Infrared image structure perception representation method based on polarization residual enhancement

Abstract

The application discloses an infrared image structure perception representation method based on polarization residual enhancement, relates to the technical field of infrared image representation, and comprises the following steps: firstly, a mapping graph TCS based on the difference between the maximum polarization channel and other channels is constructed; then, an infrared radiation representation graph LOD based on pixel reverse modulation and feature aggregation is constructed; finally, the difference information of the TCS and the LOD is fused, a structure polarization divergence HPD is constructed, and the polarization distinguishing capability between the target and the background is strengthened. The infrared image structure perception representation method based on polarization residual enhancement provided by the application solves the problems of false detection, missed detection, weak detection and poor detection caused by the existing polarization modeling.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infrared image representation, in particular to an infrared image structure perception representation method based on polarization residual enhancement. BACKGROUND

[0002] The infrared polarization imaging system can not only obtain the intensity information of the target, but also obtain the polarization information related to the surface material, microstructure, geometric shape and the like, which significantly improves the distinguishability of the target and the background. In the field of polarization image analysis, the construction of polarization feature components has always been an important research direction to improve the perceptibility of images and the distinguishability of targets. Effective representation of the polarization characteristics in the polarization image is also a key prerequisite for target recognition, enhancement and detection.

[0003] In 1852, George Gabriel Stokes first proposed the Stokes vector model, which describes the total intensity, linear polarization characteristics, and circular polarization properties of light with four components S0, S1, S2, and S3. This model is the foundation of modern polarization imaging theory. In infrared polarization imaging systems, due to hardware limitations and actual imaging environments, only the first three components S0, S1, and S2 are usually obtained, and then the degree of polarization (DoLP) and angle of polarization (AoP) are derived to enhance structural edges, distinguish surface materials, and perceive weak targets. To address the problem of insufficient structural response of traditional polarization components in target detail expression, researchers have explored the polarization biological vision mechanism. Martin J et al. developed the concept of "polarization distance" based on the insect polarization receptor dynamics model proposed by Bernard and Wehner in 1977, which is used to quantify the distinguishability of objects in polarization imaging and has achieved good experimental results in underwater polarization perception. In 2019, Wu et al. proposed a polarization feature construction strategy based on dominant polarization direction response and high-order structure difference modeling. This method combines physical interpretability and structural sensitivity to construct discriminative polarization features, effectively improving the separability of detail areas in polarization images, and improving the detection accuracy by about 17% in typical infrared target detection tasks. In the direction of polarization information enhancement and fusion, Zhang et al. designed a multi-component feature fusion model that fuses polarization structure residual information and gradient response in 2021. This method effectively improves image contrast and edge clarity by introducing a structure enhancement factor to nonlinearly combine polarization features. In the infrared-polarization image fusion task, this model improves the signal-to-noise ratio of the target area by more than 2.3 times, significantly enhancing the visual saliency of weak targets. With the development of deep learning technology, scholars have begun to combine data-driven models with polarization physical properties to further improve the feature modeling capability of polarization images. In 2022, Li et al. proposed a polarization attention mechanism-based deep network structure Pol-AttNet, which can automatically extract discriminative polarization features and adapt to structural changes in complex scenes. In the multi-scene polarization image classification task, this method improves the accuracy by more than 12%, verifying its robustness and generalization ability in target recognition.

[0004] Although the existing polarization feature component representation method has made certain progress in target enhancement and structure discrimination, there are still many defects and deficiencies. First, the traditional components such as DoLP and AoP have information redundancy in representing the difference of polarization structure, and are sensitive to background noise, which can easily lead to false response. Second, although the newly constructed composite polarization components enhance the structure expression ability, the parameter composition depends on empirical design, lacks unified theoretical support, and has insufficient universality and interpretability. In addition, the polarization feature modeling based on deep learning improves the recognition accuracy, but faces problems such as dependence on training samples and poor model migration, which is difficult to adapt to complex imaging environment.

[0005] Overall, the construction of current polarization feature components has not yet achieved a high degree of unification of physical modeling, structure perception and task adaptability. How to effectively obtain and utilize polarization information and reduce background response interference error by using image polarization characteristics, and develop new representation strategies with the advantages of physical constraints and computational efficiency, is a problem that needs to be solved by those skilled in the art. SUMMARY

[0006] The purpose of the present application is to provide an infrared image structure perception representation method based on polarization residual enhancement, which solves the problems of false detection, missed detection, weak detection and poor detection easily caused in the existing polarization modeling.

[0007] To achieve the above purpose, the present application provides an infrared image structure perception representation method based on polarization residual enhancement, comprising the following steps:

[0008] Step 1, multi-angle polarization image acquisition and preprocessing; using a long-wave infrared polarization camera, infrared images in three polarization directions of 0°, 60° and 120° are collected respectively, and then pixel-level statistical analysis is performed on the collected infrared images in three polarization directions to construct a minimum overlap pixel map OD and a maximum overlap pixel map HOD;

[0009] Step 2, constructing a counter-modulation guide light modulation factor MOD according to the minimum overlap pixel map OD and the maximum overlap pixel map HOD constructed in step 1;

[0010] Step 3, obtaining a high-quality infrared radiation representation map LOD according to the counter-modulation guide light modulation factor MOD constructed in step 2;

[0011] Step 4, constructing a representation model to extract target polarization structure information TCS in the light field;

[0012] Step 5, constructing an infrared polarization characteristic representation physical quantity structure polarization divergence HPD to represent the difference and divergence degree between the target and the background in the polarization structure response in the infrared polarization image from the physical level.

[0013] The expression of the minimum overlap pixel map OD built in step 1 is as follows:

[0014] OD(x, y) = min{I 0° (x, y), I 60° (x, y), I 120° (x, y)} (1)

[0015] The expression of the maximum overlap pixel map HOD built in step 1 is as follows: 0° (x, y), I 60° (x, y), I 120° (x, y)} (2)

[0016] In the formula, I 0° (x, y) represents the light intensity distribution in the 0-degree polarization direction, I 60° (x, y) represents the light intensity distribution in the 60-degree polarization direction, and I 120° (x, y) represents the light intensity distribution in the 120-degree polarization direction.

[0017] The expression of the anti-modulation guide light modulation factor MOD built in step 2 is as follows:

[0018] MOD(x, y) = min{(1-HOD(x, y)), OD(x, y)} (3).

[0019] The process of obtaining the high-quality infrared radiation representation map LOD in step 3 is as follows:

[0020] S31, modulate the light intensity model represented by the Stokes parameter, and the expression is as follows:

[0021]

[0022] In the formula, S0 represents the total light intensity, which is the sum of light intensities in all polarization states, S1 represents the difference between the horizontal polarization light intensity and the vertical polarization light intensity, S2 represents the difference between the +45° polarization light intensity and the -45° polarization light intensity, and θ represents the polarization angle.

[0023] S32, modulate the Lambert's law to obtain the expression of the polarization light intensity changing with the detection angle θ, which is as follows:

[0024]

[0025] In the formula, DoLP represents the linear polarization degree, AOP represents the polarization angle, the first term "1 / 2(1-DoLP)S0" represents the non-polarized light intensity component, reflecting the isotropic part; when DoLP = 0, the light intensity is I(θ) = 1 / 2·S0, which is independent of the angle; the second term "DoLP·S0·cos2 AoP-θ)” corresponds to the polarized light intensity component, showing cos 2 Modulation characteristics: maximum when θ=AoP±π, minimum when θ=AoP±π / 2;

[0026] S33, extract pure non-polarized radiation information, the expression is as follows:

[0027]

[0028] In the formula, I D represents the non-polarized radiation parameter;

[0029] S34, obtain high-quality infrared radiation characterization map LOD according to formula (3)-(6).

[0030] Preferably, the expression of the high-quality infrared radiation characterization map LOD obtained in S34 is as follows:

[0031]

[0032] Preferably, the expression of the target polarization structure information TCS in the light field extracted in step 4 is as follows:

[0033] TCS=1-(I 0° (x,y)∪I 60° (x,y)∪I 120° (x,y))-3*HOD(x,y)(8).

[0034] Preferably, the expression of the infrared polarization characteristic constructed in step 5 is as follows:

[0035]

[0036] In the formula, the difference between |LOD-TCS| reflects the inconsistency of the target and the background in the polarization response behavior, indicating the divergence of the structure; and the value |LOD+TCS| amplifies the region with common polarization response, emphasizing the superposition of the structure strength; the exponential term γ introduces a nonlinear regulation mechanism, enhances the contrast of the high-response region, and suppresses the low-amplitude noise; the denominator S0 as the Stokes zero-order component, characterizes the total intensity information.

[0037] Therefore, the present application adopts the above-mentioned infrared image structure perception characterization method based on polarization residual enhancement, which has the following beneficial effects:

[0038] (1) Enhanced the characterization ability of the target polarization response; by constructing the difference map TCS between the maximum polarization response channel and the remaining channels, the polarization characteristics of the target region are effectively highlighted, overcoming the weak response and direction insensitivity problems in the traditional DoLP method for polarization information extraction, improving the separability and saliency of the target features;

[0039] (2) Improved the robustness and noise suppression ability in complex background; designed the infrared radiation representation graph LOD based on pixel-level reverse modulation and feature aggregation, which can adaptively suppress the background clutter interference of non-target regions, realize robust modeling of fine structures without relying on priori, and enhance the adaptability of the representation algorithm in complex environment;

[0040] (3) Achieved effective enhancement expression of polarization structure difference; the structure polarization divergence HPD constructed by fusing TCS and LOD can comprehensively reflect the structure and energy difference between the target and the background in the infrared polarization image, improve the structure fidelity and perception quality of the infrared intensity image, and show excellent representation effect and universality in various typical natural scenes.

[0041] The technical solutions of the present application will be described in further detail below with the help of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is the overall flowchart of the infrared image structure perception representation method based on polarization residual enhancement of the present application;

[0043] Figure 2 is the detection result graph of different physical parameters of the embodiment of the present application, wherein (a) is the detection result graph of I 0° ; (b) is the detection result graph of I 60° ; (c) is the detection result graph of I 120° ; (d) is the detection result graph of OD; (e) is the detection result graph of HOD; (f) is the detection result graph of MOD;

[0044] Figure 3 is the infrared polarization characteristic multi-parameter representation graph of the embodiment of the present application, wherein (a) is the S0 representation graph; (b) is the pseudo-color graph of S0; (c) is the LOD representation graph; (d) is the pseudo-color graph of LOD; (e) is the DoLP representation graph; (f) is the pseudo-color graph of DoLP; (g) is the TCS representation graph; (h) is the pseudo-color graph of TCS;

[0045] Figure 4 is the 0° polarized graph of the metal hub of the embodiment of the present application;

[0046] Figure 5 is the 60° polarized graph of the metal hub of the embodiment of the present application;

[0047] Figure 6 A 120° deflection map of a metal wheel hub for an embodiment of the present application;

[0048] Figure 7 A LOD map of a metal wheel hub for an embodiment of the present application;

[0049] Figure 8 A TCS map of a metal wheel hub for an embodiment of the present application;

[0050] Figure 9 A structure polarization dispersion HPD map for an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following detailed description of embodiments of the application provided in the accompanying drawings is not intended to limit the scope of the application as claimed, but merely represents selected embodiments of the application. Based upon the embodiments in the present application, all other embodiments that a person of ordinary skill in the art obtains without creative work, are within the scope of protection of the present application.

[0052] Referring to Figures 1-9 A polarization residual enhancement-based infrared image structure perception representation method, comprising the following steps:

[0053] Step 1, multi-angle polarization image acquisition and preprocessing; using a long-wave infrared polarization camera, infrared images in three polarization directions of 0°, 60° and 120° are collected respectively, then pixel-level statistical analysis is performed on the collected infrared images in three polarization directions, and minimum overlap pixel map OD and maximum overlap pixel map HOD are constructed to enhance the stability of the background area and suppress clutter interference; the specific expression is as follows:

[0054] OD(x,y)=min{I 0° (x,y),I 60° (x,y),I 120° (x,y)} (1)

[0055] HOD(x,y)=max{I 0° (x,y),I 60° (x,y),I 120° (x,y)} (2)

[0056] In the formula, I 0° (x,y) represents the light intensity distribution in the 0° polarization direction, I 60° (x,y) represents the light intensity distribution in the 60° polarization direction, and I 120° (x,y) represents the light intensity distribution in the 120° polarization direction.

[0057] Step 2, in the high-light or strong clutter interference scene, the infrared image is easy to produce pixel saturation due to the background being too bright and the limited dynamic range of the sensor, and then cause deviation in subsequent intensity calculation. In order to suppress such errors, according to the minimum overlap pixel map OD and the maximum overlap pixel map HOD constructed in step 1, a demodulation guide light intensity modulation factor MOD is constructed, which effectively suppresses the clutter interference in the non-target area by constraining the intensity of the overlapping pixels at different observation angles, and improves the robustness and accuracy of the reference light intensity estimation; the specific expression is as follows:

[0058] MOD(x,y)=min{(1-HOD(x,y)),OD(x,y)} (3).

[0059] Step 3, according to the demodulation guide light intensity modulation factor MOD constructed in step 2, a high-quality infrared radiation representation map LOD is obtained; the specific process is as follows:

[0060] S31, in order to obtain more stable light intensity feature representation, the light intensity model expressed by the Stokes parameter is modulated, and the expression is as follows:

[0061]

[0062] In the formula, S0 represents the total light intensity, which is the sum of the light intensity of all polarization states, S1 represents the difference between the horizontal polarization light intensity and the vertical polarization light intensity, S2 represents the difference between the +45° polarization light intensity and the -45° polarization light intensity, and θ represents the polarization angle;

[0063] S32, the Malus law is modulated to obtain the expression of the polarization light intensity changing with the detection angle θ, which is as follows:

[0064]

[0065] In the formula, DoLP represents the linear polarization degree, AOP represents the polarization angle, the first term "1 / 2(1-DoLP)S0" represents the non-polarized light intensity component, reflecting the isotropic part; when DoLP=0, the light intensity is I(θ)=1 / 2·S0, which is independent of the angle; the second term "DoLP·S0·cos 2 (AoP-θ)" corresponds to the polarization light intensity component, which shows the cos 2 Modulation characteristics: when θ=AoP±π, it reaches the maximum value, and when θ=AoP±π / 2, it reaches the minimum value;

[0066] In which, the calculation expression of the linear polarization degree DoLP and the polarization angle AOP is as follows:

[0067]

[0068] S33, the orthogonal angle of AOP is selected to eliminate angle dependence, and a more pure non-polarized radiation information is extracted by using the following formula:

[0069]

[0070] In the formula, I D represents the non-polarized radiation parameter;

[0071] S34, considering that the background clutter interference and imaging saturation may cause distortion of the target contrast, direct normalization processing of the three polarized images will face significant challenges. In view of the excellent performance of the anti-modulation guided light modulation factor MOD in improving the information representation accuracy and the image visualization quality, it provides a new idea for efficient perception and downstream processing tasks of infrared images.

[0072] Figure 2 The visualization effect comparison of six different ways in typical scenes is shown, including three-channel polarization images (0°, 60°, 120°), OD, HOD and MOD images. As can be seen from the figure, the polarization channel image has the problems of unclear structure and insufficient contrast. Although the OD image suppresses the background, the information loss is obvious, and the HOD image enhances the target brightness but the clutter interference is still significant. In contrast, the MOD image combines the minimum and maximum overlap features, and balances well between background suppression and structure enhancement. Through the analysis of the image visual results, it can be found that in the scenes shown in the 3rd and 4th rows, the metal structure of the vehicle hub is very sensitive to infrared response, which is an important reference point for evaluating the outline clarity and detail retention capability. The compression problem of light and dark levels of OD and HOD images is more serious. While the MOD image clearly outlines the structure boundary of the metal parts in the hub, retains the infrared response characteristics of the metal material, and avoids the visual whitening or detail submergence phenomenon caused by global brightness enhancement. Overall, in complex background, mountain and metal detail scenes, the MOD image shows superior edge retention and target recognition, and has stronger perception performance.

[0073] Therefore, high-quality infrared radiation representation image LOD is obtained according to formulas (3)-(6); the specific expression is as follows:

[0074]

[0075] Step 4, construct a representation model to extract the target polarization structure information TCS in the light field; the model enhances the discrimination ability of different materials and surface characteristics by emphasizing the change range of light intensity at different polarization angles; the specific expression is as follows:

[0076] TCS = 1-(I 0° (x,y)∪I 60° (x,y)∪I 120°(x, y)) - 3 * HOD(x, y) (8).

[0077] Figure 3 The comparative effects of traditional S0, DoLP and LOD, TCS proposed in this paper in typical scenes are shown, where columns (a) (c) are quantified by labeling the contrast index, columns (e) (g) are quantified by labeling the spatial frequency index, the structural saliency and texture expression ability. Experiments show that LOD enhances the key target outline and brightness change while suppressing high-light interference; while TCS maintains stable polarization response in weak texture areas by constraining non-structural noise. Taking the "building-vegetation-drone" mixed scene as an example, S0 and DoLP have obvious structural blur and noise pollution, while LOD and TCS significantly improve the clarity of the target.

[0078] This method can realize efficient extraction of polarization feature images. The obtained TCS image can not only retain sharp edges, but also stably enhance the structural features of weak texture areas, thereby improving the perception consistency of target structures at different scales.

[0079] Step 5, construct the infrared polarization characteristic physical quantity structure polarization divergence HPD to represent the difference and divergence degree of the target and the background in the polarization structure response in the infrared polarization image from the physical level; wherein the expression of the constructed infrared polarization characteristic physical quantity structure polarization divergence HPD is as follows:

[0080]

[0081] In the formula, the difference between |LOD-TCS| reflects the inconsistency of the target and the background in the polarization response behavior, indicating the divergence in structure; and the value |LOD+TCS| amplifies the area with common polarization response, emphasizing the superposition of structural strength; the exponential term γ introduces a nonlinear regulation mechanism to enhance the contrast of high-response areas and suppress low-amplitude noise; the denominator S0 as the Stokes zero-order component represents the total intensity information, which plays a role in energy constraint and background suppression in the normalization process, ensuring the stability of the physical scale of the polarization divergence.

[0082] Figures 4 to 9 The polarization images of the metal hub at 0°, 60°, and 120°, the LOD image, the TCS image, and the HPD image are shown. The visual display effect of the algorithm processing comprehensively shows that HPD, as a structure polarization divergence representation, integrates polarization directionality difference, channel response structure, and intensity distribution non-uniformity, can effectively depict the separation degree of the target and the background in the polarization space in the infrared polarization image, and provides physical support for weak target detection and high-contrast image enhancement.

[0083] Therefore, the application adopts the above-mentioned infrared image structure perception representation method based on polarization residual enhancement, first constructs a mapping graph TCS based on the difference between the maximum polarization channel and other channels, significantly enhances the polarization response characteristics of the target area, and effectively breaks through the limitations of the traditional DoLP method in polarization information extraction; then constructs an infrared radiation representation graph LOD based on pixel reverse modulation and feature aggregation, effectively suppresses the background clutter interference in the non-target area, and improves the expression stability and recognition degree of the infrared image in the complex background; finally, the difference information of TCS and LOD is fused to construct a structure polarization divergence HPD, and the polarization distinguishing ability between the target and the background is strengthened, so as to show excellent structure fidelity and visual quality in the complex observation environment such as cloud layer, sky, vegetation and mountain.

[0084] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or equivalently replace the technical solutions of the present application, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for polarization residual enhancement based structural-aware representation of infrared images, characterized in that, Comprise the following steps: Step 1, multi-angle polarization image acquisition and pretreatment; using long-wave infrared polarization camera, respectively collect 0, 60 and 120 three polarization direction infrared image, then carry out pixel level statistical analysis to the three polarization direction infrared image collected, construct minimum overlap pixel diagram OD and maximum overlap pixel diagram HOD; Step 2, according to the minimum overlap pixel diagram OD and maximum overlap pixel diagram HOD constructed in step 1, construct anti-modulation guide light intensity modulation factor MOD; Step 3, according to the demodulation of the light guide constructed in step 2, the modulation factor MOD of the light emphasis is obtained, and the high-quality infrared radiation representation map is obtained ; Step 4, construct representation model, extract target polarization structure information TCS in light field; Step 5, construct infrared polarization characteristic representation physical quantity structure polarization divergence HPD, from the physical level, represent the difference and divergence degree between target and background in polarization structure response in infrared polarization image; The expression of target polarization structure information TCS extracted in step 4 in light field is as follows: (8)； wherein represents the light intensity distribution in the 0 degree polarization direction, represents the light intensity distribution in the 60 degree polarization direction, represents the light intensity distribution in the 120 degree polarization direction; The expression of infrared polarization characteristic representation physical quantity structure polarization divergence HPD constructed in step 5 is as follows: (9) wherein, The difference between the two reflects the inconsistency of the target and the background in the polarization response behavior, indicating the divergence of the structure; and the value Amplifying the area with common polarization response, emphasizing the superposition of structural strength; the exponential term Introducing a nonlinear control mechanism to enhance the contrast of high response areas and suppress low amplitude noise; the denominator As the zeroth-order component of Stokes, it represents the total intensity information.

2. The method according to claim 1, wherein, The expression of minimum overlap pixel diagram OD and maximum overlap pixel diagram HOD constructed in step 1 is as follows: (1) (2)。 3. The method of claim 2, wherein the method is a polarization residual enhancement based structural perception representation method for infrared images. The expression of anti-modulation guide light intensity modulation factor MOD constructed in step 2 is as follows: (3)。 4. The method according to claim 3, wherein, High quality infrared radiation signature is obtained in step 3 The process is as follows: S31, modulate the light intensity model expressed by Stokes parameter, expression as follows: (4) wherein I0represents the total light intensity, which is the sum of the light intensity of all polarization states, Ihrepresents the difference between the light intensity of horizontally polarized light and the light intensity of vertically polarized light, I45represents the difference between the light intensity of +45° polarized light and the light intensity of -45° polarized light, represents the polarization angle; S32, modulate Malus law, obtain the intensity of polarized light with the detection angle The expression of the change is as follows: (5) wherein represents the linear polarization degree, represents the polarization angle, the first term represents the non-polarized light intensity component, reflecting the isotropic part; when the light intensity is independent of the angle; the second term corresponds to the polarized light intensity component, exhibiting a modulation characteristic: reaching a maximum value when and a minimum value when . S33, extract pure non-polarized radiation information, expression as follows: (6) wherein denotes unpolarized radiation parameters; S34. Obtain high quality infrared radiation signature according to equations (3) - (6) .

5. The method according to claim 4, wherein, High quality infrared radiation signature obtained in S34 The expression is as follows: (7)。

Images