AI-based embedded image correction and stitching processing method, system and medium

Abstract

The present application provides an AI-based embedded image correction and stitching processing method, system and medium. It includes: obtaining an image correction model according to the captured images in different scenes of the construction site and their corresponding corrected images, obtaining a GAN generator model according to the captured images to be stitched in different scenes of the construction site and their corresponding stitched images, obtaining a system performance evaluation index according to the embedded system monitoring data processing, determining the optimized image correction model and the optimized GAN generator model according to the system performance evaluation index, and processing the current captured image to generate the current stitched image, detecting and further processing the current stitched image to obtain the image quality evaluation index, and performing image processing qualification judgment. The present application obtains an adaptive AI model through training and adjusts the model according to the operating performance of the embedded system, and accurately evaluates the image quality after model processing, which is helpful to optimize the image correction and stitching processing technology.

Description

Technical Field

[0001] The present application relates to the field of big data and image processing technology, and more specifically, to an AI-based embedded image correction and stitching processing method, system and medium. Background Art

[0002] The construction site environment is complex and changeable, with a large number of obstructions and building structures of different heights, resulting in uneven light distribution, causing color distortion and reduced contrast in the images taken by the camera. In addition, the construction site environment is relatively noisy and vibrates frequently. The cameras installed on site are easily affected by external vibrations and cause image jitter, affecting image quality. Traditional image correction and stitching methods may have problems with poor correction quality or obvious stitching marks due to factors such as lighting changes, object obstructions, and geometric deformation. AI-based image correction and image stitching methods can reasonably infer and process deformation, color distortion and other parts by learning a large number of complex scene image features, making the processed images more natural. However, there is currently a lack of technology that processes images through AI models and accurately evaluates the quality of image processing.

[0003] In view of the above problems, effective technical solutions are urgently needed. Summary of the invention

[0004] The purpose of this application is to provide an AI-based embedded image correction and stitching processing method, system and medium, obtain an adaptive AI model through training and adjust the model according to the operating performance of the embedded system, and accurately evaluate the image quality after model processing, which helps to optimize the image correction and stitching processing technology.

[0005] This application also provides an AI-based embedded image correction and stitching processing method, comprising the following steps:

[0006] Obtain the captured images of different scenes on the construction site and their corresponding corrected images and perform feature extraction, perform model training based on the feature extraction data, and obtain an image correction model;

[0007] Obtain the images to be stitched and their corresponding stitched images in different scenes of the construction site and perform feature extraction. Perform model training based on the feature extraction data to obtain the GAN generator model.

[0008] Acquire embedded system monitoring data, obtain a system performance evaluation index based on the embedded system monitoring data, and determine an optimized image correction model and an optimized GAN generator model based on the system performance evaluation index;

[0009] Acquire the current captured image and process it according to the optimized image correction model to obtain a current corrected image, and process the current corrected image according to the optimized GAN generator model to obtain a corresponding current stitched image;

[0010] Detect the current spliced ​​image to obtain spatial structure detection data, color detection data and splicing seam detection data, obtain a spatial structure evaluation index based on the spatial structure detection data, and obtain a color evaluation index based on the color detection data;

[0011] The stitching quality evaluation index is obtained according to the stitching seam detection data processing, and the image quality evaluation index is obtained by combining the spatial structure evaluation index and the color evaluation index processing, and the image processing qualification is judged.

[0012] Optionally, in the AI-based embedded image correction and stitching processing method described in the present application, the acquisition of captured images under different scenes of the construction site and their corresponding corrected images and feature extraction, model training based on the feature extraction data, and obtaining the image correction model include:

[0013] Acquire captured images and their corresponding corrected images under different scenes of the construction site, perform feature extraction on the captured images and their corresponding corrected images respectively, and obtain feature data of the captured images and their corresponding corrected images;

[0014] The captured image feature data includes geometric feature data, texture feature data, pixel-level feature data and topological feature data, and the corrected image feature data includes corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data and corrected topological feature data;

[0015] Model training is performed based on the geometric feature data, texture feature data, pixel-level feature data, topological feature data and the corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data, and corrected topological feature data to obtain an image correction model.

[0016] Optionally, in the AI-based embedded image correction and stitching processing method described in the present application, the acquisition of the to-be-stitched photographed images and their corresponding stitched images in different scenes of the construction site and feature extraction, model training based on the feature extraction data, and obtaining the GAN generator model include:

[0017] Acquire the photographed images to be stitched and their corresponding stitched images under different scenes of the construction site, perform feature extraction on the photographed images to be stitched and their corresponding stitched images respectively, and obtain feature data of the images to be stitched and their corresponding stitched images;

[0018] The image feature data to be spliced ​​includes boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data and image semantic feature data, and the spliced ​​image feature data includes splicing boundary geometric feature data, splicing boundary texture feature data, splicing boundary pixel-level feature data and splicing image semantic feature data;

[0019] GAN model training is performed according to the boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data, image semantic feature data and the spliced ​​boundary geometric feature data, spliced ​​boundary texture feature data, spliced ​​boundary pixel-level feature data, and spliced ​​image semantic feature data to obtain a GAN generator model.

[0020] Optionally, in the AI-based embedded image correction and stitching processing method described in the present application, acquiring embedded system monitoring data, obtaining a system performance evaluation index according to embedded system monitoring data processing, and determining an optimized image correction model and an optimized GAN generator model according to the system performance evaluation index include:

[0021] Obtain embedded system monitoring data, including model processing time, response speed, power consumption data, and memory usage;

[0022] Obtaining a system performance evaluation index based on the model processing time, response speed, power consumption data and memory occupancy rate;

[0023] The system performance evaluation index is compared with a preset system performance evaluation index threshold. If the system performance evaluation index is less than the preset system performance evaluation index threshold, the image correction model and the GAN generator model are quantized by using a preset model quantization algorithm to generate an optimized image correction model and an optimized GAN generator model;

[0024] If the system performance evaluation index is greater than or equal to a preset system performance evaluation index threshold, the image correction model is used as the optimized image correction model, and the GAN generator model is used as the optimized GAN generator model.

[0025] Optionally, in the AI-based embedded image correction and stitching processing method described in the present application, the acquiring the current captured image and processing it according to the optimized image correction model to obtain the current corrected image, and processing the current corrected image according to the optimized GAN generator model to obtain the corresponding current stitched image include:

[0026] Acquire the current captured image and perform feature extraction to obtain feature data of the current captured image, including current geometric feature data, current texture feature data, current pixel-level feature data, and current topological feature data;

[0027] Inputting the current geometric feature data, current texture feature data, current pixel-level feature data and current topological feature data into the optimized image correction model for processing to obtain current correction feature data and its corresponding current correction image;

[0028] Extract features of the current corrected image to obtain feature data to be stitched, including content feature data of the current image, spatial feature data of the current image, and semantic feature data of the current image;

[0029] The current image content feature data, the current image spatial feature data and the current image semantic feature data are input into the optimized GAN generator model for processing to obtain current splicing feature data and its corresponding current splicing image.

[0030] Optionally, in the AI-based embedded image correction and stitching processing method described in the present application, the current stitched image is detected to obtain spatial structure detection data, color detection data and stitching seam detection data, and a spatial structure evaluation index is obtained according to the spatial structure detection data processing, and a color evaluation index is obtained according to the color detection data processing, including:

[0031] The spatial structure detection data includes a structural similarity index, shape accuracy and position error, the color detection data includes a color difference value, a color histogram difference value and local color consistency data, and the seam detection data includes a seam contrast, a seam correlation data, a seam energy value, a seam entropy value and a seam gradient value;

[0032] Obtaining a spatial structure evaluation index according to the structural similarity index, shape accuracy and position error processing;

[0033] A color evaluation index is obtained according to the color difference value, the color histogram difference value and the local color consistency data.

[0034] Optionally, in the AI-based embedded image correction and stitching processing method described in the present application, the stitching quality evaluation index is obtained according to the stitching seam detection data processing, and the image quality evaluation index is obtained by combining the spatial structure evaluation index and the color evaluation index processing, and the image processing qualification judgment is performed, including:

[0035] Obtaining a splicing quality evaluation index according to the splicing seam contrast, the splicing seam correlation data, the splicing seam energy value, the splicing seam entropy value and the splicing seam gradient value;

[0036] Obtaining an image quality assessment index according to the spatial structure assessment index, the color assessment index and the stitching quality assessment index;

[0037] The image quality assessment index is compared with a preset image quality assessment index threshold, and if the threshold comparison result does not meet the preset threshold comparison result requirement, an image processing failure determination is performed.

[0038] In a second aspect, the present application provides an AI-based embedded image correction and stitching processing system, the system comprising: a memory and a processor, the memory storing a program of an AI-based embedded image correction and stitching processing method, the program of the AI-based embedded image correction and stitching processing method being executed by the processor to implement the following steps:

[0039] Obtain the captured images of different scenes on the construction site and their corresponding corrected images and perform feature extraction, perform model training based on the feature extraction data, and obtain an image correction model;

[0040] Obtain the images to be stitched and their corresponding stitched images in different scenes of the construction site and perform feature extraction. Perform model training based on the feature extraction data to obtain the GAN generator model.

[0041] Acquire embedded system monitoring data, obtain a system performance evaluation index based on the embedded system monitoring data, and determine an optimized image correction model and an optimized GAN generator model based on the system performance evaluation index;

[0042] Acquire the current captured image and process it according to the optimized image correction model to obtain a current corrected image, and process the current corrected image according to the optimized GAN generator model to obtain a corresponding current stitched image;

[0043] Detect the current spliced ​​image to obtain spatial structure detection data, color detection data and splicing seam detection data, obtain a spatial structure evaluation index based on the spatial structure detection data, and obtain a color evaluation index based on the color detection data;

[0044] The stitching quality evaluation index is obtained according to the stitching seam detection data processing, and the image quality evaluation index is obtained by combining the spatial structure evaluation index and the color evaluation index processing, and the image processing qualification is judged.

[0045] Optionally, in the AI-based embedded image correction and stitching processing system described in the present application, the steps of acquiring captured images of different scenes of the construction site and their corresponding corrected images and performing feature extraction, performing model training according to the feature extraction data, and obtaining an image correction model include:

[0046] Acquire captured images and their corresponding corrected images under different scenes of the construction site, perform feature extraction on the captured images and their corresponding corrected images respectively, and obtain feature data of the captured images and their corresponding corrected images;

[0047] The captured image feature data includes geometric feature data, texture feature data, pixel-level feature data and topological feature data, and the corrected image feature data includes corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data and corrected topological feature data;

[0048] Model training is performed based on the geometric feature data, texture feature data, pixel-level feature data, topological feature data and the corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data, and corrected topological feature data to obtain an image correction model.

[0049] In a third aspect, the present application also provides a computer-readable storage medium, which stores an AI-based embedded image correction and stitching processing method program. When the AI-based embedded image correction and stitching processing method program is executed by a processor, the steps of the AI-based embedded image correction and stitching processing method as described in any one of the above items are implemented.

[0050] From the above, it can be seen that the AI-based embedded image correction and stitching processing method, system and medium provided in this application obtain an adaptive AI model through training and adjust the model according to the operating performance of the embedded system, and accurately evaluate the image quality after model processing, which helps to optimize the image correction and stitching processing technology.

[0051] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or understood by practicing the embodiments of the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 A flowchart of an AI-based embedded image correction and stitching processing method provided in an embodiment of the present application;

[0054] Figure 2 A flowchart of obtaining an image correction model for the AI-based embedded image correction and stitching processing method provided in an embodiment of the present application;

[0055] Figure 3A flowchart of obtaining a GAN generator model for the AI-based embedded image correction and stitching processing method provided in an embodiment of the present application;

[0056] Figure 4 A flowchart of determining an optimized image correction model and an optimized GAN generator model for the AI-based embedded image correction and stitching processing method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction 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, rather than all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0058] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0059] Please refer to Figure 1 , Figure 1 The following is a flowchart of an AI-based embedded image correction and stitching processing method in some embodiments of the present application. The AI-based embedded image correction and stitching processing method is used in a terminal device, such as a computer, a mobile phone terminal, etc. The AI-based embedded image correction and stitching processing method comprises the following steps:

[0060] S11, obtaining the captured images of different scenes of the construction site and their corresponding corrected images and performing feature extraction, performing model training based on the feature extraction data, and obtaining an image correction model;

[0061] S12, obtaining the images to be spliced ​​and their corresponding spliced ​​images under different scenes of the construction site and performing feature extraction, performing model training based on the feature extraction data, and obtaining a GAN generator model;

[0062] S13, acquiring embedded system monitoring data, obtaining a system performance evaluation index according to the embedded system monitoring data, and determining an optimized image correction model and an optimized GAN generator model according to the system performance evaluation index;

[0063] S14, acquiring the current captured image and processing it according to the optimized image correction model to obtain a current corrected image, and processing the current corrected image according to the optimized GAN generator model to obtain a corresponding current stitched image;

[0064] S15, detecting the current spliced ​​image to obtain spatial structure detection data, color detection data, and splicing seam detection data, and obtaining a spatial structure evaluation index based on the spatial structure detection data and a color evaluation index based on the color detection data;

[0065] S16, obtaining a stitching quality evaluation index according to the stitching seam detection data, and obtaining an image quality evaluation index in combination with the spatial structure evaluation index and the color evaluation index, and performing image processing qualification determination.

[0066] It should be noted that the present application obtains an adaptive AI model through training and adjusts the model according to the operating performance of the embedded system, and accurately evaluates the image quality after model processing, which is helpful to optimize the image correction and stitching processing technology. Specifically, first, an image correction model is obtained according to the captured images in different scenes of the construction site and their corresponding corrected image processing, and a GAN generator model is obtained according to the captured images to be stitched in different scenes of the construction site and their corresponding stitched image processing. Then, a system performance evaluation index is obtained according to the embedded system monitoring data processing, that is, whether the model training process causes a large operating burden on the embedded system is judged. If it has a large impact on the system performance, the model is quantified to reduce the computational burden, that is, the optimized image correction model and the optimized GAN generator model are determined according to the system performance evaluation index. Finally, the current captured image is processed according to the optimized model to generate the current stitched image, and the spatial structure, color and stitching quality of the current stitched image are detected to achieve an accurate evaluation of the image quality, and the image processing eligibility is determined according to the image quality evaluation result.

[0067] Please refer to Figure 2 , Figure 2 The flowchart of the embedded image correction and stitching processing method based on AI in some embodiments of the present application is to obtain an image correction model. According to an embodiment of the present invention, the captured images of different scenes of the construction site and their corresponding corrected images are obtained and feature extraction is performed, and model training is performed according to the feature extraction data to obtain the image correction model, including:

[0068] S21, acquiring captured images and their corresponding corrected images under different scenes of the construction site, performing feature extraction on the captured images and their corresponding corrected images respectively, and obtaining feature data of the captured images and their corresponding corrected images;

[0069] S22, the captured image feature data includes geometric feature data, texture feature data, pixel-level feature data and topological feature data, and the corrected image feature data includes corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data and corrected topological feature data;

[0070] S23, performing model training according to the geometric feature data, texture feature data, pixel-level feature data, topological feature data and the corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data, and corrected topological feature data to obtain an image correction model.

[0071] It should be noted that the geometric feature data include object shape data, object position data and object angle data, and the corrected geometric feature data include corrected object shape data, corrected object position data and corrected object angle data; the texture feature data include contrast, correlation data, energy value, entropy value and pixel LBP value, and the corrected texture feature data include corrected contrast, corrected correlation data, corrected energy value, corrected entropy value and corrected pixel LBP value; the pixel-level feature data include grayscale value, color value, pixel gradient value, pixel gradient direction data and pixel position data, and the corrected pixel-level feature data include corrected grayscale value, corrected color value, corrected pixel gradient value, corrected pixel gradient direction data and corrected pixel position data; the topological feature data include connectivity feature data and hole feature data, and the corrected topological feature data include corrected connectivity feature data and corrected hole feature data.

[0072] Among them, contrast, correlation data, energy value, entropy value and pixel LBP value are all data reflecting texture characteristics. Contrast reflects the clarity of the texture and the degree of grayscale change. Correlation data reflects the similarity and spatial distribution correlation between the grayscale values ​​of different pixels. The energy value is used to indicate the uniformity of the image texture. The entropy value measures the randomness or complexity of the texture. The pixel LBP (Local Binary Pattern) value is data used to measure local texture characteristics.

[0073] Among them, the grayscale value represents the brightness, the color value represents the color of the pixel, the pixel position data represents the coordinate position of the pixel in the image, the pixel gradient direction data depicts the contour shape of the object, and the pixel gradient value reflects the intensity of the pixel change in a certain direction, which is used for image edge detection.

[0074] Among them, the topological feature data is used to reflect the basic structure of the scene in the image, the connectivity feature data is used to describe the connection status of objects or areas in the image, and the hole feature data includes the number of holes, hole size data, hole shape data and hole position data. In this embodiment, the layout and structure of the building can be inferred by analyzing the topological features of the building in the image (such as the connectivity of the various parts of the building and the hole relationship between rooms).

[0075] Please refer to Figure 3 , Figure 3 The flowchart of the AI-based embedded image correction and stitching processing method for obtaining the GAN generator model in some embodiments of the present application is as follows. According to an embodiment of the present invention, the method of obtaining the images to be stitched and their corresponding stitched images under different scenes of the construction site and performing feature extraction, and performing model training based on the feature extraction data to obtain the GAN generator model includes:

[0076] S31, obtaining the photographed images to be stitched and their corresponding stitched images under different scenes of the construction site, performing feature extraction on the photographed images to be stitched and their corresponding stitched images respectively, and obtaining feature data of the images to be stitched and their corresponding stitched images;

[0077] S32, the image feature data to be stitched includes boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data and image semantic feature data, and the stitched image feature data includes stitching boundary geometric feature data, stitching boundary texture feature data, stitching boundary pixel-level feature data and stitching image semantic feature data;

[0078] S33, performing GAN model training according to the boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data, image semantic feature data and the spliced ​​boundary geometric feature data, spliced ​​boundary texture feature data, spliced ​​boundary pixel-level feature data, and spliced ​​image semantic feature data to obtain a GAN generator model.

[0079] It should be noted that the boundary geometric feature data includes boundary object shape data, boundary object position data and boundary object angle data, and the splicing geometric feature data includes splicing boundary object shape data, splicing boundary object position data and splicing boundary object angle data; the boundary texture feature data includes boundary contrast, boundary correlation data, boundary energy value, boundary entropy value and boundary pixel LBP value, and the splicing texture feature data includes splicing boundary contrast, splicing boundary correlation data, splicing boundary energy value, splicing boundary entropy value and splicing boundary pixel LBP value; the boundary pixel-level feature data includes boundary grayscale value, boundary color value, boundary pixel gradient value, boundary pixel gradient direction data and boundary pixel position data, and the splicing boundary pixel-level feature data includes splicing boundary grayscale value, splicing boundary color value, splicing boundary pixel gradient value, splicing boundary pixel gradient direction data and splicing boundary pixel position data; the image semantic feature data includes object category label data and scene type data, and the splicing image semantic feature data includes splicing object category label data and splicing scene type data.

[0080] Among them, image semantic feature data is data used to represent object categories and scenes in images. Pre-trained category classification models and scene classification models can be used to identify and obtain object category label data and scene type data in images. This can provide semantic guidance for the GAN (Generative Adversarial Networks) generator model, allowing it to generate spliced ​​images that are more in line with scene logic.

[0081] Please refer to Figure 4 , Figure 4 The present invention is a flowchart of determining an optimized image correction model and an optimized GAN generator model in an AI-based embedded image correction and stitching processing method in some embodiments of the present application. According to an embodiment of the present invention, the embedded system monitoring data is acquired, a system performance evaluation index is obtained according to the embedded system monitoring data processing, and the optimized image correction model and the optimized GAN generator model are determined according to the system performance evaluation index, including:

[0082] S41, obtaining embedded system monitoring data, including model processing time, response speed, power consumption data and memory occupancy rate;

[0083] S42, obtaining a system performance evaluation index according to the model processing time, response speed, power consumption data and memory occupancy rate;

[0084] S43, comparing the system performance evaluation index with a preset system performance evaluation index threshold, if the system performance evaluation index is less than the preset system performance evaluation index threshold, using a preset model quantization algorithm to perform model quantization on the image correction model and the GAN generator model respectively, to generate an optimized image correction model and an optimized GAN generator model;

[0085] S44. If the system performance evaluation index is greater than or equal to a preset system performance evaluation index threshold, the image correction model is used as the optimized image correction model, and the GAN generator model is used as the optimized GAN generator model.

[0086] It should be noted that if the system performance evaluation index is less than the preset system performance evaluation index threshold, it means that the model training is relatively complex, resulting in poor system performance. In order to reduce the amount of calculation, memory usage and power consumption and improve the system operation speed, the model needs to be quantized to reduce the calculation and storage overhead of the model by reducing the number of representation bits of the model parameters.

[0087] The calculation formula of the system performance evaluation index is:

[0088] ;

[0089] in, is the system performance evaluation index, , , and They are response speed, memory usage, power consumption data and model processing time. , , and is the preset feature coefficient (which can be obtained by querying the preset computer vision processing platform).

[0090] According to an embodiment of the present invention, the step of acquiring the current captured image and processing the current corrected image according to the optimized image correction model, and processing the current corrected image according to the optimized GAN generator model to obtain the corresponding current stitched image includes:

[0091] Acquire the current captured image and perform feature extraction to obtain feature data of the current captured image, including current geometric feature data, current texture feature data, current pixel-level feature data, and current topological feature data;

[0092] Inputting the current geometric feature data, current texture feature data, current pixel-level feature data and current topological feature data into the optimized image correction model for processing to obtain current correction feature data and its corresponding current correction image;

[0093] Extract features of the current corrected image to obtain feature data to be stitched, including content feature data of the current image, spatial feature data of the current image, and semantic feature data of the current image;

[0094] The current image content feature data, the current image spatial feature data and the current image semantic feature data are input into the optimized GAN generator model for processing to obtain current splicing feature data and its corresponding current splicing image.

[0095] It should be noted that the current captured image is processed according to the optimized image correction model and then processed according to the optimized GAN generator model to obtain the corresponding current stitched image.

[0096] According to an embodiment of the present invention, the detecting of the current stitched image to obtain spatial structure detection data, color detection data and stitching seam detection data, obtaining a spatial structure evaluation index according to the spatial structure detection data, and obtaining a color evaluation index according to the color detection data includes:

[0097] The spatial structure detection data includes a structural similarity index, shape accuracy and position error, the color detection data includes a color difference value, a color histogram difference value and local color consistency data, and the seam detection data includes a seam contrast, a seam correlation data, a seam energy value, a seam entropy value and a seam gradient value;

[0098] Obtaining a spatial structure evaluation index according to the structural similarity index, shape accuracy and position error processing;

[0099] A color evaluation index is obtained according to the color difference value, the color histogram difference value and the local color consistency data.

[0100] It should be noted that the structural similarity index is an indicator to measure the similarity between the two images before and after processing. The larger the structural similarity index, the more similar the two images are in structure. The shape accuracy can be represented by the Hausdorff distance. The more accurate the shape of the object in the processed image, the smaller the Hausdorff distance. The position error is calculated by calculating the distance between the true center position of the target object and the detection center position. The larger the error, the greater the deviation of the object position in the corrected spliced ​​image. The color difference is used to measure the color difference between the pixels of the two images before and after processing. The color histogram difference is used to measure the difference in the overall color distribution of the two images before and after processing. The smaller the color histogram difference, the smaller the color difference between the two images. The local color consistency data is used to describe the uniformity and similarity of the color of the spliced ​​area, which can be represented by calculating the color variance of the spliced ​​area.

[0101] The calculation formula of the spatial structure evaluation index is:

[0102] ;

[0103] in, is the spatial structure evaluation index, , and are the structural similarity index, shape accuracy and position error, respectively. and is the preset characteristic coefficient (which can be obtained by querying the preset computer vision processing platform);

[0104] The calculation formula of the color evaluation index is:

[0105] ;

[0106] in, is the color evaluation index, is the color difference, is the color histogram difference value, is the local color consistency data, , and is the preset feature coefficient (which can be obtained by querying the preset computer vision processing platform).

[0107] According to an embodiment of the present invention, the processing of the splicing seam detection data to obtain a splicing quality evaluation index, combining the spatial structure evaluation index and the color evaluation index to obtain an image quality evaluation index, and performing image processing qualification determination includes:

[0108] Obtaining a splicing quality evaluation index according to the splicing seam contrast, the splicing seam correlation data, the splicing seam energy value, the splicing seam entropy value and the splicing seam gradient value;

[0109] Obtaining an image quality assessment index according to the spatial structure assessment index, the color assessment index and the stitching quality assessment index;

[0110] The image quality assessment index is compared with a preset image quality assessment index threshold, and if the threshold comparison result does not meet the preset threshold comparison result requirement, an image processing failure determination is performed.

[0111] It should be noted that the seam gradient value is a data used to measure the degree of difference between pixels on both sides of the seam. If the seam gradient value is too large, it means that the pixels on both sides of the seam are significantly different and the stitching effect is poor. The seam contrast, seam correlation data, seam energy value, and seam entropy value are all data that reflect the texture characteristics of the seam. The seam contrast is data that reflects the difference in the grayscale values ​​of pixels on both sides of the seam. The seam correlation data is data that reflects the degree of correlation between the grayscale values ​​of pixels around the seam. The smaller the seam correlation data, the more incoherent the texture is on both sides of the seam. The seam energy value is used to indicate the uniformity of the grayscale distribution of the image. The larger the seam energy value, the more uniform the distribution of the grayscale values ​​of pixels around the seam. The seam entropy value is used to describe the randomness of the grayscale distribution in the image. The larger the seam entropy value, the more complex the texture at the seam or the more noise there is, and the stitching effect is not ideal.

[0112] The calculation formula of the splicing quality evaluation index is:

[0113] ;

[0114] in, is the splicing quality evaluation index, is the gradient value of the seam. , , and They are respectively the seam contrast, seam correlation data, seam energy value and seam entropy value. , , , and is the preset characteristic coefficient (which can be obtained by querying the preset computer vision processing platform);

[0115] The calculation formula of the image quality evaluation index is:

[0116] ;

[0117] in, is the spatial structure evaluation index, is the color evaluation index, is the splicing quality evaluation index, , and is the preset weight coefficient (which can be obtained by querying the preset computer vision processing platform).

[0118] The present invention also discloses an AI-based embedded image correction and stitching processing system, comprising a memory and a processor, wherein the memory stores an AI-based embedded image correction and stitching processing method program, and when the AI-based embedded image correction and stitching processing method program is executed by the processor, the following steps are implemented:

[0119] Obtain the captured images of different scenes on the construction site and their corresponding corrected images and perform feature extraction, perform model training based on the feature extraction data, and obtain an image correction model;

[0120] Obtain the images to be stitched and their corresponding stitched images in different scenes of the construction site and perform feature extraction. Perform model training based on the feature extraction data to obtain the GAN generator model.

[0121] Acquire embedded system monitoring data, obtain a system performance evaluation index based on the embedded system monitoring data, and determine an optimized image correction model and an optimized GAN generator model based on the system performance evaluation index;

[0122] Acquire the current captured image and process it according to the optimized image correction model to obtain a current corrected image, and process the current corrected image according to the optimized GAN generator model to obtain a corresponding current stitched image;

[0123] Detect the current spliced ​​image to obtain spatial structure detection data, color detection data and splicing seam detection data, obtain a spatial structure evaluation index based on the spatial structure detection data, and obtain a color evaluation index based on the color detection data;

[0124] The stitching quality evaluation index is obtained according to the stitching seam detection data processing, and the image quality evaluation index is obtained by combining the spatial structure evaluation index and the color evaluation index processing, and the image processing qualification is judged.

[0125] It should be noted that the present application obtains an adaptive AI model through training and adjusts the model according to the operating performance of the embedded system, and accurately evaluates the image quality after model processing, which is helpful to optimize the image correction and stitching processing technology. Specifically, first, an image correction model is obtained according to the captured images in different scenes of the construction site and their corresponding corrected image processing, and a GAN generator model is obtained according to the captured images to be stitched in different scenes of the construction site and their corresponding stitched image processing. Then, a system performance evaluation index is obtained according to the embedded system monitoring data processing, that is, whether the model training process causes a large operating burden on the embedded system is judged. If it has a large impact on the system performance, the model is quantified to reduce the computational burden, that is, the optimized image correction model and the optimized GAN generator model are determined according to the system performance evaluation index. Finally, the current captured image is processed according to the optimized model to generate the current stitched image, and the spatial structure, color and stitching quality of the current stitched image are detected to achieve an accurate evaluation of the image quality, and the image processing eligibility is determined according to the image quality evaluation result.

[0126] According to an embodiment of the present invention, the method of acquiring captured images of different scenes of a construction site and their corresponding corrected images and performing feature extraction, and performing model training according to the feature extraction data to obtain an image correction model includes:

[0127] Acquire captured images and their corresponding corrected images under different scenes of the construction site, perform feature extraction on the captured images and their corresponding corrected images respectively, and obtain feature data of the captured images and their corresponding corrected images;

[0128] The captured image feature data includes geometric feature data, texture feature data, pixel-level feature data and topological feature data, and the corrected image feature data includes corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data and corrected topological feature data;

[0129] Model training is performed based on the geometric feature data, texture feature data, pixel-level feature data, topological feature data and the corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data, and corrected topological feature data to obtain an image correction model.

[0130] It should be noted that the geometric feature data include object shape data, object position data and object angle data, and the corrected geometric feature data include corrected object shape data, corrected object position data and corrected object angle data; the texture feature data include contrast, correlation data, energy value, entropy value and pixel LBP value, and the corrected texture feature data include corrected contrast, corrected correlation data, corrected energy value, corrected entropy value and corrected pixel LBP value; the pixel-level feature data include grayscale value, color value, pixel gradient value, pixel gradient direction data and pixel position data, and the corrected pixel-level feature data include corrected grayscale value, corrected color value, corrected pixel gradient value, corrected pixel gradient direction data and corrected pixel position data; the topological feature data include connectivity feature data and hole feature data, and the corrected topological feature data include corrected connectivity feature data and corrected hole feature data.

[0131] Among them, contrast, correlation data, energy value, entropy value and pixel LBP value are all data reflecting texture characteristics. Contrast reflects the clarity of the texture and the degree of grayscale change. Correlation data reflects the similarity and spatial distribution correlation between the grayscale values ​​of different pixels. The energy value is used to indicate the uniformity of the image texture. The entropy value measures the randomness or complexity of the texture. The pixel LBP value is the data used to measure local texture characteristics.

[0132] Among them, the grayscale value represents the brightness, the color value represents the color of the pixel, the pixel position data represents the coordinate position of the pixel in the image, the pixel gradient direction data depicts the contour shape of the object, and the pixel gradient value reflects the intensity of the pixel change in a certain direction, which is used for image edge detection.

[0133] Among them, the topological feature data is used to reflect the basic structure of the scene in the image, the connectivity feature data is used to describe the connection status of objects or areas in the image, and the hole feature data includes the number of holes, hole size data, hole shape data and hole position data. In this embodiment, the layout and structure of the building can be inferred by analyzing the topological features of the building in the image (such as the connectivity of the various parts of the building and the hole relationship between rooms).

[0134] According to an embodiment of the present invention, the method of acquiring the images to be spliced ​​and their corresponding spliced ​​images in different scenes of the construction site and performing feature extraction, and performing model training according to the feature extraction data to obtain the GAN generator model includes:

[0135] Acquire the photographed images to be stitched and their corresponding stitched images under different scenes of the construction site, perform feature extraction on the photographed images to be stitched and their corresponding stitched images respectively, and obtain feature data of the images to be stitched and their corresponding stitched images;

[0136] The image feature data to be spliced ​​includes boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data and image semantic feature data, and the spliced ​​image feature data includes splicing boundary geometric feature data, splicing boundary texture feature data, splicing boundary pixel-level feature data and splicing image semantic feature data;

[0137] GAN model training is performed according to the boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data, image semantic feature data and the spliced ​​boundary geometric feature data, spliced ​​boundary texture feature data, spliced ​​boundary pixel-level feature data, and spliced ​​image semantic feature data to obtain a GAN generator model.

[0138] It should be noted that the boundary geometric feature data includes boundary object shape data, boundary object position data and boundary object angle data, and the splicing geometric feature data includes splicing boundary object shape data, splicing boundary object position data and splicing boundary object angle data; the boundary texture feature data includes boundary contrast, boundary correlation data, boundary energy value, boundary entropy value and boundary pixel LBP value, and the splicing texture feature data includes splicing boundary contrast, splicing boundary correlation data, splicing boundary energy value, splicing boundary entropy value and splicing boundary pixel LBP value; the boundary pixel-level feature data includes boundary grayscale value, boundary color value, boundary pixel gradient value, boundary pixel gradient direction data and boundary pixel position data, and the splicing boundary pixel-level feature data includes splicing boundary grayscale value, splicing boundary color value, splicing boundary pixel gradient value, splicing boundary pixel gradient direction data and splicing boundary pixel position data; the image semantic feature data includes object category label data and scene type data, and the splicing image semantic feature data includes splicing object category label data and splicing scene type data.

[0139] Among them, image semantic feature data is data used to represent object categories and scenes in images. Pre-trained category classification models and scene classification models can be used to identify and obtain object category label data and scene type data in images. This can provide semantic guidance for the GAN (Generative Adversarial Networks) generator model, allowing it to generate spliced ​​images that are more in line with scene logic.

[0140] According to an embodiment of the present invention, the method of acquiring embedded system monitoring data, obtaining a system performance evaluation index according to the embedded system monitoring data, and determining an optimized image correction model and an optimized GAN generator model according to the system performance evaluation index includes:

[0141] Obtain embedded system monitoring data, including model processing time, response speed, power consumption data, and memory usage;

[0142] Obtaining a system performance evaluation index based on the model processing time, response speed, power consumption data and memory occupancy rate;

[0143] The system performance evaluation index is compared with a preset system performance evaluation index threshold. If the system performance evaluation index is less than the preset system performance evaluation index threshold, the image correction model and the GAN generator model are quantized by using a preset model quantization algorithm to generate an optimized image correction model and an optimized GAN generator model;

[0144] If the system performance evaluation index is greater than or equal to a preset system performance evaluation index threshold, the image correction model is used as the optimized image correction model, and the GAN generator model is used as the optimized GAN generator model.

[0145] It should be noted that if the system performance evaluation index is less than the preset system performance evaluation index threshold, it means that the model training is relatively complex, resulting in poor system performance. In order to reduce the amount of calculation, memory usage and power consumption and improve the system operation speed, the model needs to be quantized to reduce the calculation and storage overhead of the model by reducing the number of representation bits of the model parameters.

[0146] The calculation formula of the system performance evaluation index is:

[0147] ;

[0148] in, is the system performance evaluation index, , , and They are response speed, memory usage, power consumption data and model processing time. , , and is the preset feature coefficient (which can be obtained by querying the preset computer vision processing platform).

[0149] According to an embodiment of the present invention, the step of acquiring the current captured image and processing the current corrected image according to the optimized image correction model, and processing the current corrected image according to the optimized GAN generator model to obtain the corresponding current stitched image includes:

[0150] Acquire the current captured image and perform feature extraction to obtain feature data of the current captured image, including current geometric feature data, current texture feature data, current pixel-level feature data, and current topological feature data;

[0151] Inputting the current geometric feature data, current texture feature data, current pixel-level feature data and current topological feature data into the optimized image correction model for processing to obtain current correction feature data and its corresponding current correction image;

[0152] Extract features of the current corrected image to obtain feature data to be stitched, including content feature data of the current image, spatial feature data of the current image, and semantic feature data of the current image;

[0153] The current image content feature data, the current image spatial feature data and the current image semantic feature data are input into the optimized GAN generator model for processing to obtain current splicing feature data and its corresponding current splicing image.

[0154] It should be noted that the current captured image is processed according to the optimized image correction model and then processed according to the optimized GAN generator model to obtain the corresponding current stitched image.

[0155] According to an embodiment of the present invention, the detecting of the current stitched image to obtain spatial structure detection data, color detection data and stitching seam detection data, obtaining a spatial structure evaluation index according to the spatial structure detection data, and obtaining a color evaluation index according to the color detection data includes:

[0156] The spatial structure detection data includes a structural similarity index, shape accuracy and position error, the color detection data includes a color difference value, a color histogram difference value and local color consistency data, and the seam detection data includes a seam contrast, a seam correlation data, a seam energy value, a seam entropy value and a seam gradient value;

[0157] Obtaining a spatial structure evaluation index according to the structural similarity index, shape accuracy and position error processing;

[0158] A color evaluation index is obtained according to the color difference value, the color histogram difference value and the local color consistency data.

[0159] It should be noted that the structural similarity index is an indicator to measure the similarity between the two images before and after processing. The larger the structural similarity index, the more similar the two images are in structure. The shape accuracy can be represented by the Hausdorff distance. The more accurate the shape of the object in the processed image, the smaller the Hausdorff distance. The position error is calculated by calculating the distance between the true center position of the target object and the detection center position. The larger the error, the greater the deviation of the object position in the corrected spliced ​​image. The color difference is used to measure the color difference between the pixels of the two images before and after processing. The color histogram difference is used to measure the difference in the overall color distribution of the two images before and after processing. The smaller the color histogram difference, the smaller the color difference between the two images. The local color consistency data is used to describe the uniformity and similarity of the color of the spliced ​​area, which can be represented by calculating the color variance of the spliced ​​area.

[0160] The calculation formula of the spatial structure evaluation index is:

[0161] ;

[0162] in, is the spatial structure evaluation index, , and are the structural similarity index, shape accuracy and position error, respectively. and is the preset characteristic coefficient (which can be obtained by querying the preset computer vision processing platform);

[0163] The calculation formula of the color evaluation index is:

[0164] ;

[0165] in, is the color evaluation index, is the color difference, is the color histogram difference value, is the local color consistency data, , and is the preset feature coefficient (which can be obtained by querying the preset computer vision processing platform).

[0166] According to an embodiment of the present invention, the processing of the splicing seam detection data to obtain a splicing quality evaluation index, combining the spatial structure evaluation index and the color evaluation index to obtain an image quality evaluation index, and performing image processing qualification determination includes:

[0167] Obtaining a splicing quality evaluation index according to the splicing seam contrast, the splicing seam correlation data, the splicing seam energy value, the splicing seam entropy value and the splicing seam gradient value;

[0168] Obtaining an image quality assessment index according to the spatial structure assessment index, the color assessment index and the stitching quality assessment index;

[0169] The image quality assessment index is compared with a preset image quality assessment index threshold, and if the threshold comparison result does not meet the preset threshold comparison result requirement, an image processing failure determination is performed.

[0170] It should be noted that the seam gradient value is a data used to measure the degree of difference between pixels on both sides of the seam. If the seam gradient value is too large, it means that the pixels on both sides of the seam are significantly different and the stitching effect is poor. The seam contrast, seam correlation data, seam energy value, and seam entropy value are all data that reflect the texture characteristics of the seam. The seam contrast is data that reflects the difference in the grayscale values ​​of pixels on both sides of the seam. The seam correlation data is data that reflects the degree of correlation between the grayscale values ​​of pixels around the seam. The smaller the seam correlation data, the more incoherent the texture is on both sides of the seam. The seam energy value is used to indicate the uniformity of the grayscale distribution of the image. The larger the seam energy value, the more uniform the distribution of the grayscale values ​​of pixels around the seam. The seam entropy value is used to describe the randomness of the grayscale distribution in the image. The larger the seam entropy value, the more complex the texture at the seam or the more noise there is, and the stitching effect is not ideal.

[0171] The calculation formula of the splicing quality evaluation index is:

[0172] ;

[0173] in, is the splicing quality evaluation index, is the gradient value of the seam. , , and They are respectively the seam contrast, seam correlation data, seam energy value and seam entropy value. , , , and is the preset characteristic coefficient (which can be obtained by querying the preset computer vision processing platform);

[0174] The calculation formula of the image quality evaluation index is:

[0175] ;

[0176] in, is the spatial structure evaluation index, is the color evaluation index, is the splicing quality evaluation index, , and is the preset weight coefficient (which can be obtained by querying the preset computer vision processing platform).

[0177] A third aspect of the present invention provides a readable storage medium, which stores an AI-based embedded image correction and stitching processing method program. When the AI-based embedded image correction and stitching processing method program is executed by a processor, the steps of the AI-based embedded image correction and stitching processing method as described in any one of the above items are implemented.

[0178] The AI-based embedded image correction and stitching processing method, system and medium disclosed in the present invention obtain an adaptive AI model through training and adjust the model according to the operating performance of the embedded system, and accurately evaluate the image quality after model processing, which is helpful to optimize the image correction and stitching processing technology.

[0179] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0180] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0181] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0182] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, the aforementioned program can be stored in a readable storage medium, and when the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0183] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

Claims

1. An AI-based embedded image correction and stitching processing method, characterized in that: The following steps are involved: Obtain the captured images of different scenes on the construction site and their corresponding corrected images and perform feature extraction, perform model training based on the feature extraction data, and obtain an image correction model; Obtain the images to be stitched and their corresponding stitched images in different scenes of the construction site and perform feature extraction. Perform model training based on the feature extraction data to obtain the GAN generator model. Acquire embedded system monitoring data, obtain a system performance evaluation index based on the embedded system monitoring data, and determine an optimized image correction model and an optimized GAN generator model based on the system performance evaluation index; Acquire the current captured image and process it according to the optimized image correction model to obtain a current corrected image, and process the current corrected image according to the optimized GAN generator model to obtain a corresponding current stitched image; Detect the current spliced ​​image to obtain spatial structure detection data, color detection data and splicing seam detection data, obtain a spatial structure evaluation index based on the spatial structure detection data, and obtain a color evaluation index based on the color detection data; Processing the splicing seam detection data to obtain a splicing quality evaluation index, and combining the spatial structure evaluation index and the color evaluation index to obtain an image quality evaluation index, and performing image processing qualification determination; Acquire the photographed images to be stitched and their corresponding stitched images under different scenes of the construction site, perform feature extraction on the photographed images to be stitched and their corresponding stitched images respectively, and obtain feature data of the images to be stitched and their corresponding stitched images; The image feature data to be spliced ​​includes boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data and image semantic feature data, and the spliced ​​image feature data includes splicing boundary geometric feature data, splicing boundary texture feature data, splicing boundary pixel-level feature data and splicing image semantic feature data; Performing GAN model training according to the boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data, image semantic feature data, and the spliced ​​boundary geometric feature data, spliced ​​boundary texture feature data, spliced ​​boundary pixel-level feature data, and spliced ​​image semantic feature data to obtain a GAN generator model; The spatial structure detection data includes a structural similarity index, shape accuracy and position error, the color detection data includes a color difference value, a color histogram difference value and local color consistency data, and the seam detection data includes a seam contrast, a seam correlation data, a seam energy value, a seam entropy value and a seam gradient value; Obtaining a spatial structure evaluation index according to the structural similarity index, shape accuracy and position error processing; Obtaining a color evaluation index according to the color difference value, the color histogram difference value and the local color consistency data processing; Obtaining a splicing quality evaluation index according to the splicing seam contrast, the splicing seam correlation data, the splicing seam energy value, the splicing seam entropy value and the splicing seam gradient value; Obtaining an image quality assessment index according to the spatial structure assessment index, the color assessment index and the stitching quality assessment index; The image quality assessment index is compared with a preset image quality assessment index threshold. If the threshold comparison result does not meet the preset threshold comparison result requirement, a determination of image processing failure is made.

2. The AI-based embedded image correction and stitching processing method according to claim 1, characterized in that: The method of acquiring the captured images of different scenes of the construction site and the corresponding corrected images and performing feature extraction, and performing model training according to the feature extraction data to obtain the image correction model includes: Acquire captured images and their corresponding corrected images under different scenes of the construction site, perform feature extraction on the captured images and their corresponding corrected images respectively, and obtain feature data of the captured images and their corresponding corrected images; The captured image feature data includes geometric feature data, texture feature data, pixel-level feature data and topological feature data, and the corrected image feature data includes corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data and corrected topological feature data; Model training is performed based on the geometric feature data, texture feature data, pixel-level feature data, topological feature data and the corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data, and corrected topological feature data to obtain an image correction model.

3. The AI-based embedded image correction and stitching processing method according to claim 2, characterized in that: The method of acquiring embedded system monitoring data, obtaining a system performance evaluation index according to the embedded system monitoring data, and determining an optimized image correction model and an optimized GAN generator model according to the system performance evaluation index includes: Obtain embedded system monitoring data, including model processing time, response speed, power consumption data, and memory usage; Obtaining a system performance evaluation index based on the model processing time, response speed, power consumption data and memory occupancy rate; The system performance evaluation index is compared with a preset system performance evaluation index threshold. If the system performance evaluation index is less than the preset system performance evaluation index threshold, the image correction model and the GAN generator model are quantized by using a preset model quantization algorithm to generate an optimized image correction model and an optimized GAN generator model; If the system performance evaluation index is greater than or equal to a preset system performance evaluation index threshold, the image correction model is used as the optimized image correction model, and the GAN generator model is used as the optimized GAN generator model.

4. The AI-based embedded image correction and stitching processing method according to claim 3, characterized in that: The step of acquiring the current captured image and processing the image according to the optimized image correction model to obtain the current corrected image, and processing the current corrected image according to the optimized GAN generator model to obtain the corresponding current stitched image includes: Acquire the current captured image and perform feature extraction to obtain feature data of the current captured image, including current geometric feature data, current texture feature data, current pixel-level feature data, and current topological feature data; Inputting the current geometric feature data, current texture feature data, current pixel-level feature data and current topological feature data into the optimized image correction model for processing to obtain current correction feature data and its corresponding current correction image; Extract features of the current corrected image to obtain feature data to be stitched, including content feature data of the current image, spatial feature data of the current image, and semantic feature data of the current image; The current image content feature data, the current image spatial feature data and the current image semantic feature data are input into the optimized GAN generator model for processing to obtain current splicing feature data and its corresponding current splicing image.

5. AI-based embedded image correction and stitching processing system, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a program of an embedded image correction and stitching processing method based on AI, and when the embedded image correction and stitching processing method program based on AI is executed by the processor, the following steps are implemented: Obtain the captured images of different scenes on the construction site and their corresponding corrected images and perform feature extraction, perform model training based on the feature extraction data, and obtain an image correction model; Obtain the images to be stitched and their corresponding stitched images in different scenes of the construction site and perform feature extraction. Perform model training based on the feature extraction data to obtain the GAN generator model. Acquire embedded system monitoring data, obtain a system performance evaluation index based on the embedded system monitoring data, and determine an optimized image correction model and an optimized GAN generator model based on the system performance evaluation index; Acquire the current captured image and process it according to the optimized image correction model to obtain a current corrected image, and process the current corrected image according to the optimized GAN generator model to obtain a corresponding current stitched image; Detect the current spliced ​​image to obtain spatial structure detection data, color detection data and splicing seam detection data, obtain a spatial structure evaluation index based on the spatial structure detection data, and obtain a color evaluation index based on the color detection data; Processing the splicing seam detection data to obtain a splicing quality evaluation index, and combining the spatial structure evaluation index and the color evaluation index to obtain an image quality evaluation index, and performing image processing qualification determination; Acquire the photographed images to be stitched and their corresponding stitched images under different scenes of the construction site, perform feature extraction on the photographed images to be stitched and their corresponding stitched images respectively, and obtain feature data of the images to be stitched and their corresponding stitched images; The image feature data to be spliced ​​includes boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data and image semantic feature data, and the spliced ​​image feature data includes splicing boundary geometric feature data, splicing boundary texture feature data, splicing boundary pixel-level feature data and splicing image semantic feature data; Performing GAN model training according to the boundary geometric feature data, boundary texture feature data, boundary pixel-level feature data, image semantic feature data, and the spliced ​​boundary geometric feature data, spliced ​​boundary texture feature data, spliced ​​boundary pixel-level feature data, and spliced ​​image semantic feature data to obtain a GAN generator model; The spatial structure detection data includes a structural similarity index, shape accuracy and position error, the color detection data includes a color difference value, a color histogram difference value and local color consistency data, and the seam detection data includes a seam contrast, a seam correlation data, a seam energy value, a seam entropy value and a seam gradient value; Obtaining a spatial structure evaluation index according to the structural similarity index, shape accuracy and position error processing; Obtaining a color evaluation index according to the color difference value, the color histogram difference value and the local color consistency data processing; Obtaining a splicing quality evaluation index according to the splicing seam contrast, the splicing seam correlation data, the splicing seam energy value, the splicing seam entropy value and the splicing seam gradient value; Obtaining an image quality assessment index according to the spatial structure assessment index, the color assessment index and the stitching quality assessment index; The image quality assessment index is compared with a preset image quality assessment index threshold. If the threshold comparison result does not meet the preset threshold comparison result requirement, a determination of image processing failure is made.

6. The AI-based embedded image correction and stitching processing system according to claim 5, characterized in that: The method of acquiring the captured images of different scenes of the construction site and the corresponding corrected images and performing feature extraction, and performing model training according to the feature extraction data to obtain the image correction model includes: Acquire captured images and their corresponding corrected images under different scenes of the construction site, perform feature extraction on the captured images and their corresponding corrected images respectively, and obtain feature data of the captured images and their corresponding corrected images; The captured image feature data includes geometric feature data, texture feature data, pixel-level feature data and topological feature data, and the corrected image feature data includes corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data and corrected topological feature data; Model training is performed based on the geometric feature data, texture feature data, pixel-level feature data, topological feature data and the corrected geometric feature data, corrected texture feature data, corrected pixel-level feature data, and corrected topological feature data to obtain an image correction model.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an AI-based embedded image correction and stitching processing program. When the AI-based embedded image correction and stitching processing program is executed by the processor, the steps of the AI-based embedded image correction and stitching processing method as described in any one of claims 1 to 4 are implemented.

Images