaries in the gradient amplitudes of the sea surface, sea surface clutter and ship targets in the infrared image. In this invention, the minimum gradient threshold is set to If it is set to 20, the area below the gradient threshold is considered to be sea surface background / clutter noise, and the area above the threshold is considered to be the target area.

[0082] Gradient domain processing: Set the gradient of the area below the minimum gradient threshold to 0 to filter out most of the noise. For the remaining areas, perform dynamic range compression, use logarithmic transformation to enhance the medium gradient area, use the logarithmic function to naturally compress the large gradient value, and retain the linear characteristics of the small gradient. Set the maximum gradient threshold is 100. Among them, the gradient domain transformation equation is:

[0083] ;

[0084] in, is the gradient amplitude The new gradient amplitude after gradient domain transformation, is the maximum gradient amplitude, is the minimum gradient threshold, is the maximum gradient threshold, is a logarithmic function. After gradient transformation, the gradient range is from Transform to .

[0085] Then, histogram equalization is implemented based on the Cumulative Distribution Function (CDF). The gradient amplitude after gradient domain transformation is mapped through CDF to ensure that the enhanced gradient distribution is more uniform and avoid the gradient values ​​in certain areas being too prominent after logarithmic transformation.

[0086] Step S2022: After transforming the gradient amplitude through the cumulative distribution function mapping, the transformed gradient amplitude is used to construct a new gradient field. Finally, the Poisson equation is used to convert the gradient domain of the new gradient field into a spatial domain image to obtain a gradient field restoration image.

[0087] Step S2022 specifically synthesizes a new gradient field first, and then performs gradient field restoration.

[0088] Synthesize a new gradient field: retain the original gradient field direction, use the gradient amplitude after gradient domain processing to construct a new gradient field. The expression is:

[0089] ;

[0090] in, is the new gradient field, is the transverse gradient value of the new gradient field, is the longitudinal gradient value of the new gradient field, is the adjusted gradient amplitude, is the original gradient unit direction vector.

[0091] Gradient field restoration: Based on the Poisson equation, the new gradient field is restored, that is, the gradient domain is converted into a spatial domain image, and the edges and details enhanced by gradient processing are retained.

[0092] Step S203: Using a guided filtering equation, perform guided filtering processing on the gradient field restoration image to obtain a guided filtering processed image.

[0093] Step S203 specifically involves first determining a guide image and an input image, determining the guide image as a historical infrared sea surface image, and the input image as a gradient field restored image. Then, based on the guide image, the input image is filtered using a guided filtering equation to obtain a guided filtered image. The present invention utilizes guided filtering to avoid loss of target contour and texture detail during gradient domain processing, where the gradient values ​​of some ship target contours or internal structures are small and fall below the minimum gradient domain value before being filtered.

[0094] The guided filter is an adaptive weight filter based on a local linear model. The present invention uses the historical infrared sea surface image as the guided image and the gradient field restored image as the input image. The input image is filtered using the guided filter equation. The guided filter equation is:

[0095] ;

[0096] in, is the pixel position, Pixel A local window centered on is any pixel in the local window, The pixel values ​​of the image are processed for guided filtering. is the pixel value of the guidance image, the coefficient and are the coefficients of the linear function when the window center is at pixel k.

[0097] The guided filtering equation is solved by minimizing the cost function:

[0098] ;

[0099] in, is the pixel value of the guidance image, is the pixel value of the input image, is the regularization parameter.

[0100] The present invention uses historical infrared sea surface images as a guide to ensure that the edges of the final guided filter processed image are consistent with the original scene. and regularization parameter Balances smoothing intensity with edge sharpness.

[0101] Step S204: using an atmospheric scattering model to enhance the contrast of the guided filter processed image, and obtaining a historical ship target recognition result.

[0102] Step S204 specifically involves first constructing an atmospheric scattering equation based on the historical infrared sea surface image, the guided filter processed image, the atmospheric light intensity and transmittance to obtain an atmospheric scattering model, and then inputting the guided filter processed image into the atmospheric scattering model to obtain the historical ship target recognition result.

[0103] The guided filter processed image generated in step S203 is generally dark or whitish. The present invention uses an atmospheric scattering model to improve the contrast between the target and the background.

[0104] The atmospheric scattering equation is:

[0105] ;

[0106] in, is the pixel position, is a contrast-enhanced infrared sea surface image. is a historical infrared sea surface image. To process the image for guided filtering, is the atmospheric light intensity, is the transmittance, is the ambient light item.

[0107] Solving for ambient light terms :because , and the transmittance is related to the scene depth. The depth of field difference between adjacent pixels in the local area of ​​the infrared image is small, and the transmittance formula can be expressed by mean filtering.

[0108] ;

[0109] in, average (▪) is the mean value function, Processing images for guided filtering The mean image obtained by mean filtering is is the adjustment coefficient used to control the degree of contrast enhancement. .

[0110] Solving for atmospheric light intensity :Because atmospheric light has Characteristics, where max(▪) is the maximum value function, so the atmospheric light intensity estimation formula is:

[0111] ;

[0112] in, is the light intensity parameter, the present invention takes .

[0113] Based on the above solution process, the contrast-enhanced infrared sea surface image can be obtained , normalize it to [0,1] and convert it to 8-bit image to obtain the final contrast enhanced image.

[0114] Step S3: Based on the sea surface fish scale wave suppression algorithm and detection model based on the gradient domain, an infrared ship target detection model resistant to fish scale wave interference is constructed. The architecture of the infrared ship target detection model resistant to fish scale wave interference is as follows: Figure 4 shown.

[0115] Step S3 specifically includes:

[0116] Step S301: Construct a training set and a test set based on historical infrared sea surface images and historical ship target recognition results.

[0117] Step S302: Connect the gradient domain-based sea surface fish-scale wave suppression algorithm and the detection model to obtain an untrained infrared ship target detection model that is resistant to fish-scale wave interference.

[0118] Step S303: input the training set into the untrained infrared ship target detection model resistant to fish-scale wave interference, and train the untrained infrared ship target detection model resistant to fish-scale wave interference using a stochastic gradient descent optimizer to obtain a trained infrared ship target detection model resistant to fish-scale wave interference.

[0119] Step S304: input the test set into the trained infrared ship target detection model resistant to fish-scale wave interference, adjust the trained infrared ship target detection model resistant to fish-scale wave interference, and obtain a trained infrared ship target detection model resistant to fish-scale wave interference.

[0120] Step S305: Determine the trained infrared ship target detection model resistant to fish-scale wave interference as the infrared ship target detection model resistant to fish-scale wave interference.

[0121] In order to train and test the infrared ship target detection model that is resistant to fish scale wave interference and verify the effect of the fish scale wave suppression algorithm on improving the model recognition accuracy, this paper is based on the Arrow Optoelectronics infrared maritime ship dataset.

[0122] The Arrow Optoelectronics infrared marine vessel dataset uses infrared equipment with varying resolutions and focal lengths to collect 8,402 infrared data images in various scenarios. The dataset includes seven types of targets: cruise ships, bulk carriers, warships, sailboats, kayaks, container ships, and fishing vessels. During the experiments, the dataset was divided into a training set: validation set: test set with a ratio of 8:1:1, resulting in a training set: validation set: test set ratio of 6,721:840:841. Furthermore, to demonstrate the algorithm's ability to suppress fish-scale wave noise, 326 representative fish-scale wave interference samples were selected from the test set to form the fish-scale wave test set.

[0123] The present invention uses the Windows 11 operating system and utilizes the PyCharm integrated development environment (IDE) for programming. The development environment is configured with Python 3.8, the PyTorch 1.8 deep learning framework, and CUDA 11.1. The hardware resources relied on by the experiment include an Intel Core i7-13700KF model central processing unit (CPU) and an NVIDIA RTX 4090 model graphics processing unit (GPU). During the model training phase, a total of 300 cycles were executed, with a batch size of 32 samples per cycle. The Stochastic Gradient Descent (SGD) optimizer was used for training, with a learning rate of 0.02.

[0124] Step S4: Input the target infrared sea surface image into the infrared ship target detection model that is resistant to fish scale wave interference to obtain a ship target recognition result, wherein the ship target recognition result includes: target category, location, and predicted probability.

[0125] Based on the fish-scale wave suppression algorithm in the gradient domain and the YOLOv11n detection model, the present invention constructs an infrared ship target detection model that is resistant to fish-scale wave interference. That is, after the model acquires the infrared image, it first performs image enhancement preprocessing and then performs detection reasoning.

[0126] The present invention conducted an ablation experiment on each architectural component of the infrared ship target detection model that is resistant to fish scale wave interference on the ship dataset. As shown in Table 1, the original model is the YOLOv11n model. It can be seen that in the gradient domain-based sea surface fish scale wave suppression algorithm proposed in the present invention, the gradient domain processing, guided filter processing and introduction of the atmospheric scattering model improved the recognition accuracy mAP50 of the verification set by 0.5%, 0.6% and 0.9%, respectively, and the accuracy of the fish scale wave test set by 0.9%, 1.0% and 1.5%, respectively, which proves the effectiveness of the method of the present invention.

[0127] Table 1 Ablation experiment data of each architecture component of the infrared ship target detection model against fish scale wave interference

[0128] ;

[0129] The processing effect of the fish scale wave suppression algorithm based on the gradient domain of the present invention is as follows: Figure 5 As shown in Figure 2, it can be seen that the original infrared images in the three scenes can eliminate most of the sea surface clutter after gradient domain processing and restoration, and the best effect is in scene 1, as shown in Figure 2. Figure 5 (a) Figure 5 (b) Figure 5 (c) and Figure 5 As shown in (d) in , for scenarios 2 and 3, Figure 5 (e) in Figure 5 (f) in Figure 5 (g) in Figure 5 (h) in Figure 5 (i) Figure 5 (g) and Figure 5 As shown in (k), in a scene with large fish-scale wave noise, although the processed image still has obvious noise and poor visual effect, and the texture information of the target is lost, it does not affect the processing effect of the fish-scale wave suppression algorithm based on the gradient domain.

[0130] Because the guided filtering process in this invention uses historical infrared ocean surface images (original image information) as a guide, the target's contour information is improved, and the fish-scale noise is restored to the form of waves rather than raised dots. After processing with the atmospheric scattering model, the scene contrast is effectively enhanced, making the target and its contour details more prominent. Therefore, the gradient-domain-based ocean fish-scale noise suppression algorithm of this invention is feasible.

[0131] The fish scale wave suppression algorithm based on gradient domain of the present invention is compared with other typical fish scale wave suppression algorithms (sharp mask method, unsharp mask method and gradient histogram-based enhanced equalization algorithm) combined with the detection effect of YOLOv11n model. Figure 6 Compared with the detection effect of YOLOv11n model, as shown in Figure 6 (a) Figure 6 (f) in Figure 6 (k) and Figure 6 As shown in (p), it can be seen that the unsharp masking method and the unsharp masking method, because they belong to the spatial domain image enhancement algorithm, have weak ability to suppress fish-scale wave noise, and the visual image quality improvement effect is weak, such as Figure 6 (b) Figure 6 (c) Figure 6 (g) in Figure 6 (h) in Figure 6 (l) in Figure 6 (m) in Figure 6 (q) and Figure 6 The gradient histogram-based enhanced equalization algorithm belongs to the gradient domain image enhancement algorithm. Although it can effectively improve the contrast of the target area, it also enhances the fish scale noise, resulting in a decrease in confidence or even false detection. Figure 6 (d) Figure 6 (i) Figure 6 (n) and Figure 6 The fish-scale wave suppression algorithm based on the gradient domain of the present invention, due to the introduction of gradient domain processing restoration, guided filtering and atmospheric scattering model, can not only effectively improve the target prediction confidence in scenes with different fish-scale wave interference levels, but also has a significant visual improvement effect, such as Figure 6 (e) in Figure 6 (j) in Figure 6 (o) and Figure 6 The fish scale wave suppression algorithm based on the gradient domain of the present invention, except for the scene 1 where the fish scale wave interference is extremely large, the detection effect diagrams obtained according to scenes 2, 3 and 4 are as follows: Figure 6 (j) in Figure 6 (o) and Figure 6 From (t), we can see that the fish-scale wave noise has been completely suppressed and the target ship is highlighted, which proves that the fish-scale wave suppression algorithm based on the gradient domain can effectively improve the fish-scale wave interference and enhance the accuracy of ship target recognition.

[0132] The present invention also provides an infrared sea surface ship target detection system that is resistant to fish scale wave interference, such as Figure 7 Shown, including:

[0133] The target acquisition module is used to obtain target infrared sea surface images.

[0134] The algorithm building module is used to construct a gradient domain-based fish-scale wave suppression algorithm based on gradient domain processing, gradient field restoration, guided filter processing and atmospheric scattering model.

[0135] The model building module is used to build an infrared ship target detection model that is resistant to fish scale wave interference based on the sea surface fish scale wave suppression algorithm based on the gradient domain.

[0136] The target recognition module is used to input the target infrared sea surface image into the infrared ship target detection model that is resistant to fish scale interference to obtain the ship target recognition result. The ship target recognition result includes: target category, location and predicted probability.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical method of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical method to deviate from the spirit and scope of the technical method of the present invention.

Claims

1. A method for detecting infrared sea surface ship targets with resistance to fish scale wave interference, characterized in that: The following steps are involved: S1. Acquire target infrared sea surface image; S2. Based on gradient domain processing, gradient field restoration, guided filter processing and atmospheric scattering model, a fish-scale wave suppression algorithm based on gradient domain is constructed; S3. Based on the sea surface fish scale wave suppression algorithm and detection model based on the gradient domain, an infrared ship target detection model that is resistant to fish scale wave interference is constructed; S4. Inputting the target infrared sea surface image into the infrared ship target detection model resistant to fish scale wave interference to obtain a ship target recognition result; The ship target recognition result includes: target category, location and predicted probability; In S2, a fish-scale wave suppression algorithm based on gradient domain processing, gradient field restoration, guided filter processing and atmospheric scattering model is constructed, which specifically includes: S201, obtaining historical infrared sea surface images and historical ship target recognition results; S202, performing gradient domain processing and gradient field restoration on the historical infrared sea surface image in sequence according to the Scharr operator, gradient domain transformation equation, cumulative distribution function, and Poisson equation to obtain a gradient field restored image; S203, using a guided filtering equation to perform guided filtering processing on the gradient field restoration image to obtain a guided filtering processed image; S204, using an atmospheric scattering model to enhance the contrast of the guided filter processed image, and obtaining a historical ship target recognition result; In S202, the historical infrared sea surface image is subjected to gradient domain processing and gradient field restoration in sequence according to the Scharr operator, gradient domain transformation equation, cumulative distribution function, and Poisson equation to obtain a gradient field restored image, which specifically includes: The Scharr operator is used to calculate the gradient of the historical infrared sea surface image to obtain the gradient amplitude of the original gradient field; Based on the gradient amplitude of the original gradient field, the minimum gradient threshold is set to determine the target area; The gradient domain transformation equation is used to compress the dynamic range of the target area and obtain the transformation gradient amplitude; After mapping the transformed gradient amplitude through the cumulative distribution function, the transformed gradient amplitude is used to construct a new gradient field; Using Poisson's equation, the gradient domain of the new gradient field is converted into a spatial domain image to obtain the gradient field restoration image; The expression of the gradient amplitude of the original gradient field is: ； in, Represents the gradient magnitude of the image, that is, the gradient vector The size of is the horizontal direction, is the longitudinal direction, is the gradient component in the horizontal direction, is the gradient component in the vertical direction; The gradient domain transformation equation is: ； in, is the gradient amplitude The new gradient amplitude after gradient domain transformation, is the maximum gradient amplitude, is the minimum gradient threshold, is the maximum gradient threshold, is a logarithmic function.

2. The infrared sea surface ship target detection method resistant to fish scale wave interference according to claim 1 is characterized in that: In step S203, guided filtering is performed on the gradient field restoration image using a guided filtering equation to obtain a guided filtering processed image, specifically including: Determine a guide image and an input image; the guide image is a historical infrared sea surface image; the input image is a gradient field restoration image; Based on the guided image, the input image is filtered using the guided filtering equation to obtain a guided filtering processed image.

3. The infrared sea surface ship target detection method resistant to fish scale wave interference according to claim 2 is characterized in that: The guided filtering equation is: ； in, is the pixel position, Pixel A local window centered on is any pixel in the local window, The pixel values ​​of the image are processed for guided filtering. is the pixel value of the guidance image, the coefficient and are the coefficients of the linear function when the window center is at pixel k.

4. The infrared sea surface ship target detection method resistant to fish scale wave interference according to claim 1 is characterized in that: In S204, the atmospheric scattering model is used to enhance the contrast of the guided filter processed image and obtain the historical ship target recognition results, which include: Based on historical infrared sea surface images, guided filter processed images, atmospheric light intensity and transmittance, an atmospheric scattering equation is constructed to obtain an atmospheric scattering model. The guided filter processed image is input into the atmospheric scattering model to obtain the historical ship target recognition result.

5. The infrared sea surface ship target detection method resistant to fish scale wave interference according to claim 4 is characterized in that: The atmospheric scattering equation is: ； in, is the pixel position, is a contrast-enhanced infrared sea surface image. is a historical infrared sea surface image. To process the image for guided filtering, is the atmospheric light intensity, is the transmittance, is the ambient light item.

6. The infrared sea surface ship target detection method resistant to fish scale wave interference according to claim 1 is characterized in that: In S3, an infrared ship target detection model resistant to fish-scale wave interference is constructed based on the gradient domain-based sea surface fish-scale wave suppression algorithm and detection model. Specifically, it includes: Construct training and test sets based on historical infrared sea surface images and historical ship target recognition results; The gradient domain-based sea surface fish scale wave suppression algorithm and the detection model are connected to obtain an untrained infrared ship target detection model that is resistant to fish scale wave interference. The training set is input into an untrained infrared ship target detection model resistant to fish scale wave interference, and a stochastic gradient descent optimizer is used to train the untrained infrared ship target detection model resistant to fish scale wave interference to obtain a trained infrared ship target detection model resistant to fish scale wave interference; Inputting the test set into the trained infrared ship target detection model resistant to fish scale wave interference, adjusting the trained infrared ship target detection model resistant to fish scale wave interference, and obtaining a trained infrared ship target detection model resistant to fish scale wave interference; The trained infrared ship target detection model resistant to fish scale wave interference is determined as the infrared ship target detection model resistant to fish scale wave interference.

7. An infrared sea surface ship target detection system resistant to fish scale wave interference, characterized in that: include: Target acquisition module, used to acquire target infrared sea surface images; An algorithm building module is used to construct a gradient domain-based fish-scale wave suppression algorithm based on gradient domain processing, gradient field restoration, guided filter processing, and atmospheric scattering model; Based on gradient domain processing, gradient field restoration, guided filter processing and atmospheric scattering model, a fish-scale wave suppression algorithm based on gradient domain is constructed, which specifically includes: Obtain historical infrared sea surface images and historical ship target recognition results; According to the Scharr operator, gradient domain transformation equation, cumulative distribution function and Poisson equation, the historical infrared sea surface images are processed in gradient domain and restored in gradient field to obtain the gradient field restored image. Using the guided filter equation, the gradient field restoration image is processed by guided filtering to obtain a guided filter processed image; The atmospheric scattering model is used to enhance the contrast of the guided filter processed image and obtain the historical ship target recognition results; According to the Scharr operator, gradient domain transformation equation, cumulative distribution function and Poisson equation, the historical infrared sea surface images are processed in gradient domain and restored in gradient field to obtain the gradient field restored image, which includes: The Scharr operator is used to calculate the gradient of the historical infrared sea surface image to obtain the gradient amplitude of the original gradient field; Based on the gradient amplitude of the original gradient field, the minimum gradient threshold is set to determine the target area; The gradient domain transformation equation is used to compress the dynamic range of the target area and obtain the transformation gradient amplitude; After mapping the transformed gradient amplitude through the cumulative distribution function, the transformed gradient amplitude is used to construct a new gradient field; Using Poisson's equation, the gradient domain of the new gradient field is converted into a spatial domain image to obtain the gradient field restoration image; The expression of the gradient amplitude of the original gradient field is: ； in, Represents the gradient magnitude of the image, that is, the gradient vector The size of is the horizontal direction, is the longitudinal direction, is the gradient component in the horizontal direction, is the gradient component in the vertical direction; The gradient domain transformation equation is: ； in, is the gradient amplitude The new gradient amplitude after gradient domain transformation, is the maximum gradient amplitude, is the minimum gradient threshold, is the maximum gradient threshold, is a logarithmic function; A model building module is used to build an infrared ship target detection model that is resistant to fish-scale wave interference based on a sea surface fish-scale wave suppression algorithm based on a gradient domain; The target recognition module is used to input the target infrared sea surface image into the infrared ship target detection model that is resistant to fish scale wave interference to obtain the ship target recognition result; the ship target recognition result includes: target category, position and predicted probability.

Images