Image data analysis processing system and method based on artificial intelligence

Abstract

The invention provides an artificial intelligence-based image data analysis processing system and method. The method comprises the following steps of: determining a damage analysis sequence of a damaged image in a form damage evaluation process through pixel distribution characteristics of image data; determining a pixel missing domain at the crack in the damaged image, and determining loss edge features of all missing pixels at the crack in the damaged image through the damaged analysis sequence and the pixel missing domain; performing regularization constraint on all missing pixels in the pixel missing domain based on a pre-trained image restoration model to obtain a constrained reconstructed image, and determining restoration loss of the damaged image during restoration according to the constrained reconstructed image, the loss edge features and the pixel distribution features; and performing damage analysis on the target damaged image according to the repair loss, and further visualizing an analysis result through a damaged image data analysis system. By adopting the scheme of the invention, self-adaptive analysis of loss in a damaged image restoration process can be realized, so that the restoration precision of the damaged image is improved.

Description

Technical Field

[0001] The present application relates to the technical field of image data analysis and processing, and more specifically, to an image data analysis and processing system and method based on artificial intelligence. Background Art

[0002] With the rapid development of computer vision and artificial intelligence technologies, image data analysis and processing have been widely used in many fields, including medical image analysis, autonomous driving, security monitoring, industrial inspection, etc. Traditional image processing methods mainly rely on manually designed feature extraction and algorithm implementation. However, these methods often have limitations when dealing with complex and diverse image data.

[0003] Image data analysis and processing based on artificial intelligence mainly relies on deep learning models such as convolutional neural networks (CNN) and generative adversarial networks (GAN). Convolutional neural networks perform well in tasks such as image restoration, image classification, target detection, and semantic segmentation. In the field of image restoration, existing image restoration models can automatically fill in missing parts in images to restore the integrity of images. Although image data analysis methods based on artificial intelligence have made significant progress, existing technologies are difficult to accurately identify and repair subtle damage such as cracks and noise in old images, resulting in low quality of restored images. By adaptively analyzing the loss in the process of repairing damaged images, the repair parameters of the image repair model can be adaptively adjusted, and then the repair process of damaged images can be optimized, thereby improving the repair accuracy of damaged images. Therefore, how to achieve adaptive analysis of the loss in the process of repairing damaged images, thereby improving the repair accuracy of damaged images, is a problem faced by the industry. Summary of the invention

[0004] The present application provides an image data analysis and processing system and method based on artificial intelligence, which can realize adaptive analysis of losses in the process of damaged image restoration, thereby improving the restoration accuracy of damaged images.

[0005] In a first aspect, the present application provides an image data analysis and processing method based on artificial intelligence, comprising the following steps: Acquire image data of a target damaged image; Determining a damage analysis sequence of the damaged image in a morphological damage assessment process according to pixel distribution characteristics of the image data; Determine a pixel missing domain at the crack in the damaged image, and determine the loss edge features of all missing pixels at the crack in the damaged image through the damaged parsing sequence and the pixel missing domain; Regularizing and constraining all missing pixels in the pixel missing domain based on a pre-trained image restoration model, thereby obtaining a constrained reconstructed image, and determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature, and the pixel distribution feature; A damage analysis is performed on the target damaged image according to the repair loss, and the analysis result is visualized through the damaged image data analysis system.

[0006] In some embodiments, determining the damage analysis sequence of the damaged image in the morphological damage assessment process according to the pixel distribution characteristics of the image data specifically includes: Determining pixel distribution characteristics of each pixel in the damaged image according to the image data; Performing morphological corrosion on the damaged image based on all pixel distribution features to obtain a corrosion image corresponding to the damaged image; Performing morphological expansion on the eroded image to obtain an expanded image corresponding to the eroded image; The damaged image is subjected to damage assessment through the dilated image and the grayscale image of the damaged image to obtain a damaged parsing sequence.

[0007] In some embodiments, determining the pixel missing region at the crack in the damaged image specifically includes: Determining a plurality of missing pixels of the damaged image; Determine different crack positions of the damaged image through all missing pixels; Determining missing pixel sets at different cracks of the damaged image based on each crack position and each missing pixel point; All pixel missing sets are combined to obtain a pixel missing domain at the crack in the damaged image.

[0008] In some embodiments, determining the loss edge features of all missing pixels at the crack in the damaged image through the damaged parsing sequence and the pixel missing domain specifically includes: Obtaining each missing pixel point in the pixel missing domain; Determine the image entropy corresponding to the pixel missing domain part in the damaged image according to the gray value of each missing pixel point; Determine a normal pixel domain at a non-crack location in the damaged image from the missing pixel domain; The loss edge features of all missing pixels at the crack in the damaged image are determined by using the pixel normal domain and the image entropy.

[0009] In some embodiments, regularizing all missing pixels in the pixel missing domain based on a pre-trained image restoration model to obtain a constrained reconstructed image specifically includes: Determine the pixel difference amount of all missing pixels in the pixel missing domain according to the loss edge feature; The resolution of all missing pixels in the pixel missing domain is constrained by the loss edge feature, so as to obtain reinforced pixels and weakened pixels of all missing pixels in the pixel missing domain; Based on the enhanced pixels, the weakened pixels and the pixel difference, a pre-trained image restoration model is used to perform regular convolution on all missing pixels in the pixel missing domain to obtain a pixel reconstruction domain; Acquire a normal domain of pixels at non-crack locations in the damaged image; A constrained reconstructed image is determined by the pixel reconstruction domain and the pixel normal domain.

[0010] In some embodiments, determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature and the pixel distribution feature specifically includes: Acquiring pixel distribution characteristics of the image data; Determining an estimated restoration value of the damaged image based on the pixel distribution characteristics; determining an aggregated feature value of the damaged image using the image data; Smoothing the damaged image according to the aggregated feature value to obtain a smoothed feature value; The restoration loss when restoring the damaged image is determined based on the smoothed feature value, the aggregated feature value, the estimated restoration value and the constrained reconstructed image.

[0011] In some embodiments, the image restoration model refers to a generative adversarial network model for image restoration.

[0012] In a second aspect, the present application provides an image data analysis and processing system based on artificial intelligence, comprising: An acquisition module, used for acquiring image data of a target damaged image; A processing module, configured to determine a damage analysis sequence of the damaged image in a morphological damage assessment process according to pixel distribution characteristics of the image data; The processing module is further used to determine a pixel missing domain at the crack in the damaged image, and determine the loss edge features of all missing pixels at the crack in the damaged image through the damaged parsing sequence and the pixel missing domain; The processing module is further used to perform regularization constraints on all missing pixels in the pixel missing domain based on a pre-trained image restoration model, thereby obtaining a constrained reconstructed image, and determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature and the pixel distribution feature; An execution module is used to perform damage analysis on the target damaged image according to the repair loss, and then visualize the analysis result through the damaged image data analysis system.

[0013] In a third aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned artificial intelligence-based image data analysis and processing method.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned artificial intelligence-based image data analysis and processing method.

[0015] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects: In the artificial intelligence-based image data analysis and processing system and method provided in the present application, first, image data of a target damaged image is acquired, and a damaged parsing sequence of the damaged image in a morphological damage assessment process is determined according to the pixel distribution characteristics of the image data; secondly, a pixel missing domain at a crack in the damaged image is determined, and the loss edge characteristics of all missing pixels at the crack in the damaged image are determined through the damaged parsing sequence and the pixel missing domain; then, regularization constraints are applied to all missing pixels in the pixel missing domain based on a pre-trained image restoration model to obtain a constrained reconstructed image, and the restoration loss of the damaged image during restoration is determined according to the constrained reconstructed image, the loss edge characteristics, and the pixel distribution characteristics; finally, damage analysis is performed on the target damaged image according to the restoration loss, and the analysis results are visualized through the damaged image data analysis system.

[0016] It can be seen that the present application determines the restoration loss of the damaged image during restoration through the constrained reconstructed image, the loss edge feature and the pixel distribution feature; first, by determining the shape, size and position of the boundary of the missing pixel area in the damaged image, the loss edge feature is obtained, wherein the boundary is extracted by edge detection, which can help determine the shape and distribution of the crack, and the image entropy of the local area of ​​the image is calculated to help analyze the local texture features; secondly, regularization constraints are imposed on all missing pixels in the pixel missing domain to obtain a constrained reconstructed image, which can ensure that the texture, color and brightness of the constrained reconstructed image match the surrounding normal area, thereby avoiding obvious boundaries or faults, wherein, by constraining the resolution of the missing pixels, the noise existing in all missing pixels can be accurately identified and filtered out, and the important texture details in the damaged image are retained, so that the restored image is more accurate; finally, by determining the accumulated error in the process of repairing the damaged image, the difference between the restored image and the ideal situation can be judged, so that the restoration parameters of the image restoration model can be adaptively adjusted to improve the restoration effect of the damaged image; in summary, the present application solution can realize the adaptive analysis of the loss in the process of repairing the damaged image, thereby improving the restoration accuracy of the damaged image. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is an exemplary flow chart of an image data analysis and processing method based on artificial intelligence according to some embodiments of the present application; Figure 2 is a schematic diagram of a process for determining loss edge features according to some embodiments of the present application; Figure 3 is a schematic diagram of a process of determining a constrained reconstructed image according to some embodiments of the present application; Figure 4 is a schematic diagram of exemplary hardware and / or software of an artificial intelligence-based image data analysis and processing system according to some embodiments of the present application; Figure 5 It is a structural schematic diagram of a computer device for implementing an artificial intelligence-based image data analysis and processing method as shown in some embodiments of the present application. DETAILED DESCRIPTION

[0018] The core of this application is to determine the damaged analysis sequence of the damaged image in the process of morphological damage assessment through the pixel distribution characteristics of the image data; determine the pixel missing domain at the crack in the damaged image, and determine the loss edge characteristics of all missing pixels at the crack in the damaged image through the damaged analysis sequence and the pixel missing domain; regularize all missing pixels in the pixel missing domain based on a pre-trained image restoration model to obtain a constrained reconstructed image, and determine the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge characteristics and the pixel distribution characteristics; perform damage analysis on the target damaged image based on the restoration loss, and then visualize the analysis results through a damaged image data analysis system; this scheme can realize adaptive analysis of the loss in the process of damaged image restoration, thereby improving the restoration accuracy of the damaged image.

[0019] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 , which is an exemplary flow chart of an image data analysis and processing method based on artificial intelligence according to some embodiments of the present application. The image data analysis and processing method 100 based on artificial intelligence mainly includes the following steps: In step 101, image data of a target damaged image is acquired.

[0020] In specific implementation, the damaged image data analysis system is first started, and the image data of the target damaged image can be obtained in the following manner, namely: the image restoration model in the image damage analysis and processing system is started, the image data of the target damaged image is extracted through the feature extraction mechanism in the image restoration model, and the image data is input into the damaged image data analysis system, wherein the damaged image refers to the image texture loss caused by fading of the physical image over a long period of time, mechanical damage or other external factors, resulting in cracks in the originally clear image, and the image data refers to the collection of grayscale values ​​of each pixel in the damaged image.

[0021] It should be noted that the image restoration model in the present application refers to a generative adversarial network model for image restoration, and the feature extraction mechanism refers to a method for extracting image data in the image restoration model. The feature extraction mechanism includes a feature extraction model and a feature extraction function. As a preferred embodiment, the feature extraction mechanism can adopt the generator in the generative adversarial network. In other embodiments, other feature extraction mechanisms may also be used, which is not limited here.

[0022] In step 102, a damage analysis sequence of the damaged image in a morphological damage assessment process is determined according to pixel distribution characteristics of the image data.

[0023] In some embodiments, determining the damage analysis sequence of the damaged image in the morphological damage assessment process according to the pixel distribution characteristics of the image data can be implemented by the following steps: Determining pixel distribution characteristics of each pixel in the damaged image according to the image data; Performing morphological corrosion on the damaged image based on all pixel distribution features to obtain a corrosion image corresponding to the damaged image; Performing morphological expansion on the eroded image to obtain an expanded image corresponding to the eroded image; The damaged image is subjected to damage assessment through the dilated image and the grayscale image of the damaged image to obtain a damaged parsing sequence.

[0024] In a specific implementation, the pixel distribution characteristics of each pixel in the damaged image are determined according to the image data in the following manner, namely: first, the image data is converted into a grayscale distribution map using OpenCV in the prior art, which is conducive to analyzing the overall pixel distribution of the damaged image through the image data; secondly, the mean grayscale values ​​of all pixels in a 3×3 area adjacent to each pixel in the grayscale distribution map are calculated respectively. Calculating the mean grayscale values ​​of all pixels in the adjacent 3×3 area can retain more detailed information in the subsequent corrosion process. It should be noted that the adjacent 3×3 area refers to the area adjacent to a certain pixel point with the pixel point as the center. The adjacent 3×3 region refers to the region composed of 9 pixels (including the pixel) of the damaged image. If a pixel is located at the 4 edges of the damaged image, the adjacent 3×3 region refers to the region composed of 6 pixels (including the pixel) adjacent to the pixel. If a pixel is located at the 4 corners of the damaged image, the adjacent 3×3 region refers to the region composed of 4 pixels (including the pixel) adjacent to the pixel. Finally, each mean value calculated is used as the pixel distribution feature of the corresponding pixel in the damaged image, and all the pixel distribution features are stored in a temporary library for subsequent use, wherein the temporary library refers to a memory data structure used to temporarily store intermediate calculation results.

[0025] In addition, in the specific implementation, the damaged image is morphologically eroded by all pixel distribution characteristics, and the eroded image corresponding to the damaged image can be obtained in the following way, namely: the pixel distribution characteristics of each pixel point are obtained from a temporary library, and each pixel distribution characteristic is used as the corrosion core for each morphological corrosion, and the grayscale values ​​of all pixel points in the 3×3 area adjacent to each pixel point are replaced by the corrosion core, so as to obtain the eroded image corresponding to the damaged image.

[0026] It should be noted that in the present application, determining the pixel distribution characteristics of each pixel in the damaged image according to the image data and performing morphological corrosion on the damaged image based on all the pixel distribution characteristics are performed alternately and synchronously, that is, each time the pixel distribution characteristics of a pixel are determined, a morphological corrosion is performed on the pixels in the 3×3 area adjacent to the pixel, and after a morphological corrosion is performed on the pixels in the 3×3 area adjacent to the pixel, the pixel distribution characteristics of the next pixel are determined. Such alternating synchronization can avoid excessive corrosion and damage to the original image.

[0027] In specific implementation, the morphological expansion of the eroded image to obtain the expanded image corresponding to the eroded image can be implemented in the following manner, namely: the morphological expansion processing of the eroded image can be performed using Numpy in the prior art, and the processed image is used as the expanded image corresponding to the eroded image, wherein the convolution kernel of the morphological expansion processing can be set to 3, which is conducive to restoring the original pixels of the damaged image; the damage assessment of the damaged image is performed through the grayscale image of the expanded image and the damaged image, and the damaged analysis sequence can be implemented in the following manner, namely: the Pillow library in the prior art is used to convert the damaged image into a grayscale image, and the difference between two pixel points in the grayscale image corresponding to the same position in the expanded image is calculated, and then an array composed of all non-zero differences in the order of calculation is used as the damaged analysis sequence.

[0028] It should be noted that the damaged analysis sequence in the present application is an indicator for evaluating the degree of damage to pixels in a damaged image and a dilated image, that is, the more values ​​in the damaged analysis sequence, the greater the difference between the pixels in the damaged image and the dilated image, and the fewer values ​​in the damaged analysis sequence, the smaller the difference between the pixels in the damaged image and the dilated image, wherein the dilated image refers to the image after the damaged image is subjected to morphological erosion and morphological dilation processing, and the morphological erosion and morphological dilation processing of the damaged image is to remove the small differences in pixels in the damaged image.

[0029] In step 103, a pixel missing domain at the crack in the damaged image is determined, and loss edge features of all missing pixels at the crack in the damaged image are determined through the damaged parsing sequence and the pixel missing domain.

[0030] In some embodiments, determining the pixel missing region at the crack in the damaged image may be achieved by using the following steps: Determining a plurality of missing pixels of the damaged image; Determine different crack positions of the damaged image through all missing pixels; Determining missing pixel sets at different cracks of the damaged image based on each crack position and each missing pixel point; All pixel missing sets are combined to obtain a pixel missing domain at the crack in the damaged image.

[0031] In specific implementation, determining multiple missing pixel points of the damaged image can be achieved in the following manner, namely: obtaining a damaged parsing sequence of the damaged image in the process of morphological damage assessment, and using an edge detection algorithm in the prior art to detect pixel points corresponding to each difference in the damaged parsing sequence (the pixel points are pixel points in the damaged image), and then using all detected pixel points as missing pixel points; determining different crack positions of the damaged image through all missing pixel points can be achieved in the following manner, namely: using an edge detection algorithm to detect connected positions of connected missing pixel points in all missing pixels, and using each connected position as a different crack position of the damaged image, wherein the connected position refers to a common position of connected missing pixel points in all missing pixel points, and the connected missing pixel points refer to a set composed of all adjacent edge and oblique edge missing pixel points; determining pixel missing sets at different crack locations of the damaged image from each crack position and each missing pixel point can be achieved in the following manner, namely: using a set composed of all connected missing pixel points corresponding to each crack position as a pixel missing set at different crack locations of the damaged image; and then combining all pixel missing sets into a pixel missing domain at the crack location in the damaged image.

[0032] It should be noted that the pixel missing domain in the present application refers to the area obtained by combining all pixel missing sets, wherein the pixel missing set refers to a set of all missing pixel points corresponding to different crack positions. Pixel points with differences are regarded as missing pixel points, and the damaged position of the damaged image can be determined, which is conducive to the subsequent analysis of the damage degree of the damaged image.

[0033] In some embodiments, reference Figure 2 As shown, this figure is a schematic diagram of a process for determining loss edge features shown in some embodiments of the present application. In this embodiment, the loss edge features of all missing pixels at the crack in the damaged image are determined by the damaged parsing sequence and the pixel missing domain, which can be implemented by the following steps: In step 1031, each missing pixel point in the pixel missing domain is obtained; In step 1032, the image entropy corresponding to the pixel missing domain portion in the damaged image is determined according to the grayscale value of each missing pixel point; In step 1033, a normal pixel domain at a non-crack location in the damaged image is determined from the missing pixel domain; In step 1034, the loss edge features of all missing pixels at the cracks in the damaged image are determined using the pixel normal domain and the image entropy.

[0034] In specific implementation, obtaining each missing pixel point in the pixel missing domain can be achieved in the following manner, namely: using OpenCV in the prior art to extract all missing pixel points corresponding to the position of the pixel missing domain; determining the image entropy corresponding to the pixel missing domain part in the damaged image according to the grayscale value of each missing pixel point can be achieved in the following manner, namely: substituting the grayscale values ​​of all missing pixels into the image entropy calculation formula, and using the calculation result as the image entropy corresponding to the pixel missing domain part in the damaged image, the image entropy refers to an indicator to measure the aggregation characteristics of the grayscale distribution of an image, wherein entropy refers to the degree of disorder of the system, the larger the image entropy, the more disordered the pixel values ​​of the pixel missing domain part in the damaged image, and the smaller the image entropy, the more stable the pixel values ​​of the pixel missing domain part in the damaged image.

[0035] In addition, in a specific implementation, determining the normal pixel domain at the non-crack location in the damaged image from the pixel missing domain can be achieved in the following manner, namely: copying a layer in the damaged image analysis interface, and using the Canny algorithm in the prior art to detect the pixel missing domain in the copied layer, and then removing the detected pixel missing domain, and using the remaining part of the damaged image copy layer as the normal pixel domain at the non-crack location in the damaged image.

[0036] In specific implementation, determining the loss edge features of all missing pixels at the crack in the damaged image through the normal pixel domain and the image entropy can be achieved in the following manner, namely: first, calculating the image entropy of the normal pixel domain; second, taking the average of the image entropy of the normal pixel domain and the image entropy corresponding to the missing pixel domain as a contour analysis index for analyzing the contour of the missing pixels, wherein the contour analysis index refers to a comprehensive value of the difference in edge details of the missing pixels in the damaged image, and the contour analysis index can help identify the detail area in the damaged image; then, using the contour detection algorithm in the prior art to detect the edge of the missing pixel domain, and using the contour analysis index as a threshold for contour detection before detection; finally, extracting the loss edge features of all missing pixels at the crack in the damaged image through the contour detection algorithm after setting the threshold, wherein by setting the threshold, important contour features in the damaged image can be retained and enhanced, avoiding ignoring the edge area with rich details in the process of extracting the loss edge features, so that the analysis process focuses on the crack area in the damaged image.

[0037] It should be noted that the loss edge feature in the present application refers to the shape, size and position of the boundary of the missing pixel area in the damaged image. The loss edge feature is beneficial to subsequent image restoration. For example, edge detection is used to extract the boundary, which can help determine the shape and distribution of the crack; the image entropy of the local area of ​​the image is calculated to help analyze the local texture features; in the process of extracting the loss edge feature in the present embodiment, the image entropy of the normal domain of the pixel and the image entropy corresponding to the pixel missing domain are calculated, and the image entropy is combined with the existing contour detection algorithm, which is beneficial to analyze the damage caused by external factors in the damaged image, thereby judging the degree of damage of the damaged image.

[0038] In step 104, regularization constraints are performed on all missing pixels in the pixel missing domain based on a pre-trained image restoration model to obtain a constrained reconstructed image. The restoration loss of the damaged image during restoration is determined by the constrained reconstructed image, the loss edge features, and the pixel distribution features.

[0039] In some embodiments, reference Figure 3 As shown, this figure is a schematic diagram of the process of determining the constrained reconstructed image shown in some embodiments of the present application. In this embodiment, all missing pixels in the pixel missing domain are regularized and constrained based on the pre-trained image restoration model, and then the constrained reconstructed image is obtained, which can be implemented by the following steps: In step 1041, the pixel difference amounts of all missing pixels in the pixel missing domain are determined according to the loss edge feature; In step 1042, the resolution of all missing pixels in the missing pixel domain is constrained by the loss edge feature, so as to obtain reinforced pixels and weakened pixels of all missing pixels in the missing pixel domain; In step 1043, a regularized convolution is performed on all missing pixels in the pixel missing domain based on the enhanced pixels, the weakened pixels and the pixel difference through a pre-trained image restoration model to obtain a pixel reconstruction domain; In step 1044, a normal domain of pixels at non-crack locations in the damaged image is obtained; In step 1045, a constrained reconstructed image is determined from the pixel reconstruction domain and the pixel normal domain.

[0040] In a specific implementation, determining the pixel difference amount of all missing pixels in the pixel missing domain according to the loss edge feature can be implemented in the following manner, namely: the loss edge feature refers to describing the shape, size and position of the boundary of the missing pixel region in the damaged image. Through the loss edge feature, each pixel point of the boundary of the missing pixel region in the pixel missing domain is used as a center point, and each center point is used to divide a plurality of 3×3 pixel regions. The division of the plurality of 3×3 pixel regions can integrate the missing pixel region and the pixel normal region in the damaged image. The 3×3 pixel region can be divided with reference to the adjacent 3×3 region in the aforementioned embodiment, which will not be described in detail here. Then, the variance of all pixels in each 3×3 pixel region is calculated, and the mean of all variances is used as the pixel difference amount of all missing pixels in the pixel missing domain. The pixel difference amount represents an indicator of the degree of difference between the missing pixel region and the pixel normal region in the damaged image, that is, the larger the pixel difference amount, the more obvious the difference between the missing pixel region and the pixel normal region in the damaged image, and the smaller the pixel difference amount, the smaller the difference between the missing pixel region and the pixel normal region in the damaged image.

[0041] In addition, in a specific implementation, the resolution of all missing pixels in the pixel missing domain is constrained by the loss edge feature, and the reinforcement pixels and weakened pixels of all missing pixels in the pixel missing domain are obtained. This can be achieved in the following manner, namely: all missing pixels in the pixel missing domain are upsampled and downsampled respectively, and the reinforcement pixels and weakened pixels of all missing pixels in the pixel missing domain can be obtained. As a preferred embodiment, the loss edge feature can be used as an indicator for averaging processing during upsampling and downsampling, and cv2 in the prior art can be used to upsample all missing pixels in the pixel missing domain, and all missing pixels after upsampling to enhance the resolution are used as reinforcement pixels. All missing pixels in the pixel missing domain are then downsampled, and all missing pixels after downsampling to reduce the resolution are used as reinforcement pixels.

[0042] In a specific implementation, based on the enhanced pixels, the weakened pixels and the pixel difference amount, a regular convolution is performed on all the missing pixels in the pixel missing domain through a pre-trained image restoration model to obtain a pixel reconstruction domain, which can be implemented in the following manner, namely, the image pyramid technology in the prior art can be used to perform secondary training on the image restoration model in the aforementioned embodiment, and then the trained image restoration model is used to perform regular convolution on all the missing pixels in the pixel missing domain to obtain a pixel reconstruction domain, namely, all the missing pixels in the pixel missing domain are used as the bottom layer of the pyramid, the enhanced pixels and the weakened pixels are used as the upper two layers of the pyramid in turn, and the pixel difference amount is used as the convolution kernel in the convolution process, and then the pixel area obtained by convolving the trained image restoration model is used as the pixel reconstruction domain; determining the constrained reconstructed image from the pixel reconstruction domain and the pixel normal domain can be implemented in the following manner, namely, an image combined from the pixel normal domain and the pixel reconstruction domain is used as the constrained reconstructed image.

[0043] It should be noted that in the present application, the purpose of performing regularization constraints on all missing pixels in the pixel missing domain is to fill the missing areas of pixels in the target damaged image. After regularization constraints and reconstruction of the missing pixels, it can be ensured that the texture, color and brightness of the constrained reconstructed image match the surrounding normal areas, thereby avoiding obvious boundaries or faults. Among them, by constraining the resolution of the missing pixels, it is possible to filter out the details and noise in all missing pixels, and retain important texture details in the damaged image, making the repaired image clearer and more accurate.

[0044] In some embodiments, determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature and the pixel distribution feature can be achieved by using the following steps: Acquiring pixel distribution characteristics of the image data; Determining an estimated restoration value of the damaged image based on the pixel distribution characteristics; determining an aggregated feature value of the damaged image using the image data; Smoothing the damaged image according to the aggregated feature value to obtain a smoothed feature value; The restoration loss when restoring the damaged image is determined based on the smoothed feature value, the aggregated feature value, the estimated restoration value and the constrained reconstructed image.

[0045] It should be noted that the restoration loss in the present application refers to the accumulated error in the process of restoring a damaged image. The restoration loss represents the loss error between the actual restoration effect and the ideal restoration effect in the process of restoring a damaged image. The restoration loss is used to evaluate the degree of error generated in the restoration process. The restoration loss can be determined based on the smoothing eigenvalue, the aggregated eigenvalue, the estimated restoration value and the constrained reconstructed image, wherein the smoothing eigenvalue refers to the eigenvalue obtained by smoothing the damaged image, which is used to measure the image smoothing effect; the aggregated eigenvalue refers to the eigenvalue calculated by the standard deviation of the image data, which is used to analyze the degree of aggregation of the image data; the estimated restoration value refers to the mean of the pixel distribution characteristics in the damaged image after processing by the isolation forest algorithm, which can be used as a restoration value for estimating the restoration effect; as an important indicator for evaluating the restoration effect, the restoration loss can not only be used to judge the difference between the restored image and the ideal situation, but the calculation of the restoration loss can also select and adjust the image features to improve the restoration effect.

[0046] In specific implementation, determining the estimated restoration value of the damaged image by the pixel distribution feature can be achieved in the following manner, namely: after processing the image data of the damaged image using the isolation forest algorithm in the prior art, the standard deviation of all the means in the pixel distribution feature is used as the estimated restoration value, wherein the isolation forest algorithm is used for processing in order to filter out abnormal image data caused by external interference factors; determining the aggregate feature value of the damaged image by the image data can be achieved in the following manner, namely: calculating the standard deviation of the image data, and using the standard deviation as the aggregate feature value of the damaged image, and the degree of aggregation of the image data can be analyzed by calculating the standard deviation.

[0047] In addition, in a specific implementation, the damaged image is smoothed according to the aggregated eigenvalues, and the smoothed eigenvalues ​​can be obtained in the following manner, namely: the damaged image is smoothed using the bilateral filtering technology in the prior art, and the aggregated eigenvalues ​​are used as a direction reference for image smoothing, the maximum pixel difference in the smoothing process is extracted, and the value is used as a smoothed eigenvalue. Using the aggregated eigenvalues ​​as a direction reference is beneficial for increasing the protection of the missing edges of the pixels of the damaged image while smoothing the image. The main reason for using the bilateral filtering technology is that the bilateral filter consists of two parts, one part is related to the pixel space distance, and the other part is related to the pixel difference of the pixel point.

[0048] In specific implementation, the restoration loss when repairing the damaged image is determined based on the smoothing eigenvalue, the aggregated eigenvalue, the estimated restoration value and the constrained reconstructed image, which can be achieved in the following manner, namely: taking the difference between the smoothing eigenvalue and the estimated restoration value as the smoothing deviation, and taking the difference between the aggregated eigenvalue and the estimated restoration value as the aggregation deviation, and then taking the pixel average value of all pixels in the constrained reconstructed image as the reconstruction mean, and finally taking the absolute difference between the sum of the smoothing deviation and the aggregated deviation and the reconstruction mean as the restoration loss when repairing the damaged image.

[0049] In step 105, a damage analysis is performed on the target damaged image according to the repair loss, and the analysis result is visualized through the damaged image data analysis system.

[0050] In some embodiments, performing damage analysis on the target damaged image according to the repair loss may be implemented by the following steps: Acquire the surface roughness of the target damaged image based on the pre-trained feature model; determining, according to the surface roughness, guiding features for repairing the damaged image; The loss of the damaged image is estimated by using the guiding feature and the restoration loss to obtain a loss compensation parameter in the process of restoring missing pixels; Obtain historical damage assessment data; The damage degree of the target damaged image is determined by using the damage degree assessment data and the loss compensation parameter.

[0051] In specific implementation, the surface roughness of the target damaged image can be obtained based on the pre-trained feature model in the following manner, namely: write code for secondary training of the VGG model, perform secondary training on the VGG model through a pre-made image training set, and use the trained VGG model mainly to analyze image features, such as: color features for image retrieval and image classification, texture features for object recognition and image segmentation, shape features for object detection and recognition, etc., wherein the surface roughness belongs to texture features, and the surface roughness is an indicator for measuring the slight deviation and irregularity of pixels on the image surface. The surface roughness can also describe the microscopic unevenness of the image surface, and the VGG model after secondary training can be used to extract the surface roughness of the target damaged image.

[0052] In some embodiments, determining the guiding features for repairing the damaged image according to the surface roughness may be implemented by the following steps: Obtaining surface roughness standards from historical restoration images; Initializing a guiding feature mapping model using the surface roughness standard as an initialization parameter; Using the surface roughness as a constraint parameter of the guiding feature mapping model; The damaged image is repaired through the guiding feature mapping model to obtain the guiding features for repair.

[0053] In specific implementation, obtaining the surface roughness standard from the historical repaired image data can be achieved in the following manner, namely: obtaining all the historical damaged images after repair, and using the VGG model after secondary training to extract the historical surface roughness of all the historical damaged images, and taking the average of all the historical surface roughness as the surface roughness standard, which can be used to guide the repair of the current damaged image; using the surface roughness standard as the initialization parameter, initializing a guiding feature mapping model can be achieved in the following manner, namely: using the surface roughness standard as the initialization parameter, setting the convolution kernel, bias and other parameters of the guiding feature mapping model, using the surface roughness standard extracted from the historical repair data, so that the guiding feature mapping model can have a certain ability to identify and process the surface features of the damaged image, and the surface roughness standard enables the model to consider the flatness of the damaged image during the repair process, so as to perform the repair more accurately and improve the overall quality of the damaged image after repair.

[0054] In addition, using the surface roughness as a constraint parameter of the guiding feature mapping model can help the guiding feature mapping model better capture and restore the details of the damaged image during the repair process. The surface roughness can also dynamically adjust the repair strategy to avoid excessive smoothing and distortion when repairing the damaged image. Repairing the damaged image through the guiding feature mapping model to obtain the guiding features for repair can be achieved in the following manner, namely: passing the damaged image and the surface roughness as inputs into the guiding feature mapping model, extracting image features using the convolutional neural network in the guiding feature mapping model, adjusting the repair strategy based on the extracted features and the surface roughness, and then extracting repair parameters from the adjusted repair strategy, so that the repair parameters are used as guiding features for repair.

[0055] It should be noted that the guiding features in this application are guiding values ​​when repairing the damaged image based on the artificial intelligence model. The guiding values ​​include several aspects, such as: color, texture, shape, grayscale value, etc. Applying the guiding features to the repair results can maintain the texture and details of the image, making the repaired image more natural and realistic.

[0056] In some embodiments, the loss of the damaged image is estimated by using the guiding features and the restoration loss, and the loss compensation parameter in the process of restoring the missing pixels is obtained by the following steps: Acquiring image data of the damaged image; Tracing the image data of the damaged image according to the guiding feature to obtain tracing guiding data; Roughly estimating the damaged image using the image data and the traceability guidance data, thereby obtaining a rough compensation amount when repairing the damaged image; The loss compensation parameter in the process of repairing the missing pixels is determined according to the roughness compensation amount and the repair loss.

[0057] In specific implementation, the image data of the damaged image is traced according to the guiding features, and the tracing guidance data can be obtained in the following manner, namely: the guiding features can be set as the tracing parameters of the image restoration model in the prior art, and the damaged image can be preliminarily restored, and then the image data of the image after the preliminary restoration can be extracted using a feature extraction mechanism, and the extracted image data can be used as the tracing guidance data; the feature extraction mechanism can use the generator in the generative adversarial network, which will not be repeated here.

[0058] In addition, in a specific implementation, a rough estimation of the damaged image is performed using the image data and the tracing guidance data, and then a rough compensation amount for repairing the damaged image is obtained, which can be achieved in the following manner, namely: calculating the standard deviation of the tracing guidance data, and then calculating the standard deviation of the image data, and using the absolute value of the difference between the two standard deviations as the rough compensation amount for repairing the damaged image, wherein the roughness refers to the compensation value required to repair the surface texture of the damaged image.

[0059] In specific implementation, the loss compensation parameter in the process of repairing missing pixels is determined according to the roughness compensation amount and the repair loss, which can be achieved in the following manner, namely: the roughness compensation amount is to repair the damaged image according to the surface roughness, so as to obtain the difference in the degree of aggregation between the repaired image data and the original image data; the repair loss refers to the error accumulation caused by various factors in the repair process when repairing the damaged image; the loss compensation parameter in the process of repairing missing pixels can be obtained by combining the roughness compensation amount and the repair loss; as a preferred embodiment, the loss compensation parameter can be determined by the sum of the roughness compensation amount and the repair loss.

[0060] It should be noted that the loss compensation parameter in the present application refers to the compensation amount required to compensate for the errors and detail loss generated in the repair process during the repair of missing pixels. The loss compensation parameter is determined by comprehensively analyzing the roughness compensation amount and the repair loss in the repair process. The loss compensation parameter reflects the additional adjustment required when repairing missing pixels. The calculation of the loss compensation parameter takes into account the roughness compensation amount and the repair loss, and can play an adjustment and optimization role in the process of repairing the damaged image, so that the repaired image is closer to the original image and reduces unnecessary errors and detail loss caused by the repair.

[0061] In specific implementation, obtaining historical damage assessment data can be achieved in the following manner, namely: obtaining the historical damage degree at each repair analysis, and using all historical damage degrees as damage assessment data; initializing an assessment model based on the damage assessment data can be achieved in the following manner, namely: setting a historical assessment form for the assessment model based on the damage assessment data, wherein the historical assessment form is a display module in the assessment model, which can select the historical damage degrees at the time of nearly 20 repair analyses, and display the selected historical damage degrees in the historical assessment form in chronological order, so as to facilitate judgment of the reliability of the assessment model.

[0062] In addition, in a specific implementation, determining the degree of damage of the target damaged image through the damage degree assessment data and the loss compensation parameter can be achieved in the following manner, namely: the degree of damage refers to the degree of pixel damage and detail loss in the target damaged image, the loss compensation parameter can be used as the input of the aforementioned assessment model, the reciprocal of the loss compensation parameter is calculated by the calculation module of the assessment model, and then the difference between the standard constant and the reciprocal is used as the degree of damage of the target damaged image, wherein the value of the standard constant is usually 1, so that the loss compensation parameter is linearly related to the degree of damage of the target damaged image, that is: the larger the loss compensation parameter, the greater the degree of damage of the target damaged image, and the smaller the loss compensation parameter, the smaller the degree of damage of the target damaged image, and the value of the degree of damage is between 0-1.

[0063] It should be noted that in the present application, visualizing the analysis results through the damaged image data analysis system means using the visualization library of the damaged image data analysis system (for example: Matplotlib) to display the damage degree after the damage degree assessment data, so as to facilitate intuitive analysis of the damage degree of the target damaged image.

[0064] On the other hand, in some embodiments, the present application provides an image data analysis and processing system based on artificial intelligence, referring to Figure 4, which is a schematic diagram of exemplary hardware and / or software of an image data analysis and processing system based on artificial intelligence according to some embodiments of the present application. The image data analysis and processing system based on artificial intelligence 400 includes: an acquisition module 401, a processing module 402 and an execution module 403, which are described as follows: Acquisition module 401, in this application, acquisition module 401 is mainly used to acquire image data of a target damaged image; Processing module 402, in the present application, the processing module 402 is used to determine the damage analysis sequence of the damaged image in the morphological damage assessment process according to the pixel distribution characteristics of the image data; In the present application, the processing module 402 is further used to determine a pixel missing domain at the crack in the damaged image, and determine the loss edge features of all missing pixels at the crack in the damaged image through the damaged parsing sequence and the pixel missing domain; The processing module 402 in the present application is also used to perform regularization constraints on all missing pixels in the pixel missing domain based on a pre-trained image restoration model, thereby obtaining a constrained reconstructed image, and determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature and the pixel distribution feature; Execution module 403, in the present application, the execution module 403 is mainly used to perform damage analysis on the target damaged image according to the repair loss, and then visualize the analysis result through the damaged image data analysis system.

[0065] In addition, the present application also provides a computer device, which includes a memory and a processor, the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned artificial intelligence-based image data analysis and processing method.

[0066] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing an image data analysis and processing method based on artificial intelligence according to some embodiments of the present application. The image data analysis and processing method based on artificial intelligence in the above embodiments can be Figure 5 The computer device 500 shown in the figure is implemented, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503 and at least one communication interface 504.

[0067] The processor 501 may be a general-purpose central processing unit (CPU) or an application specific integrated circuit (ASIC).

[0068] The communication bus 502 may be used to transmit information between the above-mentioned components.

[0069] The memory 503 may be a read only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read only memory (EEPROM), a compact disc read only memory (CD ROM) or other optical disc storage, an optical disc storage (including a compressed optical disc, a laser disc, an optical disc, a digital versatile disc, a Blu-ray disc, etc.), a magnetic disk or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 503 may exist independently and be connected to the processor 501 via the communication bus 502. The memory 503 may also be integrated with the processor 501.

[0070] The memory 503 is used to store the program code for executing the solution of the present application, and the execution is controlled by the processor 501. The processor 501 is used to execute the program code stored in the memory 503. The program code may include one or more software modules. The image data analysis and processing method based on artificial intelligence in the above embodiment can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0071] The communication interface 504 uses any transceiver or other device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0072] In a specific implementation, as an embodiment, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions).

[0073] The above-mentioned computer device may be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the computer device may be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device or an embedded device. The embodiment of the present application does not limit the type of computer device.

[0074] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned artificial intelligence-based image data analysis and processing method.

[0075] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0076] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. An image data analysis and processing method based on artificial intelligence, characterized in that: The steps include: Acquire image data of a target damaged image; Determining a damage analysis sequence of the damaged image in a morphological damage assessment process according to pixel distribution characteristics of the image data; Determine a pixel missing domain at the crack in the damaged image, and determine the loss edge features of all missing pixels at the crack in the damaged image through the damaged parsing sequence and the pixel missing domain; Regularizing and constraining all missing pixels in the pixel missing domain based on a pre-trained image restoration model, thereby obtaining a constrained reconstructed image, and determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature, and the pixel distribution feature; A damage analysis is performed on the target damaged image according to the repair loss, and the analysis result is visualized through the damaged image data analysis system.

2. The method according to claim 1, characterized in that Determining the damage analysis sequence of the damaged image in the morphological damage assessment process according to the pixel distribution characteristics of the image data specifically includes: Determining pixel distribution characteristics of each pixel in the damaged image according to the image data; Performing morphological corrosion on the damaged image based on all pixel distribution features to obtain a corrosion image corresponding to the damaged image; Performing morphological expansion on the eroded image to obtain an expanded image corresponding to the eroded image; The damaged image is subjected to damage assessment through the dilated image and the grayscale image of the damaged image to obtain a damaged parsing sequence.

3. The method according to claim 1, characterized in that Determining the pixel missing region at the crack in the damaged image specifically includes: Determining a plurality of missing pixels of the damaged image; Determine different crack positions of the damaged image through all missing pixels; Determining missing pixel sets at different cracks of the damaged image based on each crack position and each missing pixel point; All pixel missing sets are combined to obtain a pixel missing domain at the crack in the damaged image.

4. The method according to claim 1, characterized in that Determining the loss edge features of all missing pixels at the crack in the damaged image through the damaged parsing sequence and the pixel missing domain specifically includes: Obtaining each missing pixel point in the pixel missing domain; Determine the image entropy corresponding to the pixel missing domain part in the damaged image according to the gray value of each missing pixel point; Determine a normal pixel domain at a non-crack location in the damaged image from the missing pixel domain; The loss edge features of all missing pixels at the crack in the damaged image are determined by using the pixel normal domain and the image entropy.

5. The method according to claim 1, characterized in that Regularizing and constraining all missing pixels in the pixel missing domain based on the pre-trained image restoration model, and then obtaining a constrained reconstructed image specifically includes: Determine the pixel difference amount of all missing pixels in the pixel missing domain according to the loss edge feature; The resolution of all missing pixels in the pixel missing domain is constrained by the loss edge feature to obtain reinforced pixels and weakened pixels of all missing pixels in the pixel missing domain; Based on the enhanced pixels, the weakened pixels and the pixel difference, a pre-trained image restoration model is used to perform regular convolution on all missing pixels in the pixel missing domain to obtain a pixel reconstruction domain; Acquire a normal domain of pixels at non-crack locations in the damaged image; A constrained reconstructed image is determined by the pixel reconstruction domain and the pixel normal domain.

6. The method according to claim 1, characterized in that Determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature and the pixel distribution feature specifically includes: Acquiring pixel distribution characteristics of the image data; Determining an estimated restoration value of the damaged image based on the pixel distribution characteristics; determining an aggregated feature value of the damaged image using the image data; Smoothing the damaged image according to the aggregated feature value to obtain a smoothed feature value; The restoration loss when restoring the damaged image is determined based on the smoothed feature value, the aggregated feature value, the estimated restoration value and the constrained reconstructed image.

7. The method according to claim 1, characterized in that The image restoration model refers to a generative adversarial network model used for image restoration.

8. An image data analysis and processing system based on artificial intelligence, characterized in that: include: An acquisition module, used for acquiring image data of a target damaged image; A processing module, configured to determine a damage analysis sequence of the damaged image in a morphological damage assessment process according to pixel distribution characteristics of the image data; The processing module is further used to determine a pixel missing domain at the crack in the damaged image, and determine the loss edge features of all missing pixels at the crack in the damaged image through the damaged parsing sequence and the pixel missing domain; The processing module is further used to perform regularization constraints on all missing pixels in the pixel missing domain based on a pre-trained image restoration model, thereby obtaining a constrained reconstructed image, and determining the restoration loss of the damaged image during restoration based on the constrained reconstructed image, the loss edge feature and the pixel distribution feature; An execution module is used to perform damage analysis on the target damaged image according to the repair loss, and then visualize the analysis result through the damaged image data analysis system.

9. A computer device, comprising a memory and a processor, wherein the memory stores a code, characterized in that: The processor is configured to obtain the code and execute the artificial intelligence-based image data analysis and processing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the image data analysis and processing method based on artificial intelligence as described in any one of claims 1 to 7 is implemented.

Images