A space-based infrared air target detection method, storage medium and computer device

Abstract

The present invention belongs to the fields of infrared remote sensing and infrared space technology, and discloses a space-based infrared air target detection method, a storage medium, and a computer device, and determines a reference frame I b and a reference frame I b+l , performs local normalization processing on the b frame, performs local normalization processing on the (b + j)th frame, and performs inter-frame correlation difference; after the inter-frame correlation difference is completed, performs dipole enhancement and spatial difference; multiplies the data after spatial background suppression, dipole enhancement, and spatial difference to obtain an intermediate result; performs global normalization processing, local feature contrast, and adaptive threshold segmentation on the intermediate result data, and performs detection results. The present invention realizes space-based infrared air target detection by constructing a local inter-frame matching difference model and a spatial local contrast model, reduces the processing complexity of infrared air target detection and the resource requirements for hardware implementation, and effectively improves the efficiency of target detection.

Description

Technical Field

[0001] The present invention belongs to the fields of infrared remote sensing and infrared space technology, and particularly relates to a space-based infrared air target detection method, a storage medium, and a computer device. Background Technique

[0002] At present, the distance between an air target and a space-based infrared detector is greater than 300 km. Due to the low resolution of infrared remote sensing and the influence of factors such as atmospheric interference, optical scattering, and diffraction, air targets mostly exhibit the characteristics of small size and weak energy in infrared images, meeting the criteria of infrared dim small targets. Weak energy means that the contrast between the target and the background is not obvious enough, and small size means that the target does not have certain shape and texture features. In addition, in the actual application scenario, the background of space-based infrared images consists of clouds, land, ocean, and various random noises, and there are strong background clutters in these backgrounds, whose gray values are much larger than the gray values of air targets. Moreover, the computing resources available on the space-based platform are limited. Therefore, it is difficult for the space-based platform to fully preprocess the original infrared image and narrow the intensity gap between the target and the clutter; at the same time, the limited computing resources also limit the performance of space-based detection algorithms and reduce the detection efficiency. Due to the above reasons, effective space-based detection of air targets is still a difficult task. Infrared small target detection is a hot issue, and a large number of detection methods have been proposed. However, most of the existing detection methods are for ground or air detection, rather than space-based detection, because the conditions of the space-based platform are very different from those of the ground and air platforms.

[0003] In recent years, a variety of space-based infrared target detection methods have been proposed. Due to the limitation of computing resources, these space-based detection methods are all based on the local contrast (LCM) method, and the single-frame detection method is applicable to infrared images generated by both staring and scanning modes. Zhao et al. proposed a single-frame spatial detection method to suppress complex backgrounds through multi-directional filter fusion. The method of Lv et al. is also a single-frame detection method, which can detect weak targets with a signal-to-clutter ratio (SCR) less than 1. However, the single-frame detection method has deficiencies in suppressing clutter, so the multi-frame detection method is the mainstream in the field of space-based detection. However, these methods are only applicable to the staring mode that produces a stable or slightly moving background. Deng et al. constructed a spatio-temporal local contrast filter (STLCF) and obtained good detection results. On the basis of the single-frame method, Zhao further proposed another spatio-temporal local contrast method based on space (STLCM), which has good clutter suppression ability. Lv et al. proposed a detection method based on TDLMS, neighborhood gray difference, and connected domain processing in 2018 and a spatio-temporal joint processing model (TS-RGL) in 2019, and both methods showed good detection performance.

[0004] However, the infrared images used in the above method have undergone preprocessing of contrast stretching and histogram equalization, which means that these images are not the original images from the space-based platform, but the secondary images from post-processing. In the space-based images that have only undergone preprocessing of background subtraction and non-uniformity correction, the intensity of clutter is much greater than that of the target. Existing methods have a good suppression effect on most backgrounds, but are prone to enhancing clutter and ignoring real targets, resulting in many missed detections and false alarms.

[0005] Through the above analysis, the problems and defects existing in the prior art are as follows: The prior art is prone to enhancing clutter and ignoring real targets, resulting in many missed detections and false alarms.

[0006] The difficulty in solving the above problems and defects is as follows: In summary, the space-based infrared air target detection method based on inter-frame matching difference and local spatial contrast proposed in the present invention has important significance in the field of space-based target detection.

[0007] The significance of solving the above problems and defects is as follows: Aiming at the characteristic that the clutter intensity in space-based infrared images is much greater than that of air targets, this method can effectively enhance dim targets while suppressing strong clutter, and finally realize the effective detection of space-based infrared air targets. Summary of the Invention

[0008] Aiming at the problems existing in the prior art, the present invention provides a space-based infrared air target detection method, a storage medium, and a computer device. In the first stage of the present invention, a reference frame I b and a reference frame I b+l are determined, and a local inter-frame matching difference model is constructed to traverse the two frames of images to obtain an intermediate result map, so as to realize the suppression of strong clutter and the enhancement of dim targets in the reference frame. In the second stage, the intermediate result is globally normalized to between [0, 255], and a spatial local contrast model is introduced to process the intermediate result to obtain a saliency map, so as to realize the re-enhancement of the target and the suppression of background residuals; finally, an adaptive segmentation threshold is set, and those with gray values higher than this threshold in the saliency map can be considered as air targets.

[0009] The present invention is implemented as follows. A space-based infrared air target detection method, the space-based infrared air target detection method comprising:

[0010] Step 1, determine a reference frame I b and a reference frame I b+1, perform local normalization on the B-frame, and perform spatial background suppression or inter-frame correlation difference. Perform local normalization on the (B + J)-th frame, and calculate the local correlation registration coefficient between the B-frame and the (B + J)-th frame to confirm the degree of correlation, and then perform local inter-frame difference between the B-frame and the (B + J)-th frame. Among them, local normalization can normalize the intensities of the target and clutter to the same interval to avoid the situation where the target is missed. Inter-frame difference is conducive to achieving preliminary suppression of the background and strong clutter.

[0011] Step 2, after completing the inter-frame correlation difference, perform dipole enhancement and spatial difference; perform operations on the data after completing spatial background suppression, dipole enhancement, and spatial difference to obtain an intermediate result. Among them, dipole enhancement can enhance the target when the target appears and otherwise suppress the background, and spatial difference can suppress non-uniformity and noise.

[0012] Step 3, perform global normalization processing, local feature contrast, and self-adaptive segmentation on the intermediate result data, and perform detection results. Local feature contrast can further achieve background suppression and target enhancement, while adaptive threshold segmentation can extract the real target from the background.

[0013] Furthermore, in the above Step 1, determine the reference frame I b and the reference frame I b+l , and the process of local normalization is as follows:

[0014] At the position (x, y) in I b , set a local slice I b (x, y) with a size 3 times that of the target according to the target size, denoted as R 11 , the local slice moves from left to right and from top to bottom. B represents the B-th frame of the image sequence, and the positions of the elements in R 11 are located in the neighborhood of (x, y) :

[0015]

[0016] where (i, j) is the position of the element in the local slice R 11 in I b , s is the target size radius, which is determined by the actual target size in the original image. When the actual target size is 3×3, s = 1; when the target size is 5×5, s = 2; when the target size is 7×7, s = 3;

[0017] Extract the 3×3 neighborhood at the position (x, y) in the reference frame I b+l from I b+l , and represent it with Ω local , and its definition is as follows:

[0018] Ω local={(p, q)|max(|p - x|, |y - q|) ≤ 1};

[0019] where (p, q) are the pixel coordinates in the reference frame I b+l in the

[0020] Extract the local slices centered on each pixel in the neighborhood Ω local respectively, denoted as R 2m , m = 1, 2, 3,..., 9, and the sizes of these 9 local slices are the same as R 11 ;

[0021] Normalize the gray values of the pixels in the above local slices to [0, 1] through the local normalization function respectively; Since R 11 and R 2m have the same dimension, the coordinates in both types of local slices are represented by the coordinate system (g, h), where g, h ∈ {1, 2,..., 2×s + 1}; The definition of the local normalization function is as follows:

[0022] R nor1 (g, h) = {[R1(g, h) - min(R 11 )] / [max(R 11 ) - min(R 11 )]};

[0023]

[0024] where R nor1 (g, h) represents the value of the pixel at (g, h) in R 11 after normalization, similarly.

[0025] Furthermore, the specific process of spatial background suppression in step 1 is as follows:

[0026] Spatial background suppression is achieved by locally subtracting the background to suppress the spatial background at the position (x, y);

[0027]

[0028] Define the local correlation function and calculate the correlation degree between R 11 and R 2m ; The definition of the correlation function is as follows:

[0029]

[0030] where |*| represents the absolute value operation, represents taking the average of the gray values of the local slice;

[0031] Obtain the matching coefficient r(x, y) of the reference frame and the reference frame at the position (x, y):

[0032] r(x, y) = max(r m ).

[0033] Furthermore, the specific process of the inter-frame correlation difference in the first step is as follows:

[0034] Perform local inter-frame difference to suppress the background and strong clutter:

[0035]

[0036]

[0037] where R dif is the difference slice obtained after local difference;

[0038] Suppress the non-uniformity stripes. The neighborhood where R 11 is located is divided into an internal region and an external region. In the neighborhood of R dif is also correspondingly divided into an internal region Ω and an external region Ω int and the external region Ω ext , and the relationship between the two regions is as follows:

[0039] Ω int = {(g, h)|max(|g - x|, |h - j|) ≤ s}, s = 1, 2, 3, 4;

[0040]

[0041]

[0042] where represents the empty set; due to the characteristic that the remaining non-uniformity stripes show small-range fluctuations in gray value, the suppression of the remaining non-uniformity stripes can be achieved through the following formula:

[0043] d dif2 (x, y) = max(R int ) - max(R ext );

[0044] where R int represents the matrix composed of Ω int pixels, and R ext is the matrix composed of Ω ext pixels.

[0045] Furthermore, the specific process of the dipole enhancement in the second step is as follows:

[0046] Extract the target dipole and enhance the target:

[0047] d dipole (x, y) = [max(R int ) - min(R int )] 2 ;

[0048] According to the above formula, if there is no target at the point (x, y), the background and strong clutter are suppressed by the above formula. When there is a target, due to the target motion characteristics, there are dipoles after difference, then the above formula is used to extract the dipoles and enhance the target;

[0049] Obtain the intermediate result value I med (x, y):

[0050] I med (x, y) = d dif1 (x, y) × [1 - r(x, y)] × d dif2 (x, y) × d dipole (x, y);

[0051] After traversing the entire image through the above formula, the intermediate result matrix I can be obtained med .

[0052] Furthermore, the specific process of the global normalization process in step three is as follows:

[0053] Introduce the spatial local contrast model for target re-enhancement and background residual suppression; considering that there is a non-linear amplification process in the SLC model, in order to maximize the gap between the target and the background, first globally normalize I med to [0, 255] to obtain the normalized matrix I med1 ; the global normalization formula is as follows:

[0054] I med1 (x, y) = 255 × {[I med (x, y) - min(I med )] / [max(I med ) - min(I med )]};

[0055] Furthermore, the specific process of the local feature contrast in step three is as follows:

[0056] In I med1 take any pixel point (x, y) as the center, extract the local slice R loc with the same dimension as R1, and extract the sub-slice R0 with the same dimension as R int ;

[0057] Subtract the background to generate the sub-slice R s centered on the pixel point (x, y), Rs The value R of each point in s (g, h) is obtained by the following formula, and then the value of the point (x, y) is represented by S tar (x, y) represents:

[0058]

[0059] where R0(g, h) represents the gray value at the point (g, h) in the local sub-slice R0;

[0060] Perform target energy accumulation to obtain the target energy accumulation value E(x, y) at the point (x, y):

[0061]

[0062] where s is the target size radius;

[0063] Calculate the value L of the local statistical value contrast factor of the center point (x, y) of the local slice con (x, y):

[0064] L con (x, y) = E(x, y) / std(R loc );

[0065] where std(R loc ) represents the standard deviation of the local slice R loc ;

[0066] Calculate the spatial local contrast value I of the center point (x, y) of the local slice space (x, y):

[0067] I space (x, y) = L con (x, y) × S tar (x, y);

[0068] Finally, traverse I med1 All pixels to obtain the final saliency map I space .

[0069] Furthermore, the specific process of the self-adaptive segmentation in the third step is as follows:

[0070] Calculate the adaptive segmentation threshold T for I according to the following formula space Perform binary segmentation to determine the target position:

[0071]

[0072] where k is the segmentation coefficient, and the empirical value is 20 - 30; when I spaceWhen the value of the element in it is greater than T, set it to 1; otherwise, set it to 0. The points set to 1 are the airborne target components.

[0073] Another object of the present invention is to provide a program storage medium for receiving user input. The stored computer program enables an electronic device to execute the space-based infrared airborne target detection method, which includes the following steps:

[0074] Step 1: Perform local normalization on the b-frame, and perform spatial background suppression or inter-frame correlation difference. Perform local normalization on the (b + J)-th frame, and calculate the local correlation registration coefficient between the b-frame and the (b + j)-th frame to confirm the correlation degree, and then perform local inter-frame difference between the b-frame and the (b + j)-th frame;

[0075] Step 2: After the inter-frame correlation difference is completed, perform dipole enhancement and spatial difference; perform operations on the data after spatial background suppression, dipole enhancement and spatial difference to obtain an intermediate result;

[0076] Step 3: Perform global normalization, local feature contrast, and self-adaptive segmentation on the intermediate result data, and perform detection results.

[0077] Another object of the present invention is to provide a computer device. The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the space-based infrared airborne target detection method.

[0078] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows: In view of the high false alarm rate and low detection rate of the existing infrared target detection method in space-based infrared target detection, the space-based infrared airborne target detection method proposed by the present invention is particularly suitable for suppressing the background and strong clutter in infrared images in the space-based staring mode, enhancing airborne targets, and finally realizing the effective detection of space-based infrared airborne targets. The present invention realizes the detection of space-based infrared airborne targets by constructing a local inter-frame matching difference model and a spatial local contrast model, with a simple structure, reducing the processing complexity of infrared airborne target detection and the resource requirements for hardware implementation, and effectively improving the efficiency of target detection. Description of the Drawings

[0079] Figure 1 is a flowchart of the space-based infrared airborne target detection method provided by an embodiment of the present invention.

[0080] Figure 2 is a schematic diagram of the space-based infrared airborne target detection process provided by an embodiment of the present invention.

[0081] Figure 3 is the local slice center point in the reference frame I provided by an embodiment of the present invention b in and the reference frame Ib+j Schematic diagram of the positional relationship of the reference neighborhood center point in

[0082] Figure 4 is the reference frame I provided by an embodiment of the present invention b+j Schematic diagram of the positional relationship of nine local slices at the position (x, y) in

[0083] Figure 5 is the input reference frame image and its three-dimensional view provided by an embodiment of the present invention;

[0084] Figure 5 Among them: Figure a is a grayscale image, the target is located in the lower left red frame, and the enlarged view of the target neighborhood is in the red frame; Figure b is the three-dimensional view of the reference frame, and the target is located in the red frame.

[0085] Figure 6 is the input reference frame image and its three-dimensional view provided by an embodiment of the present invention;

[0086] Figure 6 Among them: Figure a is a grayscale image, the target is located in the lower left red frame, and the enlarged view of the target neighborhood is in the red frame; Figure b is the three-dimensional view of the reference frame, and the target is located in the red frame.

[0087] Figure 7 is the intermediate result and its three-dimensional view provided by an embodiment of the present invention;

[0088] Figure 7 Among them: Figure a is a grayscale image, the target is located in the lower left red frame, and the enlarged view of the target neighborhood is in the red frame; Figure b is the three-dimensional view of the intermediate result, and the target is located in the red frame.

[0089] Figure 8 is the result after further enhancing the spatial local contrast and its three-dimensional view provided by an embodiment of the present invention;

[0090] Figure 8 Among them: Figure a is a grayscale image, the target is located in the lower left red frame, and the enlarged view of the target neighborhood is in the red frame; Figure b is the three-dimensional view of the result after further enhancement, and the target is located in the red frame.

[0091] Figure 9 is the obtained detection result and its three-dimensional view provided by an embodiment of the present invention;

[0092] Figure a is a grayscale image, the target is located in the lower left red frame, and the enlarged view of the target neighborhood is in the red frame; Figure b is the three-dimensional view of the detection result, and the target is located in the red frame.

[0093] Figure 10 is the schematic diagram of the comparison result provided by an embodiment of the present invention. Detailed implementation manners

[0094] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0095] In view of the problems existing in the prior art, the present invention provides a space-based infrared air target detection method, a storage medium and a computer device. The present invention will be described in detail below with reference to the accompanying drawings.

[0096] Ordinary technical personnel in the industry can also implement the space-based infrared air target detection method provided by the present invention using other steps. Figure 1 The space-based infrared air target detection method provided by the present invention is only a specific embodiment.

[0097] As Figure 1 shown, the space-based infrared air target detection method provided by the embodiment of the present invention includes:

[0098] S101: Determine the reference frame I b and the reference frame I b+l , perform local normalization on the b-frame, and perform spatial background suppression or inter-frame correlation difference. Perform local normalization on the (b + j)-th frame and perform inter-frame correlation difference;

[0099] After the inter-frame correlation difference is completed, perform dipole enhancement and spatial difference; perform operations on the data after spatial background suppression, dipole enhancement and spatial difference to obtain an intermediate result.

[0100] Perform global normalization on the intermediate result data, local feature contrast and self-adaptive segmentation, and perform detection results.

[0101] In S101 provided by the embodiment of the present invention, when determining the reference frame I b and the reference frame I b+l , the process of local normalization is as follows:

[0102] At the position (x, y) in I b , set a local slice I b (x, y) with a size three times that of the target according to the target size, denoted as R 11 . The local slice moves from left to right and from top to bottom. b represents that the reference frame is the b-th frame of the image sequence. The positions of the elements in R 11 are located in the neighborhood of (x, y) :

[0103]

[0104] where (i, j) is the local slice R 11The inner element is at I b The position in it, s is the target size radius, which is determined by the actual target size in the original image. For example, when the actual target size is 3×3, s = 1; when the target size is 5×5, s = 2; when the target size is 7×7, s = 3;

[0105] Extract I at the reference frame I b+l Extract I b+l The 3×3 neighborhood at (x, y), denoted by Ω 1ocal It is defined as follows:

[0106] Ω local = {(p, q)|max(|p - x|, |y - q|) ≤ 1};

[0107] Where (p, q) are the pixel coordinates in the reference frame I b+l

[0108] Extract the local slices centered on each pixel in the neighborhood Ω local respectively, denoted as R 2m , m = 1, 2, 3,..., 9, and the sizes of these 9 local slices are the same as R 11 ;

[0109] Normalize the gray values of the pixels in the above local slices to [0, 1] respectively through the local normalization function; Since R 11 and R 2m have the same dimension, the coordinates in both types of local slices are represented by the coordinate system (g, h), g, h ∈ {1, 2,..., 2×s + 1}; The definition of the local normalization function is as follows:

[0110] R nor1 (g, h) = {[R1(g, h) - min(R 11 )] / [max(R 11 ) - min(R 11 )]};

[0111]

[0112] Where R nor1 (g, h) represents the value of the pixel at (g, h) in R 11 after normalization, similarly.

[0113] The specific process of spatial background suppression in S101 provided by the embodiment of the present invention is as follows:

[0114] For spatial background suppression, local background subtraction is used to achieve spatial background suppression at the position (x, y);

[0115] ​

[0116] Define the local correlation function and calculate the correlation degree between R 11 and R 2m ; the correlation function is defined as follows:

[0117]

[0118] where |*| represents the absolute value operation, represents taking the average value of the local slice gray value;

[0119] Obtain the matching coefficient r(x, y) of the reference frame and the reference frame at the position (x, y):

[0120] r(x, y) = max(r m ).

[0121] The specific process of the inter-frame correlation difference in S101 provided by the embodiment of the present invention is as follows:

[0122] Perform local inter-frame difference to suppress the background and strong clutter:

[0123]

[0124]

[0125] where R dif is the difference slice obtained after local difference;

[0126] Suppress the non-uniformity stripes. The neighborhood where R 11 is located is divided into an internal area and an external area. In the neighborhood of R dif is also correspondingly divided into an internal area Ω and an external area Ω int and an external area Ω ext , and the relationship between the two areas is as follows:

[0127] Ω int = {(g, h)|max(|g - x|, |h - j|) ≤ s}, s = 1, 2, 3, 4;

[0128]

[0129]

[0130] where represents the empty set; due to the characteristic that the residual non-uniformity stripes show small-range fluctuations in gray value, the suppression of the residual non-uniformity stripes can be achieved through the following formula:

[0131] d dif2 (x, y) = max(R int)-max(R ext );

[0132] where R int represents a matrix composed of Ω int pixels, and R ext is a matrix composed of Ω ext pixels.

[0133] The specific process of dipole enhancement in S102 provided by the embodiment of the present invention is as follows:

[0134] Extract the target dipole and enhance the target:

[0135] d dipole (x, y) = [max(R int ) - min(R int )] 2 ;

[0136] According to the above formula, if there is no target at the point (x, y), the background and strong clutter are suppressed by the above formula. When there is a target, due to the target motion characteristics, there is a dipole after differentiation, then the dipole is extracted by the above formula and the target is enhanced;

[0137] Obtain the intermediate result value I med (x, y) at the pixel (x, y):

[0138] I med (x, y) = d dif1 (x, y) × [1 - r(x, y)] × d dif2 (x, y) × d dipole (x, y);

[0139] After traversing the entire image through the above formula, the intermediate result matrix I med can be obtained.

[0140] The specific process of global normalization processing in S103 provided by the embodiment of the present invention is as follows:

[0141] Introduce a spatial local contrast model for target re-enhancement and background residual suppression; considering that there is a non-linear amplification process in the SLC model, in order to maximize the gap between the target and the background, first globally normalize I med to [0, 255] to obtain the normalized matrix I med1 ; the global normalization formula is as follows:

[0142] I med1 (x, y) = 255 × {[I med (x, y) - min(I med )] / [max(I med ) - min(I med)]};

[0143] In S103 provided by the embodiments of the present invention, the specific process of local feature contrast is as follows:

[0144] In I med1 Taking any pixel point (x, y) as the center, extract a local slice R with the same dimension as R1 loc , and extract a sub-slice R0 with the same dimension as R int ;

[0145] Subtract the background to generate a sub-slice R with the pixel point (x, y) as the center s , R s The value R of each point in s (g, h) is obtained by the following formula, and the value of the point (x, y) is represented by S tar (x, y):

[0146]

[0147] Where R0(g, h) represents the gray value at the point (g, h) in the local sub-slice R0;

[0148] Perform target energy accumulation to obtain the target energy accumulation value f(x, y) at the point (x, y):

[0149]

[0150] Where s is the target size radius;

[0151] Calculate the value L of the local statistical value contrast factor of the center point (x, y) of the local slice con (x, y):

[0152] L con (x, y) = E(x, y) / std(R loc );

[0153] Where std(R loc ) represents the standard deviation of the local slice R loc ;

[0154] Calculate the spatial local contrast value I of the center point (x, y) of the local slice space (x, y):

[0155] I space (x, y) = L con (x, y) × S tar (x, y);

[0156] Finally, traverse all pixels of I med1 to obtain the final saliency map I space .

[0157] The specific process of self - adaptive segmentation in S103 provided by the embodiments of the present invention is as follows:

[0158] Calculate the self - adaptive segmentation threshold T for I according to the following formula space Perform binary segmentation to determine the target position:

[0159]

[0160] where k is the segmentation coefficient, and the empirical value is 20 - 30; when the value of the element in I space is greater than T, set it to 1, otherwise set it to 0. The points set to 1 are the components of the aerial target.

[0161] The technical solution of the present invention will be described in detail below in conjunction with specific embodiments.

[0162] The first stage includes the following 11 steps:

[0163] 1. At the position (x, y) in I b , set a local slice I b (x, y) with a size 3 times that of the target, denoted as R 11 . The local slice moves from left to right and from top to bottom. b represents that the reference frame is the b - th frame of the image sequence, as Figure 2 shown. The positions of all elements in R 11 are within the neighborhood of (x, y):

[0164]

[0165] where (i, j) is the position of the element in the local slice R 11 in I b . s is the radius of the target size, which is determined by the actual target size in the original image. For example, when the actual target size is 3×3, s = 1; when the target size is 5×5, s = 2; when the target size is 7×7, s = 3.

[0166] 2. Extract the 3×3 neighborhood at the position (x, y) in the reference frame I b+l in I b+l , denoted as Ω local , and its definition is as follows:

[0167] Ω local ={(p, q)|max(|p - x|, |y - q|)≤1};

[0168] where (p, q) are the pixel coordinates in the reference frame I b+l .

[0169] 3. Respectively, with the neighborhood Ωlocal Extract the local slices centered on each pixel inside, denoted as R 2m , where m = 1, 2, 3, ..., 9. The sizes of these 9 local slices are the same as that of R 11 , and the included neighborhoods are as shown in Figure 3 .

[0170] 4. Normalize the gray values of the pixels in the above local slices to [0, 1] respectively through the local normalization function. Since R 11 and R 2m have the same dimension, the coordinates in both types of local slices are represented by the coordinate system (g, h), where g, h ∈ {1, 2, ..., 2×s + 1}. The definition of the local normalization function is as follows:

[0171] R nor1 (g, h) = {[R1(g, h) - min(R 11 )] / [max(R 11 ) - min(R 11 )]};

[0172]

[0173] where R nor1 (g, h) represents the value of the pixel at (g, h) in R 11 after normalization, and similarly for others. Similarly.

[0174] 6. Spatial background suppression is achieved by locally subtracting the background for the spatial background suppression at the position (x, y);

[0175]

[0176] 5. Define the local correlation function to calculate the correlation degree between R 11 and R 2m . The definition of the correlation function is as follows:

[0177]

[0178] where |*| represents the absolute value operation, represents taking the mean value of the gray values of the local slice.

[0179] 7. Obtain the matching coefficient r(x, y) at the position (x, y) of the reference frame and the reference frame:

[0180] r(x, y) = max(r m ).

[0181] 8. Perform local inter-frame difference to suppress the background and strong clutter:

[0182]

[0183]

[0184] where R dif is the differential slice obtained after local differentiation.

[0185] 9. Suppress non-uniformity stripes. In Figure 2 the neighborhood where R 11 is located is divided into an internal region and an external region. In the neighborhood of R dif is also correspondingly divided into an internal region Ω and an external region Ω int and the external region Ω ext , the relationship between the two regions is as follows:

[0186] Ω int = {(g, h)|max(|g - x|, |h - j|) ≤ s}, s = 1, 2, 3, 4;

[0187]

[0188]

[0189] where represents the empty set. Since the residual non-uniformity stripes exhibit the characteristic of small-range fluctuations in gray value, the suppression of the residual non-uniformity stripes can be achieved through the following formula:

[0190] d dif2 (x, y) = max(R int ) - max(R ext );

[0191] where R int represents the matrix composed of Ω int pixels, and R ext is the matrix composed of Ω ext pixels.

[0192] 10. Extract the target dipole and enhance the target:

[0193] d dipole (x, y) = [max(R int ) - min(R int )] 2 ;

[0194] According to the above formula, if there is no target at the point (x, y), the background and strong clutter can be suppressed through the above formula. When there is a target, due to the target motion characteristics, there are dipoles after differentiation, then the dipoles can be extracted and the target can be enhanced using the above formula.

[0195] 11. Obtain the intermediate result value I at pixel (x, y). med (x, y):

[0196] I med (x, y) = d dif1 (x, y) × [1 - r(x, y)] × d dif2 (x, y) × d dipole (x, y);

[0197] After traversing the entire image through the above formula, the intermediate result matrix I can be obtained. med .

[0198] The second stage includes the following 7 steps:

[0199] 1) In the second stage, the Spatial Local Contrast (SLC) model is introduced for target re-enhancement and background residual suppression. Considering the non-linear amplification process in the SLC model, to maximize the gap between the target and the background, in the first step, I med is globally normalized to [0, 255] to obtain the normalized matrix I med1 . The global normalization formula is as follows:

[0200] I med1 (x, y) = 255 × {[I med (x, y) - min(I med )] / [max(I med ) - min(I med )]}.

[0201] 2) In I med1 , with any pixel point (x, y) as the center, extract a local slice R with the same dimension as R1 loc , and extract a sub-slice R0 with the same dimension as R int .

[0202] 3) Subtract the background to generate a sub-slice R centered on the pixel point (x, y). s , and the value of each point R s in R s (g, h) is obtained by the following formula, and at this time, the value of the point (x, y) is represented by S tar (x, y):

[0203]

[0204] where R0(g, h) represents the gray value at the point (g, h) within the local sub-slice R0.

[0205] 4) Perform target energy accumulation to obtain the target energy accumulation value E(x, y) at the point (x, y):

[0206]

[0207] where s is the target size radius.

[0208] 5) Calculate the value L of the local statistical value contrast factor of the local slice center point (x, y) con (x, y):

[0209] L con (x, y) = f(x, y) / std(R loc );

[0210] where std(R loc ) represents the standard deviation of the local slice R loc .

[0211] 6) Calculate the spatial local contrast value I of the local slice center point (x, y) space (x, y):

[0212] I space (x, y) = L con (x, y) × S tar (x, y);

[0213] Finally, traverse I med1 All pixels to obtain the final saliency map I space .

[0214] 7) Calculate the adaptive segmentation threshold T for I according to the following formula space Perform binary segmentation to determine the target position:

[0215]

[0216] where k is the segmentation coefficient, and the empirical value is 20 - 30. When the value of the element in I space is greater than T, set it to 1, otherwise set it to 0. The points set to 1 are the airborne target components.

[0217] The technical solution of the present invention will be described in detail below in combination with simulation experiments.

[0218] Simulation environment: Matlab2020b;

[0219] Test input: Space-based medium-wave image sequence, size 200×256, background is sea-land background, target is an airplane, and the target size is 7×7.

[0220] The technical effect of the present invention will be described in detail below in combination with experiments.

[0221] Table 1 Test data

[0222]

[0223] Table 2 Comparison method

[0224]

[0225]

[0226] Figure 10 Comparison result. The target position, the magnified details of which are located at the lower left corner of the image.

[0227] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated designed hardware. Those of ordinary skill in the art can understand that the above devices and methods can be implemented using computer-executable instructions and / or included in processor control code, such as provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits of programmable hardware devices such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above hardware circuits and software such as firmware.

[0228] The above are only the specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention shall be covered by the protection scope of the present invention.

Claims

1. A space-based infrared air target detection method, characterized in that, The above-mentioned space-based infrared air target detection method includes: Step 1, determine the reference frame I b and the reference frame I b+l , perform local normalization on the b-frame, and perform spatial background suppression or inter-frame correlation difference. Perform local normalization on the (b + j)-th frame, calculate the local correlation registration coefficient between the b-frame and the (b + j)-th frame, confirm the correlation degree, and then perform local inter-frame difference between the b-frame and the (b + j)-th frame; After the inter-frame correlation difference is completed in Step 2, dipole enhancement and spatial difference are performed; the data after spatial background suppression, dipole enhancement, and spatial difference are operated to obtain an intermediate result; In Step 3, global normalization processing, local feature contrast, and self-adaptive segmentation are performed on the intermediate result data, and the detection result is obtained.

2. The space-based infrared air target detection method according to claim 1, characterized in that, Determine the reference frame I in the first step b and the reference frame I b+l , and the local normalization process is as follows: At I b At the position (x, y) in it, set a local slice I with a size 3 times the target according to the target size b (x, y), denoted as R11, the local slice moves from left to right and from top to bottom, b represents the b-th frame of the image sequence, and the positions of the elements in R11 are in the neighborhood of (x, y) Inside: where (i, j) is the position of the element in the local slice R11 within I b where s is the radius of the target size, which is determined by the actual target size in the original image. When the actual target size is 3×3, s = 1; when the target size is 5×5, s = 2; when the target size is 7×7, s = 3; In reference frame I b+l Extract I b+l The 3×3 neighborhood at (x, y), denoted as Ωlocal, is defined as follows: Ω local = {(p, q) | max(|p - x|, |y - q|) ≤ 1}; where (p, q) are the pixel coordinates in reference frame I b+l in; Taking each pixel in the neighborhood Ω local as the center respectively to extract its local slice, denoted as R 2m , where m = 1, 2, 3,..., 9, and the sizes of these 9 local slices are the same as that of R11; The gray values of the above-mentioned local slice pixels are normalized to [0, 1] respectively through the local normalization function; since the dimensions of R11 and R2m are the same, the coordinates in both types of local slices are represented by the coordinate system (g, h), where g, h ∈ {1, 2,..., 2×s + 1}; the definition of the local normalization function is as follows: R nor1 (g, h) = {[R1(g, h) - min(R 11 )] / [max(R 11 ) - min(R 11 )]}; where R nor1 (g, h) represents the value of the pixel at (g, h) in R 11 after normalization, Similarly.

3. The space-based infrared air target detection method according to claim 1, characterized in that, The specific process of spatial background suppression in Step 1 is as follows: Spatial background suppression realizes the spatial background suppression at the position (x, y) through local background subtraction; Define the local correlation function and calculate the correlation degree between R 11 and R 2m ; the correlation function is defined as follows: where |*| represents the absolute value operation, represents taking the average of the grayscale values of the local slice; Obtain the matching coefficient r(x, y) at the position (x, y) of the reference frame and the reference frame: r(x, y) = max(r m ).

4. The space-based infrared air target detection method according to claim 1, characterized in that, The specific process of inter-frame correlation difference in Step 1 is as follows: Perform local inter-frame difference to suppress the background and strong clutter: where R dif is a differential slice obtained after local differentiation; Suppress non-uniformity stripes, R 11 The neighborhood where it is located is divided into an inner region and an outer region. At R dif 's neighborhood is also correspondingly divided into an inner region Ω int and an outer region Ω ext , and the relationship between the two regions is as follows: Ω int = {(g, h) | max(|g - x|, |h - j|) ≤ s}, s = 1, 2, 3, 4; Among them represents the empty set; due to the characteristic that the residual non-uniformity stripes show small-range fluctuations in gray value, the suppression of the residual non-uniformity stripes is achieved through the following formula: d dif2 (x, y) = max(R int ) - max(R ext ); where R int represents a matrix composed of Ω int pixels, and R ext is a matrix composed of Ω ext pixels.

5. The space-based infrared air target detection method according to claim 1, wherein, The specific process of dipole enhancement in Step 2 is as follows: Extract the target dipole to enhance the target: d dipole (x, y) = [max(R int ) - min(R int )] 2 ; According to the above formula, if there is no target at the point (x, y), the background and strong clutter are suppressed through the above formula. When there is a target, due to the target motion characteristics, there is a dipole after the difference, and then the dipole is extracted and the target is enhanced using the above formula; Obtain the intermediate result value I at pixel (x, y) med (x, y): I med (x, y) = d dif1 (x, y) × [1 - r(x, y)] × d dif2 (x, y) × d dipole (x, y); After traversing the entire image through the above formula, the intermediate result matrix can be obtained Imed .

6. The space-based infrared air target detection method according to claim 1, wherein, The specific process of global normalization processing in Step 3 is as follows: Introduce a spatial local contrast model for target re-enhancement and background residual suppression; considering the non-linear amplification process in the SLC model, in order to maximize the gap between the target and the background, first normalize I med globally to [0, 255] to obtain the normalized matrix I med1 ; The global normalization formula is as follows: I med1 (x, y) = 255 × {[I med (x, y) - min(I med )] / [max(I med ) - min(I med )]}。 7. The space-based infrared air target detection method according to claim 1, wherein, The specific process of local feature contrast in Step 3 is as follows: In I med1 Taking any pixel point (x, y) as the center, a local slice R with the same dimension as R1 is extracted loc , and a sub-slice R0 with the same dimension as R int is extracted; Subtract the background to generate a sub-slice R centered on the pixel point (x, y) s , R s The values of each point in R s (g, h) are obtained by the following formula, and the value of the point (x, y) is represented by S tar (x, y): Where R0(g, h) represents the gray value at the point (g, h) in the local sub-slice R0; Perform target energy accumulation to obtain the target energy accumulation value E(x, y) at the point (x, y): Where s is the target size radius; Calculate the value L of the local statistical value contrast factor at the center point (x, y) of the local slice con (x, y): L con (x, y) = E(x, y) / std(R loc ); where std(R loc ) represents the standard deviation of the local slice R loc ; Calculate the spatial local contrast value I of the center point (x, y) of the local slice space (x, y): I space (x, y) = L con (x, y) × S tar (x, y); Final traversal I med1 All pixels obtain the final saliency map I space .

8. The space-based infrared air target detection method according to claim 1, wherein, The specific process of self-adaptive segmentation in Step 3 is as follows: Calculate the adaptive segmentation threshold T for I according to the following formula space Perform binary segmentation to determine the target position: where k is the segmentation coefficient, and the empirical value is taken as 20 - 30; when the value of the element in I space is greater than T, set it to 1, otherwise set it to 0, and the points set to 1 are the airborne target components.

9. A computer device, wherein, The computer device includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the space-based infrared air target detection method according to any one of claims 1 to 8.

Images