Satellite-borne remote sensing infrared image small target detection method based on step contrast and step gradient fusion

Abstract

A kind of small target detection method of satellite-borne remote sensing infrared image based on step contrast and step gradient fusion belongs to small target detection technical field.The present application is aimed at the problems of low detection rate and high false alarm rate of existing single-frame satellite-borne remote sensing infrared image small target detection.The present application comprises: designing a square step gradient detection model, using the step gradient detection model to traverse each pixel point in the satellite-borne remote sensing infrared image as the first layer center pixel point, performing pixel gray step gradient detection of each pixel point, and obtaining a pixel gray step gradient image of small targets; designing a convolution template with the same size as the step gradient detection model and having a corresponding layer, using the convolution template to traverse each pixel point in the satellite-borne remote sensing infrared image as the first layer center pixel point, performing step contrast detection of each pixel point, and obtaining a pixel gray step contrast image; and performing fusion to obtain a small target region map.The present application is used for small target detection of infrared images.

Description

Technical Field

[0001] This invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion, belonging to the field of small target detection technology. Background Technology

[0002] Humans use their eyes to capture images and perceive their environment through their five senses; in this perception process, the use of the eyes accounts for more than 80%. For high-altitude satellites, optical sensors are the satellite's eyes, playing a crucial role.

[0003] Currently, satellite technology is developing rapidly, with the number of satellites launched gradually increasing and the launch frequency also rising. Besides the number and frequency of launches, the types of satellites launched are also diverse, with remote sensing satellites making up a large portion. Remote sensing satellites are space-based Earth observation devices that use onboard remote sensing equipment to collect information on land, hydrology, atmosphere, and other targets across various wavelengths. Depending on the sensors carried by the remote sensing satellite, the obtained remote sensing images also differ, with infrared images being an important component of remote sensing images.

[0004] Spaceborne infrared remote sensing images lie outside the visible light band, possessing a longer wavelength. They can detect the thermal radiation of objects, rather than visible light reflection. This allows spaceborne infrared remote sensing images to provide information different from visible light images. Furthermore, spaceborne infrared remote sensing images can sense the thermal radiation of objects, thus revealing their temperature distribution. Objects at different temperatures will appear with different brightness or hues in infrared images, which is very useful for thermal zone analysis and heat source detection. Moreover, infrared radiation can penetrate atmospheric interference such as clouds and haze relatively well, making spaceborne infrared remote sensing images relatively strong in observational capabilities under adverse weather conditions. For these reasons, spaceborne infrared remote sensing images have unique advantages in the field of target detection and are widely used.

[0005] Traditional single-frame satellite-borne remote sensing infrared image small target detection algorithms have the advantages of low computational complexity and high real-time performance, but they also suffer from drawbacks such as insufficient detection rate and high false alarm rate. Therefore, improving the performance of traditional single-frame detection algorithms is of great research significance. Summary of the Invention

[0006] To address the issues of low detection rate and high false alarm rate in existing single-frame spaceborne remote sensing infrared image small target detection methods, this invention provides a small target detection method for spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0007] This invention provides a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion, comprising:

[0008] Small target detection based on stepped gradient in spaceborne remote sensing infrared images: A square stepped gradient detection model is designed, consisting of three layers expanding outward from the center. The first layer corresponds to the center of small targets in the spaceborne remote sensing infrared image, the second layer corresponds to the edges of small targets, and the third layer corresponds to the background of small targets. The stepped gradient detection model iterates through each pixel in the spaceborne remote sensing infrared image, using each pixel as the center pixel of the first layer, and performs pixel grayscale stepped gradient detection for each pixel. The stepped gradient detection model is selected so that the four corners point towards the center of the first layer. The direction of each pixel is used as one of the four gradient directions. The gray-level gradient values ​​of adjacent layers in the spaceborne remote sensing infrared image are calculated in each gradient direction. Then, the mean square sum of the gray-level gradient values ​​in each gradient direction is calculated, and the mean square value of the gray-level values ​​in each gradient direction is also calculated. The maximum and minimum values ​​of the mean square values ​​of the gray-level values ​​are determined. The step gradient of the current center pixel is calculated by combining the ratio of the maximum and minimum values ​​of the mean square values ​​of the gray-level values ​​and the mean square values ​​of the four pixels. The gray-level step gradient map of the pixels representing small targets in the spaceborne remote sensing infrared image is obtained from the calculation results of all step gradients.

[0009] Small target detection in spaceborne remote sensing infrared images based on stepped contrast: A convolutional template with the same size and corresponding layers as the stepped gradient detection model is designed. Each pixel in the spaceborne remote sensing infrared image is traversed using the convolutional template as the center pixel of the first layer, and stepped contrast detection is performed on each pixel. Convolution operation is performed on the spaceborne remote sensing infrared image to obtain the maximum pixel grayscale value of the second and third layers. The stepped contrast of the current center pixel is calculated by combining the pixel grayscale value of the center pixel of the first layer. The pixel grayscale stepped contrast map representing small targets in the spaceborne remote sensing infrared image is obtained from the calculation results of all stepped contrasts.

[0010] The pixel grayscale gradient map and the pixel grayscale contrast map are fused to determine the small target region map in the spaceborne remote sensing infrared image.

[0011] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0012] The method for calculating the pixel grayscale gradient value of adjacent layers in each gradient direction is as follows:

[0013] g1_1 = I0 - I1,

[0014] g1_2 = I1 - I5,

[0015] g2_1 = I0 - I2,

[0016] g2_2 = I2 - I6,

[0017] g3_1 = I0 - I3,

[0018] g3_2 = I3 - I7,

[0019] g4_1 = I0 - I4,

[0020] g4_2 = I4 - I8,

[0021] In the formula, g1_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I1 of the second layer along the first gradient direction ψ1; g1_2 is the pixel grayscale gradient value of the corresponding pixel I1 of the second layer and the corresponding pixel I5 of the third layer along the first gradient direction ψ1; g2_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I2 of the second layer along the second gradient direction ψ2; g2_2 is the pixel grayscale gradient value of the corresponding pixel I2 of the second layer and the corresponding pixel I6 of the third layer along the second gradient direction ψ2. The values ​​are as follows: g3_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I3 of the second layer on the third gradient direction ψ3; g3_2 is the pixel grayscale gradient value of the corresponding pixel I3 of the second layer and the corresponding pixel I7 of the third layer on the third gradient direction ψ3; g4_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I4 of the second layer on the fourth gradient direction ψ4; g4_2 is the pixel grayscale gradient value of the corresponding pixel I4 of the second layer and the corresponding pixel I8 of the third layer on the fourth gradient direction ψ4.

[0022] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0023] The method for calculating the mean square value of pixel grayscale in each gradient direction is as follows:

[0024]

[0025] In the formula G k For ψ k Mean square value of pixel grayscale in the direction;

[0026] The method for determining the maximum and minimum values ​​of the mean square of pixel grayscale is as follows:

[0027] G max =maxG k ,

[0028] G min =minG k ,

[0029] In the formula G max G represents the maximum value of the mean square value of pixel grayscale.min It represents the minimum value of the mean square value of pixel grayscale.

[0030] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0031] The current step gradient of the center pixel is:

[0032]

[0033] In the formula, G0 is the step gradient of the current center pixel, and η is the step gradient threshold value; the pixel grayscale step gradient map G representing the small target in the spaceborne remote sensing infrared image is formed by the step gradients G0 of all pixels.

[0034] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0035] In the process of small target detection based on stepped contrast in spaceborne remote sensing infrared images, the method for calculating the maximum pixel grayscale value of the second and third layers is as follows:

[0036]

[0037] In the formula, Im0 is the maximum pixel grayscale value of the second layer, I in To input spaceborne remote sensing infrared images, S loops The second layer is the convolutional template; In0 is the maximum pixel grayscale value of the third layer, S loopt This is the third layer of the convolution template.

[0038] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0039] The method for calculating the stepped contrast of the current center pixel is as follows:

[0040] HTCM=H[(I0-Im0)]·H[(I0-In0)],

[0041] In the formula, HTCM is the step contrast of the center pixel, and H[x] is the step function:

[0042] x = I0 - Im0 or x = I0 - In0;

[0043]

[0044] Let I0 - Im0 = ΔIm0, I0 - In0 = ΔIn0, then:

[0045] HTCM = H[ΔIm0]·H[ΔIn0];

[0046] The grayscale gradient contrast map I, representing small targets in a spaceborne remote sensing infrared image, is formed by the gradient contrast of all pixels (HTCM).

[0047] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0048] The pixel grayscale gradient map G and the pixel grayscale contrast map I are fused to obtain the small target region map M in the spaceborne remote sensing infrared image:

[0049] M = I × G.

[0050] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0051] A four-scale stepped gradient detection model and convolutional templates of the four scales are used to detect pixel grayscale stepped gradient maps and pixel grayscale stepped contrast maps of small targets in spaceborne remote sensing infrared images, respectively. The first layer size corresponding to the four scales is 3×3, 5×5, 7×7, and 9×9, respectively. The second layer size is the area range of one pixel outside the first layer, and the third layer size is the area range of two pixels outside the second layer.

[0052] The fusion results obtained from detection at four different scales are calculated to obtain four small target region maps; the maximum pixel gray value of each pixel in the four small target region maps is selected to form the final small target region map.

[0053] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0054] The pixel grayscale gradient maps of small targets detected by the first-layer stepped gradient detection models with sizes of 3×3, 5×5, 7×7, and 9×9 are represented as G1, G2, G3, and G4, respectively:

[0055] G1 = f grad_1 (I in ),

[0056] G2 = f grad_2 (I in ),

[0057] G3 = f grad_3 (I in ),

[0058] G4 = f grad_4 (I in ),

[0059] In the formula f grad_1For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 3×3, f grad_2 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 5×5, f grad_3 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 7×7, f grad_4 Pixel grayscale step gradient detection for the first layer step gradient detection model with a size of 9×9;

[0060] The pixel grayscale contrast maps of small targets detected by the first-layer convolutional templates with sizes of 3×3, 5×5, 7×7, and 9×9 are represented as I1, I2, I3, and I4, respectively:

[0061] I1=f htcm_1 (I in ),

[0062] I2=f htcm_2 (I in ),

[0063] I3=f htcm_3 (I in ),

[0064] I4=f htcm_4 (I in )

[0065] In the formula f htcm_1 For pixel grayscale contrast detection of the first layer's 3×3 scale convolutional template, f htcm_2 For pixel grayscale contrast detection of the first layer's 5×5 scale convolutional template, f htcm_3 For pixel grayscale step contrast detection of the first layer convolutional template with a size of 7×7, f htcm_4 Pixel grayscale contrast detection of the first layer convolutional template with a size of 9×9.

[0066] The present invention relates to a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion.

[0067] Calculate the small target region maps obtained under four different scales of detection:

[0068] M1 = I1 × G1,

[0069] M2 = I2 × G2,

[0070] M3 = I3 × G3,

[0071] M4 = I4 × G4,

[0072] The final small target area map M is then:

[0073] M = max(M1, M2, M3, M4).

[0074] The beneficial effects of this invention are as follows: This invention's method, targeting the radiation characteristics of small targets in spaceborne remote sensing infrared images, designs a stepped detection window. The small target is divided into three layers from the inside out, with the central portion as the first layer and the third layer extending outwards. The first layer is considered the center of the small target, the second layer as the gradually attenuating edge, and the third layer as the surrounding background. This serves as the detection framework for designing gradient characteristics. To simplify gradient data calculation, four directions are selected as gradient characteristics, and the gradients between layers in adjacent detection windows are calculated. This yields the four-directional gradient values ​​of the small target in the spaceborne remote sensing infrared image.

[0075] The method of this invention can improve the detection effect of small targets in spaceborne remote sensing infrared images while reducing the computational load, thereby increasing the detection speed.

[0076] This method improves upon traditional single-frame detection algorithms by extracting gradient characteristics in a single gradient calculation. This increases the detection rate and reduces the false alarm rate. Attached Figure Description

[0077] Figure 1 This is a gradient distribution model diagram of small targets in infrared images;

[0078] Figure 2 This is a diagram of the stepped gradient detection model;

[0079] Figure 3 This is a schematic diagram of a convolution template;

[0080] Figure 4 This is a schematic diagram illustrating the fusion process of the pixel grayscale gradient map and the pixel grayscale contrast map.

[0081] Figure 5 This is a schematic diagram of the first layer of the detection template, which has a size of 3×3.

[0082] Figure 6 This is a schematic diagram of the first layer of the inspection template, which has a size of 5×5.

[0083] Figure 7 This is a schematic diagram of the first layer of the inspection template, which has a size of 7×7.

[0084] Figure 8 This is a schematic diagram of the first layer of the inspection template, which has a size of 9×9.

[0085] Figure 9This is a comparison diagram between the original spaceborne remote sensing infrared image of the sea surface in the verification experiment and the detection results of the method of this invention;

[0086] Figure 10 This is a comparison between the original grayscale 3D image of the original spaceborne remote sensing infrared image of the sea surface and the grayscale 3D image detected by the method of this invention; in the figure, Col is the vertical coordinate of the image pixel, row is the horizontal coordinate of the image pixel, and value is the pixel value of the pixel.

[0087] Figure 11 This is a comparison diagram between the original satellite-borne remote sensing infrared image of the cloud layer and the detection results of the method of this invention;

[0088] Figure 12 This is a comparison image of the original grayscale 3D image of the original spaceborne remote sensing infrared image of the cloud layer and the grayscale 3D image detected by the method of this invention.

[0089] Figure 13 This is a comparison diagram of the original satellite-borne remote sensing infrared image of the cloud layer and the detection results of the method of this invention;

[0090] Figure 14 This is a comparison image of the original grayscale 3D image of the original spaceborne remote sensing infrared image of the cloud layer and the grayscale 3D image detected by the method of this invention.

[0091] Figure 15 This is a multi-scale processing flowchart of the small target detection method for spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion, as described in this invention. Detailed Implementation

[0092] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0093] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0094] The present invention will be further described below with reference to the accompanying drawings, but this should not be construed as limiting the invention.

[0095] Combination Figures 1 to 4 As shown, this invention provides a method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion, including:

[0096] Small target detection based on stepped gradient in spaceborne remote sensing infrared images: A square stepped gradient detection model is designed, consisting of three layers expanding outward from the center. The first layer corresponds to the center of small targets in the spaceborne remote sensing infrared image, the second layer corresponds to the edges of small targets, and the third layer corresponds to the background of small targets. The stepped gradient detection model iterates through each pixel in the spaceborne remote sensing infrared image, using each pixel as the center pixel of the first layer, and performs pixel grayscale stepped gradient detection for each pixel. The stepped gradient detection model is selected so that the four corners point towards the center of the first layer. The direction of each pixel is used as one of the four gradient directions. The gray-level gradient values ​​of adjacent layers in the spaceborne remote sensing infrared image are calculated in each gradient direction. Then, the mean square sum of the gray-level gradient values ​​in each gradient direction is calculated, and the mean square value of the gray-level values ​​in each gradient direction is also calculated. The maximum and minimum values ​​of the mean square values ​​of the gray-level values ​​are determined. The step gradient of the current center pixel is calculated by combining the ratio of the maximum and minimum values ​​of the mean square values ​​of the gray-level values ​​and the mean square values ​​of the four pixels. The gray-level step gradient map of the pixels representing small targets in the spaceborne remote sensing infrared image is obtained from the calculation results of all step gradients.

[0097] Small target detection in spaceborne remote sensing infrared images based on stepped contrast: A convolutional template with the same size and corresponding layers as the stepped gradient detection model is designed. Each pixel in the spaceborne remote sensing infrared image is traversed using the convolutional template as the center pixel of the first layer, and stepped contrast detection is performed on each pixel. Convolution operation is performed on the spaceborne remote sensing infrared image to obtain the maximum pixel grayscale value of the second and third layers. The stepped contrast of the current center pixel is calculated by combining the pixel grayscale value of the center pixel of the first layer. The pixel grayscale stepped contrast map representing small targets in the spaceborne remote sensing infrared image is obtained from the calculation results of all stepped contrasts.

[0098] The pixel grayscale gradient map and the pixel grayscale contrast map are fused to determine the small target region map in the spaceborne remote sensing infrared image.

[0099] Small targets in infrared images have isotropic Gaussian shapes, so their gradients are uniformly distributed in all directions. The gradient direction of detected targets points towards the center of the target, while the gradient direction of other high-brightness objects, such as clouds, points inward along the edges of the object, lacking the gradient characteristic of small targets concentrating towards the center. To address this gradient characteristic, this implementation method, based on the original local gradient detection method, uses a new detection window and redesigns the gradient characteristic detection method. The aim is to improve the detection effect of small targets in spaceborne remote sensing infrared images while reducing the computational load and thus increasing the detection speed.

[0100] In order to simplify the calculation of gradient data, this embodiment will... Figure 2The four directions indicated by the red arrows are used as gradient characteristics to calculate the gradient between layers in adjacent detection windows.

[0101] Furthermore, combined with Figure 2 As shown, the method for calculating the pixel grayscale gradient value of adjacent layers in each gradient direction is as follows:

[0102] g1_1 = I0 - I1,

[0103] g1_2 = I1 - I5,

[0104] g2_1 = I0 - I2,

[0105] g2_2 = I2 - I6,

[0106] g3_1 = I0 - I3,

[0107] g3_2 = I3 - I7,

[0108] g4_1 = I0 - I4,

[0109] g4_2 = I4 - I8,

[0110] In the formula, g1_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I1 of the second layer along the first gradient direction ψ1; g1_2 is the pixel grayscale gradient value of the corresponding pixel I1 of the second layer and the corresponding pixel I5 of the third layer along the first gradient direction ψ1; g2_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I2 of the second layer along the second gradient direction ψ2; g2_2 is the pixel grayscale gradient value of the corresponding pixel I2 of the second layer and the corresponding pixel I6 of the third layer along the second gradient direction ψ2. The values ​​are as follows: g3_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I3 of the second layer on the third gradient direction ψ3; g3_2 is the pixel grayscale gradient value of the corresponding pixel I3 of the second layer and the corresponding pixel I7 of the third layer on the third gradient direction ψ3; g4_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I4 of the second layer on the fourth gradient direction ψ4; g4_2 is the pixel grayscale gradient value of the corresponding pixel I4 of the second layer and the corresponding pixel I8 of the third layer on the fourth gradient direction ψ4.

[0111] Then, the mean square sum of the two gradient values ​​in the four directions is calculated, followed by the mean of the mean square of the gradient in each direction.

[0112] The method for calculating the mean square value of pixel grayscale in each gradient direction is as follows:

[0113]

[0114] In the formula G k For ψ k Mean square value of pixel grayscale in the direction;

[0115] After calculating the mean of the gradient mean square for each direction, calculate the maximum and minimum values ​​of the four values.

[0116] The method for determining the maximum and minimum values ​​of the mean square of pixel grayscale is as follows:

[0117] G max =maxG k ,

[0118] G min =minG k ,

[0119] In the formula G max G represents the maximum value of the mean square value of pixel grayscale. min It represents the minimum value of the mean square value of pixel grayscale.

[0120] After performing gradient calculations on small targets in spaceborne remote sensing infrared images, a threshold η is set. Since small targets in infrared images have isotropic Gaussian shapes, their gradient directions are uniformly distributed in all directions. However, background clutter is typically localized, while small targets in spaceborne remote sensing infrared images are centered, and the differences in step gradient values ​​across the four directions are small. Therefore, by utilizing G... min and G max The ratio can suppress localized clutter while appropriately enhancing small targets in spaceborne remote sensing infrared images. To achieve better background clutter suppression, a value lower than G will be used. min and G max The ratio of pixels is assigned to 0. To better enhance the effect of small targets in spaceborne remote sensing infrared images, the ratio is higher than G. min and G max The ratio of pixels is the sum of the mean square values ​​of the four-way gradient.

[0121] The current step gradient of the center pixel is:

[0122]

[0123] In the formula, G0 is the step gradient of the current center pixel, and η is the step gradient threshold value; the pixel grayscale step gradient map G representing the small target in the spaceborne remote sensing infrared image is formed by the step gradients G0 of all pixels.

[0124] Furthermore, input image I in At that time, a convolution template with a ring structure is set; combined with Figure 3As shown, in the process of small target detection based on stepped contrast in spaceborne remote sensing infrared images, the calculation method for the maximum pixel grayscale value of the second and third layers is as follows:

[0125]

[0126] In the formula, Im0 is the maximum pixel grayscale value of the second layer, I in To input spaceborne remote sensing infrared images, S loops The second layer is the convolutional template; In0 is the maximum pixel grayscale value of the third layer, S loopt This is the third layer of the convolution template.

[0127] The maximum pixel grayscale value in the second layer of the current pixel's loop region is represented as Im0, and In0 represents the highest pixel grayscale value in the third layer of the loop region; S loops and S loopt These refer to the second and third layers of the loop structure template, respectively. loops The size selection is related to the target size; it is usually chosen to be equal to or greater than the size of the maximum target. loopt The size is S loops +2

[0128] In this embodiment, the step contrast of the current center pixel is calculated as follows:

[0129] HTCM=H[(I0-Im0)]·H[(I0-In0)],

[0130] In the formula, HTCM is the step contrast of the center pixel, and H[x] is the step function:

[0131] x = I0 - Im0 or x = I0 - In0;

[0132]

[0133] Let I0 - Im0 = ΔIm0, I0 - In0 = ΔIn0, then:

[0134] HTCM = H[ΔIm0]·H[ΔIn0];

[0135] The grayscale gradient contrast map I, representing small targets in a spaceborne remote sensing infrared image, is formed by the gradient contrast of all pixels (HTCM).

[0136] ΔIm0 and ΔIn0 are considered as the differences between the second, third, and center pixels, reflecting the size relationship between neighboring pixels and the center pixel. The principle behind this is based on the radiation characteristics of small infrared targets: their grayscale values ​​gradually decrease from the center outwards, with the highest grayscale value at the center. Therefore, within the current detection area, if the grayscale value of the center pixel exceeds the maximum value of its second and third-level neighbors, it must also exceed the values ​​of all other pixels in those neighbors. This method aligns with the radiation characteristics of small infrared targets, and unlike calculating the grayscale difference in all directions for each level of neighborhood, it only requires a single calculation using the maximum value within the annular region.

[0137] Combination Figure 4 As shown, different detection methods target different characteristics of small targets in spaceborne remote sensing infrared images. In order to make full use of multiple characteristics for detection and improve detection performance, this embodiment integrates the step gradient detection method and the step contrast detection method for small targets in spaceborne remote sensing infrared images.

[0138] The pixel grayscale gradient map G and the pixel grayscale contrast map I are fused to obtain the small target region map M in the spaceborne remote sensing infrared image:

[0139] M = I × G.

[0140] The small target region map M not only enhances the target using two methods, but also further suppresses the background after the two methods are fused, thus improving the detection effect.

[0141] Furthermore, besides targets of 3×3 and 5×5 sizes, the size of the target may be uncertain in actual spaceborne remote sensing infrared image small target detection. If a spaceborne remote sensing infrared image small target detection method with known target size is used, the detection effect will be affected, leading to a decrease in detection performance. Therefore, this implementation incorporates multi-scale detection to improve detection performance when the target size is unknown.

[0142] Multiscale detection is a method for processing images of targets of unknown size using multiple detection templates of different sizes.

[0143] Combination Figures 5 to 8 and Figure 15 As shown, a four-scale stepped gradient detection model and convolutional templates of the four scales are used to detect the pixel grayscale stepped gradient map and pixel grayscale stepped contrast map of small targets in spaceborne remote sensing infrared images, respectively; wherein the first layer size corresponding to the four scales is 3×3, 5×5, 7×7, and 9×9, the second layer size is the area range of one pixel outside the first layer, and the third layer size is the area range of two pixels outside the second layer;

[0144] The fusion results obtained from detection at four different scales are calculated to obtain four small target region maps; the maximum pixel gray value of each pixel in the four small target region maps is selected to form the final small target region map.

[0145] This implementation method is designed for detecting small targets in spaceborne remote sensing infrared images with dimensions of 3×3, 5×5, 7×7, and 9×9. Typically, a small target is defined as one smaller than 9×9 pixels.

[0146] Figures 5 to 8 In the middle, the first white area represents the size of the small target.

[0147] The pixel grayscale gradient maps of small targets detected by the first-layer stepped gradient detection models with sizes of 3×3, 5×5, 7×7, and 9×9 are represented as G1, G2, G3, and G4, respectively:

[0148] G1 = f grad_1 (I in ),

[0149] G2 = f grad_2 (I in ),

[0150] G3 = f grad_3 (I in ),

[0151] G4 = f grad_4 (I in ),

[0152] In the formula f grad_1 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 3×3, f grad_2 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 5×5, f grad_3 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 7×7, f grad_4 Pixel grayscale step gradient detection for the first layer step gradient detection model with a size of 9×9;

[0153] The pixel grayscale contrast maps of small targets detected by the first-layer convolutional templates with sizes of 3×3, 5×5, 7×7, and 9×9 are represented as I1, I2, I3, and I4, respectively:

[0154] I1=f htcm_1 (I in ),

[0155] I2=f htcm_2 (I in ),

[0156] I3=f htcm_3 (I in ),

[0157] I4=f htcm_4 (I in )

[0158] In the formula f htcm_1 For pixel grayscale contrast detection of the first layer's 3×3 scale convolutional template, f htcm_2 For pixel grayscale contrast detection of the first layer's 5×5 scale convolutional template, f htcm_3 For pixel grayscale step contrast detection of the first layer convolutional template with a size of 7×7, f htcm_4 Pixel grayscale contrast detection of the first layer convolutional template with a size of 9×9.

[0159] Finally, the small target region maps obtained under four different scales of detection are calculated:

[0160] M1 = I1 × G1,

[0161] M2 = I2 × G2,

[0162] M3 = I3 × G3,

[0163] M4 = I4 × G4,

[0164] Calculate the maximum value at each pixel in the four fused images to obtain the detection results of multi-scale stepped contrast and stepped gradient fusion. The final small target region map M is then:

[0165] M = max(M1, M2, M3, M4).

[0166] Verification experiment:

[0167] Combination Figures 9 to 14 As shown, the processing results of three original spaceborne remote sensing infrared images demonstrate that the method of this invention exhibits good detection performance in conditions such as sea surfaces and clouds, effectively detecting small targets in spaceborne remote sensing infrared images. The grayscale 3D images before and after processing show that the method of this invention has good background suppression and target enhancement capabilities for spaceborne remote sensing infrared images. Even in complex backgrounds where targets are obscured by clouds, it maintains good detection performance. Furthermore, the method has strong target enhancement capabilities; the grayscale value of the target is significantly increased after algorithm processing, which aids in subsequent image segmentation and effectively reduces the probability of false detections.

[0168] In addition to the visualization analysis of the high-speed detection of small targets in spaceborne remote sensing infrared images, other indicators are introduced to conduct a more comprehensive evaluation of the methods. For each method, the image segmentation results with better performance are selected for comparison. The comparison results are shown in Table 1.

[0169] Table 1

[0170] algorithm SNRg BSF Pd Fa LCM 6.106 0.527 0.66 0.015 RLCM 2.473 2.738 0.75 0.01 TTLCM 9.112 4.279 0.923 0.01 ASMD 37.572 10.138 0.913 0.01 LIG 25.9555 0.3977 0.956 0.03 HTCM 30.9097 30.6421 0.936 0.003 Method of the present invention 28.9676 0.0072 0.97 0.0011

[0171] As can be seen from the comparison in Table 1, the method of this invention performs well, especially in terms of detection rate and false alarm rate, significantly reducing the false alarm rate of small target detection in spaceborne remote sensing infrared imaging and improving the reliability of the method. It also shows excellent performance in the SNRg index. However, its BSF index is poor, being very low. Analysis shows that BSF is calculated as the ratio of the standard deviation of the input image to the standard deviation of the output image. A low BSF indicates a large standard deviation in the spaceborne remote sensing infrared image processed by this detection algorithm. Analysis of the three-dimensional grayscale values ​​of the processed image reveals that the actual background suppression effect is very good, with almost all background suppressed. The large standard deviation is due to the excessively high grayscale values ​​after target enhancement, i.e., an overly strong enhancement effect. Therefore, although the BSF index is low, the actual background suppression capability is good, and the target enhancement effect is also good.

[0172] In summary, the method of the present invention has good detection performance and practical value.

[0173] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion, characterized in that... include, Small target detection based on stepped gradient in spaceborne remote sensing infrared images: A square stepped gradient detection model is designed, consisting of three layers extending outward from the center. The first layer corresponds to the center of small targets in the spaceborne remote sensing infrared image, the second layer corresponds to the edge of small targets, and the third layer corresponds to the background of small targets. The stepped gradient detection model traverses each pixel in the spaceborne remote sensing infrared image as the center pixel of the first layer, and performs pixel grayscale stepped gradient detection for each pixel. The directions from the four corners of the stepped gradient detection model to the center pixel of the first layer are selected as four gradient directions. The pixel grayscale gradient values ​​of adjacent layers in each gradient direction are calculated. Then, the mean square sum of the pixel grayscale gradient values ​​in each gradient direction is calculated, and the mean square value of the pixel grayscale values ​​in each gradient direction is also calculated. Determine the maximum and minimum values ​​of the mean square value of pixel grayscale; The step gradient of the current center pixel is calculated by combining the ratio of the maximum and minimum values ​​of the mean square grayscale values ​​of the pixels and the mean square grayscale values ​​of the four pixels; the grayscale step gradient map of the pixels representing small targets in the spaceborne remote sensing infrared image is obtained from the calculation results of all step gradients. Small target detection in spaceborne remote sensing infrared images based on stepped contrast: A convolutional template with the same size and corresponding layers as the stepped gradient detection model is designed. Each pixel in the spaceborne remote sensing infrared image is traversed using the convolutional template as the center pixel of the first layer, and stepped contrast detection is performed on each pixel. Convolution operation is performed on the spaceborne remote sensing infrared image to obtain the maximum pixel grayscale value of the second and third layers. The stepped contrast of the current center pixel is calculated by combining the pixel grayscale value of the center pixel of the first layer. The pixel grayscale stepped contrast map representing small targets in the spaceborne remote sensing infrared image is obtained from the calculation results of all stepped contrasts. The pixel grayscale gradient map and the pixel grayscale contrast map are fused to determine the small target region map in the spaceborne remote sensing infrared image.

2. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion as described in claim 1, characterized in that, The method for calculating the pixel grayscale gradient value of adjacent layers in each gradient direction is as follows: g1_1 = I0 - I1, g1_2 = I1 - I5, g2_1 = I0 - I2, g2_2 = I2 - I6, g3_1 = I0 - I3, g3_2 = I3 - I7, g4_1 = I0 - I4, g4_2 = I4 - I8, In the formula, g1_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I1 of the second layer along the first gradient direction ψ1; g1_2 is the pixel grayscale gradient value of the corresponding pixel I1 of the second layer and the corresponding pixel I5 of the third layer along the first gradient direction ψ1; g2_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I2 of the second layer along the second gradient direction ψ2; g2_2 is the pixel grayscale gradient value of the corresponding pixel I2 of the second layer and the corresponding pixel I6 of the third layer along the second gradient direction ψ2. The values ​​are as follows: g3_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I3 of the second layer on the third gradient direction ψ3; g3_2 is the pixel grayscale gradient value of the corresponding pixel I3 of the second layer and the corresponding pixel I7 of the third layer on the third gradient direction ψ3; g4_1 is the pixel grayscale gradient value of the center pixel I0 of the first layer and the corresponding pixel I4 of the second layer on the fourth gradient direction ψ4; g4_2 is the pixel grayscale gradient value of the corresponding pixel I4 of the second layer and the corresponding pixel I8 of the third layer on the fourth gradient direction ψ4.

3. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 2, characterized in that, The method for calculating the mean square value of pixel grayscale in each gradient direction is as follows: In the formula G k For ψ k Mean square value of pixel grayscale in the direction; The method for determining the maximum and minimum values ​​of the mean square of pixel grayscale is as follows: G max =maxG k , G min =minG k , In the formula G max G represents the maximum value of the mean square value of pixel grayscale. min It represents the minimum value of the mean square value of pixel grayscale.

4. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 3, characterized in that, The current step gradient of the center pixel is: In the formula, G0 is the step gradient of the current center pixel, and η is the step gradient threshold value; the pixel grayscale step gradient map G representing the small target in the spaceborne remote sensing infrared image is formed by the step gradients G0 of all pixels.

5. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 4, characterized in that, In the process of small target detection based on stepped contrast in spaceborne remote sensing infrared images, the method for calculating the maximum pixel grayscale value of the second and third layers is as follows: In the formula, Im0 is the maximum pixel grayscale value of the second layer, I in To input spaceborne remote sensing infrared images, S loops The second layer is the convolutional template; In0 is the maximum pixel grayscale value of the third layer, S loopt This is the third layer of the convolution template.

6. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 5, characterized in that, The method for calculating the stepped contrast of the current center pixel is as follows: HTCM=H[(I0-Im0)]·H[(I0-In0)], In the formula, HTCM is the step contrast of the center pixel, and H[x] is the step function: x = I0 - Im0 or x = I0 - In0; Let I0 - Im0 = ΔIm0, I0 - In0 = ΔIn0, then: HTCM = H[ΔIm0]·H[ΔIn0]; The grayscale gradient contrast map I, representing small targets in a spaceborne remote sensing infrared image, is formed by the gradient contrast of all pixels (HTCM).

7. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 6, characterized in that, The pixel grayscale gradient map G and the pixel grayscale contrast map I are fused to obtain the small target region map M in the spaceborne remote sensing infrared image: M = I × G.

8. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 7, characterized in that, A four-scale stepped gradient detection model and convolutional templates of the four scales are used to detect pixel grayscale stepped gradient maps and pixel grayscale stepped contrast maps of small targets in spaceborne remote sensing infrared images, respectively. The first layer size corresponding to the four scales is 3×3, 5×5, 7×7, and 9×9, respectively. The second layer size is the area range of one pixel outside the first layer, and the third layer size is the area range of two pixels outside the second layer. The fusion results obtained from detection at four different scales are calculated to obtain four small target region maps; the maximum pixel gray value of each pixel in the four small target region maps is selected to form the final small target region map.

9. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 8, characterized in that, The pixel grayscale gradient maps of small targets detected by the first-layer stepped gradient detection models with sizes of 3×3, 5×5, 7×7, and 9×9 are represented as G1, G2, G3, and G4, respectively: G1=f grad_1 (I in ), G2=f grad_2 (I in ), G3=f grad_3 (I in ), G4=f grad_4 (I in ), In the formula f grad_1 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 3×3, f grad_2 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 5×5, f grad_3 For pixel grayscale gradient detection of the first layer of the stepped gradient detection model with a size of 7×7, f grad_4 Pixel grayscale step gradient detection for the first layer step gradient detection model with a size of 9×9; The pixel grayscale contrast maps of small targets detected by the first-layer convolutional templates with sizes of 3×3, 5×5, 7×7, and 9×9 are represented as I1, I2, I3, and I4, respectively: I1=f htcm_1 (I in ), I2=f htcm_2 (I in ), I3=f htcm_3 (I in ), I4=f htcm_4 (I in ) In the formula f htcm_1 For pixel grayscale contrast detection of the first layer's 3×3 scale convolutional template, f htcm_2 For pixel grayscale contrast detection of the first layer's 5×5 scale convolutional template, f htcm_3 For pixel grayscale step contrast detection of the first layer convolutional template with a size of 7×7, f htcm_4 Pixel grayscale contrast detection of the first layer convolutional template with a size of 9×9.

10. The method for small target detection in spaceborne remote sensing infrared images based on stepped contrast and stepped gradient fusion according to claim 9, characterized in that, Calculate the small target region maps obtained under four different scales of detection: M1 = I1 × G1, M2 = I2 × G2, M3 = I3 × G3, M4 = I4 × G4, The final small target area map M is then: M = max(M1, M2, M3, M4).

Images