Ship target detection method, device and electronic equipment

Abstract

The application provides a ship target detection method, device and electronic equipment, and relates to the technical field of target detection.The method comprises the following steps: determining a ship area where a target ship contained in an original remote sensing image is located, and determining a cloud area where a cloud layer contained in the original remote sensing image is located; determining an axis of the target ship in the ship area, and determining a ship body neighborhood of the target ship based on the ship area and the axis of the target ship; determining a cloud layer neighborhood of the cloud area based on the cloud area, and determining an image boundary area of the original remote sensing image based on the original remote sensing image; determining a first ship wake area where a wake of the target ship contained in the ship body neighborhood, the cloud layer neighborhood and the image boundary area is located respectively; and detecting the target ship based on the first ship wake area.The application can effectively realize the detection of the ships covered by the cloud layer and invisible outside the image boundary in the visible light remote sensing image, and improves the accuracy of target ship detection.

Description

Technical Field

[0001] This invention relates to the field of target detection technology, and in particular to a method, apparatus and electronic equipment for detecting ship targets. Background Technology

[0002] The automatic detection of ship targets has wide applications in the maritime and commercial sectors. Remote sensing images play a crucial role in detecting maritime ship targets due to their wide field of view and high resolution.

[0003] Based on the type of payload carried by remote sensing satellites, remote sensing images can be broadly classified into two categories: Synthetic Aperture Radar (SAR) images and visible light remote sensing images. Ship detection methods using SAR images can be divided into direct detection methods and indirect detection methods. Direct detection methods directly utilize the pixel differences between the ship and the ocean background to detect the ship. This method is mainly based on the statistical characteristics of SAR images, with the Constant False Alarm Rate (CFAR) being the most representative direct detection method. Indirect detection methods detect ships by detecting auxiliary features (such as ship wakes). When the ship's hull features are not obvious in the SAR image, the detection of auxiliary features becomes particularly important. Ship wakes can persist over a large area on the sea surface for a considerable period of time; therefore, the accuracy of observing ship wakes is greater than that of observing the ship itself, and the detected features are more obvious than the ship target itself. Furthermore, combining ship wakes with the ship's position in the image can more accurately predict the ship's speed, heading, and actual location.

[0004] Compared to SAR images, visible light remote sensing images offer advantages such as visualization, high resolution, and low cost, and can capture richer local detail features. Traditional ship detection methods based on visible light images typically rely on visual saliency and visual perception principles to extract ship regions from the ocean background. However, because the quality of visible light remote sensing images is easily affected by weather factors, traditional detection methods do not perform ideally in complex ocean environments. Ship wakes can be used to assist in ship detection.

[0005] However, in high-resolution visible light remote sensing images, ship wakes exhibit significant local details and do not appear as simple straight lines, but rather as regions with a certain shape. This makes it difficult to apply traditional SAR image processing methods to visible light remote sensing images.

[0006] Therefore, how to accurately extract ship wakes from visible light remote sensing images and use the extracted ship wakes as an aid to ship detection, thereby improving the accuracy of ship target detection, has become an urgent problem to be solved in the industry. Summary of the Invention

[0007] To address the problems existing in the prior art, the present invention provides a method, apparatus and electronic equipment for ship target detection.

[0008] In a first aspect, the present invention provides a method for detecting ship targets, comprising:

[0009] The ship region containing the target ship in the original remote sensing image is determined, and the cloud region containing the cloud layer in the original remote sensing image is determined.

[0010] Determine the axis of the target ship in the ship region, and determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship;

[0011] Based on the cloud region, determine the cloud neighborhood of the cloud region, and based on the original remote sensing image, determine the image boundary region of the original remote sensing image;

[0012] The first ship wake region, which contains the wake of the target ship within the hull neighborhood, the cloud neighborhood, and the image boundary region, is determined respectively.

[0013] The target ship is detected based on the wake region of the first ship;

[0014] Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

[0015] Optionally, according to a ship target detection method provided by the present invention, determining the first ship wake region containing the wake of the target ship in the hull neighborhood, the cloud neighborhood, and the image boundary region includes:

[0016] The hull neighborhood, the cloud neighborhood, and the image boundary region are taken as the target region, and the image corresponding to the target region is converted from the RGB color space to the LAB color space;

[0017] In the LAB color space, the luminance component of the image corresponding to the target region is modeled, and based on the idea of ​​constant false alarm rate (CFAR), the first ship wake region where the wake of the target ship is located is extracted from the target region.

[0018] Optionally, according to a ship target detection method provided by the present invention, the step of modeling the luminance component of the image corresponding to the target region in the LAB color space, and extracting the first ship wake region where the wake of the target ship is located from the target region based on the idea of ​​constant false alarm rate (CFAR), includes:

[0019] In the LAB color space, the luminance component of the image corresponding to the target region is modeled, and based on the idea of ​​constant false alarm rate (CFAR), the second ship wake region where the wake of the target ship is located is extracted from the target region.

[0020] The second ship wake region is clustered using the density-based clustering algorithm DBSCAN to obtain the first ship wake region.

[0021] Optionally, according to a ship target detection method provided by the present invention, the step of using the density-based clustering algorithm DBSCAN to cluster the second ship wake region to obtain the first ship wake region includes:

[0022] The second ship wake region was clustered using the density-based clustering algorithm DBSCAN to obtain the third ship wake region.

[0023] Using a binary normal distribution, all the obtained wake regions of the third ship are simulated to obtain the equal probability curves of the binary normal distribution;

[0024] If the ellipticity of the equal probability curve is greater than a preset ellipticity, and the area of ​​the third ship wake region is greater than a preset area, then the third ship wake region is taken as the first ship wake region.

[0025] Optionally, according to the ship target detection method provided by the present invention, the step of modeling the brightness component of the image corresponding to the target region includes:

[0026] The luminance component of the image corresponding to the target region is modeled based on the Burr XII distribution.

[0027] Optionally, according to a ship target detection method provided by the present invention, determining the ship region containing the target ship in the original remote sensing image includes:

[0028] Based on the YOLO detection network, target ships contained in the original remote sensing image are detected, and the first YOLO detection region of the YOLO detection network in the original remote sensing image is obtained.

[0029] The length and width of the first YOLO detection region are both enlarged by a preset multiple in the original remote sensing image to obtain the second YOLO detection region;

[0030] If the ratio of the area occupied by the ocean background in the second YOLO detection area to the area of ​​the second YOLO detection area is greater than a preset value, the ship area is determined based on the first YOLO detection area.

[0031] Optionally, according to a ship target detection method provided by the present invention, determining the cloud region containing the clouds in the original remote sensing image includes:

[0032] The original remote sensing image is converted from the RGB color space to the LAB color space;

[0033] In the LAB color space, the reference average value of the ocean region color is used as the cluster center, and the original remote sensing image is clustered based on the k-means++ algorithm to obtain the first ocean region contained in the original remote sensing image.

[0034] A second ocean region is obtained by performing morphological opening and closing operations on the first ocean region.

[0035] Based on the principle of dark channel prior, the cloud region contained in the second ocean region is obtained.

[0036] Optionally, according to a ship target detection method provided by the present invention, determining the axis of the target ship in the ship region includes:

[0037] Based on the principal component analysis (PCA) method, the axis of the target ship in the ship region is determined.

[0038] Secondly, the present invention also provides a ship target detection device, comprising:

[0039] The first determining module is used to determine the ship area where the target ship is located in the original remote sensing image, and to determine the cloud area where the cloud layer is located in the original remote sensing image.

[0040] The second determining module is used to determine the axis of the target ship in the ship region, and to determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship.

[0041] The third determining module is used to determine the cloud neighborhood of the cloud region based on the cloud region, and to determine the image boundary region of the original remote sensing image based on the original remote sensing image.

[0042] The fourth determining module is used to determine the first ship wake region containing the wake of the target ship in the hull neighborhood, the cloud neighborhood, and the image boundary region, respectively.

[0043] The ship detection module is used to detect the target ship based on the wake area of ​​the first ship;

[0044] Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

[0045] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the ship target detection method as described in the first aspect.

[0046] The ship target detection method, apparatus, and electronic equipment provided by this invention first determine the hull neighborhood of the target ship contained in the original remote sensing image, the cloud region containing the clouds contained in the original remote sensing image, and the image boundary region of the original remote sensing image. Then, the hull neighborhood, cloud region, and image boundary region are used as the search area for the ship's wake. The first ship wake region containing the wake of the target ship contained in the hull neighborhood, cloud neighborhood, and image boundary region are determined respectively. Then, the target ship is detected based on the first ship wake region. This can effectively realize the detection of ships covered by clouds and invisible outside the image boundary in visible light remote sensing images, thus improving the accuracy of target ship detection. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0048] Figure 1 This is a flowchart illustrating the ship target detection method provided by the present invention;

[0049] Figure 2 This is one of the pixel grayscale distribution diagrams of the ocean background provided by the present invention;

[0050] Figure 3This is the second schematic diagram of pixel grayscale distribution in an ocean background provided by the present invention;

[0051] Figure 4 This is a schematic diagram of the structure of the ship target detection device provided by the present invention;

[0052] Figure 5 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0054] It should be noted that in the description of this invention, the terms "first," "second," etc., are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, the first object can be one or more.

[0055] The ship target detection method, apparatus, and electronic equipment provided by the present invention will be described exemplarily below with reference to the accompanying drawings.

[0056] Figure 1 This is a flowchart illustrating the ship target detection method provided by the present invention, as shown below. Figure 1 As shown, the method includes:

[0057] Step 100: Determine the ship region where the target ship is located in the original remote sensing image, and determine the cloud region where the cloud layer is located in the original remote sensing image.

[0058] Step 110: Determine the axis of the target ship in the ship region, and determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship;

[0059] Step 120: Based on the cloud region, determine the cloud neighborhood of the cloud region, and based on the original remote sensing image, determine the image boundary region of the original remote sensing image;

[0060] Step 130: Determine the first ship wake region containing the wake of the target ship in the hull neighborhood, the cloud neighborhood, and the image boundary region, respectively.

[0061] Step 140: Detect the target ship based on the wake region of the first ship;

[0062] Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

[0063] It should be noted that the execution subject of the ship target detection method provided in this embodiment of the invention can be an electronic device, a component in the electronic device, an integrated circuit, or a chip. The electronic device can be a mobile electronic device or a non-mobile electronic device. For example, a mobile electronic device can be a mobile phone, tablet computer, laptop computer, PDA, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc., while a non-mobile electronic device can be a server, network attached storage (NAS), personal computer (PC), television (TV), or self-service machine, etc. This embodiment of the invention does not specifically limit the specific implementation of these devices.

[0064] The following describes the technical solution of the present invention in detail, taking the computer execution of the ship target detection method provided by the present invention as an example.

[0065] Specifically, to overcome the shortcomings of existing technologies in accurately extracting ship wakes from visible light remote sensing images to assist in ship target detection, this invention first determines the hull neighborhood of the target ship in the original remote sensing image, the cloud region containing the clouds in the original remote sensing image, and the image boundary region of the original remote sensing image. The hull neighborhood, cloud region, and image boundary region are then used as the search area for ship wakes. A first ship wake region is determined within the hull neighborhood, cloud neighborhood, and image boundary region, respectively. Based on this first ship wake region, the target ship is detected. This effectively enables the detection of ships covered by clouds or invisible outside the image boundary in visible light remote sensing images, improving the accuracy of target ship detection.

[0066] It should be noted that the quality of the acquired raw remote sensing image has a significant impact on the analysis of image features. Subsequent ship wake detection is only meaningful when the quality value of the acquired raw remote sensing image is greater than the significance threshold.

[0067] For quantitative analysis, the quality of the original remote sensing image can be described based on the quality criteria shown in Equation (1):

[0068]

[0069] Among them, the larger the QI value, the better the quality of the original remote sensing image. i ∈[0,1] represents the normalization index generated from the i-th image feature, w i The weights for the corresponding indicators are given by n, which represents the number of image features.

[0070] Alternatively, image entropy H can also be utilized. e Tenengrad gradient H T Gabor characteristics H G and reblur performance H R Image quality is calculated using four features.

[0071] Wherein, image entropy H e The calculation formula is shown in equation (2):

[0072] H e =-∑(p×log2(p)) (2)

[0073] Where p represents the normalized histogram count of the image, H e A higher value indicates better image quality.

[0074] Tenengrad gradient H T The calculation formula is shown in equation (3):

[0075]

[0076] Among them, G x (x, y), G y (x, y) represent the directional gradients in the x and y directions, respectively, H T A higher value indicates better image quality.

[0077] The Gabor filter is a linear bandpass filter widely used in image processing for edge detection, texture classification, feature extraction, and disparity estimation. Gabor features H GIt is the maximum image entropy of the response matrix obtained by convolving the original image Im0 with the Gabor filter {g(x, y|λ, θ, ψ, σ, γ)}:

[0078]

[0079]

[0080]

[0081] And have

[0082] R G =g(x,y|λ,θ,ψ,σ,γ)*Im0 (7)

[0083] H G =max(H e ({R G})) (8)

[0084] Here, you can select the direction. The wavelength is λ = [2, 4, 8], x represents the x-coordinate of the pixel, y represents the y-coordinate of the pixel, ψ represents the phase offset, σ represents the variance parameter of the Gaussian envelope function, γ represents the aspect ratio of the spatial scale, and H G A higher value indicates better image quality.

[0085] Reblur is based on the idea that smoothing a blurred image results in less information loss. Reblur performance H R Calculate the similarity between the original image Im0 and the blurred image Im′:

[0086] H R =SSIM(Im′,Im0) (9)

[0087] Where, Im′=G f (σ)*Im0 is a Gaussian filter G with variance σ. f The resulting blurred image; SSIM() is the Structural Similarity Index, used to measure the similarity between two images taken at the same time. This measure is based on the idea that the human visual system is sensitive to structural information in a scene:

[0088] SSIM(Im′,Im0)=[l(Im′,Im0)] α ·[c(Im′,Im0)] β ·[s(Im′,Im0)] γ (10)

[0089]

[0090]

[0091]

[0092] Where, μ x μ y , σ x , σ y , σ xy Let be the local mean, standard deviation, and cross covariance of images Im0 and Im′; C1, C2, and C3 are small regularization constants; α, β, and γ are weighting coefficients. H R The closer the value is to 0, the better the image quality.

[0093] Correspondingly, based on the aforementioned image entropy H e Tenengrad gradient H T Gabor characteristics H G and reblur performance H R Four features can be used to construct corresponding quality assessment indicators I i (i = 1, 2, 3, 4), the calculation formula is:

[0094] I1=(1+exp(a1(H e -b1))) -1 (14)

[0095] I2=(1+exp(a2(H T -b2))) -1 (15)

[0096] I3=(1+exp(a3(H G -b3))) -1 (16)

[0097] I4=1-(1+exp(a4(H R -b4))) -1 (17)

[0098] Where a1, a2, a3, a4 and b1, b2, b3, b4 are activation parameters.

[0099] Optionally, after determining that the obtained original remote sensing image meets the quality assessment requirements, the original remote sensing image can be denoised using a small variance Gaussian filter before subsequent ship wake detection.

[0100] Optionally, in an embodiment of the present invention, a wreath search region S is determined. rIt includes the ship hull neighborhood S1, the cloud neighborhood S2, and the image boundary region S3. The ship hull neighborhood S1 is generated along the axis of the target ship by the minimum boundary rectangle of the region twice the length of the ship before and after the target ship. The cloud neighborhood S2 is generated based on the minimum boundary rectangle of the cloud region, with a spacing of approximately a first preset length (50-100m). The image boundary region S3 consists of the edge region at a second preset length (50-100m) from the boundary of the original remote sensing image.

[0101] It should be noted that the first preset length and the second preset length can be adapted to the actual application, and the embodiments of the present invention do not impose specific limitations on this.

[0102] Optionally, determining the axis of the target ship in the ship region includes:

[0103] Based on the Principal Component Analysis (PCA) method, the axis of the target ship in the ship region is determined.

[0104] It should be noted that, since the target ship has a large length-to-width ratio, the axis v1 of the target ship can be obtained using principal component analysis (PCA).

[0105] Understandably, before obtaining the hull neighborhood, it is necessary to first determine the ship region containing the target ship in the original remote sensing image, then determine the axis of the target ship in the ship region, and finally determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship.

[0106] Understandably, before obtaining the cloud neighborhood, it is necessary to first determine the cloud region containing the clouds in the original remote sensing image, and then determine the cloud neighborhood based on the cloud region.

[0107] Optionally, after obtaining the hull neighborhood, the hull neighborhood and the image boundary region, the first ship wake region containing the target ship's wake contained in the hull neighborhood, the hull neighborhood and the image boundary region is determined respectively, and then the target ship is detected based on the first ship wake region.

[0108] Alternatively, after obtaining the first ship wake area, the detection of ship targets can be guided by the following principles:

[0109] (1) For the first ship wake region detected in the image boundary region and the cloud neighborhood, the length of the wake fitting curve belonging to the image boundary region and the cloud neighborhood can be calculated, and the confidence of ship detection is shown in Equation (18):

[0110]

[0111] Where, p i The coefficient of determination is the goodness of fit adjusted based on the goodness of fit, where l0 is a given length threshold. i P1 represents the length of the fitted curve for the i-th wake, and P1 represents the confidence level of the ship detection.

[0112] (2) For the hull neighborhood, the first ship wake region with a confidence level greater than a certain confidence level threshold can be selected, and the hull direction from the wake to the target ship can be selected as the heading of the target ship.

[0113] The ship target detection method provided by this invention first determines the hull neighborhood of the target ship in the original remote sensing image, the cloud region containing the clouds in the original remote sensing image, and the image boundary region of the original remote sensing image. The hull neighborhood, cloud region, and image boundary region are then used as the search area for the ship's wake. The first ship wake region containing the wake of the target ship in the hull neighborhood, cloud neighborhood, and image boundary region is determined respectively. Based on the first ship wake region, the target ship is detected. This method can effectively detect ships that are covered by clouds or invisible outside the image boundary in visible light remote sensing images, thus improving the accuracy of target ship detection.

[0114] Optionally, determining the first ship wake region containing the wake of the target ship within the hull neighborhood, the cloud neighborhood, and the image boundary region includes:

[0115] The hull neighborhood, the cloud neighborhood, and the image boundary region are taken as the target region, and the image corresponding to the target region is converted from the RGB color space to the LAB color space;

[0116] In the LAB color space, the luminance component of the image corresponding to the target region is modeled, and based on the idea of ​​constant false alarm rate (CFAR), the first ship wake region where the wake of the target ship is located is extracted from the target region.

[0117] Specifically, in this embodiment of the invention, in order to determine the first ship wake region containing the wake of the target ship in the hull neighborhood, cloud neighborhood, and image boundary region respectively, the hull neighborhood, cloud neighborhood, and image boundary region can be used as target regions. The image corresponding to the target region is converted from RGB color space to LAB color space. Then, in LAB color space, the luminance component of the image corresponding to the target region is modeled. Based on the constant false alarm rate (CFAR) concept, the first ship wake region containing the wake of the target ship is extracted from the target region.

[0118] It should be noted that among ship wake patterns, Kelvin wakes and turbulent wakes are the most likely to be observed in remote sensing images. Compared to Kelvin wakes, turbulent wakes typically have a larger area and a more concentrated grayscale distribution, making their detection more robust. Therefore, this embodiment of the invention primarily assists in ship detection by detecting turbulent wakes.

[0119] For each wake search region obtained above (hull neighborhood, cloud neighborhood, and image boundary region), the corresponding image can first be converted from RGB color space to LAB color space. Since the brightness L component directly reflects the important characteristic brightness for identifying turbulent wakes in LAB color space, the brightness L component can be modeled. Based on the idea of ​​constant false alarm rate (CFAR), the turbulent wake region, i.e., the first ship wake region, can be extracted.

[0120] It is understood that the embodiments of the present invention model the luminance component of the image corresponding to the target area in the LAB color space, and based on the idea of ​​constant false alarm rate (CFAR), effectively realize the extraction of the first ship wake region where the wake of the target ship is located from the target area, that is, achieve the purpose of accurately extracting the ship wake from the visible light remote sensing image.

[0121] Optionally, the step of modeling the luminance component of the image corresponding to the target region in the LAB color space, and extracting the first ship wake region where the target ship's wake is located from the target region based on the concept of constant false alarm rate (CFAR), includes:

[0122] In the LAB color space, the luminance component of the image corresponding to the target region is modeled, and based on the idea of ​​constant false alarm rate (CFAR), the second ship wake region where the wake of the target ship is located is extracted from the target region.

[0123] The second ship wake region is clustered using a density-based spatial clustering of applications with noise (DBSCAN) algorithm to obtain the first ship wake region.

[0124] Specifically, in this embodiment of the invention, in order to model the luminance component of the image corresponding to the target region in the LAB color space, and extract the first ship wake region where the target ship's wake is located from the target region based on the idea of ​​constant false alarm rate (CFAR), the luminance component of the image corresponding to the target region can first be modeled in the LAB color space, and the second ship wake region where the target ship's wake is located can be extracted from the target region based on the idea of ​​constant false alarm rate (CFAR). Then, the density-based clustering algorithm DBSCAN is used to cluster the second ship wake region to obtain the first ship wake region.

[0125] It should be noted that, in this embodiment of the invention, considering that environmental noise will also be extracted as ship wake, the extracted second ship wake region is clustered using a density-based spatial clustering algorithm (DBSCAN) to filter out environmental noise, thereby obtaining the first ship wake region.

[0126] It should be noted that the DBSCAN algorithm uses a neighborhood density threshold M. ε To discover that the dataset contains at least M ε Clusters at the center point, where M ε The value can be determined based on the image resolution.

[0127] The DBSCAN algorithm constructs the ε-neighborhood of data points as follows:

[0128] N ε (p)={q∈X c |dist(p, q)≤ε} (19)

[0129] Where dist is the distance function, and the L1 norm can be chosen as the distance function; p represents the data points to be validated; q represents the data points to be clustered; and X... c Let N represent the set of cluster centroids, ε represent the maximum radius of the cluster, and N represent the maximum radius of the cluster. ε (p) represents the constructed ε-neighborhood.

[0130] It is understood that the embodiments of the present invention use the density-based clustering algorithm DBSCAN to filter out environmental noise in the extracted second ship wake region to obtain the first ship wake region, which can improve the accuracy of the extracted ship wake region and thus improve the accuracy of ship target detection.

[0131] Optionally, the step of using the density-based clustering algorithm DBSCAN to cluster the second ship wake region to obtain the first ship wake region includes:

[0132] The second ship wake region was clustered using the density-based clustering algorithm DBSCAN to obtain the third ship wake region.

[0133] Using a binary normal distribution, all the obtained wake regions of the third ship are simulated to obtain the equal probability curves of the binary normal distribution;

[0134] If the ellipticity of the equal probability curve is greater than a preset ellipticity, and the area of ​​the third ship wake region is greater than a preset area, then the third ship wake region is taken as the first ship wake region.

[0135] Specifically, in this embodiment of the invention, in order to implement the density-based clustering algorithm DBSCAN to cluster the second ship wake region and obtain the first ship wake region, the density-based clustering algorithm DBSCAN can be used first to cluster the second ship wake region to obtain the third ship wake region. Then, a binary normal distribution is used to simulate all the obtained third ship wake regions to obtain the equal probability curve of the binary normal distribution. Then, if it is determined that the ellipticity of the equal probability curve is greater than the preset ellipticity and the area of ​​the third ship wake region is greater than the preset area, the third ship wake region is used as the first ship wake region.

[0136] It should be noted that, since the wake of a ship has a strip-shaped spatial distribution, this embodiment of the invention uses a bivariate normal distribution N(μ, ∑) to simulate the wake region. The covariance matrix ∑ reflects the shape of the extracted wake region, and is represented by the ellipticity J of the isoprobability curve. W The third ship wake region obtained as a standard test by DBSCAN:

[0137]

[0138] Where, ∑ -1 =UΛU T , U is an upper triangular matrix, λ max λ represents the largest eigenvalue of the matrix. min This represents the smallest eigenvalue of the matrix.

[0139] Only the area of ​​the third ship's wake region is greater than the preset area A0 and the ellipticity J of the equal probability curve. W Only regions with an ellipticity greater than the preset ellipticity J0 are identified as true ship wake regions, i.e., the first ship wake region. Furthermore, a quadratic curve can be used to fit each first ship wake region, and these regions can be merged with similar fitted curves.

[0140] It is understood that the embodiments of the present invention take into account that the wakes of ships have a strip-shaped spatial distribution. Therefore, by using the third ship wake region obtained after simulation clustering using a binary normal distribution, an equal probability curve of the binary normal distribution is obtained. When it is determined that the ellipticity of the equal probability curve is greater than a preset ellipticity and the area of ​​the third ship wake region is greater than a preset area, the third ship wake region is used as the first ship wake region for ship detection. This can further improve the accuracy of the extracted ship wake region, thereby improving the accuracy of ship target detection.

[0141] Optionally, modeling the luminance components of the image corresponding to the target region includes:

[0142] The luminance component of the image corresponding to the target region is modeled based on the Burr x II distribution.

[0143] Specifically, in this embodiment of the invention, the hull neighborhood, cloud neighborhood, and image boundary region can be used as target regions, and the image corresponding to the target region can be converted from RGB color space to LAB color space. Then, in LAB color space, the brightness component of the image corresponding to the target region is modeled based on Burr XII distribution, and based on the idea of ​​constant false alarm rate CFAR, the first ship wake region where the wake of the target ship is located is extracted from the target region.

[0144] It should be noted that the embodiments of the present invention only focus on the quality of the ocean portion in the original remote sensing image, defining the ocean background C. bg It is marine area C sea and cloud region C cloud Difference set:

[0145] C bg =C sea -C cloud (twenty one)

[0146] Figure 2 This is one of the pixel grayscale distribution diagrams of the ocean background provided by the present invention. Figure 3 This is the second schematic diagram of pixel grayscale distribution in an ocean background provided by the present invention, as shown below. Figure 2 and Figure 3 As shown, the grayscale distribution of the ocean background has a unimodal form and is skewed to the right. Therefore, this embodiment of the invention uses a Burr XII type distribution to model the L component of the wake search region, wherein... Figure 2 and Figure 3In this context, "sea region" represents the ocean region, "normal" represents the normal distribution, "lognormal" represents the log-normal distribution, "gamma" represents the gamma distribution, and "burr" represents the Burr distribution. The cumulative distribution function (CDF) and probability density function (PDF) of the Burr distribution are shown in equations (22) and (23), respectively:

[0147]

[0148]

[0149] Where α, c, k > 0 are distribution parameters.

[0150] The log-likelihood function of the Burr type XII distribution is shown in equation (24):

[0151]

[0152] Among them, L B Let x represent the probability density function of the L component of the selected wreath search region, n represent the number of pixels in the selected wreath search region, and x represent the probability density function of the L component. i This represents the L component value of the selected trail search region pixels.

[0153] Optionally, the distribution parameters (α, c, k) can be obtained based on the Maximum Likelihood Estimate (MLE) method. However, since a non-degenerate limiting distribution may exist, a finite maximum likelihood estimate for the Burr XII type distribution does not always exist. When the Burr XII type distribution cannot be solved, the Gamma distribution can be used as an alternative distribution form. The probability density function and cumulative distribution function of the Gamma distribution are shown in equations (25) and (26), respectively:

[0154]

[0155]

[0156] Wherein, the Γ(a) function is t represents the transformed Gamma integral variable, L represents the shape factor of the Gamma distribution, and μ represents the scale factor of the Gamma distribution.

[0157] The maximum likelihood of the Gamma distribution parameters is shown in equations (27) and (28):

[0158]

[0159]

[0160] Before performing distribution fitting, the brightness L component needs to be standardized to obtain better fitting results, as shown in equation (29):

[0161]

[0162] Where ε is a small regularization constant, and x represents the vector formed by the L component sets. This represents a vector consisting of the standardized L-components.

[0163] Then, based on the concept of constant false alarm rate (CFAR), bright and dark turbulent wake regions can be extracted. The definition of bright and dark here is based on the average brightness of the local sea area. Wake regions darker than the average brightness of the neighboring sea area are defined as dark turbulent wake regions, and conversely, wake regions brighter than the average brightness of the neighboring sea area are defined as bright turbulent wake regions.

[0164] For the Burr type XII distribution, we have

[0165]

[0166]

[0167] For the Gamma distribution, we have

[0168]

[0169]

[0170] Where p1 and p2 are the probabilities of a given region for a bright or dark turbulent wake, respectively, and T b and T d These are the calculated detection thresholds for bright and dark turbulent wakes, T. b T represents the detection threshold for the bright turbulent wake region. d This represents the detection threshold for the dark turbulent wake region.

[0171] The wake regions of light and dark turbulence can be obtained using the following equations (34) and (35):

[0172] W b ={x L ∶x L ≥T b} (34)

[0173] W d ={x L ∶x L ≤T d} (35)

[0174] Among them, W b W represents the extracted bright turbulent wake region. d Indicates the extracted dark turbulent wake region, x L This represents the brightness value of the pixels in the area.

[0175] Optionally, determining the ship region containing the target ship in the original remote sensing image includes:

[0176] Based on the YOLO (You Only Look Once) detection network, target ships contained in the original remote sensing image are detected, and the first YOLO detection region of the YOLO detection network in the original remote sensing image is obtained.

[0177] The length and width of the first YOLO detection region are both enlarged by a preset multiple in the original remote sensing image to obtain the second YOLO detection region;

[0178] If the ratio of the area occupied by the ocean background in the second YOLO detection area to the area of ​​the second YOLO detection area is greater than a preset value, the ship area is determined based on the first YOLO detection area.

[0179] Specifically, in this embodiment of the invention, in order to determine the ship region containing the target ship in the original remote sensing image, the target ship in the original remote sensing image can first be detected based on the YOLO detection network to obtain the first YOLO detection region in the original remote sensing image. Then, the length and width of the first YOLO detection region in the original remote sensing image are both enlarged by a preset multiple to obtain the second YOLO detection region. If the ratio of the area occupied by the ocean background in the second YOLO detection region to the area of ​​the second YOLO detection region is greater than a preset value, the ship region containing the target ship can be determined based on the first YOLO detection region.

[0180] Optionally, the preset multiplier can be adaptively set based on actual applications, and this embodiment of the invention does not impose specific limitations on it. Preferably, in this embodiment of the invention, the preset multiplier is 2 times.

[0181] It should be noted that traditional ship detection methods based on visible light images typically rely on visual saliency and visual perception principles. Itti, Ft, SR, AC, and GBVS models are widely used to extract ship regions from ocean backgrounds, while HOG (Histogram of Oriented Gradient) features are commonly used feature descriptors in ship detection. In general, these traditional algorithms do not perform ideally in complex background environments, while methods based on Convolutional Neural Networks (CNNs) are particularly effective for ship detection in visible light remote sensing images. Deep learning methods can effectively overcome the shortcomings of traditional algorithms, and due to their large model capacity, they can provide detailed descriptions of targets, significantly improving the accuracy of ship detection. Based on their processing steps, deep learning networks can be divided into two-stage detection methods and single-stage detection methods. Single-stage detection methods predict multiple bounding boxes and their classification probabilities in a single evaluation; the YOLO network is a representative real-time detection method.

[0182] Ship detection aims to detect ship instances with specific features, involving both localization and classification operations. Generally, ship regions occupy relatively small spaces in remote sensing images; for example, in a 1920×1080 image, a ship typically occupies only a small area of ​​less than 200×200 pixels. This small size makes it difficult for traditional methods to achieve satisfactory results. In recent years, deep learning has shown great potential in image processing and has been widely applied in object detection. YOLO is a single-stage real-time object detection method that can train a YOLOv5 network with one class using a dataset with image augmentation to detect ship hulls. After obtaining the YOLO detection region, it can be widened to twice the side length of the YOLO detection region. If more than 50% of the widened region belongs to the ocean background, the YOLO detection region is retained.

[0183] It is understood that the embodiments of the present invention, through the YOLO detection network based on deep learning methods, can accurately determine the ship area where the target ship is located, thereby improving the accuracy of ship target detection.

[0184] Optionally, determining the cloud region containing the clouds in the original remote sensing image includes:

[0185] The original remote sensing image is converted from the RGB color space to the LAB color space;

[0186] In the LAB color space, the reference average value of the ocean region color is used as the cluster center, and the original remote sensing image is clustered based on the k-means++ algorithm to obtain the first ocean region contained in the original remote sensing image.

[0187] A second ocean region is obtained by performing morphological opening and closing operations on the first ocean region.

[0188] Based on the principle of dark channel prior, the cloud region contained in the second ocean region is obtained.

[0189] Specifically, in this embodiment of the invention, in order to determine the cloud region containing the clouds in the original remote sensing image, the original remote sensing image can first be converted from the RGB color space to the LAB color space. Then, in the LAB color space, the reference average value of the ocean region color is used as the cluster center. The original remote sensing image is clustered based on the k-means++ algorithm to obtain the first ocean region contained in the original remote sensing image. Then, morphological opening and closing operations are performed on the first ocean region to obtain the second ocean region. Furthermore, based on the dark channel prior principle, the cloud region contained in the second ocean region is obtained.

[0190] It should be noted that in visible light remote sensing images, the ocean region C sea The colors are usually close to blue, so the original remote sensing images can be converted from the RGB color space to the LAB color space first for more effective color contrast.

[0191] A reference mean color for ocean regions can be calculated using several pre-segmented remote sensing images, represented by A and B components in the LAB color space. Once a new raw remote sensing image is obtained, the reference mean is used as the cluster centers for ocean regions, and k-means++ is used for clustering. To improve efficiency, only two cluster centers can be selected, corresponding to whether a pixel belongs to an ocean region. After clustering, morphological opening and closing operations are performed on the extracted first ocean region to remove snowflake noise and fill holes, obtaining the second ocean region. Then, the second ocean region of interest is labeled based on the results of region connectivity component analysis.

[0192] Cloud cover will make ship targets invisible in remote sensing images. The cloud region C can be marked based on the dark channel prior principle. cloud The dark channel of the image is shown in equation (36):

[0193]

[0194] Among them, J cLet Ω(x) be the color channel of the remote sensing image, and let x be the local neighborhood centered at x. Let x represent the coordinates of a point in the image, and let y represent the neighboring pixels of x.

[0195] Except for cloud areas, the values ​​of the dark channel in the image are close to zero.

[0196] Optionally, the first 0.1% of pixels in the dark channel image can be marked as cloud pixels in order of brightness, and then morphological opening, morphological closing and connectivity component analysis can be performed on the marked cloud pixels in sequence to obtain the cloud region in the ocean.

[0197] It is understood that the embodiments of the present invention effectively realize a ship detection scheme based on visible light remote sensing images by combining deep learning and feature-based image processing techniques. The YOLOv5 network is used to effectively detect ship hulls, model the ocean area, and quantitatively evaluate image quality. Then, the wake detection results are used to assist in ship detection. This solves the problem of detecting invisible ship targets covered by clouds or located outside the image boundary in visible light remote sensing images, and improves the confidence of ship detection. It has advantages such as high detection accuracy, robustness, and strong adaptability.

[0198] The ship target detection method provided by this invention first determines the hull neighborhood of the target ship in the original remote sensing image, the cloud region containing the clouds in the original remote sensing image, and the image boundary region of the original remote sensing image. The hull neighborhood, cloud region, and image boundary region are then used as the search area for the ship's wake. The first ship wake region containing the wake of the target ship in the hull neighborhood, cloud neighborhood, and image boundary region is determined respectively. Based on the first ship wake region, the target ship is detected. This method can effectively detect ships that are covered by clouds or invisible outside the image boundary in visible light remote sensing images, thus improving the accuracy of target ship detection.

[0199] The ship target detection device provided by the present invention is described below. The ship target detection device described below can be referred to in correspondence with the ship target detection method described above.

[0200] Figure 4 This is a schematic diagram of the ship target detection device provided by the present invention, as shown below. Figure 4 As shown, the device includes: a first determining module 410, a second determining module 420, a third determining module 430, a fourth determining module 440, and a ship detection module 450; wherein:

[0201] The first determining module 410 is used to determine the ship area where the target ship is located in the original remote sensing image, and to determine the cloud area where the cloud layer is located in the original remote sensing image.

[0202] The second determining module 420 is used to determine the axis of the target ship in the ship region, and to determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship.

[0203] The third determining module 430 is used to determine the cloud neighborhood of the cloud region based on the cloud region, and to determine the image boundary region of the original remote sensing image based on the original remote sensing image.

[0204] The fourth determining module 440 is used to determine the first ship wake region containing the wake of the target ship in the hull neighborhood, the cloud neighborhood, and the image boundary region, respectively.

[0205] The ship detection module 450 is used to detect the target ship based on the wake region of the first ship;

[0206] Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

[0207] The ship target detection device provided by this invention first determines the hull neighborhood of the target ship contained in the original remote sensing image, the cloud region containing the clouds contained in the original remote sensing image, and the image boundary region of the original remote sensing image. The hull neighborhood, cloud region, and image boundary region are used as the search area for the ship's wake. The device then determines the first ship wake region containing the wake of the target ship contained in the hull neighborhood, cloud neighborhood, and image boundary region, and then detects the target ship based on the first ship wake region. This device can effectively detect ships that are covered by clouds or invisible outside the image boundary in visible light remote sensing images, thus improving the accuracy of target ship detection.

[0208] It should be noted that the ship target detection device provided in this embodiment of the invention can implement all the method steps implemented in the above-mentioned ship target detection method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0209] Figure 5 This is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as... Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute the ship target detection method provided by the above methods, which includes:

[0210] The ship region containing the target ship in the original remote sensing image is determined, and the cloud region containing the cloud layer in the original remote sensing image is determined.

[0211] Determine the axis of the target ship in the ship region, and determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship;

[0212] Based on the cloud region, determine the cloud neighborhood of the cloud region, and based on the original remote sensing image, determine the image boundary region of the original remote sensing image;

[0213] The first ship wake region, which contains the wake of the target ship within the hull neighborhood, the cloud neighborhood, and the image boundary region, is determined respectively.

[0214] The target ship is detected based on the wake region of the first ship;

[0215] Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

[0216] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0217] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the ship target detection method provided by the above methods, the method comprising:

[0218] The ship region containing the target ship in the original remote sensing image is determined, and the cloud region containing the cloud layer in the original remote sensing image is determined.

[0219] Determine the axis of the target ship in the ship region, and determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship;

[0220] Based on the cloud region, determine the cloud neighborhood of the cloud region, and based on the original remote sensing image, determine the image boundary region of the original remote sensing image;

[0221] The first ship wake region, which contains the wake of the target ship within the hull neighborhood, the cloud neighborhood, and the image boundary region, is determined respectively.

[0222] The target ship is detected based on the wake region of the first ship;

[0223] Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

[0224] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the ship target detection methods provided above, the method comprising:

[0225] The ship region containing the target ship in the original remote sensing image is determined, and the cloud region containing the cloud layer in the original remote sensing image is determined.

[0226] Determine the axis of the target ship in the ship region, and determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship;

[0227] Based on the cloud region, determine the cloud neighborhood of the cloud region, and based on the original remote sensing image, determine the image boundary region of the original remote sensing image;

[0228] The first ship wake region, which contains the wake of the target ship within the hull neighborhood, the cloud neighborhood, and the image boundary region, is determined respectively.

[0229] The target ship is detected based on the wake region of the first ship;

[0230] Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

[0231] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0232] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0233] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting ship targets, characterized in that, include: The ship region containing the target ship in the original remote sensing image is determined, and the cloud region containing the cloud layer in the original remote sensing image is determined. Determine the axis of the target ship in the ship region, and determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship; Based on the cloud region, determine the cloud neighborhood of the cloud region, and based on the original remote sensing image, determine the image boundary region of the original remote sensing image; The hull neighborhood, the cloud neighborhood, and the image boundary region are taken as the target region, and the image corresponding to the target region is converted from the RGB color space to the LAB color space; In the LAB color space, the luminance component of the image corresponding to the target region is modeled based on the Burr XII distribution, and based on the idea of ​​constant false alarm rate (CFAR), the second ship wake region where the wake of the target ship is located is extracted from the target region. The second ship wake region was clustered using the density-based clustering algorithm DBSCAN to obtain the third ship wake region. Using a binary normal distribution, all the obtained wake regions of the third ship are simulated to obtain the equal probability curves of the binary normal distribution; If the ellipticity of the equal probability curve is greater than a preset ellipticity and the area of ​​the third ship wake region is greater than a preset area, the third ship wake region is taken as the first ship wake region. The target ship is detected based on the wake region of the first ship; Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

2. The ship target detection method according to claim 1, characterized in that, Determining the ship region containing the target ship in the original remote sensing image includes: Based on the YOLO detection network, target ships contained in the original remote sensing image are detected, and the first YOLO detection region of the YOLO detection network in the original remote sensing image is obtained. The length and width of the first YOLO detection region are both enlarged by a preset multiple in the original remote sensing image to obtain the second YOLO detection region; If the ratio of the area occupied by the ocean background in the second YOLO detection area to the area of ​​the second YOLO detection area is greater than a preset value, the ship area is determined based on the first YOLO detection area.

3. The ship target detection method according to claim 1, characterized in that, Determining the cloud region containing the clouds in the original remote sensing image includes: The original remote sensing image is converted from the RGB color space to the LAB color space; In the LAB color space, the reference average value of the ocean region color is used as the cluster center, and the original remote sensing image is clustered based on the k-means++ algorithm to obtain the first ocean region contained in the original remote sensing image. A second ocean region is obtained by performing morphological opening and closing operations on the first ocean region. Based on the principle of dark channel prior, the cloud region contained in the second ocean region is obtained.

4. The ship target detection method according to claim 1, characterized in that, Determining the axis of the target ship in the ship region includes: Based on the principal component analysis (PCA) method, the axis of the target ship in the ship region is determined.

5. A ship target detection device, characterized in that, include: The first determining module is used to determine the ship area where the target ship is located in the original remote sensing image, and to determine the cloud area where the cloud layer is located in the original remote sensing image. The second determining module is used to determine the axis of the target ship in the ship region, and to determine the hull neighborhood of the target ship based on the ship region and the axis of the target ship. The third determining module is used to determine the cloud neighborhood of the cloud region based on the cloud region, and to determine the image boundary region of the original remote sensing image based on the original remote sensing image. The fourth determining module is used to take the hull neighborhood, the cloud neighborhood and the image boundary region as the target region, and convert the image corresponding to the target region from the RGB color space to the LAB color space; In the LAB color space, the luminance component of the image corresponding to the target region is modeled based on the Burr XII distribution, and based on the idea of ​​constant false alarm rate (CFAR), the second ship wake region where the wake of the target ship is located is extracted from the target region; the second ship wake region is clustered using the density-based clustering algorithm DBSCAN to obtain the third ship wake region. Using a binary normal distribution, all the obtained third ship wake regions are simulated to obtain the equal probability curve of the binary normal distribution; if it is determined that the ellipticity of the equal probability curve is greater than a preset ellipticity and the area of ​​the third ship wake region is greater than a preset area, the third ship wake region is taken as the first ship wake region. The ship detection module is used to detect the target ship based on the wake area of ​​the first ship; Wherein, the hull neighborhood is on the axis of the target ship and adjacent to the ship area; the outer boundary of the cloud neighborhood is a first preset distance from the inner boundary of the cloud neighborhood; the inner boundary of the cloud neighborhood is the boundary of the minimum boundary rectangle of the cloud area; the inner boundary of the image boundary area is a second preset distance from the outer boundary of the image boundary area; and the outer boundary of the image boundary area is the boundary of the original remote sensing image.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the ship target detection method as described in any one of claims 1 to 4.

Images