Polarization phase significant map and focal plane temperature drift self-correction long-distance weak object enhancement system

Abstract

The application discloses a polarized phase saliency map and focal plane temperature drift self-correction long-distance weak object enhancement system, relates to the fields of photoelectric imaging, polarized imaging, infrared or low-light image processing, focal plane array correction and long-distance target detection, and aims to solve the problems of few target pixels, low signal-to-noise ratio, false targets caused by polarized focal plane temperature drift, traditional non-uniformity correction easily weakening real targets and ordinary saliency enhancement difficult to distinguish polarized targets from temperature drift noise in long-distance weak target enhancement. The application takes the focal plane polarized image or the time-sharing polarized image as input, forms a complete weak target enhancement process through target elimination type temperature drift field estimation, polarized channel consistency correction, complex polarized phase quantity construction, multi-scale polarized phase saliency enhancement and temperature drift-saliency feedback update. The application outputs quantifiable target parameters such as an enhanced image, a signal-to-noise ratio, a candidate region and a confidence degree, and can directly support photoelectric monitoring, security alarm, review display and other scenes.

Description

Technical Field

[0001] This invention relates to the fields of photoelectric imaging, polarization imaging, infrared or low-light image processing, focal plane array correction, and long-range target detection technology, specifically to a long-range weak body enhancement system with polarization phase saliency map and focal plane temperature drift self-correction. Background Technology

[0002] In long-range photoelectric detection, the target typically occupies only a few pixels on the imaging surface, and the target's radiation or reflection intensity is easily obscured by the sky, sea surface, ground features, clouds, thermal noise, readout noise, and platform jitter. Traditional light intensity image enhancement methods mainly utilize grayscale differences, local contrast, or spatial filtering to improve target visibility. However, when the background structure is complex, the difference in radiation between the target and the background is small, or the focal plane temperature drifts, problems such as background edges being enhanced, fixed pattern noise being amplified, and truly weak targets being smoothed and weakened can easily occur.

[0003] Polarization imaging can capture the differences between a target and the background in terms of polarization direction, degree of linear polarization, and polarization angle. For artificial targets, small targets on the sea surface, low-light targets, or weak targets in complex backgrounds, polarization features can suppress background clutter and improve the separability of the target from the background to some extent. However, existing polarization target enhancement methods typically directly threshold or detect polarization degree, polarization angle, or polarization image fusion results using neural networks, rarely considering channel offset drift, gain drift, and row and column fringe drift caused by the focal plane polarization detector under different temperatures, gains, and integration times. If temperature drift is not sufficiently suppressed, channel differences between the four polarization directions can be misinterpreted as target polarization differences, leading to false targets in the significantly polarized results.

[0004] On the other hand, focal plane non-uniformity correction and temperature drift compensation have been extensively studied and applied in engineering. These methods typically focus on fixed pattern noise, gain bias, or offset drift in intensity images, aiming to obtain images with uniform grayscale. However, enhancing distant weak targets requires not only suppressing temperature drift noise but also avoiding the subtraction of real small targets, short-lived weak targets, or target regions with polarization differences as background drift. Therefore, simply using ordinary scene non-uniformity correction or ordinary temperature lookup table correction is insufficient to simultaneously ensure polarization channel consistency and the preservation of weak targets.

[0005] Furthermore, existing saliency enhancement methods often use Fourier phase spectrum, local contrast, region entropy, or deep network response as saliency sources. While these methods are effective for ordinary infrared images of weak targets, they do not fully utilize the periodicity of the polarization angle, the local consistency of the polarization phase, and the influence of focal plane temperature drift on polarization calculation. Directly applying traditional saliency methods to polarized focal plane images may result in pseudo-saliency regions caused by channel mismatch, temperature drift, and polarization blind cells.

[0006] Therefore, it is urgent to propose a technical solution that differs from simple polarization degree enhancement, simple focal plane temperature drift correction, and simple phase spectrum saliency detection. This solution couples focal plane temperature drift self-correction with polarization phase saliency map generation. By eliminating target candidate regions, feeding back temperature drift residuals, and feeding back polarization phase residuals, the temperature drift field update in the next frame is constrained. This eliminates the pseudo-polarization differences caused by temperature drift while preserving the local abrupt change characteristics of distant weak targets in the polarization phase space. Summary of the Invention

[0007] To address the challenges of low target pixel count, low signal-to-noise ratio, false targets caused by polarization focal plane temperature drift, weakening of real targets by traditional non-uniformity correction, and difficulty in distinguishing polarization targets from temperature drift noise in ordinary saliency enhancement, this invention proposes a long-range weak object enhancement system with polarization phase saliency map and focal plane temperature drift self-correction. This system takes a focal plane polarization image or a time-division polarization image as input and forms a complete weak target enhancement process through target elimination temperature drift field estimation, polarization channel consistency correction, complex polarization phase quantity construction, multi-scale polarization phase saliency enhancement and temperature drift, and saliency feedback update.

[0008] To achieve the above objectives, the present invention provides the following technical solution: This invention proposes a long-range weak object enhancement system with polarization phase saliency map and focal plane temperature drift self-correction. The system includes an imaging optical lens, a polarization acquisition unit, a temperature acquisition unit, an image acquisition module, a polarization reconstruction module, a temperature drift self-correction module, a polarization phase saliency map generation module, a weak target enhancement module, a candidate target output module, and a parameter feedback module. The imaging optical lens is used to receive radiation light from a distant scene and image the scene onto the polarization acquisition unit; The polarization acquisition unit is used to acquire original polarization images in multiple polarization directions; The temperature acquisition unit is used to acquire focal plane temperature-related state parameters; The image acquisition module is used to read the original polarization image output by the polarization acquisition unit and transmit it to the polarization reconstruction module; The polarization reconstruction module is used to complete image reconstruction, blind pixel processing and polarization image registration based on the original polarization image, and output the initial polarization image group to the temperature drift self-correction module. The temperature drift self-correction module is used to receive the temperature state parameters of the temperature acquisition unit and the initial polarization image group output by the polarization reconstruction module, construct the focal plane temperature drift field by combining the temperature parameters, and realize target elimination temperature drift self-correction based on the background sample weight, and output the corrected polarization image to the polarization phase saliency map generation module. The polarization phase saliency map generation module is used to receive the corrected polarization image output by the temperature drift self-correction module, calculate the polarization parameters, construct the complex polarization phase quantity and generate the polarization phase saliency map, and send the polarization phase saliency map and the corrected polarization image synchronously into the weak target enhancement module. The weak target enhancement module performs multi-scale processing on the polarization phase saliency map and fuses multi-source polarization images to output a weak target enhancement image to the candidate target output module. The candidate target output module extracts candidate regions of targets within the enhanced image and outputs target information; The parameter feedback module receives the target candidate region output by the candidate target output module, the temperature drift residual output by the temperature drift self-correction module, and the saliency map residual output by the polarization phase saliency map generation module, and sends the above information back to the temperature drift self-correction module to realize online updating of correction parameters.

[0009] Furthermore, the polarization acquisition unit of the present invention is a focal plane polarization detector or a time-division polarization component, which acquires images in four polarization directions: 0°, 45°, 90°, and 135°.

[0010] Furthermore, the temperature acquisition unit of the present invention includes at least one of a focal plane temperature sensor, a readout circuit temperature sensor, a cooling end temperature sensor, or a lens barrel temperature sensor; the status parameters acquired by the temperature acquisition unit include focal plane temperature, temperature change rate, readout circuit temperature, cooling end temperature, integration time, gain level, and frame number.

[0011] Furthermore, the polarization reconstruction module of the present invention reconstructs multi-directional polarization sub-images on a unit basis of polarization superpixels, and identifies bad pixels, polarization blind pixels, and instantaneous anomalies based on the grayscale consistency and polarization response consistency within the neighborhood of the superpixel. Among them, bad pixels and polarization blind pixels are replaced by neighborhood interpolation in the same direction, cross-directional constraint interpolation, or temporal median interpolation. After the polarization reconstruction module completes sub-pixel registration, the spatial position error of the same target in different polarization directions is less than 0.1 to 0.5 pixels.

[0012] Furthermore, the temperature drift self-correction module of the present invention predicts the prior temperature drift field in each polarization direction based on the focal plane temperature, temperature change rate, integration time, gain level, and pre-stored temperature drift basis function.

[0013] Furthermore, the temperature drift self-correction module of the present invention also extracts observation residuals through dark reference pixels, edge shading pixels and low saliency background areas; and integrates the target candidate mask after motion compensation, polarization phase saliency map residuals and temperature drift field residuals of the previous frame to construct background sample weights, removes pixels containing real weak targets to avoid the target being incorrectly corrected and deducted, and configures multiple temperature and residual trigger thresholds to control the temperature drift update frequency, and increases the estimation weight of background reference pixels when the temperature changes drastically or the gain switches.

[0014] Furthermore, the polarization phase saliency map generation module of the present invention constructs a complex polarization phase quantity based on the polarization image after temperature drift correction, and estimates the background phase center and phase dispersion of the complex polarization phase quantity in a local background window, thereby generating a polarization phase saliency map.

[0015] Furthermore, the weak target enhancement module of the present invention is provided with at least three processing scales. The first scale is used to highlight point-like or small-spot-like weak targets, the second scale is used to suppress the elongated background response caused by cloud edges, sea-line edges, ground object edges or platform jitter, and the third scale is used to calculate the position stability of candidate targets by combining the sub-pixel displacement compensation results of consecutive frames.

[0016] Furthermore, the weak target enhancement module of the present invention adjusts the fusion weights of the light intensity image, linear polarization image, polarization angle stability image, polarization phase saliency map and temporal stability map according to the temperature drift residual, linear polarization degree reliability and candidate target confidence, to complete adaptive fusion.

[0017] Based on the same inventive concept, this invention also provides a method for enhancing long-range weak polarization phase saliency maps and focal plane temperature drift self-correction, the method comprising the following steps: Step 1: Image and Imaging Status Acquisition: Four sets of images of a distant scene with different polarization directions are acquired using a polarization acquisition unit. Simultaneously, the focal plane, readout circuit, cooling end temperature, integration time, gain level, frame number, and imaging status parameters are acquired and recorded. Step 2: Polarization image preprocessing: Perform superpixel reconstruction, polarization channel registration, bad pixel replacement, polarization blind pixel processing and basic non-uniformity correction on the original polarization sub-image in sequence to obtain the initial polarization image group; Step 3: Construction of focal plane temperature drift field: The focal plane temperature drift field is constructed by combining temperature parameters, dark reference pixels, edge shading pixels, polarization channel residuals and statistical information of low significant background areas. The focal plane temperature drift field includes at least one of the following: bias drift field, gain drift field, row and column fringe drift field and polarization channel relative drift field. Step 4: Target shielding temperature drift self-correction: The target candidate region, polarization phase saliency map residual, and temperature drift field residual of the previous frame are fused to generate background sample weights, so that the region containing the real weak target has reduced weight or does not participate in temperature drift parameter estimation, thus completing the image temperature drift correction. Step 5: Construction of polarization parameters and complex polarization phase: Solve for Stokes parameters based on the polarization image after temperature drift correction, calculate the degree of linear polarization and polarization angle, and construct the complex polarization phase quantity; Step 6: Polarization phase saliency map generation: Statistically analyze the background phase center and phase dispersion of complex polarization phase quantities in the local background window, and generate a polarization phase saliency map by combining angular difference, phase abrupt change degree, polarization angle stability, light intensity and polarization residual difference. Step 7: Multi-scale target enhancement weight generation: Background suppression, point target enhancement, and elongated background suppression are performed on the polarization phase saliency map at multiple spatial scales. Combined with the target temporal stability of sub-pixel displacement compensation in consecutive frames, a target enhancement weight map is generated. Step 8: Adaptive Fusion Output of Multi-Source Images: Adaptively fuse the target enhancement weight map with the multi-polarization images after temperature drift correction to output a weak target enhancement image and extract target-related information; Step 9: Closed-loop parameter feedback update: Feedback the temperature drift field residual, polarization phase saliency map residual, target candidate region, and fusion weight to the next frame of the process, and update the temperature drift self-correction and target enhancement parameters online.

[0018] The beneficial effects of this invention are as follows: 1. This invention does not simply use local contrast of light intensity, linear polarization threshold or Fourier phase spectrum to generate significant regions, but uses complex polarization phase quantity after temperature drift correction to construct polarization phase saliency map, which can reduce pseudo-saliency response caused by polarization channel drift.

[0019] 2. This invention couples focal plane temperature drift field estimation with the target enhancement process. By jointly updating the temperature drift parameters through the target candidate region elimination mechanism, polarization phase saliency map residual feedback and temperature drift field residual feedback, the true weak targets can be avoided from being subtracted during the background correction process.

[0020] 3. This invention simultaneously considers bias drift, gain drift, row and column fringe drift, and polarization channel relative drift, and can adapt to imaging fluctuations of long-distance optoelectronic devices under temperature changes, gain switching, and integration time changes.

[0021] 4. This invention describes the periodicity of the polarization angle by using the complex polarization phase quantity C=ρ·exp(j·2φ), thus avoiding the erroneous enhancement caused by numerical jumps in the polarization angle near 0° and 180°.

[0022] 5. This invention employs multi-scale point target enhancement, slender background structure suppression, and temporal position stability constraints to reduce false alarms caused by cloud edges, sea-sky lines, building edges, ground object thermal boundaries, and platform jitter.

[0023] 6. This invention does not require a large number of labeled samples to train the depth detection network as a necessary condition. It can adapt online under different focal plane temperatures, different detector gains and different background conditions, making it suitable for engineering deployment and real-time processing.

[0024] 7. The present invention outputs quantifiable target parameters such as enhanced image, signal-to-noise ratio, candidate region and confidence level, which can directly support photoelectric monitoring, security alarm, verification display and subsequent target tracking.

[0025] This invention is a method that combines polarization phase saliency mapping, focal plane temperature drift self-correction, target candidate region elimination feedback update, and multi-scale weak target enhancement. It can be used for security monitoring, sea surface target observation, border and perimeter monitoring, long-distance observation of foreign objects on airport runways, low-light photoelectric pod imaging, and other scenarios that require enhancement of point or small-scale weak targets in complex backgrounds. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a block diagram of the polarization phase saliency map and focal plane temperature drift self-correction long-distance weak body enhancement system of the present invention; Figure 2 This is a flowchart of the polarization phase saliency map and focal plane temperature drift self-correction long-distance weak body enhancement method of the present invention; Figure 3 This is a block diagram of the focal plane temperature drift self-correction process of the present invention; Figure 4 This is a flowchart of the polarization phase saliency map generation and target enhancement process of the present invention. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments. These embodiments are used to explain the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, any adaptations or adjustments to the detector type, polarization angle, temperature sampling position, image scale, filtering window, threshold parameters, and output interface without departing from the concept of the present invention should fall within the scope of protection of the present invention.

[0029] Implementation Method 1: Combination Figure 1 , Figure 3 and Figure 4This invention provides a detailed description of the polarization phase saliency map and focal plane temperature drift self-correction long-range weak object enhancement system proposed in this invention. By fully disclosing the method execution flow through this embodiment, those skilled in the art can clearly reproduce the technical solution of this invention. Furthermore, this invention clarifies the problems encountered in long-range weak target enhancement, such as few target pixels, low signal-to-noise ratio, easy generation of false targets due to polarization focal plane temperature drift, easy weakening of the real target by traditional non-uniformity correction, and difficulty in distinguishing polarization targets from temperature drift noise by ordinary saliency enhancement. The polarization phase saliency map and focal plane temperature drift self-correction long-range weak object enhancement system proposed in this invention includes: an imaging optical lens, a polarization acquisition unit, a temperature acquisition unit, an image acquisition module, a polarization reconstruction module, a temperature drift self-correction module, a polarization phase saliency map generation module, a weak target enhancement module, a candidate target output module, and a parameter feedback module. The imaging optical lens is used to receive radiation light from a distant scene and image the scene onto the polarization acquisition unit; The polarization acquisition unit is used to acquire original polarization images in multiple polarization directions; The temperature acquisition unit is used to acquire focal plane temperature-related state parameters; The image acquisition module is used to read the original polarization image output by the polarization acquisition unit and transmit it to the polarization reconstruction module; The polarization reconstruction module is used to complete image reconstruction, blind pixel processing and polarization image registration based on the original polarization image, and output the initial polarization image group to the temperature drift self-correction module. The temperature drift self-correction module is used to receive the temperature state parameters of the temperature acquisition unit and the initial polarization image group output by the polarization reconstruction module, construct the focal plane temperature drift field by combining the temperature parameters, and realize target elimination temperature drift self-correction based on the background sample weight, and output the corrected polarization image to the polarization phase saliency map generation module. The polarization phase saliency map generation module is used to receive the corrected polarization image output by the temperature drift self-correction module, calculate the polarization parameters, construct the complex polarization phase quantity and generate the polarization phase saliency map, and send the polarization phase saliency map and the corrected polarization image synchronously into the weak target enhancement module. The weak target enhancement module performs multi-scale processing on the polarization phase saliency map and fuses multi-source polarization images to output a weak target enhancement image to the candidate target output module. The candidate target output module extracts candidate regions of targets within the enhanced image and outputs target information; The parameter feedback module receives the target candidate region output by the candidate target output module, the temperature drift residual output by the temperature drift self-correction module, and the saliency map residual output by the polarization phase saliency map generation module, and sends the above information back to the temperature drift self-correction module to realize online updating of correction parameters.

[0030] Furthermore, in one specific implementation, such as Figure 1 As shown, the system includes a long-distance target area, an imaging optical lens, a polarization acquisition unit, a temperature acquisition unit, an image acquisition module, a polarization reconstruction module, a temperature drift self-correction module, a polarization phase saliency map generation module, a weak target enhancement module, a candidate target output module, a parameter feedback module, and a display terminal or alarm interface. Furthermore, distant target areas can be dot-like, small-patterned, or weakly textured target areas in a sky background, sea background, complex ground background, or low-light background.

[0031] Furthermore, the polarization acquisition unit is a focal plane polarization detector or a time-division polarization assembly, acquiring images in four polarization directions: 0°, 45°, 90°, and 135°. The focal plane polarization detector is an infrared, near-infrared, visible light, or low-light polarization focal plane array. Each polarization superpixel includes four micro-polarization units at 0°, 45°, 90°, and 135°. For systems without a focal plane structure, a rotating analyzer, a liquid crystal tunable polarizer, or a multi-channel beam-splitting polarization assembly can also be used to acquire images in different polarization directions in a time-division or simultaneous manner.

[0032] Furthermore, the temperature acquisition unit includes a focal plane temperature sensor, a readout circuit temperature sensor, a cooling end temperature sensor, or a lens barrel temperature sensor, used to record state quantities related to temperature drift. The state parameters acquired by the temperature acquisition unit include focal plane temperature, temperature change rate, readout circuit temperature, cooling end temperature, integration time, gain level, and frame number.

[0033] Furthermore, the imaging optical lens and the focal plane polarization detector acquire images of distant scenes, obtaining raw polarimetric maps including at least four polarization directions: 0°, 45°, 90°, and 135°. Simultaneously, the focal plane temperature Tt, temperature change rate dTt / dt, readout circuit temperature, cooling end temperature, integration time τt, gain level gt, and frame number t are recorded. When platform jitter or time-division polarization acquisition occurs, inertial measurement information or inter-frame motion estimation results are recorded for subsequent registration.

[0034] Furthermore, the polarization reconstruction module reconstructs the four-directional polarization sub-map at the polarization superpixel level and identifies bad pixels, polarization blind pixels, and transient outliers based on the grayscale consistency and polarization response consistency within the superpixel neighborhood. For bad pixels and polarization blind pixels, they can be replaced using in-directional neighborhood interpolation, cross-directional constrained interpolation, or temporal median interpolation. Sub-pixel registration is then performed on the four-directional polarization sub-map to ensure that the positional error of the same target in different polarization directions is less than 0.1 to 0.5 pixels.

[0035] Furthermore, the temperature drift self-correction module predicts the prior temperature drift field of the current frame based on the focal plane temperature Tt, the rate of temperature change dTt / dt, the integration time τt, the gain level gt, and the pre-stored temperature drift basis functions. For polarization directions k∈{0°, 45°, 90°, 135°}, the temperature drift field can be expressed as:

[0036] Where Bk,m is the low-rank temperature drift basis function obtained from offline calibration or online learning of the background region of continuous frames, ak,m is the coefficient determined by temperature, temperature change rate, integration time and gain level, rk,row and rk,col represent the row drift term and column drift term respectively, and uk represents the local sparse disturbance term, which is used to describe local thermal shock, local drift of readout circuit or local response drift of micro-polarization unit.

[0037] Furthermore, the temperature drift self-correction module extracts the observation residual Et,k(x,y) from dark reference pixels, edge-shaded pixels, and low-saliency background regions in the current scene. Low-saliency background regions consist of pixels outside the target candidate region of the previous frame, with low local contrast and low polarization phase saliency. To avoid subtracting the real target, this invention employs the following background sample weights:

[0038] Where Mt-1 is the target candidate mask after motion compensation in the previous frame, Rt-1 is the normalized polarization phase saliency map residual, Ht is the normalized temperature drift field residual, and γ1, γ2, and γ3 are positive weighting coefficients. Preferably, γ1 is 2.0 to 5.0, γ2 is 1.0 to 3.0, and γ3 is 1.0 to 3.0. When qb,t is less than 0.20 to 0.35, the corresponding pixel is removed from the temperature drift field background sample; when qb,t is not less than 0.35, the corresponding pixel participates in the temperature drift field parameter update with qb,t.

[0039] Furthermore, the temperature drift self-correction module is also configured with multiple temperature and residual trigger thresholds to control the frequency of temperature drift updates, and to increase the estimation weight of background reference pixels when the temperature changes drastically or the gain switches.

[0040] Specifically, the temperature change rate trigger threshold is preferably between 0.03℃ / s and 0.20℃ / s; the temperature drift field residual trigger threshold is preferably between 1.5 and 3.0 times the current background residual standard deviation σE; the polarization phase significance value of the low-significance background region is less than the sum of the background mean μP and 1.0 to 1.5 times the background standard deviation σP; pixels with polarization channel residuals exceeding 2.0σE are downweighted in the current frame temperature drift update. When the focal plane temperature changes rapidly or the detector gain switches, the system temporarily increases the update frequency of the temperature drift self-correction module and increases the weight of dark reference pixels and edge-shaded pixels in the temperature drift field estimation.

[0041] Furthermore, Stokes parameters S0, S1, and S2 are calculated from the temperature-drift-corrected four-directional polarization image, where S0 = I0 + I90, or S0 = (I0 + I45 + I90 + I135) / 2, S1 = I0 - I90, and S2 = I45 - I135. The degree of linear polarization ρ and the polarization angle φ are then calculated based on S0, S1, and S2, specifically as follows:

[0042] Where ε is a positive number to prevent division by zero. The system sets a linear polarization reliability weight ηρ. When ρ is below 0.03 to 0.08, the polarization phase reliability of the pixel is determined to be low, and the weight of the pixel in the polarization phase saliency map is reduced. Furthermore, such as Figure 3 As shown, in the focal plane temperature drift self-correction process, the original polariton image first undergoes bad pixel replacement and channel registration, and then a priori temperature drift field is formed based on temperature state variables. The temperature drift self-correction module generates observation residuals using dark reference pixels, edge-shaded pixels, and low-salience background regions, and decomposes the observation residuals into global bias terms, gain terms, row and column fringe terms, and polarization channel relative drift terms. For observation residuals that coincide with the target candidate region or polarization phase salience regions, they are used with low weights in the temperature drift field update to avoid the real target being eliminated as temperature drift noise.

[0043] Furthermore, the polarization phase saliency map generation module constructs a complex polarization phase quantity based on the temperature drift corrected polarization image, and estimates the background phase center and phase dispersion of the complex polarization phase quantity in the local background window, thereby generating a polarization phase saliency map.

[0044] Specifically, the polarization phase saliency map generation module constructs the complex polarization phase quantity C = ρ·exp(j·2φ). Since the polarization angle has a 180° periodicity, using 2φ in the complex exponential operation avoids angle jumps near 0° and 180°. Then, the background phase center φB and phase dispersion σφ of the complex polarization phase quantity C are estimated in the local background window, and the polarization phase saliency map is generated according to the following formula:

[0045] Where Cs is the mean of the complex polarization phase at the neighborhood scale s, RI is the intensity residual, RP is the polarization residual, and αρ, β, and χ are non-negative coefficients. Preferably, αρ, β, and χ are normalized to the range of 0 to 1, and satisfy αρ + β + χ = 1. When the linear polarization degree ρ is lower than the reliability threshold, αρ is reduced, and the proportions of intensity residual and temporal stability in the target enhancement weight map are increased.

[0046] Furthermore, such as Figure 4As shown, during the polarization phase saliency map generation and target enhancement process, the four-directional polarization image after temperature drift correction is solved into S0, S1, S2, linear polarization degree ρ, and polarization angle φ. The polarization phase saliency map generation module constructs the complex polarization phase quantity C = ρ·exp(j·2φ) and calculates the background phase center, phase dispersion, and phase abrupt change intensity within the local background window. The weak target enhancement module further combines target pixel area constraints, neighborhood energy constraints, slender background suppression constraints, and temporal position stability constraints to generate a target enhancement weight map.

[0047] Furthermore, the weak target enhancement module processes the polarization phase saliency map at at least three spatial scales. The first scale is used to highlight point-like or small-spotted weak targets; the second scale is used to suppress elongated background responses caused by cloud edges, sea-line boundaries, ground feature edges, or platform jitter; and the third scale is used to calculate the positional stability of candidate targets by combining the sub-pixel displacement compensation results of consecutive frames. The target enhancement weight map can be represented as:

[0048] Wherein, Etarget is the energy response of the neighborhood of the point target, Stime is the temporal position stability, Eline is the response of the elongated background structure, Hnoise is the temperature drift residual or bad element residual, and λ1 to λ5 are weighting coefficients. Preferably, the candidate region area threshold is 2 to 81 pixels, the upper limit of the candidate region aspect ratio is 3 to 5, and the target confidence is improved when the positional drift in consecutive frames is less than 0.5 to 2.0 pixels.

[0049] Furthermore, the weak target enhancement module adjusts the fusion weights of the light intensity image, linear polarization image, polarization angle stability image, polarization phase saliency map, and temporal stability map based on the temperature drift residual, linear polarization degree reliability, and candidate target confidence, thereby completing adaptive fusion.

[0050] Specifically, the weak target enhancement module adjusts the fusion weights of the intensity image, linear polarization image, polarization angle stability image, polarization phase saliency map, and temporal stability map based on the temperature drift residual, linear polarization reliability, and candidate target confidence, satisfying wI+wρ+wφ+wP+wT=1. When the temperature drift residual is high, wP and wT are increased; when the linear polarization reliability is low, wP is decreased and wT and wI are increased. The fused image is then the enhanced weak target image.

[0051] Furthermore, the candidate target output module extracts candidate regions from the enhanced image of weak targets and outputs the target center coordinates, target scale, enhanced signal-to-noise ratio (SNR), confidence level, and alarm information. The enhanced SNRout is obtained by dividing the difference between the mean of the target candidate region and the mean of the neighboring background by the standard deviation of the neighboring background. When the improvement factor of SNRout relative to the unenhanced SNRin is not less than 1.8, and the target confidence level Ct is not less than 0.55 to 0.75, an alarm is output; otherwise, only the candidate region is output for subsequent tracking or manual verification.

[0052] Furthermore, in a preferred embodiment, the area constraint of the target candidate region is determined based on the focal length of the optical system, the pixel size, the estimated target distance, and the target imaging diffusion range. For distant, weak targets, their area in the image is typically smaller than a preset area upper limit, and their neighborhood energy distribution approximates a point diffusion pattern. If the candidate region is elongated or continuous with a large-scale background edge, its target confidence is reduced; if the candidate region's position is stable after motion compensation in consecutive frames, and its polarization phase value remains significantly higher than the background threshold, its target confidence is increased.

[0053] In another embodiment, when the focal plane temperature changes rapidly or the detector gain switches, the system can temporarily increase the update frequency of the temperature drift self-correction module and increase the weight of dark reference pixels and edge-shaded pixels in the temperature drift field estimation; after the temperature drift residual recovers and stabilizes, the weight of low-significant background areas is increased, so that the system can maintain a balance between temperature abrupt changes and scene changes.

[0054] In another embodiment, the system can output a temperature drift correction quality map and a polarization phase saliency map simultaneously with the output of a weak target enhancement image. Operators or subsequent algorithms can use the temperature drift correction quality map to determine whether the current frame is affected by a sudden temperature change, and use the polarization phase saliency map to determine whether candidate targets have polarization phase characteristics different from the background, thereby improving verification efficiency.

[0055] Implementation Method Two: Combination Figure 2 The present invention provides a detailed description of the polarization phase saliency map and focal plane temperature drift self-correction long-range weak target enhancement method proposed in this invention. This method is based on the polarization phase saliency map and focal plane temperature drift self-correction long-range weak target enhancement system described in Embodiment 1. The method includes the following steps: Step 1: Image and Imaging Status Acquisition: The polarization acquisition unit (focal plane polarization detector or time-division polarization component) is used to acquire four sets of images of different polarization directions of the distant scene (0°, 45°, 90° and 135° polarization direction images). The focal plane temperature Tt, temperature change rate dTt / dt, readout circuit temperature, cooling end temperature, integration time τt, gain level gt and frame number t are acquired and recorded simultaneously.

[0056] Step 2: Polarization image preprocessing: Perform superpixel reconstruction, polarization channel registration, bad pixel replacement, polarization blind pixel processing and basic non-uniformity correction on the original polarization sub-image in sequence to obtain the initial polarization image group; the spatial position error of the same target in different polarization directions after registration is preferably less than 0.1 to 0.5 pixels.

[0057] Step 3: Construction of focal plane temperature drift field: The focal plane temperature drift field is constructed by combining temperature parameters, dark reference pixels, edge shading pixels, polarization channel residuals and statistical information of low significant background areas. The focal plane temperature drift field includes at least one of the following: bias drift field, gain drift field, row and column fringe drift field and polarization channel relative drift field. Step 4: Target-masking temperature drift self-correction: The target candidate region, polarization phase saliency map residual, and temperature drift field residual of the previous frame are fused to generate background sample weights, so that the region containing the real weak target is reduced in weight or does not participate in the temperature drift parameter estimation, thereby avoiding the real target being subtracted as temperature drift noise, and completing the image temperature drift correction. Step 5: Construction of polarization parameters and complex polarization phase: Solve the Stokes parameters based on the polarization image after temperature drift correction, calculate the degree of linear polarization and polarization angle, and construct the complex polarization phase C=ρ·exp(j·2φ), where ρ represents the degree of linear polarization, φ represents the polarization angle, and j is the imaginary unit; Step 6: Polarization phase saliency map generation: Statistically analyze the background phase center and phase dispersion of complex polarization phase quantities in the local background window, and generate a polarization phase saliency map by combining angular difference, phase abrupt change degree, polarization angle stability, light intensity and polarization residual difference. Step 7: Multi-scale target enhancement weight generation: Background suppression, point target enhancement, and elongated background suppression are performed on the polarization phase saliency map at multiple spatial scales. Combined with the target temporal stability of sub-pixel displacement compensation in consecutive frames, a target enhancement weight map is generated. Step 8: Adaptive Fusion Output of Multi-Source Images: Adaptively fuse the target enhancement weight map with the multi-type polarization images (light intensity image, linear polarization degree image, polarization angle stability image and polarization phase saliency map) after temperature drift correction to output a weak target enhancement image and extract target-related information, including: weak target enhancement image as well as target candidate region, target center coordinates, target scale, enhanced signal-to-noise ratio and confidence information.

[0058] Step 9: Closed-loop parameter feedback update: Feedback the temperature drift field residual, polarization phase saliency map residual, target candidate region, and fusion weight to the next frame flow, so that the temperature drift self-correction parameters and target enhancement parameters are updated online as the imaging environment changes.

[0059] Furthermore, in one specific implementation, combined with Figure 2 The method is described in detail; S101. Polarization Image and Temperature Status Acquisition. Long-distance scene images are acquired using an imaging optical lens and a focal plane polarization detector, obtaining raw polarimetric images with at least four polarization directions: 0°, 45°, 90°, and 135°. Simultaneously, the focal plane temperature Tt, temperature change rate dTt / dt, readout circuit temperature, cooling end temperature, integration time τt, gain level gt, and frame number t are recorded. When platform jitter or time-division polarization acquisition occurs, inertial measurement information or inter-frame motion estimation results are recorded for subsequent registration.

[0060] S102. Polarization Reconstruction and Basic Correction. The polarization reconstruction module reconstructs the four-directional polarization sub-map at the polarization superpixel level and identifies bad pixels, polarization blind pixels, and transient anomalies based on the grayscale consistency and polarization response consistency within the superpixel neighborhood. For bad pixels and polarization blind pixels, they can be replaced using in-directional neighborhood interpolation, cross-directional constrained interpolation, or temporal median interpolation. Sub-pixel registration is then performed on the four-directional polarization sub-map to ensure that the positional error of the same target in different polarization directions is less than 0.1 to 0.5 pixels.

[0061] S103. Focal Plane Temperature Drift Field Prediction. The temperature drift self-correction module predicts the prior temperature drift field of the current frame based on the focal plane temperature Tt, the rate of temperature change dTt / dt, the integration time τt, the gain level gt, and the pre-stored temperature drift basis functions. For polarization directions k∈{0°, 45°, 90°, 135°}, the temperature drift field can be expressed as:

[0062] In the formula, Bk,m is the low-rank temperature drift basis function obtained by offline calibration or online learning of the background region of continuous frames, ak,m is the coefficient determined by temperature, temperature change rate, integration time and gain level, rk,row and rk,col represent the row drift term and column drift term respectively, and uk represents the local sparse disturbance term, which is used to describe local thermal shock, local drift of readout circuit or local response drift of micro-polarization unit.

[0063] S104. Temperature Drift Field Self-Correction Based on Background Sample Weights. The temperature drift self-correction module extracts the observation residual Et,k(x,y) from dark reference pixels, edge-shaded pixels, and low-saliency background regions in the current scene. Low-saliency background regions consist of pixels outside the target candidate region of the previous frame, with low local contrast and low polarization phase saliency. To avoid subtracting the real target, this invention uses the following background sample weights:

[0064] In the formula, Mt-1 is the target candidate mask after motion compensation in the previous frame, Rt-1 is the normalized polarization phase saliency map residual, Ht is the normalized temperature drift field residual, and γ1, γ2, and γ3 are positive weighting coefficients. Preferably, γ1 is 2.0 to 5.0, γ2 is 1.0 to 3.0, and γ3 is 1.0 to 3.0. When qb,t is less than 0.20 to 0.35, the corresponding pixel is removed from the temperature drift field background sample; when qb,t is not less than 0.35, the corresponding pixel participates in the temperature drift field parameter update with qb,t.

[0065] S105. Temperature drift field threshold control. The temperature change rate trigger threshold is preferably 0.03℃ / s to 0.20℃ / s; the temperature drift field residual trigger threshold is preferably 1.5 to 3.0 times the current background residual standard deviation σE; the polarization phase significance value of the low significant background region is less than the sum of the background mean μP and 1.0 to 1.5 times the background standard deviation σP; pixels with polarization channel residuals exceeding 2.0σE are downweighted in the current frame temperature drift update. When the focal plane temperature changes rapidly or the detector gain switches, the system temporarily increases the update frequency of the temperature drift self-correction module and increases the weight of dark reference pixels and edge-shaded pixels in temperature drift field estimation.

[0066] S106. Polarization Parameter Calculation. Calculate the Stokes parameters S0, S1, and S2 from the four-directional polarization image after temperature drift correction, where S0 = I0 + I90, or S0 = (I0 + I45 + I90 + I135) / 2, S1 = I0 - I90, and S2 = I45 - I135. Calculate the degree of linear polarization ρ and the polarization angle φ based on S0, S1, and S2, specifically as follows:

[0067] Where ε is a positive number to prevent division by zero. The system sets a linear polarization reliability weight ηρ. When ρ is lower than 0.03 to 0.08, the polarization phase reliability of the pixel is determined to be low, and the weight of the pixel in the polarization phase saliency map is reduced.

[0068] S107. Polarization Phase Saliency Map Generation. The polarization phase saliency map generation module constructs the complex polarization phase quantity C = ρ·exp(j·2φ). Since the polarization angle has a periodicity of 180°, using 2φ in the complex exponential operation can avoid angle jumps near 0° and 180°. Subsequently, the background phase center φB and phase dispersion σφ of C are estimated in the local background window, and the polarization phase saliency map is generated according to the following formula:

[0069] In the formula, Cs is the mean of the complex polarization phase at the neighborhood scale s, RI is the light intensity residual, RP is the polarization residual, and αρ, β, and χ are non-negative coefficients. Preferably, αρ, β, and χ are normalized to the range of 0 to 1, and satisfy αρ + β + χ = 1. When the linear polarization degree ρ is lower than the reliability threshold, αρ is reduced, and the proportion of light intensity residual and temporal stability in the target enhancement weight map is increased.

[0070] S108, Multi-scale Weak Target Enhancement. The weak target enhancement module processes the polarization phase saliency map at at least two spatial scales. The first scale is used to highlight point-like or small-patterned weak targets; the second scale is used to suppress elongated background responses caused by cloud edges, sea-line boundaries, ground feature edges, or platform jitter; and the third scale is used to calculate the positional stability of candidate targets by combining the sub-pixel displacement compensation results of consecutive frames. The target enhancement weight map can be represented as:

[0071] In the formula, Etarget represents the energy response of the neighborhood of the point target, Stime represents the temporal position stability, Eline represents the response of the elongated background structure, Hnoise represents the temperature drift residual or bad element residual, and λ1 to λ5 are weighting coefficients. Preferably, the candidate region area threshold is 2 to 81 pixels, the upper limit of the candidate region aspect ratio is 3 to 5, and the target confidence is improved when the positional drift in consecutive frames is less than 0.5 to 2.0 pixels.

[0072] S109. Adaptive Fusion Output. The weak target enhancement module adjusts the fusion weights of the intensity image, linear polarization image, polarization angle stability image, polarization phase saliency map, and temporal stability map based on the temperature drift residual, linear polarization reliability, and candidate target confidence, satisfying wI+wρ+wφ+wP+wT=1. When the temperature drift residual is high, wP and wT are increased; when the linear polarization reliability is low, wP is decreased and wT and wI are increased. The fused image is then the enhanced weak target image.

[0073] S110. Target Output and Feedback Update. The candidate target output module extracts candidate regions from the enhanced image of weak targets and outputs the target center coordinates, target scale, enhanced signal-to-noise ratio (SNR), confidence level, and alarm information. The enhanced SNRout is obtained by dividing the difference between the mean of the target candidate region and the mean of the neighborhood background by the standard deviation of the neighborhood background. When the improvement factor of SNRout relative to the unenhanced SNRin is not less than 1.8, and the target confidence level Ct is not less than 0.55 to 0.75, an alarm is output; otherwise, only the candidate region is output for subsequent tracking or manual verification.

[0074] This invention can be applied to scenarios such as infrared search and tracking, long-distance security monitoring, port and sea surface target observation, long-distance observation of foreign objects on airport runways, early weak target enhancement for forest fires, low-light industrial monitoring, and optoelectronic pod image processing. For different application scenarios, the polarization detection band, lens focal length, image scale, number of temperature drift basis functions, local background window size, target area threshold, phase significance threshold, and temporal stability threshold can be adjusted.

[0075] In summary, this invention eliminates pseudo-differences caused by polarization channel drift through focal plane temperature drift self-correction, constructs a polarization phase saliency map through complex polarization phase quantity, and generates weak target enhancement images through multi-scale background suppression, target area constraint, temporal stability constraint, and feedback background sample weighting. It can improve the visibility and detectability of weak targets at a distance under complex background and focal plane temperature fluctuation conditions.

[0076] The above description of the technical solution provided by the present invention through several specific embodiments is intended to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above-described specific embodiments are not intended to limit the present invention. Any reasonable modifications and improvements to the present invention, reasonable combinations of implementation methods and equivalent substitutions based on the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A long-range weak volume enhancement system with polarization phase saliency map and focal plane temperature drift self-correction, characterized in that, The system includes an imaging optical lens, a polarization acquisition unit, a temperature acquisition unit, an image acquisition module, a polarization reconstruction module, a temperature drift self-correction module, a polarization phase saliency map generation module, a weak target enhancement module, a candidate target output module, and a parameter feedback module. The imaging optical lens is used to receive radiation light from a distant scene and image the scene onto the polarization acquisition unit; The polarization acquisition unit is used to acquire original polarization images in multiple polarization directions; The temperature acquisition unit is used to acquire focal plane temperature-related state parameters; The image acquisition module is used to read the original polarization image output by the polarization acquisition unit and transmit it to the polarization reconstruction module; The polarization reconstruction module is used to complete image reconstruction, blind pixel processing and polarization image registration based on the original polarization image, and output the initial polarization image group to the temperature drift self-correction module. The temperature drift self-correction module is used to receive the temperature state parameters of the temperature acquisition unit and the initial polarization image group output by the polarization reconstruction module, construct the focal plane temperature drift field by combining the temperature parameters, and realize target elimination temperature drift self-correction based on the background sample weight, and output the corrected polarization image to the polarization phase saliency map generation module. The polarization phase saliency map generation module is used to receive the corrected polarization image output by the temperature drift self-correction module, calculate the polarization parameters, construct the complex polarization phase quantity and generate the polarization phase saliency map, and send the polarization phase saliency map and the corrected polarization image synchronously into the weak target enhancement module. The weak target enhancement module performs multi-scale processing on the polarization phase saliency map and fuses multi-source polarization images to output a weak target enhancement image to the candidate target output module. The candidate target output module extracts candidate regions of targets within the enhanced image and outputs target information; The parameter feedback module receives the target candidate region output by the candidate target output module, the temperature drift residual output by the temperature drift self-correction module, and the saliency map residual output by the polarization phase saliency map generation module, and sends the above information back to the temperature drift self-correction module to realize online updating of correction parameters.

2. The polarization phase saliency map and focal plane temperature drift self-correcting long-range weak volume enhancement system according to claim 1, characterized in that, The polarization acquisition unit is a focal plane polarization detector or a time-division polarization component, which acquires images in four polarization directions: 0°, 45°, 90°, and 135°.

3. The polarization phase saliency map and focal plane temperature drift self-correcting long-range weak volume enhancement system according to claim 1, characterized in that, The temperature acquisition unit includes at least one of a focal plane temperature sensor, a readout circuit temperature sensor, a cooling end temperature sensor, or a lens barrel temperature sensor; the status parameters acquired by the temperature acquisition unit include focal plane temperature, temperature change rate, readout circuit temperature, cooling end temperature, integration time, gain level, and frame number.

4. The polarization phase saliency map and focal plane temperature drift self-correcting long-range weak volume enhancement system according to claim 1, characterized in that, The polarization reconstruction module reconstructs multi-directional polarization sub-images on a unit basis of polarization superpixels, and identifies bad pixels, polarization blind pixels, and transient anomalies based on the grayscale consistency and polarization response consistency within the superpixel neighborhood. Bad pixels and polarization blind pixels are replaced by neighborhood interpolation in the same direction, cross-directional constraint interpolation, or temporal median interpolation. After the polarization reconstruction module completes sub-pixel registration, the spatial position error of the same target in different polarization directions is less than 0.1 to 0.5 pixels.

5. The polarization phase saliency map and focal plane temperature drift self-correction long-range weak volume enhancement system according to claim 1, characterized in that, The temperature drift self-correction module predicts the prior temperature drift field in each polarization direction based on the focal plane temperature, temperature change rate, integration time, gain level, and pre-stored temperature drift basis function.

6. The polarization phase saliency map and focal plane temperature drift self-correction long-range weak volume enhancement system according to claim 5, characterized in that, The temperature drift self-correction module also extracts the observation residual through dark reference pixels, edge-shaded pixels, and low-significance background areas; The background sample weights are constructed by fusing the target candidate mask after motion compensation, polarization phase saliency map residual, and temperature drift field residual from the previous frame. Pixels containing real weak targets are removed to avoid the target being incorrectly corrected and deducted. Multiple temperature and residual trigger thresholds are configured to control the temperature drift update frequency. When the temperature changes drastically or the gain switches, the estimation weight of the background reference pixels is increased.

7. The polarization phase saliency map and focal plane temperature drift self-correction long-range weak volume enhancement system according to claim 1, characterized in that, The polarization phase saliency map generation module constructs a complex polarization phase quantity based on the polarization image after temperature drift correction, and estimates the background phase center and phase dispersion of the complex polarization phase quantity in a local background window to generate a polarization phase saliency map.

8. The polarization phase saliency map and focal plane temperature drift self-correcting long-range weak volume enhancement system according to claim 1, characterized in that, The weak target enhancement module is configured with at least three processing scales. The first scale is used to highlight point-like or small-spot-like weak targets. The second scale is used to suppress the elongated background response caused by cloud edges, sea-line edges, ground object edges, or platform jitter. The third scale is used to calculate the position stability of candidate targets by combining the sub-pixel displacement compensation results of consecutive frames.

9. The polarization phase saliency map and focal plane temperature drift self-correction long-range weak volume enhancement system according to claim 8, characterized in that, The weak target enhancement module adjusts the fusion weights of the light intensity image, linear polarization image, polarization angle stability image, polarization phase saliency map, and temporal stability map based on the temperature drift residual, linear polarization degree reliability, and candidate target confidence, thereby completing adaptive fusion.

10. A method for enhancing long-range weak targets based on the polarization phase saliency map and focal plane temperature drift self-correction long-range weak object enhancement system according to any one of claims 1-9, characterized in that, The method includes the following steps: Step 1: Image and Imaging Status Acquisition: Four sets of images of a distant scene with different polarization directions are acquired using a polarization acquisition unit. Simultaneously, the focal plane, readout circuit, cooling end temperature, integration time, gain level, frame number, and imaging status parameters are acquired and recorded. Step 2: Polarization image preprocessing: Perform superpixel reconstruction, polarization channel registration, bad pixel replacement, polarization blind pixel processing and basic non-uniformity correction on the original polarization sub-image in sequence to obtain the initial polarization image group; Step 3: Construction of focal plane temperature drift field: The focal plane temperature drift field is constructed by combining temperature parameters, dark reference pixels, edge shading pixels, polarization channel residuals and statistical information of low significant background areas. The focal plane temperature drift field includes at least one of the following: bias drift field, gain drift field, row and column fringe drift field and polarization channel relative drift field. Step 4: Target shielding temperature drift self-correction: The target candidate region, polarization phase saliency map residual, and temperature drift field residual of the previous frame are fused to generate background sample weights, so that the region containing the real weak target has reduced weight or does not participate in temperature drift parameter estimation, thus completing the image temperature drift correction. Step 5: Construction of polarization parameters and complex polarization phase: Solve for the Stokes parameters based on the polarization image after temperature drift correction, calculate the degree of linear polarization and polarization angle, and construct the complex polarization phase quantity; Step 6: Polarization phase saliency map generation: Statistically analyze the background phase center and phase dispersion of complex polarization phase quantities in the local background window, and generate a polarization phase saliency map by combining angular difference, phase abrupt change degree, polarization angle stability, light intensity and polarization residual difference. Step 7: Multi-scale target enhancement weight generation: Background suppression, point target enhancement, and elongated background suppression are performed on the polarization phase saliency map at multiple spatial scales. Combined with the target temporal stability of sub-pixel displacement compensation in consecutive frames, a target enhancement weight map is generated. Step 8: Adaptive Fusion Output of Multi-Source Images: Adaptively fuse the target enhancement weight map with the multi-polarization images after temperature drift correction to output a weak target enhancement image and extract target-related information; Step 9: Closed-loop parameter feedback update: Feedback the temperature drift field residual, polarization phase saliency map residual, target candidate region, and fusion weight to the next frame of the process, and update the temperature drift self-correction and target enhancement parameters online.

Images