Image cognition interpretation information density evaluation and improvement method

Abstract

The invention discloses an image cognition interpretation information density evaluation and improvement method, and belongs to the technical field of optical imaging application. The method comprises the following steps of: starting from a human eye cognitive mechanism and an interpretation process, extracting an image interpretation quality influence factor and a key parameter and an association relationship between the two; key parameters are comprehensively considered, and a cognitive interpretation information density evaluation model is established; based on a cognitive interpretation information density evaluation model, a processing algorithm is designed to realize fine improvement of image interpretation quality with consideration of transfer function improvement, texture detail maintenance and noise artifact suppression as targets. According to the method, the problem that interpretation quality elements are not clear in the prior art is solved, the technical blank of effective information content quantitative evaluation is filled, the technical bottleneck that artifacts are likely to be caused by traditional processing is broken through, a complete technical system from information density evaluation to interpretation quality improvement is formed, the ground feature classification precision and the target recognition rate are improved, and the method is suitable for popularization and application. And a key support is provided for intelligent application of the optical remote sensing image.

Description

Technical Field

[0001] The present invention relates to an image cognitive interpretation information density evaluation and improvement method, belonging to the technical field of optical imaging applications. Background Art

[0002] In the aviation and aerospace fields, various optical remote sensing images, with their ability to quickly and accurately acquire ground object information, have become indispensable technologies for applications such as nature monitoring, geographic exploration, and environmental surveillance. However, the optical remote sensing imaging process is constrained by multiple inherent factors, including atmospheric interference, system aberrations, sensor noise, and geometric radiation distortion. These factors can lead to image artifacts such as fog, detail degradation, and noise artifacts, severely limiting the effective extraction and application of ground object information.

[0003] With the surge in demand for high-resolution, timely Earth observation, fields such as precise national land monitoring, disaster response, dynamic environmental assessment, and military target identification are placing unprecedented demands on image data clarity, accuracy, fidelity, and reliability. Currently, optical remote sensing imaging links are complex and highly coupled with influencing factors. Existing technologies lack effective information content evaluation methods and enhancements for human-in-the-loop interpretation applications. This results in processed images being susceptible to noise amplification and artifact generation. Improving and maintaining the quality of original images has become a core technical bottleneck hindering accurate interpretation.

[0004] Improving the cognitive interpretation of optical remote sensing imagery is a key prerequisite for unlocking the value of remote sensing big data, directly impacting the effectiveness of major strategic tasks such as global change research, resource management, and national security. While existing technologies have made some progress in areas such as image restoration and super-resolution reconstruction, a comprehensive technical framework, from quantitative evaluation of information density to targeted enhancements of interpretation quality, has yet to be established, making it difficult to meet the urgent need for improved application performance of optical satellite systems.

[0005] High-quality interpretation images can significantly improve object classification accuracy, target recognition rate, and change detection sensitivity. They are the core support for the transformation and upgrading of remote sensing applications from qualitative to quantitative, from static to dynamic, and from manual to intelligent. Breaking through the technical bottlenecks in evaluating and improving the effective information density of optical remote sensing images not only affects the usability and credibility of remote sensing information products, but also represents a strategic imperative for enhancing China's independent spatial information security capabilities and seizing the commanding heights of remote sensing technology. Summary of the Invention

[0006] In order to solve the problems existing in the background technology, the present invention provides a method for evaluating and improving the information density of image cognitive interpretation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for evaluating and improving the information density of image cognitive interpretation, the method comprising the following steps:

[0008] S1: from human eye cognitive mechanism and interpretation process, extract image interpretation quality influencing factors and key parameters and sort out the correlation between them;

[0009] S2: considering the key parameters, an evaluation model of cognitive interpretation information density is established;

[0010] S3: based on the cognitive interpretation information density evaluation model, taking into account the transfer function improvement, texture detail preservation and noise artifact suppression as the goal, the processing algorithm is designed to realize the fine improvement of image interpretation quality.

[0011] Further, the image interpretation quality influencing factors of S1 include the shape, texture, scale, shadow and activity characteristics of the target.

[0012] Further, the key parameters of S1 include ground resolution, dynamic range, transfer function, ringing aliasing and signal-to-noise ratio.

[0013] Further, the transfer function is the on-orbit full-link normalized transfer function H of the optical remote sensing satellite:

[0014] (1)

[0015] In formula (1):

[0016] represents the transfer function contribution of the optical system;

[0017] represents the transfer function contribution of atmospheric transmission;

[0018] represents the transfer function contribution of the detector;

[0019] represents the transfer function contribution of satellite platform motion;

[0020] The transfer function cutoff frequency is , is the pixel size.

[0021] Further, the dynamic range is the effective gray scale number of the image, that is, the total number of gray levels whose gray value pixel ratio exceeds the threshold value in the image, and the effective quantization L of the image effective gray scale number is calculated as follows:

[0022] (2)

[0023] In formula (2):

[0024] represents the effective dynamic of the image.

[0025] Furthermore, the signal-to-noise ratio includes imaging signal-to-noise ratio and quantization signal-to-noise ratio;

[0026] The imaging signal-to-noise ratio for:

[0027] (3)

[0028] In formula (3):

[0029] represents the signal standard deviation;

[0030] represents the standard deviation of imaging noise;

[0031] represents the average electron number of imaging;

[0032] represents device noise;

[0033] Indicates circuit noise;

[0034] represents the charge conversion efficiency;

[0035] The quantized signal-to-noise ratio for:

[0036] (4)

[0037] In formula (4):

[0038] K represents gain;

[0039] represents the standard deviation of quantization noise;

[0040] c represents the quantization interval width adjustment parameter.

[0041] Furthermore, the ringing artifact is distributed in the high-frequency aliasing area, which is expressed as:

[0042] (5)

[0043] In formula (5):

[0044] ALI represents the aliased ringing information power spectrum;

[0045] Indicates the ringing intensity parameter;

[0046] Represents the normalized power spectrum of the physical scene;

[0047] Represents the two-way frequency coordinates in the frequency domain;

[0048] represents an aliasing comb function:

[0049] (6)

[0050] in formula (6):

[0051] represents an impulse function;

[0052] represents a horizontal sampling interval;

[0053] represents a vertical sampling interval;

[0054] m, n represent summation count numbers.

[0055] Further, the cognitive interpretation information density evaluation model S2 is as follows:

[0056] (7)

[0057] in formula (7):

[0058] represents cognitive interpretation information density;

[0059] represents an imaging system focal length;

[0060] represents an imaging distance;

[0061] represents a horizontal pixel physical size;

[0062] represents a vertical pixel physical size;

[0063] represents a presentation symbol, and has no physical meaning;

[0064] represents an image spectral channel number;

[0065] represents a spectral adjustment parameter;

[0066] represents an effective quantization adjustment parameter;

[0067] Further, the processing algorithm S3 is an adaptive prior regularization method, and the optimization model is:

[0068] (8)

[0069] In formula (8), formula (8) is:

[0070] represents an original high-interpretation-quality image;

[0071] and p all represent scene detail fidelity prior fitting parameters, and are optimization variables;

[0072] represents a point spread function;

[0073] represents a minimum seeking;

[0074] represents a regularization parameter matrix;

[0075] represents an imaging system degradation image;

[0076] represents a smoothing 0 norm.

[0077] Compared with the prior art, the beneficial effects of the present application are:

[0078] The present application starts from the human eye cognitive mechanism and interpretation process, systematically combs the correlation between image interpretation quality influencing elements and key parameters, solves the problem of unclear interpretation quality elements in the prior art, constructs a cognitive interpretation information density evaluation model, fills the technical gap of effective information content quantitative evaluation, designs a processing algorithm to realize the collaborative control of transfer function optimization, detail fidelity and noise suppression, breaks through the technical bottleneck of traditional processing that easily causes artifacts, forms a complete technical system from information density evaluation to interpretation quality improvement, improves the feature classification precision and target recognition rate, and provides key support for intelligent application of optical remote sensing images. BRIEF DESCRIPTION OF DRAWINGS

[0079] Figure 1 is a cognitive interpretation influencing parameter decomposition schematic diagram of the present application;

[0080] Figure 2 is an interpretation quality improvement effect schematic diagram. DETAILED DESCRIPTION

[0081] The technical solutions in the present application will be described clearly and completely in the present application combined with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0082] The application discloses an image cognitive interpretation information density evaluation and improvement method, which is a remote sensing image information density evaluation and improvement method for human eye visual interpretation and machine autonomous interpretation.

[0083] S1: when interpreting high-value targets in an image, starting from human eye cognitive mechanism and interpretation process, image interpretation quality influencing factors and key parameters are extracted, and the correlation between the two is analyzed;

[0084] The interpretation quality is not only determined by the information quality (such as definition, noise and other basic attributes) of the image itself, but also a comprehensive concept integrating multiple factors. Specifically, the interpretation quality includes the following factors:

[0085] Image features: objective visual attributes of the target, such as shape, texture and scale;

[0086] Visual perception: subjective feeling and recognition rule of the human eye to the image;

[0087] Professional knowledge: judgment of the interpreter based on field experience (such as geology and military);

[0088] Analysis and calculation: processing and quantitative analysis of image data through algorithms;

[0089] Intelligence interpretation: deep mining and interpretation of image information in combination with task targets.

[0090] This means that when evaluating the interpretation quality, attention should be paid not only to the single physical property of the image, but also to the combination of subjective cognition and objective data, basic features and application targets.

[0091] An image information density evaluation model needs to be constructed, and the construction of the evaluation model needs to follow the principle of "prior information + actual demand + human eye orientation":

[0092] Prior information basis: based on physical light field (such as light propagation rule) and imaging system parameters (such as optical system and detector performance), the physical characteristics of the image itself are determined;

[0093] Combination of application demand: different scenes (such as disaster monitoring and military identification) have different focuses on the image, and the model needs to adapt to specific task targets;

[0094] Human eye interpretation orientation: finally serving the cognition and interpretation of the human eye to the image, and ensuring that the model result conforms to the human visual understanding rule.

[0095] The key parameters in the model represent the effectiveness of the image information from different dimensions:

[0096] Transfer function: determines the integrity of scene information expression, that is, whether the image can completely transfer the detailed features (such as edges and textures) of the ground object, and the better the performance of the transfer function, the less the information loss;

[0097] Dynamic range and effective quantization: determine the completeness of scene dynamic information acquisition, dynamic range reflects the presentation ability of image to light and dark difference, and effective quantization reflects the utilization efficiency of gray level, both of which guarantee the information capture of complex scenes (such as alternating light and dark areas);

[0098] Noise artifact suppression capability: represents the richness of image in retaining original scene information, noise and artifacts (such as ringing effect generated in the processing process) will interfere with the true information, and the stronger the suppression capability, the higher the fidelity of the original ground object information.

[0099] S2: Based on the information theory and the imaging mechanism of optical remote sensing, the existing information density representation model is extended to the physical domain, and the key parameters are comprehensively considered to establish a cognitive interpretation information density evaluation model, which provides theoretical support for the improvement of image interpretation quality;

[0100] S3: Based on the cognitive interpretation information density evaluation model, it is known that the ground image quality improvement is a multi-objective comprehensive processing improvement problem. When the resolution is unchanged, the processing algorithm is designed to realize the fine improvement of image interpretation quality by taking into account the transfer function improvement, texture detail preservation and noise artifact suppression, so as to improve the application ability of optical satellite image.

[0101] Further, the image interpretation quality influencing factors of S1 include the shape, texture, scale, shadow and activity characteristics of the target.

[0102] Further, the key parameters of S1 include ground resolution, dynamic range, transfer function, ringing aliasing and signal-to-noise ratio.

[0103] Further, the transfer function is the on-orbit full-link normalized transfer function H (maximum value normalization) of optical remote sensing satellite:

[0104] (1)

[0105] In formula (1):

[0106] represents the transfer function contribution of the optical system;

[0107] represents the transfer function contribution of atmospheric transmission;

[0108] represents the transfer function contribution of the detector;

[0109] represents the transfer function contribution of satellite platform motion;

[0110] The transfer function cut-off frequency needs to consider the actual pixel physical size, which is , is a pixel size.

[0111] Further, the dynamic range is an image effective gray scale number, that is, a total number of gray levels of pixels with a gray value exceeding a threshold value in the image, and the threshold value is 0.001%; an effective quantization L of the image effective gray scale number is calculated according to the following formula:

[0112] (2)

[0113] In formula (2), μ represents an image average gray value, and σ represents an image effective dynamic.

[0114] represents an image effective dynamic.

[0115] Further, the signal-to-noise ratio includes an imaging signal-to-noise ratio and a quantization signal-to-noise ratio.

[0116] The imaging signal-to-noise ratio (unit: 1) is:

[0117] (3)

[0118] In formula (3), μ represents an image average gray value, and σ represents an imaging noise standard deviation.

[0119] represents a signal standard deviation.

[0120] represents an imaging noise standard deviation.

[0121] represents an imaging average electron number, which is related to a detector area, an average quantum efficiency, a spectral range, an integral time, a target radiance, an optical transmittance, and the like.

[0122] represents a device noise.

[0123] represents a circuit noise.

[0124] represents a charge conversion efficiency.

[0125] The quantization signal-to-noise ratio (unit: 1) is:

[0126] (4)

[0127] In formula (4), μ represents an image average gray value, and σ represents a quantization noise standard deviation.

[0128] K represents a gain.

[0129] represents a quantization noise standard deviation.

[0130] c represents a quantization interval width adjustment parameter.​​

[0131] Further, the ringing artifact is distributed in the high frequency aliasing region, expressed as:

[0132] (5)

[0133] In formula (5):

[0134] ALI represents the aliasing ringing information power spectrum;

[0135] The ringing intensity parameter represents the distortion introduced by the processing algorithm, the principle of the algorithm itself, and the processing strength of the algorithm Related;

[0136] The physical scene normalized power spectrum is expressed as:

[0137] The frequency domain bidirectional frequency coordinate is expressed as:

[0138] The aliasing comb function is expressed as:

[0139] (6)

[0140] In formula (6):

[0141] The impact function is expressed as:

[0142] The horizontal direction sampling interval is expressed as:

[0143] The vertical direction sampling interval is expressed as:

[0144] m, n both represent summation count numbers, which are positive integers.

[0145] Further, the cognitive interpretation information density evaluation model of S2 is as follows:

[0146] (7)

[0147] In formula (7):

[0148] The cognitive interpretation information density is expressed as:

[0149] The imaging system focal length is expressed as:

[0150] The imaging distance is expressed as:

[0151] represents the horizontal direction pixel physical size;

[0152] represents the vertical direction pixel physical size;

[0153] represents the conversion symbol, no physical meaning;

[0154] represents the image spectral channel number;

[0155] represents the spectral adjustment parameter;

[0156] represents the effective quantization adjustment parameter;

[0157] and characterizes the influence degree of the spectral channel number and the effective quantization on the current interpretation task.

[0158] Through the cognitive interpretation information density evaluation model, it can be known that the higher the spatial resolution, the higher the effective quantization, and the higher the transfer function, the richer the interpretation information; the less the aliasing ringing, the higher the signal-to-noise ratio, the higher the interpretation information fidelity, and the stronger the image interpretation application capability, which can provide a theoretical basis for imaging system optimization and image interpretation quality improvement.

[0159] Further, the processing algorithm of S3 is an adaptive prior regularization method, and the optimization model is:

[0160] (8)

[0161] In formula (8):

[0162] represents the original high-interpretation-quality image;

[0163] and p both represent scene detail fidelity prior fitting parameters, which are optimization variables;

[0164] represents the point spread function;

[0165] represents the minimum minimization;

[0166] represents the regularization parameter matrix, which is processed differently in different frequency bands;

[0167] represents the imaging system degradation image;

[0168] represents the smoothing 0 norm.

[0169] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

[0170] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A method for evaluating and improving information density of image cognitive interpretation, characterized by: The method comprises the following steps: S1: Based on the human eye's cognitive mechanism and interpretation process, extract the factors and key parameters that affect image interpretation quality and sort out the correlation between them; S2: Comprehensively consider key parameters and establish a cognitive interpretation information density evaluation model; S3: Based on the cognitive interpretation information density evaluation model, a processing algorithm is designed to improve the image interpretation quality, with the goal of improving transfer function, preserving texture details, and suppressing noise artifacts.

2. The method for evaluating and improving information density of image cognitive interpretation according to claim 1, characterized in that: The factors affecting the image interpretation quality described in S1 include the shape, texture, scale, shadow and activity characteristics of the target.

3. The method for evaluating and improving information density of image cognitive interpretation according to claim 1, characterized in that: The key parameters mentioned in S1 include ground resolution, dynamic range, transfer function, ringing and aliasing, and signal-to-noise ratio.

4. The method for evaluating and improving information density of image cognitive interpretation according to claim 3, characterized in that: The transfer function is the normalized transfer function H of the full link of the optical remote sensing satellite on orbit: (1) In formula (1): represents the transfer function contribution of the optical system; represents the transfer function contribution of atmospheric transmission; represents the transfer function contribution of the detector; represents the transfer function contribution of the satellite platform motion; The transfer function cutoff frequency is , is the pixel size.

5. The method for evaluating and improving information density of image cognitive interpretation according to claim 4, characterized in that: The dynamic range is the effective grayscale number of the image, that is, the total number of grayscale levels in which the proportion of grayscale pixels in the image exceeds the threshold. The effective quantization L calculation formula for the effective grayscale number of the image is as follows: (2) In formula (2): Indicates that the image is effectively dynamic.

6. The method for evaluating and improving information density of image cognitive interpretation according to claim 5, characterized in that: The signal-to-noise ratio includes imaging signal-to-noise ratio and quantization signal-to-noise ratio; The imaging signal-to-noise ratio for: (3) In formula (3): represents the signal standard deviation; represents the standard deviation of imaging noise; represents the average electron number of imaging; represents device noise; Indicates circuit noise; represents the charge conversion efficiency; The quantized signal-to-noise ratio for: (4) In formula (4): K represents gain; represents the standard deviation of quantization noise; c represents the quantization interval width adjustment parameter.

7. The method for evaluating and improving information density of image cognitive interpretation according to claim 6, characterized in that: The ringing artifact is distributed in the high-frequency aliasing area and is expressed as: (5) In formula (5): ALI represents the aliased ringing information power spectrum; Indicates the ringing intensity parameter; Represents the normalized power spectrum of the physical scene; Represents the two-way frequency coordinates in the frequency domain; Represents the aliasing dressing function: (6) In formula (6): represents the impulse function; Indicates the horizontal sampling interval; Indicates the vertical sampling interval; Both m and n represent summation count numbers.

8. The method for evaluating and improving information density of image cognitive interpretation according to claim 7, characterized in that: The cognitive interpretation information density evaluation model described in S2 is as follows: (7) In formula (7): represents the information density of cognitive interpretation; represents the focal length of the imaging system; Indicates imaging distance; Indicates the physical size of the pixel in the horizontal direction; Indicates the physical size of the pixel in the vertical direction; It represents a transfer symbol and has no physical meaning; Indicates the number of image spectral channels; represents the spectrum adjustment parameter; Represents the effective quantitative adjustment parameter.

9. The method for evaluating and improving information density of image cognitive interpretation according to claim 1 or 8, characterized in that: The processing algorithm described in S3 is an adaptive prior regularization method, and the optimization model is: (8) In formula (8): Represents the original high interpretation quality image; and p both represent the scene detail fidelity prior fitting parameters and are optimization variables; represents the point spread function; represents minimization optimization; represents the regularization parameter matrix; Represents the degraded image of the imaging system; represents the smooth zero norm.

Images