A method for adaptive background suppression and range-gated imaging noise elimination
By using adaptive background modeling and dynamic network structure, high-dimensional feature vectors are generated and a background noise model is constructed, which solves the noise pollution problem in range-gated imaging technology, achieves a balance between noise suppression and signal fidelity, and improves imaging quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KAYA (BEIJING) INT PHOTOELECTRIC TECH CO LTD
- Filing Date
- 2025-08-05
- Publication Date
- 2026-07-14
Smart Images
Figure CN121008252B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of range-gated imaging technology, and particularly relates to a method for noise reduction in range-gated imaging with adaptive background suppression. Background Technology
[0002] Range-gated imaging (RGI) is an active imaging technique that uses laser pulse illumination and time-synchronized gating exposure to achieve high-contrast imaging of targets within a specific range. Its core idea is to selectively receive target reflection signals and suppress background noise by controlling the flight time of the laser pulse and the time window of the gating detector.
[0003] However, existing range-gated imaging techniques are susceptible to noise contamination from backscattered light, ambient stray light, and detector dark current. Traditional fixed-threshold background subtraction is difficult to adapt to dynamic environmental changes. Furthermore, noise intensity varies drastically with depth gradient and spatial position, and conventional global filtering leads to loss of detail. Summary of the Invention
[0004] In view of the above-mentioned deficiencies of the prior art, this invention proposes an adaptive background suppression range-gated imaging noise reduction method, which achieves a balance between noise suppression and signal fidelity through adaptive background modeling, position coding enhancement, and dynamic network structure. The technical solution designed in this invention includes the following steps:
[0005] S10: Obtain each pixel of the distance gating image and input it into the position encoder to generate a high-dimensional feature vector;
[0006] S20: Construct a background noise model and perform adaptive background subtraction on the distance gating image to output the intermediate distance gating image;
[0007] S30: Concatenate the intermediate distance gated image with the high-dimensional feature vector in the channel dimension to output the fused feature;
[0008] S40: Input fused features into a dynamic network structure, output a reconstructed distance gating image.
[0009] Preferably, the formula for generating the high-dimensional feature vector in S10 is as follows:
[0010]
[0011] In the formula, It is a high-dimensional feature vector. , For encoding dimensions, For wavelength adjustment factor, Distance gating image, These are the two-dimensional coordinates of a pixel.
[0012] Preferably, the wavelength adjustment factor is formulated as follows:
[0013]
[0014] In the formula, The normalization coefficient is... For depth gradient, This represents the total number of pixels in the image.
[0015] Preferably, the output intermediate distance gating image in S20 is formulated as follows:
[0016]
[0017] In the formula, Gating the image for intermediate distance, This is a background noise model.
[0018] Preferably, the output fusion feature in S30 is formulated as follows:
[0019]
[0020] In the formula, As a feature of fusion, For feature selection function, Gating the image for intermediate distance, For tensor splicing, It is a high-dimensional feature vector.
[0021] Preferably, the output reconstruction distance-gated image in S40 includes:
[0022] The fused features are then processed sequentially in the expansion layer. Convolution expands the number of channels to twice that of depthwise convolution. Depthwise separable convolutions extract spatial features and compress layers Convolution compresses the number of channels to their original size and connects them to the input residual, outputting a reconstructed distance-gated image.
[0023] Preferably, S10 further includes:
[0024] After acquiring each pixel of the distance-gated image, the spatial offset of each pixel is calculated and a global matching degree index is generated. The distance-gated image is then filtered based on the global matching degree index.
[0025] Preferably, the formula for calculating the spatial offset of each pixel is as follows:
[0026]
[0027] In the formula, This is the spatial offset. The x-coordinate of the center coordinate. The ordinate is the center coordinate.
[0028] Preferably, the formula for generating the global matching degree index is as follows:
[0029]
[0030] In the formula, As a global matching degree metric, This represents the total number of pixels in the image. The matching coefficients for each pixel.
[0031] Preferably, the matching coefficient of the pixel is calculated using the following formula:
[0032]
[0033] In the formula, This is the baseline offset of the sample from the gated image.
[0034] Beneficial effects:
[0035] This application provides a range-gated imaging noise reduction method with adaptive background suppression. By using adaptive background modeling, position coding enhancement, and dynamic network structure, it achieves a balance between noise suppression and signal fidelity, providing a more reliable solution for imaging in low-visibility environments. Attached Figure Description
[0036] Figure 1 This is a schematic diagram of a preferred embodiment of the present invention. Detailed Implementation
[0037] The embodiments of the present invention will be described in detail below. The embodiments described below are implemented based on the technical solution of the present invention, and detailed implementation methods and specific operation processes are given. However, the protection scope of the present invention is not limited to the embodiments described below.
[0038] This invention designs an adaptive background suppression range-gated imaging noise reduction method, the technical solution of which includes the following steps, such as... Figure 1 As shown, it specifically includes:
[0039] S10: Obtain each pixel of the distance gating image and input it into the position encoder to generate a high-dimensional feature vector;
[0040] S20: Construct a background noise model and perform adaptive background subtraction on the distance gating image to output the intermediate distance gating image;
[0041] S30: Concatenate the intermediate distance gated image with the high-dimensional feature vector in the channel dimension to output the fused feature;
[0042] S40: Input fused features into a dynamic network structure, output a reconstructed distance gating image.
[0043] Preferably, the formula for generating high-dimensional feature vectors in S10 is as follows:
[0044]
[0045] In the formula, It is a high-dimensional feature vector. , For encoding dimensions, For wavelength adjustment factor, Distance gating image, These are the two-dimensional coordinates of a pixel.
[0046] Preferably, the wavelength adjustment factor is calculated using the following formula:
[0047]
[0048] In the formula, The normalization coefficient is... For depth gradient, This represents the total number of pixels in the image.
[0049] Specifically, for As a depth gradient, a depth map of the scene is obtained using laser pulses and a synchronously gating detector. This depth map records the actual distance from each pixel to the camera; therefore, it can be obtained by performing horizontal and vertical difference operations on the depth map. .
[0050] In addition, for S10, dynamic adjustment is achieved through depth gradient. High-frequency encoding is used in areas of deep abrupt change (such as object edges) to improve detail preservation, and direct stitching is also possible. The original intensity information was preserved, avoiding feature loss.
[0051] Preferably, the output intermediate distance gating image in S20 is formulated as follows:
[0052]
[0053] In the formula, Gating the image for intermediate distance, This is a background noise model.
[0054] Specifically, for For the background noise model, multiple frames of pure background distance gated images of a targetless scene are continuously acquired, and each pixel is analyzed. The intensity distribution statistics on the time series are calculated, and the background model is dynamically updated using a sliding window mechanism. The core statistics include the exponentially weighted moving average, local regional variance, and the frequency of impulse noise occurrence, which are used to adapt to slow changes in illumination, capture spatially correlated noise, and detect sudden interference, respectively. In addition, the background noise model is automatically reset when it detects a sudden change in scene illumination, and an aging attenuation factor is set for stationary objects to prevent false absorption.
[0055] Preferably, the output fusion feature in S30 is formulated as follows:
[0056]
[0057] In the formula, As a feature of fusion, For feature selection function, Gating the image for intermediate distance, For tensor splicing, It is a high-dimensional feature vector.
[0058] Specifically, for The feature selection function is implemented using a learnable attention mechanism, including through... Convolution generates channel attention weight maps, which are then applied to... Differential weighting is applied to feature and location-encoded features; for splicing The operation requires consistent space dimensions. To prevent misalignment issues, in Multi-scale features are introduced for noise suppression.
[0059] Preferably, the output reconstruction of the distance gating image in S40 includes:
[0060] The fused features are then processed sequentially in the expansion layer. Convolution expands the number of channels to twice that of depthwise convolution. Depthwise separable convolutions extract spatial features and compress layers Convolution compresses the number of channels to their original size and connects them to the input residual, outputting a reconstructed distance-gated image.
[0061] Preferably, S10 further includes:
[0062] After acquiring each pixel of the distance-gated image, the spatial offset of each pixel is calculated and a global matching degree index is generated. The distance-gated image is then filtered based on the global matching degree index.
[0063] Preferably, the spatial offset of each pixel is calculated using the following formula:
[0064]
[0065] In the formula, This is the spatial offset. The x-coordinate of the center coordinate. The ordinate is the center coordinate.
[0066] Preferably, a global matching degree index is generated, using the following formula:
[0067]
[0068] In the formula, As a global matching degree metric, This represents the total number of pixels in the image. The matching coefficients for each pixel.
[0069] Preferably, the matching coefficient of the pixel is calculated using the following formula:
[0070]
[0071] In the formula, This is the baseline offset of the sample from the gated image.
[0072] Specifically, filtering of the distance-gated image includes:
[0073] When global matching degree When the distance gating image is greater than or equal to the preset global matching threshold, the corresponding distance gating image enters pixel-level processing to generate a filtered distance gating image, as shown in the following formula:
[0074]
[0075]
[0076] In the formula, For the filtered distance gating image, The distance gating image before filtering. It is a binary mask. This is the preset matching coefficient threshold.
[0077] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.
Claims
1. A range-gated imaging noise reduction method with adaptive background suppression, characterized in that, include: S10: Obtain each pixel of the distance gating image and input it into the position encoder to generate a high-dimensional feature vector; S20: Construct a background noise model and perform adaptive background subtraction on the distance gating image to output the intermediate distance gating image; S30: Concatenate the intermediate distance gated image with the high-dimensional feature vector in the channel dimension to output the fused feature; S40: Input fused features into a dynamic network structure, output a reconstructed distance gating image; The formula for generating high-dimensional feature vectors in S10 is as follows: In the formula, It is a high-dimensional feature vector. , For encoding dimensions, For wavelength adjustment factor, Distance gating image, Two-dimensional coordinates of a pixel; The wavelength adjustment factor is calculated as follows: In the formula, The normalization coefficient is... For depth gradient, This represents the total number of pixels in the image. The formula for the output intermediate distance gating image in S20 is as follows: In the formula, Gating the image for intermediate distance, For background noise model; The output fusion feature in S30 is defined by the following formula: In the formula, As a feature of fusion, For feature selection function, Gating the image for intermediate distance, For tensor splicing, It is a high-dimensional feature vector; The output reconstruction of the distance-gated image in S40 includes: sequentially performing an expansion layer on the fused features. Convolution expands the number of channels to twice that of depthwise convolution. Depthwise separable convolutions extract spatial features and compress layers Convolution compresses the number of channels to their original size and concatenates them with the input residual to output a reconstructed distance-gated image. S10 further includes: after acquiring each pixel of the distance gating image, calculating the spatial offset of each pixel and generating a global matching degree index, and filtering the distance gating image based on the global matching degree index.
2. The adaptive background suppression range-gated imaging noise reduction method according to claim 1, characterized in that, The formula for calculating the spatial offset of each pixel is as follows: In the formula, This is the spatial offset. The x-coordinate of the center coordinate. The ordinate of the center coordinate. It is the vertical Gaussian scaling factor of the pixel, used to constrain the spatial attenuation range in the vertical coordinate direction of the pixel and control the vertical pixel offset sensitivity.
3. The adaptive background suppression range-gated imaging noise reduction method according to claim 2, characterized in that, The formula for generating the global matching degree index is as follows: In the formula, As a global matching degree metric, This represents the total number of pixels in the image. The matching coefficients for each pixel.
4. The adaptive background suppression range-gated imaging noise reduction method according to claim 3, characterized in that, The matching coefficient of the pixel is calculated using the following formula: In the formula, This is the baseline offset of the sample from the gated image.