Wavelet transform and adaptive dual-mode based laser spot positioning method

Abstract

The application discloses a laser spot positioning method based on wavelet transform and adaptive double mode. First, wavelet decomposition and inverse transform reconstruction are performed on an infrared image to generate a denoising image and a background noise image, and then a background suppression image is obtained, and the background suppression image is processed to obtain a to-be-measured image; second, multi-scale white top hat transform and adaptive threshold segmentation are performed on the to-be-measured image, candidate spot connected domains are extracted, geometric features thereof are obtained, and the geometric features are used for preliminary screening; then, energy distribution features of the candidate spot connected domains are extracted, comprehensive evaluation is performed in combination with the geometric features, and a coarse positioning coordinate is determined; finally, the number of saturated pixels in a core region is counted and compared with a dynamic proportional threshold value, a state-adaptive double-mode fine positioning mechanism is enabled, and a final fine positioning coordinate is obtained. The application effectively suppresses complex background noise and local highlight interference, reduces errors caused by flat-top saturation distortion, and realizes stable and high-precision positioning under poor illumination.

Description

Technical Field

[0001] This invention belongs to the field of digital image processing and photoelectric measurement technology, specifically relating to a laser spot localization method based on wavelet transform and adaptive dual-mode. Background Technology

[0002] In recent years, with the rapid development of modern optoelectronic countermeasures and laser technology, various optoelectronic detection equipment faces increasingly severe laser interference threats in complex working environments. When suppressive or deceptive high-brightness laser signals emitted by the opponent illuminate our infrared imaging devices, they form high-brightness, large-area flares in the imaging system. These high-brightness interference flares not only cause local saturation or even large-area blindness in the imaging device, severely damaging the contrast and texture details of the image, but also directly obscure real, high-value, weak targets in the field of view, causing the entire optoelectronic system to temporarily lose its target acquisition and tracking capabilities. Therefore, high-precision center positioning of such high-brightness flares in complex environmental backgrounds has become an indispensable core technology in the field of optoelectronic anti-interference. Accurately extracting the center coordinates of the interference flare has extremely important value: on the one hand, these coordinates can provide accurate spatial location priors for subsequent image adaptive compensation, halo area occlusion removal, and local image restoration; on the other hand, the center coordinates of the flare directly map the azimuth vector of the opponent's laser source in space, which is an important prerequisite for guiding our accurate tracing and countermeasures against the interference source.

[0003] However, in practical photoelectric detection scenarios, high-precision extraction of high-brightness laser spots often faces numerous challenges. On the one hand, in complex environments, the combined effects of strong background thermal radiation fluctuations and the dark current noise of the infrared detector itself result in images typically filled with a large amount of complex background noise. On the other hand, high-brightness laser beams have extremely prominent energy focusing characteristics. When they enter the optical system and are imaged on the focal plane, the instantaneous irradiance in the central region of the spot is extremely high, easily exceeding the linear response range of the pixel and directly reaching the upper limit of the grayscale range, forming a typical "flat-top overexposure" phenomenon. At the same time, excess photogenerated charge will violently overflow to surrounding adjacent pixels, producing irregular halos. The superposition of the aforementioned strong background noise, irregular halo diffusion, and central flat-top saturation characteristics completely breaks the original ideal two-dimensional Gaussian energy distribution of the laser spot, causing severe distortion of the core geometric structure and gradient information of the spot, greatly increasing the difficulty of accurate coordinate extraction.

[0004] To address these issues, scholars both domestically and internationally have conducted extensive research on spot localization algorithms and proposed many classic algorithms. However, some problems still exist in spot localization under conditions of complex background noise, irregular halos, and overexposure distortion.

[0005] (1) Although the localization algorithm based on the traditional gray-scale centroid method is simple to calculate and has strong real-time performance, it relies too much on the global gray-scale distribution of the image and is extremely sensitive to background noise. When there is asymmetrical strong halo or uneven background noise, the noise signal will seriously deviate the final centroid coordinates; and when faced with flat-top light spots that are overexposed in the central area, the traditional centroid method cannot accurately utilize the effective information of the edge.

[0006] (2) The positioning algorithm based on two-dimensional Gaussian surface fitting can indeed achieve extremely high sub-pixel accuracy when the spot energy is in an ideal normal distribution. However, the calculation model of this type of algorithm is complex and time-consuming, making it difficult to meet the requirements of high speed and real-time performance. More importantly, this algorithm is highly dependent on the ideal spot energy distribution model. When the center of the laser spot is overexposed, its internal energy distribution no longer conforms to the standard Gaussian curve, and the method will produce a large error at this time.

[0007] (3) Geometric morphology algorithms based on Hough transform or edge fitting have global optimization characteristics and a certain degree of anti-interference ability against local noise. However, this algorithm relies too much on the integrity and standardization of the target boundary. In strong halo scenes, the edge of the real light spot often becomes blurred and extremely irregular (non-ideal circle) due to photon diffusion, which makes it easy for Hough transform to extract false contours and generate false detections. At the same time, due to the influence of parameter space discretization, this type of algorithm itself is extremely difficult to break through the limitation of pixel-level accuracy and cannot meet the high-precision measurement requirements.

[0008] In summary, existing classic algorithms generally struggle to balance complex noise suppression with high-precision center extraction under flat-top overexposure distortion, often falling into a technical bottleneck where noise robustness and positioning accuracy are difficult to achieve simultaneously. Summary of the Invention

[0009] This invention proposes a laser spot localization method based on wavelet transform and adaptive dual-mode, which can achieve stable and high-precision localization of the laser spot center within a single frame for infrared images under conditions of complex background noise, local high brightness interference, and center saturation distortion.

[0010] The technical solution for realizing this invention is: a laser spot localization method based on wavelet transform and adaptive dual-mode, comprising the following steps:

[0011] Step 1: Use a laser to illuminate the infrared imaging device to acquire the original infrared image. Perform wavelet decomposition on the original infrared image to obtain the low-frequency approximation component and the high-frequency detail component. Use the low-frequency approximation component and the high-frequency detail component to generate a denoised image and a background noise map in the spatial domain, thereby obtaining a background suppression image. Then, perform Gaussian smoothing on the background suppression image to obtain the image to be tested, and proceed to Step 2.

[0012] Step 2: Perform multi-scale morphological white hat transformation and adaptive threshold segmentation on the image to be tested, extract candidate spot connected components, and calculate the geometric features of each candidate spot connected component. Use the above geometric features as constraints to perform preliminary screening of the candidate spot connected components, eliminate non-spot interference, and obtain the preliminary screened candidate spot connected components. Proceed to Step 3.

[0013] Step 3: Extract the energy distribution characteristics of the connected regions of the candidate laser spots after initial screening, and comprehensively evaluate the connected regions of the candidate laser spots in combination with the above geometric characteristics to determine the coarse positioning coordinates of the laser spot, and then proceed to step 4.

[0014] Step 4: Calculate the total area of ​​the connected domains of the light spot corresponding to the coarse positioning coordinates to set a dynamic ratio threshold, and count the number of pixels that reach saturated gray level in the local window area centered on the coarse positioning coordinates. Compare this number with the dynamic ratio threshold, and combine it with the state-adaptive dual-mode fine positioning mechanism to obtain the fine positioning coordinates of the laser light spot.

[0015] Furthermore, the original infrared image mentioned in step 1 is a photoelectric detection image with complex background noise and local bright interference.

[0016] Furthermore, the wavelet decomposition and inverse transform reconstruction of the original infrared image described in step 1 includes the following steps:

[0017] S11. Use Haar wavelets to perform wavelet decomposition on the original infrared image to extract low-frequency approximation components and high-frequency detail components in multiple directions; perform threshold denoising on the high-frequency detail components, and combine them with the unprocessed low-frequency approximation components to perform inverse wavelet transform to reconstruct the denoised image; at the same time, remove the high-frequency detail components and use only the low-frequency approximation components to perform inverse wavelet transform to obtain the background noise map in the spatial domain.

[0018] S12. Perform a subtraction operation between the denoised image and the background noise map in the spatial domain to obtain the background suppressed image. Its representation is shown in the following formula:

[0019] ,

[0020] In the formula, This represents a denoised image. Represents the background noise map of the spatial domain. This represents the background suppression ratio coefficient.

[0021] S13, Background suppression image Gaussian smoothing preprocessing is performed to obtain the image to be tested.

[0022] Furthermore, the multi-scale morphological white-hat transformation and adaptive thresholding segmentation of the image to be tested described in step 2 includes the following steps:

[0023] S21. Using multiple disk structural elements of different sizes, white-hat filtering is applied to the image under test, and the maximum response value at each scale is extracted to reconstruct and enhance the image. The calculation formula is as follows:

[0024] ,

[0025] In the formula, The input image to be tested, For the first Disk structural elements with different radii. This indicates the morphological opening operation.

[0026] S22. Extract and enhance the image. Using the mean value of local high-brightness pixel groups as a reference, a decreasing adaptive threshold is applied to... Binarization segmentation is performed to extract candidate spot connected regions, and the eccentricity and fill degree of each candidate spot connected region are calculated as geometric features.

[0027] Among them, eccentricity The calculation formula is as follows:

[0028] ,

[0029] In the formula, Let be the length of the semi-major axis of the equivalent ellipse of the candidate spot's connected region. Let be the length of the minor semi-axis of the equivalent ellipse of the connected region of the candidate spot. For an ideal circular spot, Approaching 0.

[0030] Fill The calculation formula is as follows:

[0031] ,

[0032] In the formula, This represents the actual area of ​​the connected region of the candidate spot. Let be the area of ​​the connected region of the candidate laser spot with respect to the convex hull. For a laser spot with a complete shape, Approaching 1.

[0033] S23. Using the above geometric features as constraints, set eccentricity thresholds respectively. With fill threshold Candidate spot connected regions that do not meet the geometric feature constraints are eliminated to obtain the preliminary candidate spot connected regions.

[0034] Furthermore, the comprehensive evaluation of the connected regions of the candidate light spots after the initial screening, as described in step 3, includes the following steps:

[0035] S31. Extract the energy distribution characteristics of the connected regions of the candidate laser spots after initial screening. The energy distribution characteristics include the local average brightness at the center of the laser spot. and radial symmetry score .

[0036] S32. Combine geometric features to comprehensively evaluate the candidate spot connected regions and obtain the final score of the candidate spot connected regions. Its representation is shown in the following formula:

[0037] ,

[0038] In the formula, This represents the evaluation factor for roundness characteristics. This indicates the average brightness of the local area at the center of the laser spot. This indicates the radial symmetry score.

[0039] Roundness feature evaluation factor The calculation formula is as follows:

[0040] ,

[0041] Local average brightness at the center of the laser spot The calculation formula is as follows:

[0042] ,

[0043] In the formula, This represents a circle with the center of the candidate spot's connected region as the center and the equivalent radius of that candidate spot's connected region. The core pixel neighborhood with a radius of 0.5 times; This represents the total number of valid pixels in the neighborhood of the core pixel. For effective pixel grayscale values, These are the effective pixel coordinates.

[0044] equivalent radius The calculation formula is as follows:

[0045] ,

[0046] In the formula, This represents the actual area of ​​the connected region of the candidate spot.

[0047] Radial symmetry score As shown in the following formula:

[0048] ,

[0049] In the formula, This is the set of boundary cell coordinates of the currently candidate spot's connected domain. For the above set The total number of valid pixels contained Let be the two-dimensional gradient vector of the boundary pixel of the laser spot. The vector pointing from an effective pixel to the geometric center of the connected domain of the candidate spot.

[0050] S33. The center coordinates of the candidate spot with the highest final score are determined as the coarse positioning coordinates of the laser spot.

[0051] Furthermore, step 4, which describes the adaptive selection of the squared gray-level centroid method or the gradient squared centroid method based on the number of saturated pixels, includes the following steps:

[0052] S41. Using the coarse positioning coordinates of the laser spot as the center, extract a rectangular area of ​​a preset size as a local window, and count the number of saturated pixels within this local window whose grayscale value reaches a preset saturation threshold. Simultaneously calculate the total area of ​​the connected components corresponding to the coarse positioning coordinates of the light spot. And set a dynamic ratio threshold. :

[0053] ,

[0054] In the formula This is a preset area ratio constant.

[0055] S42, When the number of saturated pixels Less than or equal to the dynamic scaling threshold When using the squared gray-level centroid method, its weight operator... As shown in the following formula:

[0056] ,

[0057] S43, When the number of saturated pixels Greater than the dynamic ratio threshold When using the gradient squared centroid method, the weight operator is as follows:

[0058] ,

[0059] In the formula, The effective pixel grayscale value of the local window. This represents the minimum grayscale value of the local window. and These correspond to the horizontal and vertical gradient magnitudes calculated using the edge detection operator, with the Sobel operator being the preferred choice.

[0060] S44. Calculate the precise positioning coordinates of the laser spot using the selected weighting operator:

[0061] ,

[0062] ,

[0063] In the formula, The effective pixel coordinates within the local window. For the weight operator that is adaptively selected based on the number of saturated pixels, the variable or .

[0064] Compared with existing technologies, the significant advantages of this invention are as follows: It obtains the background noise map in the spatial domain through wavelet decomposition and inverse transform reconstruction, effectively suppressing complex background noise; it eliminates local bright non-spot interference through multi-scale morphological white-hat transformation, adaptive threshold segmentation, and geometric feature initial screening; it stably determines coarse positioning coordinates through comprehensive evaluation of energy distribution characteristics and geometric features; and it improves the robustness and accuracy of laser spot positioning under complex lighting conditions by selecting appropriate fine positioning modes under unsaturated spot and flat-top saturated spot conditions through a state-adaptive dual-mode fine positioning mechanism. Attached Figure Description

[0065] Figure 1 This is a flowchart of a laser spot localization method based on wavelet transform and adaptive dual-mode according to the present invention.

[0066] Figure 2 This is a flowchart illustrating a laser spot localization method based on wavelet transform and adaptive dual-mode according to the present invention.

[0067] Figure 3 This is an input-output diagram of a laser spot localization method based on wavelet transform and adaptive dual-mode according to the present invention. Detailed Implementation

[0068] The technical solution of the present invention will be described in detail below with reference to the embodiments and accompanying drawings.

[0069] Combination Figure 1 , Figure 2 and Figure 3 The laser spot localization method based on wavelet transform and adaptive dual-mode described in this invention uses an original infrared image with complex background noise and local bright interference as the processing object. The specific steps are as follows:

[0070] Step 1: Use a laser to illuminate an infrared imaging device to acquire the original infrared image. Perform wavelet decomposition on the original infrared image to obtain low-frequency approximation components and high-frequency detail components. Use the low-frequency approximation components and high-frequency detail components to generate a denoised image and a background noise map in the spatial domain, thereby obtaining a background suppression image. Then, perform Gaussian smoothing on the background suppression image to obtain the image to be tested.

[0071] S11. Use Haar wavelets to perform wavelet decomposition on the original infrared image to extract low-frequency approximation components and high-frequency detail components in multiple directions; perform threshold denoising on the high-frequency detail components, and combine them with the unprocessed low-frequency approximation components to perform inverse wavelet transform to reconstruct the denoised image; at the same time, remove the high-frequency detail components and use only the low-frequency approximation components to perform inverse wavelet transform to obtain the background noise map in the spatial domain.

[0072] S12, Denoise the image Background noise map of spatial domain Perform the difference operation to obtain the background-suppressed image. :

[0073] ,

[0074] In the formula, This represents a denoised image. Represents the background noise map of the spatial domain. This represents the background suppression ratio coefficient.

[0075] S13, Background suppression image Gaussian smoothing preprocessing is performed to suppress residual small-scale random noise, resulting in the image to be tested.

[0076] Step 2: Perform multi-scale morphological white hat transformation and adaptive threshold segmentation on the image to be tested, extract candidate spot connected components, and calculate the geometric features of each candidate spot connected component. Use the above geometric features as constraints to perform preliminary screening of the candidate spot connected components, eliminate non-spot interference, and obtain the preliminary screened candidate spot connected components.

[0077] S21. Using multiple disk structural elements of different sizes, white-hat filtering is applied to the image under test, and the maximum response value at each scale is extracted to reconstruct and enhance the image. The calculation formula is as follows:

[0078] ,

[0079] In the formula, The input image to be tested, For the first Disk structural elements with different radii. This indicates the morphological opening operation.

[0080] S22. Extract the mean value of local high-brightness pixel groups in the enhanced image as a reference benchmark, and use a decreasing adaptive threshold to perform binarization segmentation on the enhanced image to obtain candidate spot connected components.

[0081] S23. Calculate the eccentricity of the connected components of the candidate spot. With fill degree And the eccentricity threshold and the fill degree threshold are used as geometric feature constraints.

[0082] ,

[0083] ,

[0084] When the eccentricity of a candidate spot connected region is too high or the filling rate is too low, the candidate spot connected region is discarded, resulting in the initially screened candidate spot connected regions. This process corresponds to... Figure 2 The process includes extraction of connected domains and initial screening of geometric features for candidate light spots.

[0085] Step 3: Extract the energy distribution characteristics of the connected regions of the candidate laser spots after initial screening, and comprehensively evaluate the connected regions of the candidate laser spots in combination with the above geometric characteristics to determine the coarse positioning coordinates of the laser spot.

[0086] S31. Extract the roundness feature evaluation factor of the connected regions of the candidate light spots after initial screening. Average brightness of the center of the laser spot and radial symmetry score :

[0087] ,

[0088] ,

[0089] In the formula, This represents a circle with the center of the candidate spot's connected region as the center and the equivalent radius of that candidate spot's connected region. The core pixel neighborhood with a radius of 0.5 times; This represents the total number of valid pixels in the neighborhood of the core pixel. For effective pixel grayscale values, These are the effective pixel coordinates.

[0090] ,

[0091] In the formula, This represents the actual area of ​​the connected region of the candidate spot.

[0092] ,

[0093] In the formula, This is the set of boundary cell coordinates of the currently candidate spot's connected domain. For the above set The total number of valid pixels contained Let be the two-dimensional gradient vector of the boundary pixel of the laser spot. The vector pointing from an effective pixel to the geometric center of the connected domain of the candidate spot.

[0094] S32. Combining the above geometric features and energy distribution features, a comprehensive evaluation of the candidate spot connected regions is performed to obtain the final score of the candidate spot connected regions. :

[0095]

[0096] S33. The center coordinates of the connected region of the candidate spot with the highest final score are determined as the coarse positioning coordinates of the laser spot. This process corresponds to... Figure 2 The output section includes a comprehensive evaluation of medium energy distribution characteristics and geometric features, as well as coarse localization results.

[0097] Step 4: Calculate the total area of ​​the connected domains of the light spot corresponding to the coarse positioning coordinates to set a dynamic ratio threshold, and count the number of pixels that reach saturated gray level in the local window area centered on the coarse positioning coordinates. Compare this number with the dynamic ratio threshold, and combine it with the state-adaptive dual-mode fine positioning mechanism to obtain the fine positioning coordinates of the laser light spot.

[0098] Using the coarse positioning coordinates of the laser spot as the center, a rectangular area of ​​a preset size is extracted as a local window, and the number of saturated pixels within this local window whose grayscale value reaches a preset saturation threshold is counted. Simultaneously calculate the total area of ​​the connected components corresponding to the coarse positioning coordinates of the light spot. And set a dynamic ratio threshold. :

[0099] ,

[0100] In the formula This is a preset area ratio constant.

[0101] When the number of saturated pixels Less than or equal to the dynamic proportional threshold When it is determined that the current light spot has not undergone obvious flat-top saturation distortion, the square gray-level centroid method is used as the weight operator. :

[0102] .

[0103] When the number of saturated pixels Greater than the dynamic ratio threshold When it is determined that the current light spot has flat-top saturation distortion, the gradient squared centroid method is used as the weight operator. :

[0104] ,

[0105] In the formula, The effective pixel grayscale value of the local window. This represents the minimum grayscale value of the local window. and These correspond to the horizontal and vertical gradient magnitudes calculated using the edge detection operator, with the Sobel operator being the preferred choice.

[0106] Based on the weight operator adaptively selected according to the state, the coordinates of the effective pixels within the local window are weighted and summed to obtain the precise positioning coordinates of the laser spot. :

[0107] ,

[0108] ,

[0109] In the formula, The effective pixel coordinates within the local window. For the weight operator that is adaptively selected based on the number of saturated pixels, the variable or .

[0110] Figure 3 The precise positioning results are used to verify the feasibility of the above procedure; by Figure 3 It is known that the present invention can stably obtain the center position of the laser spot in the original infrared image with complex background noise and local bright interference.

[0111] Example

[0112] To verify the feasibility of the laser spot localization method based on wavelet transform and adaptive dual-mode described in this invention, an original infrared image with complex background noise and local bright interference was selected as the test object. The main algorithm parameters used in the processing are shown in Table 1:

[0113] Table 1 Algorithm Parameters for Example

[0114]

[0115] Given a raw infrared image, such as Figure 3 As shown in the left figure, there is a noticeable ring-shaped background noise around the laser spot, and some areas contain bright interference points. In the simulation, the original infrared image was processed according to the steps described in this invention, and the final precise positioning coordinates were superimposed on the positioning result image using crosshairs.

[0116] First, the original infrared image is reconstructed by Haar wavelet decomposition and inverse transform according to step 1. The high-frequency detail components are thresholded and denoised, and the denoised image is reconstructed by combining the low-frequency approximation components. At the same time, the high-frequency detail components are removed, and only the low-frequency approximation components are used to reconstruct the background noise map in the spatial domain. Then, the difference between the denoised image and the background noise map in the spatial domain is calculated to obtain the background suppressed image. The background suppressed image is then Gaussian smoothed to obtain the image to be tested.

[0117] Secondly, following step 2, multi-scale morphological white hat transformation and adaptive threshold segmentation are performed on the image to be tested to obtain candidate spot connected regions. Then, eccentricity and fill degree are used as geometric features to initially screen the candidate spot connected regions, retaining the candidate spot connected regions that meet the geometric feature constraints after initial screening, thereby eliminating non-spot interference.

[0118] Then, following step 3, the energy distribution characteristics of the connected domains of the candidate laser spots after initial screening are extracted, and combined with geometric features for comprehensive evaluation to determine the coarse positioning coordinates of the laser spots.

[0119] Finally, following step 4, the number of pixels reaching saturated grayscale within the local window area centered on the coarse positioning coordinates is counted and compared with the dynamic scaling threshold. When the number of saturated pixels is less than the dynamic scaling threshold, the squared grayscale centroid method is used to calculate the fine positioning coordinates; when the number of saturated pixels is greater than or equal to the dynamic scaling threshold, the gradient squared centroid method is used to calculate the fine positioning coordinates. Figure 3 As shown in the right figure, the precise positioning result is located in the core region of the laser spot energy, indicating that the present invention can still stably obtain the center position of the spot under complex background noise and local high brightness interference conditions, verifying the feasibility of the method.

Claims

1. A laser spot localization method based on wavelet transform and adaptive dual-mode, characterized in that, Includes the following steps: Step 1: Use laser to illuminate the infrared imaging device to acquire the original infrared image. Perform wavelet decomposition on the original infrared image to obtain the low-frequency approximation component and the high-frequency detail component. Use the low-frequency approximation component and the high-frequency detail component to generate a denoised image and a background noise map in the spatial domain, thereby obtaining a background suppression image. Then, perform Gaussian smoothing on the background suppression image to obtain the image to be tested, and proceed to step 2. Step 2: Perform multi-scale morphological white hat transformation and adaptive threshold segmentation on the image to be tested, extract candidate spot connected components, and calculate the geometric features of each candidate spot connected component. Use the above geometric features as constraints to perform preliminary screening of the candidate spot connected components, eliminate non-spot interference, and obtain the preliminary screened candidate spot connected components. Proceed to Step 3. Step 3: Extract the energy distribution characteristics of the connected regions of the candidate laser spots after initial screening, and comprehensively evaluate the connected regions of the candidate laser spots in combination with the above geometric characteristics to determine the coarse positioning coordinates of the laser spot, and then proceed to step 4. Step 4: Calculate the total area of ​​the connected domains of the light spot corresponding to the coarse positioning coordinates to set a dynamic ratio threshold, and count the number of pixels that reach saturated gray level in the local window area centered on the coarse positioning coordinates. Compare this number with the dynamic ratio threshold, and combine it with the state-adaptive dual-mode fine positioning mechanism to obtain the fine positioning coordinates of the laser light spot. The state-adaptive dual-mode fine localization mechanism includes the following steps: Using the coarse positioning coordinates of the laser spot as the center, a rectangular area of ​​a preset size is extracted as a local window, and the number of saturated pixels within this local window whose grayscale value reaches a preset saturation threshold is counted. Simultaneously calculate the total area of ​​the connected components corresponding to the coarse positioning coordinates of the light spot. And set a dynamic ratio threshold. : ， In the formula This is a preset area ratio constant; when When using the squared gray-level centroid method, its weight operator... As shown in the following formula: ， when When using the gradient squared centroid method, its weight operator... As shown in the following formula: ， In the formula, The effective pixel grayscale value of the local window. This represents the minimum grayscale value of the local window. and These correspond to the horizontal and vertical gradient magnitudes calculated using the edge detection operator, respectively. Precise positioning coordinates of the laser spot Calculated using the following formula: ， ， In the formula, These are the effective cell coordinates within the local window. For the weight operator that is adaptively selected based on the number of saturated pixels, the variable or .

2. The laser spot localization method based on wavelet transform and adaptive dual-mode as described in claim 1, characterized in that, In step 1, the original infrared image is a photoelectric detection image with complex background noise and local bright interference.

3. The laser spot localization method based on wavelet transform and adaptive dual-mode according to claim 2, characterized in that, In step 1, a background suppression image is generated, as follows: S11. Use Haar wavelet to perform wavelet decomposition on the original infrared image to extract low-frequency approximation components and multi-directional high-frequency detail components. Then, perform threshold denoising on the high-frequency detail components and combine them with the unprocessed low-frequency approximation components to perform inverse wavelet transform to reconstruct the denoised image. By removing high-frequency detail components and performing inverse transformation using only the low-frequency approximate components, the background noise map in the spatial domain is obtained. S12, Denoise the image Background noise map of spatial domain Perform the difference operation to obtain the background-suppressed image. : ， In the formula, This represents a denoised image. Represents the background noise map of the spatial domain. Indicates the background suppression ratio coefficient; S13, Background suppression image Gaussian smoothing preprocessing is performed to obtain the image to be tested.

4. The laser spot localization method based on wavelet transform and adaptive dual-mode according to claim 3, characterized in that, Step 2, as follows: S21. Using multiple disk structure elements of different sizes, apply white-hat filtering to the image under test, extract the maximum response value at each scale, and reconstruct and enhance the image. The calculation formula is as follows: ， In the formula, The input image to be tested, For the first Disk structural elements with different radii. Represents morphological opening operation; S22. Extract and enhance the image. Using the mean value of local high-brightness pixel groups as a reference, a decreasing adaptive threshold is applied to... Binarization segmentation is performed to extract candidate spot connected components, and the eccentricity of each candidate spot connected component is calculated. With fill degree As a geometric feature; Among them, eccentricity The calculation formula is as follows: ， In the formula, Let be the length of the major semi-axis of the equivalent ellipse of the candidate spot's connected region, and c be the length of the minor semi-axis of the equivalent ellipse of the candidate spot's connected region. For an ideal circular spot, Approaching 0; The equivalent ellipse of the candidate spot connected domain is an ideal ellipse with the same spatial distribution characteristics as the irregular connected domain. The coordinates of all effective pixels in the candidate spot connected domain are extracted, and the arithmetic mean of the effective pixel coordinates is calculated and used as the center of the equivalent ellipse. The degree of divergence of the distribution of all effective pixels in the candidate connected domain relative to the center of the equivalent ellipse is statistically analyzed, i.e., the second moment of space. The direction in which the effective pixel distribution is most dispersed and has the largest span is defined as the major axis direction of the equivalent ellipse, and the direction perpendicular to the major axis and where the pixel distribution is most compact is defined as the minor axis direction. Fill The calculation formula is as follows: ， In the formula, This represents the actual area of ​​the connected region of the candidate spot. Let be the area of ​​the connected region of the candidate laser spot with respect to the convex hull. For a laser spot with a complete shape, Approaching 1; S23. Using the above geometric features as constraints, set eccentricity thresholds respectively. With fill threshold If the eccentricity of the connected region of the candidate spot or If the non-spot interference with extremely irregular shapes is eliminated, the candidate spot connected regions after initial screening are obtained.

5. The laser spot localization method based on wavelet transform and adaptive dual-mode according to claim 4, characterized in that, Step 3 is detailed as follows: S31. Extract the energy distribution characteristics of the connected regions of the candidate laser spots after initial screening, including the local average brightness at the center of the laser spot. and radial symmetry score ; S32. Combine geometric features to comprehensively evaluate the candidate spot connected regions and obtain the final score of the candidate spot connected regions. : ， In the formula, Roundness characteristic evaluation factor; S33. The center coordinates of the connected domain of the candidate spot with the highest score are set as the coarse positioning coordinates of the laser spot.

6. The laser spot localization method based on wavelet transform and adaptive dual-mode according to claim 5, characterized in that, Roundness feature evaluation factor The calculation formula is as follows: 。 7. The laser spot localization method based on wavelet transform and adaptive dual-mode according to claim 5, characterized in that, Local average brightness at the center of the laser spot The calculation formula is as follows: ， In the formula, This represents a circle with the center of the candidate spot's connected region as the center and the equivalent radius of that candidate spot's connected region. The core pixel neighborhood with a radius of 0.5 times; This represents the total number of valid pixels in the neighborhood of the core pixel. For effective pixel grayscale values, Effective pixel coordinates; equivalent radius The calculation formula is as follows: ， In the formula, This represents the actual area of ​​the connected region of the candidate spot.

8. The laser spot localization method based on wavelet transform and adaptive dual-mode according to claim 5, characterized in that, Radial symmetry score As shown in the following formula: ， In the formula, This is the set of boundary cell coordinates of the currently candidate spot's connected domain. For the above set The total number of valid pixels contained Let be the two-dimensional gradient vector of the boundary pixel of the laser spot. The vector pointing from an effective pixel to the geometric center of the connected domain of the candidate spot.

Images