Ipsa-based ship target identification method and system for airborne infrared image

Abstract

The embodiment of the application provides an IPSA-based ship target identification method and system for an airborne infrared image, wherein the method comprises the following steps: preprocessing an acquired original airborne image; performing global threshold segmentation on the preprocessed image based on an improved PID search algorithm to obtain a global water mask, and performing processing on the preprocessed image based on a local adaptive threshold segmentation method to obtain a local water mask; generating a final water mask according to the global water mask and the local water mask; constructing a candidate target region in a non-water region constrained by the final water mask; determining all connected regions in the candidate target region, calculating the geometric features and the gray features of each connected region, and screening the connected regions according to a preset feature threshold to determine the connected regions meeting the screening condition as ship targets. The technical scheme provided by the application can adapt to a complex marine environment and efficiently and accurately extract ship targets.

Description

Technical Field

[0001] This invention belongs to the field of image processing and target recognition technology, and particularly relates to a method and system for identifying ship targets based on IPSA airborne infrared imaging. Background Technology

[0002] In the field of maritime airborne target recognition, traditional target recognition methods are often affected by factors such as image quality, complex background, and small target scale, resulting in inaccurate extraction results or low efficiency.

[0003] Traditional methods for processing maritime airborne infrared imagery, especially UAV aerial images, are limited by factors such as image quality, complex ocean backgrounds (e.g., waves, reflections), and the small scale of targets at long distances. These limitations make it difficult to effectively and accurately separate ship targets from dynamically changing seawater backgrounds. When ship targets coexist with shorelines, islands, or other obstructions, the discriminative power of existing algorithms further decreases, leading to lower recognition accuracy. This can result in misclassifying background as targets or missing targets altogether, making stable and efficient ship target identification impossible in the complex and ever-changing real-world marine environment.

[0004] Therefore, developing an algorithm that can adapt to complex marine environments and efficiently and accurately extract ship targets is of great practical significance. Summary of the Invention

[0005] The IPSA-based airborne infrared image ship target recognition method and system provided in this application can adapt to complex marine environments, accurately and stably find the optimal segmentation threshold for complex scenes, and achieve precise extraction of ship targets.

[0006] In a first aspect, embodiments of this application provide a method for identifying ship targets based on airborne infrared imaging using IPSA, including:

[0007] S1. Preprocess the acquired raw airborne images to obtain preprocessed images;

[0008] S2. Based on the improved PID search algorithm, global threshold segmentation is performed on the preprocessed image to obtain a global water mask. The preprocessed image is then processed based on the local adaptive threshold segmentation method to obtain a local water mask. Logical fusion and morphological processing are performed on the global water mask and the local water mask to generate the final water mask.

[0009] S3. In the non-water area constrained by the final water body mask, extract feature points and construct a convex hull based on the feature points. The internal region of the convex hull is the candidate target region.

[0010] S4. Within the candidate target area, identify all connected regions, calculate the geometric and grayscale features of each connected region, and filter the connected regions according to the preset feature thresholds. The connected regions that meet the filtering conditions are identified as ship targets.

[0011] In one optional implementation, in S1, the acquired raw airborne imagery is preprocessed, including:

[0012] The original airborne image is converted to grayscale and its contrast is enhanced to obtain a grayscale-enhanced image.

[0013] The grayscale enhanced image is smoothed by Gaussian filtering to obtain the preprocessed image.

[0014] In an optional implementation, in S2, global threshold segmentation is performed on the preprocessed image based on an improved PID search algorithm to obtain a global water mask, including:

[0015] S2.1 Initialize multiple solutions within the preset threshold search space;

[0016] S2.2 In each iteration, the position of each solution is updated by combining the PID control term and the Lévy flight term. The attenuation coefficient used to control the influence of the Lévy flight term decreases adaptively with the increase of the number of iterations.

[0017] S2.3. Evaluate the quality of each updated solution based on the fitness function that maximizes the inter-class variance, and determine the current optimal solution;

[0018] S2.4 After the iteration is completed, the threshold corresponding to the optimal solution of the fitness function is taken as the global best segmentation threshold, and the preprocessed image is segmented using the global best segmentation threshold to obtain the global water mask.

[0019] In one alternative implementation, the PID control term is calculated based on the historical error difference between the current solution and the current best solution.

[0020] In one optional implementation, in S2, the preprocessed image is processed based on a local adaptive threshold segmentation method to obtain a local water mask, including:

[0021] Calculate the mean and standard deviation of gray levels within a preset neighborhood centered on each pixel in the preprocessed image;

[0022] Based on the grayscale mean and grayscale standard deviation, the local adaptive segmentation threshold of the pixel is dynamically calculated to obtain the local water mask.

[0023] In one optional implementation, in step S2, the global water mask and the local water mask are logically fused and morphologically processed, including:

[0024] Perform a logical OR operation on the global water mask and the local water mask to obtain the initial fused mask;

[0025] Morphological closing and opening operations are performed sequentially on the initial fusion mask;

[0026] The morphologically processed mask is subjected to area filtering to retain only connected regions with an area not less than a preset area threshold, so as to obtain the final water body mask.

[0027] In one optional implementation, in step S3, extracting feature points includes:

[0028] Based on the Harris corner detection algorithm, the corner response value of each pixel is calculated in non-water areas;

[0029] Pixels whose corner response values ​​are greater than a preset response threshold and satisfy a preset minimum distance constraint are identified as feature points.

[0030] In one optional implementation, based on the Harris corner detection algorithm, the corner response value of each pixel is calculated in the non-water area, including:

[0031] Calculate the gradient in the horizontal direction and the gradient in the vertical direction of the preprocessed image;

[0032] The image autocorrelation matrix is ​​constructed based on the gradients in the horizontal and vertical directions;

[0033] The corner response value is calculated based on the determinant and trace of the image autocorrelation matrix.

[0034] In one optional implementation, in S4, the geometric features include: the area of ​​the connected region and the aspect ratio of the minimum bounding rectangle of the connected region;

[0035] Gray-scale features include: the average gray-scale value of pixels within a connected region and the standard deviation of gray-scale values ​​of pixels within a connected region.

[0036] Secondly, embodiments of this application provide an airborne infrared imaging ship target recognition system based on IPSA, including:

[0037] The image acquisition module is used to preprocess the acquired raw airborne images to obtain preprocessed images;

[0038] The water mask generation module is used to perform global threshold segmentation on the preprocessed image based on the improved PID search algorithm to obtain a global water mask, and to process the preprocessed image based on the local adaptive threshold segmentation method to obtain a local water mask. The global water mask and the local water mask are logically fused and morphologically processed to generate the final water mask.

[0039] The target region segmentation module is used to extract feature points in the non-water region constrained by the final water body mask, and construct a convex hull based on the feature points. The interior region of the convex hull is the candidate target region.

[0040] The output module is used to determine all connected regions within the candidate target area, calculate the geometric and grayscale features of each connected region, and filter the connected regions according to the preset feature thresholds, identifying the connected regions that meet the filtering conditions as ship targets.

[0041] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method provided in embodiments of this application.

[0042] Fourthly, embodiments of this application provide a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed in a computer, causes the computer to perform the method provided in embodiments of this application.

[0043] In the segmentation stage, this invention introduces an improved PID search algorithm (IPSA), integrating an adaptive Lévy flight mechanism and a dynamic attenuation coefficient into the traditional framework. This balances the algorithm's global exploration and local exploitation capabilities, and, guided by maximizing inter-class variance, achieves the technical effect of rapidly and stably finding the optimal global segmentation threshold, thus improving the quality of basic segmentation. Secondly, by combining a local adaptive thresholding segmentation method, dynamic threshold calculation is performed based on the local grayscale characteristics of the image to compensate for the shortcomings of the global threshold in areas of uneven grayscale. Finally, through logical fusion and morphological processing, the advantages of both methods are combined to generate a complete and accurate water mask. The technical effect of this method greatly suppresses background interference. Then, in the target localization stage, by extracting Harris corner points and constructing convex hulls in the non-water body region constrained by the mask, the potential region is delineated by utilizing the corner point feature clustering of the ship target. This achieves the technical effect of focusing the search range on the candidate target region and further eliminating sea surface noise. Finally, in the target screening stage, threshold filtering is performed based on the geometric and grayscale features of the candidate region, and the ship target is judged based on its prior morphology and statistical characteristics. This achieves the technical effect of accurately distinguishing real targets from residual interference, thus completing the complete process of robustly and accurately extracting ship targets from complex backgrounds. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the IPSA-based airborne infrared imagery-based ship target identification method provided in this application embodiment.

[0045] Figure 2 These are the original airborne images provided in the embodiments of this application;

[0046] Figure 3 This is a preprocessed image provided in the embodiments of this application;

[0047] Figure 4 This is a logic diagram of the water body segmentation operation method provided in the embodiments of this application;

[0048] Figure 5 This is a logic diagram of the operation steps of the improved PID search algorithm provided in the embodiments of this application;

[0049] Figure 6 This is a schematic diagram of the global water mask provided in the embodiments of this application;

[0050] Figure 7 This is a schematic diagram of a local water mask provided in an embodiment of this application;

[0051] Figure 8 This is a schematic diagram of the final water mask provided in the embodiments of this application;

[0052] Figure 9 This is a schematic diagram of the feature point extraction results provided in an embodiment of this application;

[0053] Figure 10 This is a schematic diagram of the convex hull region provided in an embodiment of this application;

[0054] Figure 11 This is a schematic diagram of the connected regions provided in the embodiments of this application;

[0055] Figure 12 This is a schematic diagram of the ship target extraction results provided in the embodiments of this application. Detailed Implementation

[0056] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Figure 1 This is a flowchart illustrating the IPSA-based airborne infrared imagery-based ship target recognition method provided in this application embodiment. The method can be executed by an IPSA-based airborne infrared imagery-based ship target recognition system, which can be implemented by software and / or hardware and can be configured in electronic devices such as computers.

[0058] like Figure 1 As shown, the technical solution provided in this application includes the following steps:

[0059] S1. Preprocess the acquired airborne raw images to obtain preprocessed images.

[0060] In some embodiments, such as Figure 2 and Figure 3 As shown, the preprocessing includes:

[0061] The original airborne image is converted to grayscale and its contrast is enhanced to obtain a grayscale enhanced image.

[0062] First, the original infrared image is converted to grayscale using a weighted average method. This aims to focus subsequent processing on the image's brightness information and simplify calculations.

[0063] To ensure the accuracy of image feature extraction, in addition to using a weighted average method to convert the original image to grayscale, this embodiment employs two contrast enhancement processes to stretch the grayscale range of a specific region from the original [0.3, 0.7] to [0, 1], which can highlight target features and suppress background interference.

[0064] ;

[0065] in, This represents the grayscale value of the pixel at coordinates (x, y) in the output image after transformation. This represents the original grayscale value of the pixel at coordinates (x, y) in the input image; This indicates the lower limit of the grayscale range that needs to be stretched, for example, 0.3; This indicates the upper limit of the grayscale range to be stretched, for example, 0.7; 255 represents the maximum grayscale value of an 8-bit image. The grayscale values ​​of the original image will be stretched to [...]. , Pixels within the [0, 255] range are linearly mapped to the new [0, 255] range. Pixels with grayscale values ​​lower than [0, 255] are [0, 255]. The pixels are set to 0 (black), and those higher than 0 are set to 0 (black). The pixel value is set to 255 (white). This significantly increases the grayscale difference between the target and the background, making the features more prominent.

[0066] The grayscale enhanced image is smoothed by Gaussian filtering to obtain the preprocessed image.

[0067] Specifically, to facilitate subsequent feature point extraction, the contrast-enhanced image is smoothed using a Gaussian filter to filter out noise while preserving key edge information. The filter used in this embodiment is a two-dimensional Gaussian kernel, and its formula is:

[0068] ;

[0069] Where x' and y' are the coordinate offsets of any point within the Gaussian kernel relative to the kernel center, and σ g σ is the standard deviation of the Gaussian function, used to control the smoothness of the filter. In this embodiment, σ g =1.5, G(x',y',σ) represents the Gaussian weight value of the pixel at pixel coordinates (x,y).

[0070] During processing, this two-dimensional Gaussian kernel is used as a sliding window to perform convolution operations with the image. The result of the convolution calculation is that the new gray value of each pixel in the image is a weighted average of its own gray values ​​and those of its surrounding pixels, with the weights determined by a Gaussian function. Pixels closer to the center point have a larger weight, while pixels farther away have a smaller weight. This effectively suppresses high-frequency noise, making the overall image smoother and facilitating more stable extraction of image features in subsequent processing.

[0071] Traditional histogram equalization methods enhance image contrast by redistributing pixel intensity values. However, the introduction of dynamic range adjustment and Gaussian filtering techniques allows the algorithm to perform targeted processing based on the local characteristics of the image, thereby improving the contrast between the target area and the background while better preserving key details and filtering noise.

[0072] S2. Based on the improved PID search algorithm, global threshold segmentation is performed on the preprocessed image to obtain a global water mask. Then, based on the local adaptive threshold segmentation method, the preprocessed image is processed to obtain a local water mask. Logical fusion and morphological processing are performed on the global water mask and the local water mask to generate the final water mask.

[0073] In some embodiments, see Figures 4-8 In the water segmentation stage, this embodiment proposes an improved PID-based search algorithm, IPSA. This algorithm is systematically enhanced based on the traditional PID-based search algorithm (PSA) framework, enabling it to meet the needs of metaheuristic algorithms in image recognition and realize its optimal potential. The specific steps of the algorithm are as follows:

[0074] S2.1. Initialize multiple solutions within the preset threshold search space.

[0075] Within a preset threshold search space [0.1, 0.9], N initial solutions (i.e., candidate thresholds) are generated uniformly. The position of each solution is denoted as . .

[0076] S2.2 In each iteration, the position of each solution is updated by combining the PID control term and the Lévy flight term. The attenuation coefficient used to control the influence of the Lévy flight term is adaptively attenuated as the number of iterations increases.

[0077] Specifically, in each iteration, the position update of each individual (solution) is driven by both the PID control term and the Lévy flight term to achieve a balance between local depth exploration and global breadth exploration. The update formula is:

[0078] ;

[0079] in, As a random factor, for:

[0080] ;

[0081] Among them, t i (n) represents the position of the i-th solution in the nth iteration, i.e., the threshold in the corner detection and adaptive threshold segmentation steps; r is a random factor between 0 and 1; PID is the PID control term, providing accurate local search based on historical error; Out is the global exploration term, driven by cosine annealing and Lévy flight, used to escape local optima; and T represents the maximum number of iterations. This represents the cosine annealing function, whose value gradually decreases from 1 to cos(1) as the number of iterations increases. It is used to balance global exploration and local exploitation. k This represents the difference between the current individual and the best individual in the current era, and 'a' represents the decay coefficient, which adaptively decreases with the number of iterations, allowing the algorithm to focus more on local exploitation in later stages. The decay coefficient 'a' is set as follows: Levy(N,1) represents the Lévy flight term, which injects random jumps into the iterative process and helps escape local optima.

[0082] The formula for the PID control term is as follows:

[0083] ;

[0084] in, For the current individual and the best individual of the present era The difference; These are the proportional coefficient, integral coefficient, and differential coefficient, respectively.

[0085] The Lévy flight step size for the Lévy flight term is determined by the following formula:

[0086] ;

[0087] ;

[0088] Where Levy(s) represents the Lévy flight stride, which is the size of a single random jump, and u and v are random numbers following a normal distribution, with a mean of 0 and a variance of σ. l 2 The mean of v is 0 and the variance is 1. β represents the parameter of the Lévy distribution, which controls the characteristics of the step size distribution. Γ(·) represents the gamma function.

[0089] S2.3. Evaluate the quality of each updated solution based on the fitness function that maximizes the inter-class variance, and determine the current optimal solution.

[0090] In some embodiments, the fitness function is defined as the inter-class variance in the Otsu criterion. The negative value of is used to maximize the inter-class variance. The formula for calculating the inter-class variance is as follows:

[0091] ;

[0092] ;

[0093] In the formula, and These represent the proportions of pixels whose grayscale values ​​are below and above the thresholds, respectively. and These correspond to the average grayscale values ​​of these two types of pixels, respectively.

[0094] The overall algorithmic goal of IPSA is to find a fitness function that satisfies the condition of the fitness function. The minimum threshold is expressed mathematically as follows:

[0095] ;

[0096] in, This represents the final optimal segmentation threshold.

[0097] S2.4 After the iteration is completed, the threshold corresponding to the optimal solution of the fitness function is taken as the global best segmentation threshold, and the preprocessed image is segmented using the global best segmentation threshold to obtain the global water mask.

[0098] In some embodiments, for grayscale enhanced images The algorithm uses the optimal threshold. The potential target area and the water body area are divided, and the division results are as follows: Figure 6 As shown.

[0099] Furthermore, the preprocessed image is processed using a local adaptive thresholding segmentation method to obtain a local water body mask, including:

[0100] S2.5 Calculate the mean gray value and standard deviation of gray value within a preset neighborhood centered on each pixel in the preprocessed image;

[0101] S2.6. Based on the grayscale mean and grayscale standard deviation, dynamically calculate the local adaptive segmentation threshold of the pixel to obtain the local water mask.

[0102] Specifically, for areas with uneven grayscale in infrared images, using a global threshold alone may result in incomplete or incorrect segmentation. Therefore, local adaptive threshold segmentation is introduced as a supplement. This embodiment uses an improved NiBlack algorithm to calculate the dynamic threshold within the local neighborhood of each pixel, as shown in the following formula:

[0103] ;

[0104] In the formula, This represents the local dynamic threshold of the pixel at coordinates (x, y). and They are in coordinates Centered on the pixel at that location The mean and standard deviation of gray levels within the window. Local segmentation results are shown below. Figure 7 As shown.

[0105] Furthermore, the global and local water body masks are logically fused and morphologically processed, including:

[0106] S2.7 Perform a logical OR operation on the global water mask and the local water mask to obtain the initial fusion mask;

[0107] Specifically, the global segmentation mask and the local segmentation mask are combined to ensure that all regions identified as water bodies by any method are preserved, forming the initial fusion mask. :

[0108] ;

[0109] Among them, M global M represents the global segmentation mask. local This represents a local segmentation mask.

[0110] S2.8 Perform morphological closing and opening operations sequentially on the initial fusion mask;

[0111] Specifically, firstly Morphological closing operations (imclose) are performed to fill small holes within the water area, followed by opening operations (imopen) to eliminate edge burrs and isolated noise points. All operations utilize a disk-shaped structuring element with a radius of 2. :

[0112] ;

[0113] Among them, imopen(·) represents the morphological opening operation, which first erodes and then expands, and is used to eliminate edge burrs and isolated noise points. imclose(·) represents the morphological closing operation, which first expands and then erodes, and is used to fill small holes and gaps in the water body area.

[0114] S2.9. Perform area filtering on the morphologically processed mask, retaining only connected regions with an area not less than a preset area threshold to obtain the final water body mask.

[0115] Specifically, to remove any remaining small connected regions that are not water bodies, area filtering is applied to retain only connected regions with an area of ​​at least 300 pixels, resulting in the final water body mask. :

[0116] ;

[0117] Here, `bwareaopen(M,300)` represents a binary image area filtering operation, retaining only connected regions with an area of ​​at least 300 pixels. The final water mask... This refers to the segmented, complete water area, which provides a precise non-water mask for subsequent ship target detection within the defined area (see...). Figure 8 (In the diagram, black indicates non-water body masks).

[0118] A logical OR fusion is performed on the masks generated by local adaptive and global thresholding, combining the advantages of local thresholding in preserving details and global thresholding in ensuring overall consistency. To eliminate holes and discrete noise in the fused mask, morphological closing operations are used to fill gaps; opening operations are used to remove small regions; and area filtering is used to further purify the water area. The final output water mask will be used to constrain the detection range of non-water targets and reduce background interference.

[0119] S3. In the non-water region constrained by the final water body mask, extract feature points and construct a convex hull based on the feature points. The internal region of the convex hull is the candidate target region.

[0120] For details, see Figure 9 and Figure 10 The Harris corner detection algorithm is used to identify points (i.e., corners) in non-water areas where image grayscale changes sharply. Specifically, this includes:

[0121] Based on the Harris corner detection algorithm, the corner response value of each pixel is calculated in non-water areas.

[0122] First, calculate the gradients of the preprocessed image in the horizontal (x-direction) and vertical (y-direction) directions, denoted as follows: and Then, an autocorrelation matrix M is constructed for each pixel in the image. This matrix describes the gray-level changes within a small window surrounding that pixel. The formula is:

[0123] ;

[0124] Here, w(x,y) is a Gaussian window function used to perform a weighted average of the gradient values ​​within the window, thereby enhancing robustness to image noise. The Gaussian window function is defined as follows:

[0125] ;

[0126] Where x' and y' are the coordinate offsets of a pixel within the window relative to the center of the window, and σ local It is the standard deviation of the Gaussian function, which controls the range of weighting.

[0127] Furthermore, the corner response values ​​are calculated. :

[0128] ;

[0129] In the formula, It is a matrix The determinant, It is a matrix traces, It is an empirical constant, usually taken as... .

[0130] Furthermore, pixels whose corner response values ​​are greater than a preset response threshold and satisfy a preset minimum distance constraint are identified as feature points.

[0131] Specifically, select Greater than the preset response threshold The local maxima are taken as the set of corner feature points:

[0132] ;

[0133] Where P represents the set of corner feature points, (x i ,y i ) represents the pixel coordinates of the i-th corner feature point.

[0134] To remove spatially overly dense feature points, a minimum distance threshold is set. Pixels, ensuring that the Euclidean distance between any two points satisfies the following formula:

[0135] ;

[0136] Where, d min This represents the minimum distance threshold. It is removed through iteration or matrix operations. The points are used to obtain the filtered feature point set. .

[0137] Furthermore, a convex hull is constructed based on feature points, and the interior region of the convex hull is used as the candidate target region.

[0138] Specifically, convex hull operation is performed on all extracted feature points to generate the smallest convex polygon that can enclose all feature points. The interior region of the smallest convex polygon is the interior region of the convex hull.

[0139] Generate regions of interest for the detected feature points (i.e., candidate target area):

[0140]

[0141] Where, λ i represents the weight coefficient of the i-th corner feature point, and n represents the total number of corner feature points.

[0142] S4. Within the candidate target area, identify all connected regions, calculate the geometric and grayscale features of each connected region, and filter the connected regions according to the preset feature thresholds. The connected regions that meet the filtering conditions are identified as ship targets.

[0143] In some embodiments, see Figure 11 and Figure 12 After selecting all feature points that meet the threshold conditions, the process iterates through, marks, and numbers all interconnected white pixel regions within the convex hull. Each marked, independent set of white pixels is called a connected region, and each independent connected region is considered an independent candidate ship target. For each marked connected region, a series of morphological and grayscale feature parameters are selected for description and subsequent filtering, such as region area, aspect ratio of the bounding rectangle, average grayscale, and grayscale standard deviation.

[0144] Finally, based on prior knowledge or statistical analysis of training samples, the range of ship target features, such as reasonable size ranges and aspect ratios, is used to set appropriate feature threshold filters to discriminate and filter all connected regions. Regions that clearly do not conform to ship characteristics are removed, such as noise blocks that are too small or excessively long linear interference. The remaining connected regions are then identified as the extracted ship targets, and their positions, contours, and feature vectors are recorded for target recognition result extraction. The cropped ship image is then combined with the original image to define the ship region.

[0145] This application also provides an IPSA-based airborne infrared imaging ship target identification system, which includes:

[0146] The image acquisition module is used to preprocess the acquired raw airborne images to obtain preprocessed images;

[0147] The water mask generation module is used to perform global threshold segmentation on the preprocessed image based on the improved PID search algorithm to obtain a global water mask, and to process the preprocessed image based on the local adaptive threshold segmentation method to obtain a local water mask. The global water mask and the local water mask are logically fused and morphologically processed to generate the final water mask.

[0148] The target region segmentation module is used to extract feature points in the non-water region constrained by the final water body mask, and construct a convex hull based on the feature points. The interior region of the convex hull is the candidate target region.

[0149] The output module is used to determine all connected regions within the candidate target area, calculate the geometric and grayscale features of each connected region, and filter the connected regions according to the preset feature thresholds, identifying the connected regions that meet the filtering conditions as ship targets.

[0150] The execution process of the system part of this application embodiment is the same as that of the method part of the embodiment described above, and will not be repeated here.

[0151] This application also provides an electronic device, including: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any of the IPSA-based airborne infrared imagery ship target identification methods.

[0152] This application also proposes a computer storage medium storing a computer program, which, when executed by a processor, implements any one of the IPSA-based airborne infrared imagery ship target recognition methods.

[0153] Computer storage media may be simply referred to as media. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Dual Data SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus Direct RAM (RDRAM), Direct Memory Bus Dynamic RAM (DRDRAM), and Memory Bus Dynamic RAM (RDRAM). The various embodiments described in this specification are presented in a progressive manner, with reference allowed to each other for similar or identical parts. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for apparatuses, devices, and non-volatile computer storage media are described simply because they are substantially similar to the method embodiments; relevant details can be found in the descriptions of the method embodiments.

[0154] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A method for identifying ship targets based on airborne infrared imagery using IPSA, characterized in that, include: S1. Preprocess the acquired raw airborne images to obtain preprocessed images; S2. The preprocessed image is subjected to global threshold segmentation based on the improved PID search algorithm to obtain a global water mask. The preprocessed image is then processed based on the local adaptive threshold segmentation method to obtain a local water mask. The global water mask and the local water mask are logically fused and morphologically processed to generate the final water mask. The step of performing global threshold segmentation on the preprocessed image based on an improved PID search algorithm to obtain a global water mask includes: S2.1 Initialize multiple solutions within the preset threshold search space; S2.2 In each iteration, the position of each solution is updated by combining the PID control term and the Lévy flight term, wherein the attenuation coefficient used to control the influence of the Lévy flight term adaptively decreases as the number of iterations increases; wherein the PID control term is calculated based on the historical error difference between the current solution and the current best solution. S2.

3. Evaluate the quality of each updated solution based on the fitness function that maximizes the inter-class variance, and determine the current optimal solution; S2.4 After the iteration is completed, the threshold corresponding to the optimal solution of the fitness function is taken as the global optimal segmentation threshold, and the preprocessed image is segmented using the global optimal segmentation threshold to obtain a global water mask. S3. In the non-water area constrained by the final water body mask, feature points are extracted, and a convex hull is constructed based on the feature points. The internal region of the convex hull is the candidate target region. S4. Within the candidate target area, determine all connected regions, calculate the geometric features and grayscale features of each connected region, and filter the connected regions according to a preset feature threshold, and determine the connected regions that meet the filtering conditions as ship targets.

2. The method according to claim 1, characterized in that, In step S1, the acquired raw airborne images are preprocessed, including: The original airborne image is subjected to grayscale conversion and contrast enhancement to obtain a grayscale enhanced image; The grayscale enhanced image is smoothed by Gaussian filtering to obtain the preprocessed image.

3. The method according to claim 1, characterized in that, In step S2, the process of processing the preprocessed image based on the local adaptive threshold segmentation method to obtain a local water mask includes: Calculate the mean gray level and standard deviation of gray level within a preset neighborhood centered on each pixel in the preprocessed image; Based on the mean gray level and the standard deviation of gray level, the local adaptive segmentation threshold of the pixel is dynamically calculated to obtain the local water body mask.

4. The method according to claim 3, characterized in that, In step S2, the global water mask and the local water mask are logically fused and morphologically processed, including: Perform a logical OR operation on the global water mask and the local water mask to obtain an initial fusion mask; Morphological closing and opening operations are sequentially performed on the initial fusion mask; The morphologically processed mask is subjected to area filtering to retain only connected regions with an area not less than a preset area threshold, so as to obtain the final water body mask.

5. The method according to claim 1, characterized in that, In step S3, the extraction of feature points includes: Based on the Harris corner detection algorithm, the corner response value of each pixel is calculated in the non-water body area; The pixel whose corner response value is greater than the preset response threshold and satisfies the preset minimum distance constraint is determined as the feature point.

6. The method according to claim 5, characterized in that, The Harris corner detection algorithm is used to calculate the corner response value of each pixel in the non-water body area, including: Calculate the gradient of the preprocessed image in the horizontal direction and the gradient in the vertical direction; The image autocorrelation matrix is ​​constructed based on the gradients in the horizontal and vertical directions. The corner response value is calculated based on the determinant and trace of the image autocorrelation matrix.

7. The method according to claim 1, characterized in that, In S4, the geometric features include: the area of ​​the connected region and the aspect ratio of the minimum bounding rectangle of the connected region; The grayscale features include: the average grayscale value of pixels within the connected region and the grayscale standard deviation of pixels within the connected region.

8. An airborne infrared imaging ship target recognition system based on IPSA, characterized in that, include: The image acquisition module is used to preprocess the acquired raw airborne images to obtain preprocessed images; The water mask generation module is used to perform global threshold segmentation on the preprocessed image based on an improved PID search algorithm to obtain a global water mask, and to process the preprocessed image based on a local adaptive threshold segmentation method to obtain a local water mask. The global water mask and the local water mask are then logically fused and morphologically processed to generate the final water mask. The step of performing global threshold segmentation on the preprocessed image based on an improved PID search algorithm to obtain a global water mask includes: initializing multiple solutions within a preset threshold search space; updating the position of each solution in each iteration by combining a PID control term and a Lévy flight term, wherein the attenuation coefficient used to control the influence of the Lévy flight term adaptively decreases with the increase of the number of iterations; wherein the PID control term is calculated based on the historical error difference between the current solution and the current best solution; evaluating the quality of each updated solution based on a fitness function that maximizes the inter-class variance and determining the current best solution; after the iteration is completed, using the threshold corresponding to the solution with the optimal fitness function as the global best segmentation threshold, and segmenting the preprocessed image using the global best segmentation threshold to obtain the global water mask; The target region segmentation module is used to extract feature points in the non-water region constrained by the final water body mask, and construct a convex hull based on the feature points, wherein the internal region of the convex hull is the candidate target region; The output module is used to determine all connected regions within the candidate target area, calculate the geometric features and grayscale features of each connected region, and filter the connected regions according to a preset feature threshold, and determine the connected regions that meet the filtering conditions as ship targets.

Images