Image recognition and correction system based on optical model and environmental perception

Abstract

The invention discloses an image recognition and correction system based on an optical model and environment perception, and relates to the technical field of image processing, and the system comprises an environment perception module which is used for collecting multi-medium refractive indexes, interface three-dimensional forms and initial image data through a refractive index sensor, a three-dimensional laser scanner and a visual camera; an environment perception data set is formed through dynamic correction and multi-source fusion; the optical model is constructed on the basis of environment sensing data, the light propagation path is calculated in combination with the interference compensation sub-model, and the interference compensation sub-model can quantify the interference influence by analyzing related data acquired by the environment sensing module for interference factors such as impurities and temperature gradient in a medium, so that the interference compensation precision is improved. The compensation term is generated and fused into the light propagation equation, so that the light propagation path is corrected, the calculation accuracy of the light propagation path is ensured, a firm and reliable data support is provided for subsequent image recognition and correction, and the precision and quality of image recognition and correction are improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image recognition and correction system based on optical models and environmental perception. Background Art

[0002] In the field of image processing technology, accurate image acquisition and processing are fundamental to numerous applications and crucial for subsequent image recognition, analysis, and decision-making. However, in real-world environments, light propagation is often affected by the medium in which it is transmitted. When light passes through the interface of multiple media, it refracts at this interface due to the varying refractive indices of these media. This refraction can distort the captured image, leading to deviations in the target object's shape and position, severely impacting the accuracy and reliability of subsequent image recognition and analysis.

[0003] Existing image recognition and correction methods have obvious shortcomings and fail to fully consider the complex propagation characteristics of light in a multi-media environment. Most of these methods only deal with a single medium or simple imaging distortion, and lack effective solutions to the technical problems caused by refraction at multi-media interfaces. In a single-media environment, light propagation is relatively simple. Traditional image correction methods may be based on some simplified optical models or fixed distortion parameters for adjustment. However, in a multi-media environment, the light propagation path is complex and changeable, and optical phenomena such as refraction, reflection, and scattering at different media interfaces are intertwined, making it difficult for traditional methods to accurately simulate the light propagation process, and unable to effectively eliminate image distortion caused by multi-media refraction, and it is difficult to meet the application requirements of high-precision image recognition. For example, in some image acquisition scenarios involving multi-media environments such as underwater and air interfaces, glass and air interfaces, the images processed by traditional methods still have obvious morphological and positional deviations, and cannot provide accurate and reliable data support for subsequent image recognition. Summary of the Invention

[0004] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide an image recognition and correction system based on optical models and environmental perception. It can obtain key parameters of the multi-media environment through the environmental perception module, including the refractive index of different media, the three-dimensional morphology of the multi-media interface, and the environmental texture and other information, and form an environmental perception data set. Based on these data, the multi-media light field refraction model construction module constructs an adaptive light field refraction model, adjusts the dimension according to the type of medium, and incorporates the interface morphology parameters. It uses ray tracing to simulate the light propagation process. The light propagation path calculation module uses the environmental perception data set and the multi-media light field refraction model as the basis, and uses a numerical discrete method to solve the light propagation path. , and establishes an interference compensation sub-model to correct interference factors such as impurities and temperature gradients in the medium. The target object light field reconstruction module reconstructs the target object light field distribution by inverse ray tracing based on the calculation results of the light propagation path, and optimizes it through signal optimization processing methods. Finally, the image recognition and correction module uses a recognition model based on deep learning to recognize and process the reconstructed image, and at the same time corrects the original distorted image in combination with the reconstructed image, and verifies the correction results through relevant indicators, which improves the system's adaptability to multi-media environments, ensures the accuracy of light propagation path calculation, provides reliable data support for image recognition and correction, and has wide application adaptability.

[0005] To solve the above technical problems, the present invention provides the following technical solution: an image recognition and correction system based on optical models and environmental perception, the system comprising: Environmental perception module: This module collects multi-media refractive index, interface 3D morphology, and initial image data through refractive index sensors, 3D laser scanners, and visual cameras. It then forms an environmental perception dataset through dynamic correction and multi-source fusion. Multi-medium light field refraction model construction module: Based on the environmental perception dataset, it combines Maxwell's equations and Fresnel's law to build an adaptation model. Through ray tracing, it simulates the propagation direction, intensity changes, and random optical phenomena of light, and describes the laws of light propagation in multi-medium. Light propagation path calculation module: Based on the environmental perception dataset and light field refraction model, it uses a numerical discrete method to solve the light propagation trajectory. Combined with the interference compensation sub-model, it corrects for the effects of impurities and temperature gradients to ensure accurate and continuous path calculation. Target object light field reconstruction module: This module traces the light path backwards through inverse ray tracing, integrates light parameters to construct the target's three-dimensional light field distribution, optimizes the light field data through energy attenuation and scattering interference compensation, and converts it into a standard image format; Image recognition and correction module: Utilizes a deep learning-based recognition model to perform target recognition on standard light field images. Combining the deviation between the standard image and the original distorted image, it eliminates refraction distortion through geometric transformation and grayscale correction, and verifies and outputs the correction results.

[0006] Furthermore, the environmental perception module includes a refractive index sensor, a three-dimensional laser scanner and a visual camera, which are used to collect basic data of the multi-media environment. The refractive index sensor measures the refractive index of different media in real time based on the interference principle or the surface plasmon resonance principle and transmits the measurement data. At the same time, it can correct the measurement results according to changes in ambient temperature and pressure. The three-dimensional laser scanner scans the multi-media interface to obtain three-dimensional point cloud data of the interface. By denoising, aligning and modeling the three-dimensional point cloud data, a three-dimensional morphological model including interface shape, curvature and flatness parameters is generated. The visual camera collects environmental texture and initial image information of the target object, and integrates it with the data obtained by the refractive index sensor and the three-dimensional laser scanner to form an environmental perception data set.

[0007] Furthermore, the multi-medium light field refraction model construction module constructs a multi-medium light field refraction model based on the acquired medium refractive index data and interface morphology data, combined with Maxwell's equations and Fresnel's law, adjusts the model dimension according to the type of medium, and incorporates the interface morphology parameters as boundary conditions into the model construction process. At the same time, a ray tracing method is used to simulate the light propagation process in the model, discretizes the light into several light segments, calculates the propagation direction of each light segment in different media based on the multi-medium interface light refraction direction formula, and calculates the light intensity change through the light propagation intensity attenuation formula, and establishes a probability model of light reflection, refraction, and scattering at the interface, and describes the random optical phenomena in the light propagation process through probabilistic simulation.

[0008] Furthermore, the multi-medium light field refraction model construction module calculates the propagation direction of each light segment in different media based on the multi-medium interface light refraction direction formula, and the calculation formula is: ,in is the angle of refraction of the light in the second medium, is the refractive index of the first medium, extracted from the environmental perception dataset, is the refractive index of the second medium, extracted from the environmental perception dataset, is the angle of incidence of the light in the first medium, is the distance from the incident point of the light to the center of curvature of the interface, is the curvature radius of the medium interface, is the interface curvature correction term, and its calculation formula is: , when the plane interface ; The change in light intensity is calculated using the light propagation intensity attenuation formula, which is: ,in, is the final intensity of the light after propagation, is the initial intensity of the light, is the number of layers of medium that the light traverses, It is The absorption coefficient of the layer medium, It is The scattering coefficient of the layer medium, The light is The propagation distance in the layer medium, It is The reflectivity of a medium interface is calculated based on Fresnel's law combined with the medium's refractive index. The calculation formula is: ,in 、 Respectively The refractive index of the media on both sides of the interface.

[0009] Furthermore, the light propagation path calculation module takes the environmental perception data set as input and the multi-medium light field refraction model as the basis, and adopts a numerical discrete method to solve the light propagation path. Specifically, the multi-medium space is divided into tiny grid units according to a preset accuracy, and the light propagation equation is discretized in each grid unit. After substituting the medium refractive index and interface morphology parameters, the light propagation trajectory in the unit is determined by iterative calculation through the multi-medium space light propagation trajectory discretization formula. When the light passes through the medium interface, the refraction parameters in the multi-medium light field refraction model are called to update the light propagation direction and speed. At the same time, an interference compensation sub-model is established for the impurities and temperature gradient interference factors in the medium. By analyzing the impurity distribution data and temperature field data obtained by the environmental perception module, the impact of the interference factors on the light propagation is quantified, and the compensation terms are generated and incorporated into the light propagation equation to correct the light propagation path.

[0010] Furthermore, the light propagation path calculation module determines the light propagation trajectory within the unit through iterative calculation of the multi-medium space light propagation trajectory discretization formula, which is: ,in, The light is The position vector of each discrete unit, The light is The position vector of each discrete unit, The light is The propagation speed of a discrete unit in the medium is calculated as follows: ,in is the speed of light in vacuum, is the refractive index of the medium where the unit is located, is the discrete time step, The light is The propagation direction unit vector in discrete units, is the light position offset vector caused by interference factors; at the same time, an interference compensation sub-model is established for the interference factors of impurities and temperature gradients in the medium, and its model formula is: ,in, is the light position offset vector caused by interference, is the unit vector of the interference offset direction, is the impurity interference coefficient, is the impurity concentration in the medium, is the temperature gradient interference coefficient, is the medium temperature gradient, is the distance the light travels in the current medium unit.

[0011] Furthermore, the target object light field reconstruction module is based on the calculation results of the light propagation path. It uses the inverse ray tracing method to start from the imaging plane of the image acquisition device, and reversely trace the propagation path of each imaging light until it reaches the surface of the target object. By integrating the coordinates of the reversely traced light, superimposing the intensity and calibrating the phase, the three-dimensional light field distribution of the target object containing light field intensity, phase, and direction parameters is constructed. At the same time, in view of the energy attenuation and scattering interference in the light propagation process, the reconstructed three-dimensional light field distribution is optimized using a signal optimization method. By constructing an attenuation function and a scattering kernel, the light energy loss is compensated and the phase deviation is corrected. The light field data is denoised and the optimized light field data is converted into a standard image data format.

[0012] Furthermore, the target object light field reconstruction module compensates for light energy loss and corrects phase deviation by constructing an attenuation function and a scattering kernel. The phase deviation correction formula is: ,in, is the corrected light field phase, is the raw phase value extracted from the light field data, is the number of medium layers that the light passes through from the target object to the imaging plane, It is The refractive index of the layer medium, is the wavelength of light, which is preset to the center wavelength of visible light. The light is The propagation distance in the layer medium, It is Phase deviation caused by scattering in the layer medium.

[0013] Furthermore, the image recognition and correction module adopts a recognition model based on deep learning to recognize and process the standard format image data output by the target object light field reconstruction module, and is trained based on a data set containing distorted images and corrected images of multi-media environments. By extracting target object features and realizing target classification, positioning and attribute recognition, the recognition result is output; at the same time, combined with the generated standard format image data, the original distorted image is corrected, and the original distorted image is aligned with the standard format image through image registration operation, and the geometric deviation and grayscale deviation between the two are calculated. The geometric parameters and pixel values ​​of the original distorted image are adjusted by geometric transformation and grayscale correction. After the correction is completed, the correction result is verified by peak signal-to-noise ratio and structural similarity index indicators.

[0014] Compared with existing technologies, this image recognition and correction system based on optical models and environmental perception has the following beneficial effects: 1. The present invention constructs an optical model based on environmental perception data and calculates the light propagation path in combination with an interference compensation sub-model. The interference compensation sub-model can quantify the interference effects of interference factors such as impurities and temperature gradients in the medium by analyzing the relevant data obtained by the environmental perception module, generate compensation terms and incorporate them into the light propagation equation, thereby correcting the light propagation path, ensuring the accuracy of the light propagation path calculation, providing solid and reliable data support for subsequent image recognition and correction, and helping to improve the accuracy and quality of image recognition and correction.

[0015] 2. The present invention preferentially obtains multi-media environmental data through the environmental perception module, and constructs an adaptive multi-media light field refraction model based on this data. It fully considers the complexity and diversity of the actual multi-media environment, avoids the problem of disconnection between the model and the actual environment in traditional methods, and enables the system to better adapt to different multi-media scenarios.

[0016] Other advantages, objects and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be learned from the practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0018] Figure 1 This is a structural diagram of the image recognition and correction system based on optical model and environmental perception; Figure 2 Flowchart of the image recognition and correction system based on optical model and environmental perception. DETAILED DESCRIPTION

[0019] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0020] The present invention provides an image recognition and correction system based on optical model and environmental perception, such as Figure 1 As shown in the figure, it includes an environment perception module, a multi-media light field refraction model construction module, a light propagation path calculation module, a target object light field reconstruction module and an image recognition and correction module. The environment perception module obtains key parameters of the multi-media environment, including the refractive index of different media, the three-dimensional morphology of the multi-media interface and the environmental texture and other information, and forms an environment perception data set. Based on these data, the multi-media light field refraction model construction module constructs an adaptive light field refraction model, adjusts the dimension according to the type of medium, and incorporates the interface morphology parameters, and uses ray tracing to simulate the light propagation process. The light propagation path calculation module is based on the environment perception data set and the multi-media light field refraction model, and uses a numerical discrete method to solve the light field refraction model. The light propagation path is calculated based on the image processing algorithm, and an interference compensation sub-model is established to correct the interference factors such as impurities and temperature gradients in the medium. The target object light field reconstruction module reconstructs the target object light field distribution by inverse ray tracing based on the light propagation path calculation results, and optimizes it through signal optimization processing methods. Finally, the image recognition and correction module uses a recognition model based on deep learning to recognize and process the reconstructed image, and corrects the original distorted image in combination with the reconstructed image. The correction results are verified by relevant indicators, which improves the system's adaptability to multi-media environments, ensures the accuracy of light propagation path calculation, provides reliable data support for image recognition and correction, and has wide application adaptability.

[0021] Example 1 This embodiment is applied to the detection scenario of precision equipment protected by glass covers in industrial production workshops, and solves the problem of equipment image distortion caused by refraction at the air-glass-air three-medium interface.

[0022] like Figure 2As shown in the figure, the environmental perception module is started, and the refractive index sensor is calibrated to adapt to the measurement requirements of air and glass to ensure measurement accuracy. The scanning resolution of the 3D laser scanner is configured to be 0.05mm, and the scanning range covers the glass cover and the internal equipment area. The focal length of the visual camera is adjusted to clearly capture the equipment details, and the exposure time is set to 1 / 200s to avoid interference from workshop light. The standard refractive index data of air and industrial glass are loaded, the core processing unit parameters are initialized, the grid division accuracy is set to 0.3mm, the inverse ray tracing iteration termination condition is an error ≤ 0.008mm, and the recognition model inference threshold is 0.85.

[0023] The refractive index sensor measures the refractive index of the workshop air and the glass cover at a frequency of 0.5Hz to generate a dual-medium refractive index distribution map; the 3D laser scanner scans the inner and outer surfaces of the glass cover (the air-glass and glass-air interfaces) to obtain 3D point cloud data of the interfaces, which is then processed to generate a 3D morphological model that includes the curvature and flatness of the interface; the visual camera captures the initial image of the equipment in the glass cover at a frame rate of 25fps, extracts features such as the surface texture and contour of the equipment, and integrates the three types of data through data fusion processing to form and store an environmental perception data set.

[0024] The environmental perception data set is read from the system database, and the refractive index distribution data of air and glass, as well as the morphological parameters of the internal and external interfaces of the glass cover, are extracted. Combining Maxwell's equations with Fresnel's law, a light field refraction model suitable for the air-glass-air three-medium environment is constructed. The model dimensions are adjusted according to the three-medium structure, and the morphological parameters of the internal and external interfaces of the glass cover are included as boundary conditions in the model construction process. The initial light parameters (simulating the light emission angle and initial intensity of the visual camera lens) are input, and ray tracing is used to simulate the propagation process of light in the three media: based on the multi-medium interface light refraction direction formula Calculate the refraction angle of light at the air-glass and glass-air interfaces to determine the direction of light propagation; use the light propagation intensity attenuation formula ( is the final intensity of the light after propagation, is the initial intensity of the light, is the number of dielectric layers that the light passes through, For the The absorption coefficient of the layer medium, It is The scattering coefficient of the layer medium, For the light The propagation distance in the layer medium, For the The intensity variation of light propagating in each medium is calculated based on the reflectivity of each medium interface. At the same time, a probability model of reflection, refraction, and scattering of light at the interface is established. The random optical phenomena in the process of light propagation are described through probabilistic simulation. A simulation data set containing the coordinates of the light segment, propagation direction, and intensity is generated to verify the adaptability of the model to the current industrial detection environment.

[0025] The environmental perception data set and the constructed three-medium light field refraction model are imported into the light propagation path calculation module. The air-glass-air multi-medium space is divided into small grid units according to the preset accuracy using the numerical discretization method. The light propagation equation is discretized in each grid unit. After substituting the medium refractive index and interface morphology parameters, the light propagation trajectory in the unit is determined through iterative calculation. In the process, the discretization formula of the light propagation trajectory in the multi-medium space is used. 、 The light is 、 The position vector of each discrete unit, For the light The propagation speed of a discrete unit in the medium is is the discrete time step, For the light The propagation direction unit vector in discrete units, Update the light position (the light position offset vector caused by interference factors). When the light passes through the air-glass and glass-air interfaces, the refraction parameters in the multi-medium light field refraction model are called to update the light propagation direction and speed in real time. At the same time, for the interference factors of the tiny impurity distribution inside the glass and the temperature gradient of the workshop, the impurity distribution data and temperature field data obtained by the environmental perception module are analyzed, and the interference compensation sub-model offset formula is used. ( is the unit vector of the interference offset direction, is the impurity interference coefficient, is the impurity concentration in the medium, is the temperature gradient interference coefficient, is the medium temperature gradient, The influence of interference factors on light propagation is quantified (the propagation distance of light in the current medium unit), compensation terms are generated and incorporated into the light propagation equation, the light propagation path is corrected, and through iterative calculation, the complete light propagation path from the device surface to the camera imaging plane is obtained, and the coordinate trajectory and optical parameters of each light are output.

[0026] The inverse ray tracing method starts from the imaging plane of the image acquisition device and traces the propagation path of each imaging light in reverse until it reaches the surface of the target object. By integrating the coordinates of the reversely traced light, superimposing the intensity and calibrating the phase, the three-dimensional light field distribution of the target object containing the light field intensity, phase, and direction parameters is constructed. At the same time, in view of the energy attenuation and scattering interference during the light propagation process, the signal optimization method is used to optimize the reconstructed three-dimensional light field distribution. By constructing the attenuation function and scattering kernel, the light energy loss is compensated and the phase deviation is corrected. The phase deviation correction formula is: ,in, is the corrected light field phase, is the raw phase value extracted from the light field data, is the number of medium layers that the light passes through from the target object to the imaging plane, It is The refractive index of the layer medium, is the wavelength of light, which is preset to the center wavelength of visible light. The light is The propagation distance in the layer medium, It is The phase deviation caused by layer medium scattering is eliminated, and the light field data is denoised, and the optimized light field data is converted into a standard image data format to clearly present the surface details of the device.

[0027] The standard image is input into the recognition model to identify defects such as scratches (length 2mm) and local deformation on the equipment surface, and the defect category, location and confidence level are output. The original distorted image (defect edge blur and position offset) is aligned and corrected, and the distortion is eliminated through geometric transformation and grayscale adjustment. After verification by PSNR (35dB) and SSIM (0.95), the corrected image and defect recognition results are stored to complete the inspection and correction of industrial equipment.

[0028] Example 2 This embodiment is applied to underwater biological detection scenarios in freshwater lakes. It is necessary to solve the image distortion problem caused by refraction at the air-water dual-medium interface and achieve accurate recognition and imaging correction of underwater fish targets.

[0029] First, the environmental perception module was started and the refractive index sensor was calibrated to ensure that its measurement accuracy error in freshwater environments was less than 0.0001. The scanning resolution of the 3D laser scanner was configured to 0.1mm, and the scanning range covered the underwater area of ​​1-5m. The focal length of the visual camera was adjusted to the optimal value for underwater imaging, and the exposure time was set to 1 / 100s and the white balance was set to underwater mode. At the same time, the standard refractive index data of freshwater at 25°C and standard atmospheric pressure was loaded, the parameters of the system core processing unit were initialized, the multi-media space grid division accuracy was set to 0.5mm, the iterative termination condition for inverse ray tracing was ray tracing error ≤ 0.01mm, and the inference confidence threshold of the image recognition model was 0.8.

[0030] The refractive index sensor continuously measures the refractive index of fresh water in the detection area at a frequency of 1Hz, generating a refractive index distribution map, and dynamically corrects the measured values ​​based on the lake's on-site temperature (22°C) and atmospheric pressure (100.8kPa). The 3D laser scanner performs an all-around scan of the lake surface (air-water interface), acquiring 3D point cloud data of the interface, which is then processed through denoising and registration to generate a 3D morphological model of the lake surface. The visual camera captures initial images of the underwater environment and fish at a frame rate of 30fps, extracts initial features such as texture and color from the image, and integrates the corrected refractive index data, lake surface morphological model, and initial image data through data fusion processing to form an environmental perception dataset, which is stored in the system database.

[0031] The environmental perception dataset is read from the system database, and the freshwater refractive index distribution data and lake surface morphological parameters are extracted. Combined with Maxwell's equations and Fresnel's law, an air-water dual-medium light field refraction model is constructed. The initial light parameters (simulating the light emission angle and initial intensity of the visual camera lens) are input, and ray tracing is used to simulate the propagation process of light in the air-water medium. A simulated dataset containing light segment coordinates, propagation direction, and intensity is generated to verify the adaptability of the model to the actual underwater environment.

[0032] The environmental perception dataset and the constructed dual-medium light field refraction model are imported into the light propagation path calculation module. The air-water space is divided into tiny grid cells using a numerical discretization method, and the light propagation equation is solved within each cell. When light passes from the air to the water medium interface, the refraction parameters in the model are called to update the propagation direction and speed. At the same time, the distribution of tiny impurities in the water (such as plankton) is analyzed, and the impact of impurities on light propagation is quantified by analyzing the environmental perception data. Compensation terms are generated and incorporated into the light propagation equation to correct the light trajectory. Through iterative calculation, the complete light propagation path from the underwater fish target to the camera imaging plane is obtained, and the coordinate trajectory and optical parameters of each light ray are output.

[0033] Based on the light propagation path data, the inverse ray tracing method is used to trace the light from each pixel point on the camera imaging plane back to the surface of the underwater fish, and record the intensity and phase information of each traced light. All traced light rays are integrated to construct a three-dimensional light field matrix of the fish target. In view of the energy attenuation and impurity scattering interference of light propagating in water, the three-dimensional light field matrix is ​​optimized using signal optimization processing methods to compensate for energy loss and correct phase deviation. At the same time, the light field data is denoised and converted into a standard image in RGB format.

[0034] The standard image is input into the trained deep learning-based recognition model. The model extracts features and outputs the fish category (such as crucian carp, carp), position coordinates and confidence level. If the recognition confidence level of a fish target is 0.75 (lower than the preset threshold of 0.8), the model returns to the light field reconstruction module to adjust the denoising parameters and regenerate the standard image. After the second recognition, the confidence level is increased to 0.86. The image registration process is called to align the original underwater distorted image (with fish body stretching and edge blur) with the standard image. The geometric deviation and grayscale deviation of the two are calculated. The fish body stretching distortion is corrected by geometric transformation. The image brightness deviation is adjusted through grayscale mapping to generate a corrected clear image. Finally, the correction effect is verified by PSNR (32dB) and SSIM (0.92). The corrected image and recognition results are stored in the system database to complete underwater target detection and correction.

[0035] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. Image recognition and correction system based on optical model and environmental perception, characterized by: The system comprises: Environmental perception module: This module collects multi-media refractive index, interface 3D morphology, and initial image data through refractive index sensors, 3D laser scanners, and visual cameras. It then forms an environmental perception dataset through dynamic correction and multi-source fusion. Multi-medium light field refraction model construction module: Based on the environmental perception dataset, the module combines Maxwell's equations and Fresnel's law to construct a multi-medium light field refraction model. Through ray tracing, the module simulates the direction of light propagation, intensity changes, and random optical phenomena, and describes the laws of light propagation in multi-medium. Light propagation path calculation module: Based on the environmental perception dataset and the multi-medium light field refraction model, it uses a numerical discrete method to solve the light propagation path. Combined with the interference compensation sub-model, it corrects for the effects of impurities and temperature gradients to ensure accurate and continuous path calculation. Target object light field reconstruction module: This module traces the light path backwards through inverse ray tracing, integrates light parameters to construct the target's three-dimensional light field distribution, optimizes the light field data through energy attenuation and scattering interference compensation, and converts it into a standard image data format; Image recognition and correction module: Utilizes a deep learning-based recognition model to perform target recognition on standard image data. Combining the deviation between the standard image and the original distorted image, it eliminates refraction distortion through geometric transformation and grayscale correction, and verifies and outputs the correction results.

2. The image recognition and correction system based on optical model and environmental perception according to claim 1, characterized in that: The environmental perception module includes a refractive index sensor, a three-dimensional laser scanner and a visual camera, which are used to collect basic data of a multi-media environment. The refractive index sensor measures the refractive index of different media in real time based on the principle of interference or surface plasmon resonance and transmits the measurement data. At the same time, it can correct the measurement results according to changes in ambient temperature and pressure. The three-dimensional laser scanner scans the multi-media interface to obtain three-dimensional point cloud data of the interface. By denoising, aligning and modeling the three-dimensional point cloud data, a three-dimensional morphological model containing interface shape, curvature and flatness parameters is generated. The visual camera collects environmental texture and initial image data of the target object, and integrates it with the data obtained by the refractive index sensor and the three-dimensional laser scanner to form an environmental perception data set.

3. The image recognition and correction system based on optical model and environmental perception according to claim 1, characterized in that: The multi-medium light field refraction model construction module constructs a multi-medium light field refraction model based on the acquired medium refractive index data and interface morphology data, combined with Maxwell's equations and Fresnel's law, adjusts the model dimensions according to the type of medium, and incorporates the interface morphology parameters as boundary conditions into the model construction process. At the same time, a ray tracing method is used to simulate the light propagation process in the model, discretizes the light into several light segments, calculates the propagation direction of each light segment in different media based on the multi-medium interface light refraction direction formula, calculates the light intensity change through the light propagation intensity attenuation formula, and establishes a probability model for the reflection, refraction, and scattering of light at the interface, and describes the random optical phenomena in the light propagation process through a probabilistic simulation method.

4. The image recognition and correction system based on optical model and environmental perception according to claim 3, characterized in that: The multi-medium light field refraction model construction module calculates the propagation direction of each light segment in different media based on the multi-medium interface light refraction direction formula, and the calculation formula is: ,in is the angle of refraction of the light in the second medium, is the refractive index of the first medium, extracted from the environmental perception dataset, is the refractive index of the second medium, extracted from the environmental perception dataset, is the angle of incidence of the light in the first medium, is the distance from the incident point of the light to the center of curvature of the interface, is the curvature radius of the medium interface, is the interface curvature correction term, and its calculation formula is: , when the plane interface ; The change in light intensity is calculated using the light propagation intensity attenuation formula, which is: ,in, is the final intensity of the light after propagation, is the initial intensity of the light, is the number of layers of medium that the light traverses, It is The absorption coefficient of the layer medium, It is The scattering coefficient of the layer medium, The light is The propagation distance in the layer medium, It is The reflectivity of a medium interface is calculated based on Fresnel's law combined with the medium's refractive index. The calculation formula is: ,in 、 Respectively The refractive index of the media on both sides of the interface.

5. The image recognition and correction system based on optical model and environmental perception according to claim 1, characterized in that: The light propagation path calculation module uses the environmental perception data set as input and the multi-medium light field refraction model as its basis, and solves the light propagation path using a numerical discretization method. Specifically, the multi-medium space is divided into tiny grid cells according to a preset accuracy, and the light propagation equation is discretized within each grid cell. After substituting the medium refractive index and interface morphology parameters, the light propagation path within the cell is determined by iterative calculation using the multi-medium space light propagation trajectory discretization formula. When the light passes through the medium interface, the refraction parameters in the multi-medium light field refraction model are called to update the light propagation direction and speed. At the same time, an interference compensation sub-model is established for the impurities and temperature gradient interference factors in the medium. By analyzing the impurity distribution data and temperature field data obtained by the environmental perception module, the impact of the interference factors on light propagation is quantified, and compensation terms are generated and incorporated into the light propagation equation to correct the light propagation path.

6. The image recognition and correction system based on optical model and environmental perception according to claim 5, characterized in that: The light propagation path calculation module determines the light propagation path within the unit through iterative calculation of the multi-medium space light propagation trajectory discretization formula, and the formula is: ,in, The light is The position vector of each discrete unit, The light is The position vector of each discrete unit, The light is The propagation speed of a discrete unit in the medium is calculated as follows: ,in is the speed of light in vacuum, is the refractive index of the medium where the unit is located, is the discrete time step, The light is The propagation direction unit vector in discrete units, is the light position offset vector caused by interference factors; at the same time, an interference compensation sub-model is established for the interference factors of impurities and temperature gradients in the medium, and its model formula is: ,in, is the light position offset vector caused by interference, is the unit vector of the interference offset direction, is the impurity interference coefficient, is the impurity concentration in the medium, is the temperature gradient interference coefficient, is the medium temperature gradient, is the distance the light travels in the current medium unit.

7. The image recognition and correction system based on optical model and environmental perception according to claim 1, characterized in that: The target object light field reconstruction module is based on the calculation results of the light propagation path. It uses the inverse ray tracing method to start from the imaging plane of the image acquisition device, and reversely trace the propagation path of each imaging light until it reaches the surface of the target object. By integrating the coordinates of the reversely traced light, superimposing the intensity and calibrating the phase, it constructs the three-dimensional light field distribution of the target object, which includes light field intensity, phase, and direction parameters. At the same time, in view of the energy attenuation and scattering interference in the light propagation process, the reconstructed three-dimensional light field distribution is optimized using a signal optimization method. By constructing an attenuation function and a scattering kernel, the light energy loss is compensated and the phase deviation is corrected. The light field data is denoised and converted into a standard image data format.

8. The image recognition and correction system based on optical model and environmental perception according to claim 7, characterized in that: The target object light field reconstruction module compensates for light energy loss and corrects phase deviation by constructing an attenuation function and a scattering kernel. The phase deviation correction formula is: ,in, is the corrected light field phase, is the raw phase value extracted from the light field data, is the number of medium layers that the light passes through from the target object to the imaging plane, It is The refractive index of the layer medium, is the wavelength of light, which is preset to the center wavelength of visible light. The light is The propagation distance in the layer medium, It is Phase deviation caused by scattering in the layer medium.

9. The image recognition and correction system based on optical model and environmental perception according to claim 1, characterized in that: The image recognition and correction module adopts a recognition model based on deep learning to recognize and process the standard format image data output by the target object light field reconstruction module. It is trained based on a data set containing distorted images and corrected images of a multi-media environment. It extracts target object features and realizes target classification, positioning and attribute recognition, and outputs recognition results. At the same time, it combines the generated standard format image data to correct the original distorted image, aligns the original distorted image with the standard format image through image registration operation, calculates the geometric deviation and grayscale deviation between the two, and adjusts the geometric parameters and pixel values ​​of the original distorted image through geometric transformation and grayscale correction. After the correction is completed, the correction result is verified by peak signal-to-noise ratio and structural similarity index indicators.

Images