Artificial intelligence image recognition data processing method and system

Abstract

The invention discloses an artificial intelligence image recognition data processing method and system, and relates to the technical field of artificial intelligence. The method comprises the steps of collecting a target image according to an application field of image recognition and a recognition target, performing image quality detection and restoration on the target image through an image quality detection and restoration algorithm, highlighting the target image through an image enhancement technology, generating a classification standard according to the recognition target of image recognition, and performing classification. A target image is fused through an image main body structure texture fusion algorithm, features of the fused image are extracted through a convolutional neural network, and the features are fused through an image feature fusion algorithm to generate a classification image feature fusion set. According to the method, the low-quality images are repaired to improve the image quality, the advanced fusion technology is used for fusing the images, key features of the images are reserved, potential complementary features between the images are mined, the overall expressive force of the images is enhanced, and a reliable data basis is provided for subsequent image recognition.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence image recognition data processing method and system. Background Art

[0002] In today's field of artificial intelligence image recognition, accurate and efficient data processing faces many challenges. On the one hand, the quality of images actually collected varies, and there are problems such as noise interference, uneven light, and blur, which seriously affect the extraction of effective information in the image, thereby reducing the accuracy of image recognition. For example, in security monitoring scenarios, low-quality images may lead to misjudgment of people or objects. On the other hand, with the development of multimodal image acquisition technology, such as combining visible light images with infrared images, richer image information can be obtained, but the existing image fusion technology is insufficient in fusion accuracy and efficiency, and it is difficult to fully fuse images with high precision. At the same time, different types of image features lack an effective integration strategy in the fusion process, resulting in the inability of the fused features to fully and accurately describe the image content, which restricts the improvement of image recognition performance. Summary of the invention

[0003] The present invention provides an artificial intelligence image recognition data processing method, comprising: Step S1: The cloud platform collects target images according to the application field of image recognition and the recognition target, and preprocesses the collected target images through image preprocessing technology, and generates a target image set based on the preprocessed target images; Step S2: The cloud platform performs image quality detection and repair on the target image set through an image quality detection and repair algorithm to generate a high-quality image set; the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repair the target image whose image detection quality value does not reach the preset image quality threshold; obtain the image recognition subject according to the recognition target of image recognition and highlight the image recognition subject through image enhancement technology to generate a high-quality highlighted image set; Step S3, the cloud platform generates a classification standard according to the recognition target of image recognition, and classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; the target image in the classified highlighted image subset in the classified highlighted image set is subjected to image recognition subject fusion through an image subject structure texture fusion algorithm to generate a classified subject fusion image and construct a classified fusion image set; Step S4: The cloud platform extracts the image features of the classified main fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses the features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set.

[0004] As described above, an artificial intelligence image recognition data processing method, wherein the cloud platform collects target images according to the application field of image recognition and the recognition target, and pre-processes the collected target images through image pre-processing technology, and generates a target image set based on the pre-processed target images, including the following sub-steps: Step S11: The cloud platform obtains an image acquisition object of a target image according to the application field of image recognition and the recognition target; Step S12: The cloud platform selects an image acquisition method according to the image acquisition object of the target image and acquires the target image; Step S13: The cloud platform performs image preprocessing on the collected target images through image preprocessing technology to construct a target image set.

[0005] An artificial intelligence image recognition data processing method as described above, wherein the cloud platform performs image quality detection and repair on the target image set through an image quality detection and repair algorithm to generate a high-quality image set; the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repair the target image whose image detection quality value does not reach the preset image quality threshold; the image recognition subject is obtained according to the recognition target of the image recognition and the image recognition subject is highlighted through the image enhancement technology, and the generation of the high-quality highlighted image set includes the following sub-steps: Step S21: The cloud platform performs image quality detection and restoration on the target image set through an image quality detection and restoration algorithm to generate a high-quality image set; Step S22: The cloud platform obtains the image recognition subject of the target image according to the recognition target of the image recognition, and highlights the image recognition subject of the target image in the high-quality image set through image enhancement technology to generate a high-quality highlighted image set.

[0006] As described above, an artificial intelligence image recognition data processing method, wherein the cloud platform generates a classification standard according to the recognition target of image recognition, classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; and performs image recognition subject fusion on the target image in the classified highlighted image subset in the classified highlighted image set through an image subject structure texture fusion algorithm to generate a classified subject fusion image and construct a classified fusion image set, including the following sub-steps: Step S31: The cloud platform generates a classification standard according to the recognition target of the image recognition, classifies the high-quality highlighted image set according to the classification standard, and generates a classified highlighted image set; Step S32: The cloud platform uses an image main structure texture fusion algorithm to perform image main fusion on the target image in the classified highlighted image set, generates a classified main fusion image, and constructs a classified fusion image set.

[0007] The above-mentioned artificial intelligence image recognition data processing method, wherein the cloud platform extracts the image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses the features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set, includes the following sub-steps: Step S41: The cloud platform extracts image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set; Step S42: The cloud platform fuses the image features of the classified subject fusion image in the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set.

[0008] The present invention also provides an artificial intelligence image recognition data processing system, comprising: An image acquisition and processing module, which acquires target images according to the application field of image recognition and the recognition target, pre-processes the acquired target images through image pre-processing technology, and generates a target image set based on the pre-processed target images; The image detection, repair and highlighting module performs image quality detection and repair on the target image set through an image quality detection and repair algorithm to generate a high-quality image set; the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repairs the target image whose image detection quality value does not reach the preset image quality threshold; obtains the image recognition subject according to the recognition target of image recognition and highlights the image recognition subject through image enhancement technology to generate a high-quality highlighted image set; An image classification and fusion module generates a classification standard according to the recognition target of image recognition, and classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; the target image in the classified highlighted image subset in the classified highlighted image set is subjected to image recognition subject fusion through an image subject structure texture fusion algorithm to generate a classified subject fusion image and construct a classified fusion image set; The image feature extraction and fusion module extracts the image features of the classified main fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses the features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set.

[0009] In the above-mentioned artificial intelligence image recognition data processing system, the image acquisition and processing module specifically includes: An image acquisition object acquisition submodule obtains an image acquisition object of a target image according to the application field of image recognition and the recognition target; An image acquisition submodule selects an image acquisition method according to an image acquisition object of a target image and acquires the target image; The image preprocessing submodule performs image preprocessing on the collected target images through image preprocessing technology to construct a target image set.

[0010] In the above-mentioned artificial intelligence image recognition data processing system, the image detection, repair and highlighting module specifically includes: The image detection and restoration submodule performs image quality detection and restoration on the target image set through the image quality detection and restoration algorithm to generate a high-quality image set; The image highlighting submodule obtains the image recognition subject of the target image according to the recognition target of image recognition, highlights the image recognition subject of the target image in the high-quality image set through image enhancement technology, and generates a high-quality highlighted image set.

[0011] In the above-mentioned artificial intelligence image recognition data processing system, the image classification and fusion module specifically includes: The image classification submodule generates a classification standard according to the recognition target of the image recognition, classifies the high-quality highlighted image set according to the classification standard, and generates a classified highlighted image set; The image fusion submodule uses the image main structure texture fusion algorithm to fuse the target image in the classified highlight image set, generate a classified main fusion image and construct a classified fusion image set.

[0012] In the above-mentioned artificial intelligence image recognition data processing system, the image feature extraction and fusion module specifically includes: The image feature extraction submodule extracts the image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set; The image feature fusion submodule fuses the image features of the classified subject fusion image in the classified fusion image feature set through the image feature fusion algorithm to generate a classified image feature fusion set.

[0013] The beneficial effects achieved by the present invention are as follows: the present invention can repair low-quality images to improve image quality, use advanced fusion technology to fuse images, retain the key features of the images and explore the potential complementary features between images, thereby enhancing the overall expressiveness of the images, providing a more reliable data basis for subsequent image recognition, significantly improving the accuracy of image recognition, and enabling the image recognition model to understand the image content more accurately, thereby greatly improving the performance and robustness of the image recognition system in complex scenarios, and providing more powerful technical support for image recognition applications in multiple fields such as security, medical care, and transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0015] Figure 1 This is a flow chart of an artificial intelligence image recognition data processing method provided in Example 1 of the present application; Figure 2 This is a schematic diagram of an artificial intelligence image recognition data processing system provided in Example 2 of the present application. DETAILED DESCRIPTION

[0016] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0017] Embodiment 1 like Figure 1 As shown, the first embodiment of the present application provides an artificial intelligence image recognition data processing method, the method comprising the following steps: Step S1: The cloud platform collects target images according to the application field of image recognition and the recognition target, and preprocesses the collected target images through image preprocessing technology, and generates a target image set based on the preprocessed target images; Furthermore, the cloud platform collects target images according to the application field of image recognition and the recognition target and preprocesses the collected target images through image preprocessing technology. Generating a target image set based on the preprocessed target images includes the following sub-steps: Step S11: The cloud platform obtains an image acquisition object of a target image according to the application field of image recognition and the recognition target; Specifically, the application fields of image recognition include but are not limited to the medical field, transportation field, industrial manufacturing field, security field, agricultural field, and education field; the recognition target is the image recognition result that needs to be achieved in the corresponding application field. The image acquisition object of the target image is the acquisition object that can achieve the recognition target of the corresponding application field. For example, if the application field of image recognition is the industrial manufacturing field, and the recognition target is to perform quality inspection on the appearance of the product, then the image acquisition object of its target image is the appearance of the product that needs to be inspected for appearance quality; if the image recognition field is the medical field, and the recognition target is disease detection, then the image acquisition object of its target image is the various organs and tissues of the body.

[0018] Step S12: The cloud platform selects an image acquisition method according to the image acquisition object of the target image and acquires the target image; Specifically, a corresponding image acquisition method is selected according to the image acquisition object of the target image, and the target image is acquired through the corresponding image acquisition method. For example, if the image acquisition object of the target image is the appearance of a product that needs to be inspected for appearance quality in the industrial manufacturing field, then its image acquisition method is to acquire images through image acquisition equipment and network image capture technology, and its corresponding image acquisition equipment includes but is not limited to cameras in product manufacturing factories and cameras at product sales points, and network image capture technology is used to collect images uploaded by network platform users related to products that need to be inspected for appearance quality; if the image acquisition object of the target image is disease detection in the medical field, then its image acquisition method is to acquire the target image through medical imaging equipment, and medical imaging equipment includes but is not limited to X-ray machines, CT scanners, and magnetic resonance imaging devices.

[0019] Step S13: The cloud platform performs image preprocessing on the collected target images through image preprocessing technology to construct a target image set; Specifically, image preprocessing techniques include image cropping, image scaling, image normalization and other image preprocessing methods. Image preprocessing techniques are used to unify the format of target images, improve the quality of target images, and enhance data diversity. A target image set is constructed based on the target images that have been preprocessed.

[0020] Step S2: The cloud platform performs image quality detection and repair on the target image set through an image quality detection and repair algorithm to generate a high-quality image set; the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repair the target image whose image detection quality value does not reach the preset image quality threshold; obtain the image recognition subject according to the recognition target of image recognition and highlight the image recognition subject through image enhancement technology to generate a high-quality highlighted image set; Furthermore, the cloud platform performs image quality detection and repair on the target image set through an image quality detection and repair algorithm to generate a high-quality image set; the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repairs the target image whose image detection quality value does not reach the preset image quality threshold; the image recognition subject is obtained according to the recognition target of the image recognition and the image recognition subject is highlighted through the image enhancement technology, and the generation of the high-quality highlighted image set includes the following sub-steps: Step S21: The cloud platform performs image quality detection and restoration on the target image set through an image quality detection and restoration algorithm to generate a high-quality image set; Specifically, the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repair the target image whose image detection quality value does not reach the preset image quality threshold, so that all the target images in the target image set reach the preset image quality threshold.

[0021] The specific implementation method of the image quality detection and restoration algorithm is as follows: Denote the target image set as ,in, is the target image set, The target image set target image, is the total number of target images in the target image set, The value range is .

[0022] Image quality detection formula Calculate the image detection quality value of the target image in the target image set, where, is the set of image detection quality values ​​for the target image set, is the weight coefficient of image clarity, For the The clarity score of the target image, is the weight coefficient of image contrast, For the The contrast score of the target image, is the weight coefficient of image brightness uniformity, For the The brightness uniformity score of the target image, is the weight coefficient of image color accuracy, For the The color accuracy score of the target image, is the weight coefficient of the image noise level, For the The noise level score of the target image.

[0023] The calculation formula for the clarity score of the target image is: , For the The clarity score of the target image, and Respectively The number of rows and columns of the target image, The target image Line The pixel value of the column, and are the minimum pixel value and the maximum pixel value of the target image respectively; the contrast score calculation formula of the target image is: , For the The contrast score of the target image, and Respectively The number of rows and columns of the target image, The target image Line The grayscale value of the column, and Respectively The minimum and maximum contrast of the target image, is a very small positive number; the calculation formula for the brightness uniformity score of the target image is , For the The brightness uniformity score of the target image, For the The number of sub-regions divided into the target image, For the The target image The average brightness of the sub-regions, and are the maximum and minimum values ​​of the mean brightness of all sub-regions, is a very small positive number, and Respectively The minimum brightness uniformity and maximum brightness uniformity of the target image; the calculation formula for the color accuracy score of the target image is , For the The color accuracy score of the target image, is the saturation consistency weight coefficient, and Respectively The number of rows and columns of the target image, For the The target image Line The pixel saturation weight of the column, For the The target image Line The pixel saturation difference measure of the columns, is the hue consistency weight coefficient, For the The target image Line the hue weight of the column, For the The target image Line The hue difference measure of the columns, and For the The minimum color accuracy and maximum color accuracy of the target image; the noise level score calculation formula of the target image is , For the The noise level score of the target image, For the The noise level of the target image obtained by the wavelet transform noise estimation algorithm, and Respectively The minimum and maximum noise levels of the target images.

[0024] Image detection quality value set based on target image set and the target image set Constructing target image detection quality set ,in, Detect the quality set for the target image, For the target image For the Image detection quality value of target image, is the total number of target images.

[0025] Detecting quality sets based on target images Classification function through image quality discrimination The target image is judged and classified for image quality, where: To detect the quality set based on the target image Image quality discrimination classification function, is the total number of target images, is a set of low-quality target images, is a high-quality target image subset, For the target image, For the Image detection quality value of target image, is the preset image quality threshold; the low-quality target image set is constructed through the above image quality discrimination classification function and a set of high-quality target images ,in, , For low-quality target image sets The number of target images in For high-quality target image sets The target number of images.

[0026] Based on low-quality target image set Through image restoration algorithm For low quality image sets The target image in is repaired, where is the target image restoration set, For low quality image sets The number of target images in is the fusion weight coefficient of texture restoration and local restoration, is the texture restoration weight coefficient, For the The number of texture blocks in the target image, For the The target image The texture restoration weight coefficient of the texture block, For the The target image Texture inpainting parameters for texture blocks, For the The number of local blocks of the target image, For the The target image local block repair parameters, is the weight coefficient of the structural repair item, For the The target image The weight coefficient of the corresponding points on the edge, For the The target image The edge corresponding point and The minimum distance between corresponding points on the edge. Fix parameters for the structure.

[0027] Inpainting the target image With high-quality target image set Merge to generate a high-quality image set.

[0028] Step S22: The cloud platform obtains the image recognition subject of the target image according to the recognition target of the image recognition, and highlights the image recognition subject of the target image in the high-quality image set by using the image enhancement technology to generate a high-quality highlighted image set; Specifically, the image recognition subject of the target image is the specific object to be recognized, and the image recognition subject of the target image is highlighted by image enhancement technology, and the background of the target image and the image recognition subject are clearly divided to generate a high-quality highlighted image set. For example, if the recognition target is gastric cancer detection in the medical field, the image recognition subject of the target image is the stomach, and the image recognition subject stomach in the target image is highlighted.

[0029] Step S3, the cloud platform generates a classification standard according to the recognition target of image recognition, and classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; the target image in the classified highlighted image subset in the classified highlighted image set is subjected to image recognition subject fusion through an image subject structure texture fusion algorithm to generate a classified subject fusion image and construct a classified fusion image set; Furthermore, the cloud platform generates a classification standard according to the recognition target of image recognition, and classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; the target image in the classified highlighted image subset in the classified highlighted image set is subjected to image recognition subject fusion to generate a classified subject fusion image and construct a classified fusion image set through an image subject structure texture fusion algorithm, which includes the following sub-steps: Step S31: The cloud platform generates a classification standard according to the recognition target of the image recognition, classifies the high-quality highlighted image set according to the classification standard, and generates a classified highlighted image set; Specifically, image classification elements are obtained according to the recognition target of image recognition, classification standards are generated according to the classification elements, target images in the high-quality highlighted image set are classified according to the classification standards, and a classified highlighted image set is generated based on the classified high-quality highlighted image set. The classified highlighted image set includes a classified highlighted image subset and a classification number, and the classified highlighted image subset includes target images belonging to this classification and image serial numbers.

[0030] Step S32: The cloud platform uses an image main structure texture fusion algorithm to perform image main fusion on the target image in the classified highlighted image set, generates a classified main fusion image, and constructs a classified fusion image set; Specifically, the specific implementation method of the image main structure texture fusion algorithm is as follows: The classified highlighted image set is represented as ,in Highlight the image set for classification, For classification number The classification highlights the image subset, For classification number The image numbers in the classified highlighted image subset are The target image, is the maximum number of classification numbers, For classification number The classification highlights the maximum number of image numbers in the image subset.

[0031] Highlighting image sets based on classification Through the image main structure texture fusion formula The image recognition subjects of the target images in the highlighted image subset of the classified highlighted image set are fused, where: For classification number The classification highlights the image subset The classification subject fusion image, For classification number The classification of the highlighted image subsets The number of is the main structure fusion weight coefficient, For classification number The classification highlights the number of main structural points in the image subset. For classification number The image numbers in the classified highlighted image subset are In the target image The weight coefficient of the main structure point, For classification number The image numbers in the classified highlighted image subset are In the target image The structural fusion value of the main structural points, is the main texture block fusion weight coefficient, For classification number The classification of highlights the number of main texture blocks in the image subset. For classification number The image numbers in the classified highlighted image subset are The target image The weight coefficient of the main texture block, For classification number The image numbers in the classified highlighted image subset are The target image The texture fusion value of the main texture block. Classified main fusion image based on the classified highlighted image subset in the classified highlighted image set Generate a classification fusion image set.

[0032] Step S4: The cloud platform extracts image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses the features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set; Furthermore, the cloud platform extracts the image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses the features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set, including the following sub-steps: Step S41: The cloud platform extracts image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set; Specifically, the convolutional neural network is used to extract the classified fusion image and the classified subject fusion image. The specific method of feature extraction is to first use different convolution kernels to slide convolution on the image, and then perform multiple convolutions layer by layer to extract complex features. Then, through maximum or average pooling operations, the amount of data is reduced, key features are retained and downsampling is achieved. Subsequently, activation functions such as ReLU are used to introduce nonlinearity to make the features sparse. Finally, the feature map is flattened into a one-dimensional vector and input into the fully connected layer to integrate local features to form global features.

[0033] Based on the above extracted image features, a classification fusion image feature set is generated, which includes the classification subject fusion image and its image feature set.

[0034] Step S42: The cloud platform fuses the image features of the classified subject fusion image in the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set; Specifically, the specific implementation method of the image feature fusion algorithm is as follows: The classification fusion image feature set is expressed as ,in, is the classification fusion image feature set, For the Zhang classified subject fusion image, For the The feature set of the fusion image of the classification subject, For the The first fusion image of the classified subject Features, is the number of fused images of the classified subject, For the The number of features in the fused image of the classified subject.

[0035] Based on the classification fusion image feature set through the image feature fusion algorithm The first Classification subject fusion image The image features are fused, where For the The image feature fusion value of the classified subject fusion image, For the The number of image features of the fused image of the classified subject, For the The first fusion image of the classified subject The fusion influence weight of image features, For the The first fusion image of the classified subject image features, is the fusion parameter for the classified subject fused image.

[0036] The image feature set is fused by the image feature fusion algorithm. The image features of all the classified subject fusion images are fused to obtain the image feature fusion value of the classified subject fusion image Set, according to the image feature fusion value of the classification fusion image set and the classification subject fusion image The collection generates a fusion set of classification image features.

[0037] Embodiment 2 like Figure 2 As shown, Embodiment 2 of the present application provides an artificial intelligence image recognition data processing system, comprising: An image acquisition and processing module 21 acquires target images according to the application field of image recognition and the recognition target, pre-processes the acquired target images through image pre-processing technology, and generates a target image set based on the pre-processed target images; Furthermore, the image acquisition and processing module 21 includes the following submodules: An image acquisition object acquisition submodule obtains an image acquisition object of a target image according to the application field of image recognition and the recognition target; Specifically, the application fields of image recognition include but are not limited to the medical field, transportation field, industrial manufacturing field, security field, agricultural field, and education field; the recognition target is the image recognition result that needs to be achieved in the corresponding application field. The image acquisition object of the target image is the acquisition object that can achieve the recognition target of the corresponding application field. For example, if the application field of image recognition is the industrial manufacturing field, and the recognition target is to perform quality inspection on the appearance of the product, then the image acquisition object of its target image is the appearance of the product that needs to be inspected for appearance quality; if the image recognition field is the medical field, and the recognition target is disease detection, then the image acquisition object of its target image is the various organs and tissues of the body.

[0038] An image acquisition submodule selects an image acquisition method according to an image acquisition object of a target image and acquires the target image; Specifically, a corresponding image acquisition method is selected according to the image acquisition object of the target image, and the target image is acquired through the corresponding image acquisition method. For example, if the image acquisition object of the target image is the appearance of a product that needs to be inspected for appearance quality in the industrial manufacturing field, then its image acquisition method is to acquire images through image acquisition equipment and network image capture technology, and its corresponding image acquisition equipment includes but is not limited to cameras in product manufacturing factories and cameras at product sales points, and network image capture technology is used to collect images uploaded by network platform users related to products that need to be inspected for appearance quality; if the image acquisition object of the target image is disease detection in the medical field, then its image acquisition method is to acquire the target image through medical imaging equipment, and medical imaging equipment includes but is not limited to X-ray machines, CT scanners, and magnetic resonance imaging devices.

[0039] The image preprocessing submodule performs image preprocessing on the collected target images through image preprocessing technology to construct a target image set; Specifically, image preprocessing techniques include image cropping, image scaling, image normalization and other image preprocessing methods. Image preprocessing techniques are used to unify the format of target images, improve the quality of target images, and enhance data diversity. A target image set is constructed based on the target images that have been preprocessed.

[0040] The image detection, repair and highlighting module 22 performs image quality detection and repair on the target image set through an image quality detection and repair algorithm to generate a high-quality image set; the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repairs the target image whose image detection quality value does not reach the preset image quality threshold; obtains the image recognition subject according to the recognition target of the image recognition and highlights the image recognition subject through the image enhancement technology to generate a high-quality highlighted image set; Furthermore, the image detection, restoration and highlighting module 22 includes the following submodules: The image detection and restoration submodule performs image quality detection and restoration on the target image set through the image quality detection and restoration algorithm to generate a high-quality image set; Specifically, the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repair the target image whose image detection quality value does not reach the preset image quality threshold, so that all the target images in the target image set reach the preset image quality threshold.

[0041] The specific implementation method of the image quality detection and restoration algorithm is as follows: Denote the target image set as ,in, is the target image set, The target image set target image, is the total number of target images in the target image set, The value range is .

[0042] Image quality detection formula Calculate the image detection quality value of the target image in the target image set, where, is the set of image detection quality values ​​for the target image set, is the weight coefficient of image clarity, For the The clarity score of the target image, is the weight coefficient of image contrast, For the The contrast score of the target image, is the weight coefficient of image brightness uniformity, For the The brightness uniformity score of the target image, is the weight coefficient of image color accuracy, For the The color accuracy score of the target image, is the weight coefficient of the image noise level, For the The noise level score of the target image.

[0043] Among them, the calculation formula for the clarity score of the target image is: , For the The clarity score of the target image, and Respectively The number of rows and columns of the target image, The target image Line The pixel value of the column, and are the minimum pixel value and the maximum pixel value of the target image respectively; the contrast score calculation formula of the target image is: , For the The contrast score of the target image, and Respectively The number of rows and columns of the target image, The target image Line The grayscale value of the column, and Respectively The minimum and maximum contrast of the target image, is a very small positive number; the calculation formula for the brightness uniformity score of the target image is , For the The brightness uniformity score of the target image, For the The number of sub-regions divided into the target image, For the The target image The average brightness of the sub-regions, and are the maximum and minimum values ​​of the mean brightness of all sub-regions, is a very small positive number, and Respectively The minimum brightness uniformity and maximum brightness uniformity of the target image; the calculation formula for the color accuracy score of the target image is , For the The color accuracy score of the target image, is the saturation consistency weight coefficient, and Respectively The number of rows and columns of the target image, For the The target image Line The pixel saturation weight of the column, For the The target image Line The pixel saturation difference measure of the columns, is the hue consistency weight coefficient, For the The target image Line The hue weight of the column, For the The target image Line The hue difference measure of the columns, and For the The minimum color accuracy and maximum color accuracy of the target image; the noise level score calculation formula of the target image is , For the The noise level score of the target image, For the The noise level of the target image obtained by the wavelet transform noise estimation algorithm, and Respectively The minimum and maximum noise levels of the target images.

[0044] Image detection quality value set based on target image set and the target image set Constructing target image detection quality set ,in, Detect the quality set for the target image, For the target image For the Image detection quality value of target image, is the total number of target images.

[0045] Detecting quality sets based on target images Classification function through image quality discrimination The target image is judged and classified for image quality, where: To detect the quality set based on the target image Image quality discrimination classification function, is the total number of target images, is a set of low-quality target images, is a high-quality target image subset, For the target image, For the Image detection quality value of target image, is the preset image quality threshold; the low-quality target image set is constructed through the above image quality discrimination classification function and a set of high-quality target images ,in, , For low-quality target image sets The number of target images in For high-quality target image sets The target number of images.

[0046] Based on low-quality target image set Through image restoration algorithm For low quality image sets The target image in is repaired, where is the target image restoration set, For low quality image sets The number of target images in is the fusion weight coefficient of texture restoration and local restoration, is the texture restoration weight coefficient, For the The number of texture blocks in the target image, For the The target image The texture restoration weight coefficient of the texture block, For the The target image Texture inpainting parameters for texture blocks, For the The number of local blocks of the target image, For the The target image local block repair parameters, is the weight coefficient of the structural repair item, For the The target image The weight coefficient of the corresponding points on the edge, For the The target image The edge corresponding point and The minimum distance between corresponding points on the edge. Fix parameters for the structure.

[0047] Inpainting the target image With high-quality target image set Merge to generate a high-quality image set.

[0048] The image highlighting submodule obtains the image recognition subject of the target image according to the recognition target of the image recognition, highlights the image recognition subject of the target image in the high-quality image set through the image enhancement technology, and generates a high-quality highlighted image set; Specifically, the image recognition subject of the target image is the specific object to be recognized, and the image recognition subject of the target image is highlighted by image enhancement technology, and the background of the target image and the image recognition subject are clearly divided to generate a high-quality highlighted image set. For example, if the recognition target is gastric cancer detection in the medical field, the image recognition subject of the target image is the stomach, and the image recognition subject stomach in the target image is highlighted.

[0049] The image classification and fusion module 23 generates a classification standard according to the recognition target of the image recognition, classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; performs image recognition subject fusion on the target image in the classified highlighted image subset in the classified highlighted image set through an image subject structure texture fusion algorithm to generate a classified subject fusion image and construct a classified fusion image set; Furthermore, the image classification fusion module 23 includes the following submodules: The image classification submodule generates a classification standard according to the recognition target of the image recognition, classifies the high-quality highlighted image set according to the classification standard, and generates a classified highlighted image set; Specifically, image classification elements are obtained according to the recognition target of image recognition, classification standards are generated according to the classification elements, target images in the high-quality highlighted image set are classified according to the classification standards, and a classified highlighted image set is generated based on the classified high-quality highlighted image set. The classified highlighted image set includes a classified highlighted image subset and a classification number, and the classified highlighted image subset includes target images belonging to this classification and image serial numbers.

[0050] The image fusion submodule uses the image main structure texture fusion algorithm to fuse the target image in the classified highlight image set, generate a classified main fusion image and construct a classified fusion image set; Specifically, the specific implementation method of the image main structure texture fusion algorithm is as follows: The classified highlighted image set is represented as ,in Highlight the image set for classification, For classification number The classification highlights the image subset, For classification number The image numbers in the classified highlighted image subset are The target image, is the maximum number of classification numbers, For classification number The classification highlights the maximum number of image numbers in the image subset.

[0051] Highlighting image sets based on classification Through the image main structure texture fusion formula The image recognition subjects of the target images in the highlighted image subset of the classified highlighted image set are fused, where: For classification number The classification highlights the image subset The classification subject fusion image, For classification number The classification of the highlighted image subsets The number of is the main structure fusion weight coefficient, For classification number The classification highlights the number of main structural points in the image subset. For classification number The image numbers in the classified highlighted image subset are In the target image The weight coefficient of the main structure point, For classification number The image numbers in the classified highlighted image subset are In the target image The structural fusion value of the main structural points, is the main texture block fusion weight coefficient, For classification number The classification of highlights the number of main texture blocks in the image subset. For classification number The image numbers in the classified highlighted image subset are The target image The weight coefficient of the main texture block, For classification number The image numbers in the classified highlighted image subset are The target image The texture fusion value of the main texture block. Classified main fusion image based on the classified highlighted image subset in the classified highlighted image set Generate a classification fusion image set.

[0052] An image feature extraction and fusion module 24 extracts image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set; Furthermore, the image feature extraction and fusion module 24 includes the following submodules: The image feature extraction submodule extracts the image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set; Specifically, the convolutional neural network is used to extract the classified fusion image and the classified subject fusion image. The specific method of feature extraction is to first use different convolution kernels to slide convolution on the image, and then perform multiple convolutions layer by layer to extract complex features. Then, through maximum or average pooling operations, the amount of data is reduced, key features are retained and downsampling is achieved. Subsequently, activation functions such as ReLU are used to introduce nonlinearity to make the features sparse. Finally, the feature map is flattened into a one-dimensional vector and input into the fully connected layer to integrate local features to form global features.

[0053] Based on the above extracted image features, a classification fusion image feature set is generated, which includes the classification subject fusion image and its image feature set.

[0054] The image feature fusion submodule fuses the image features of the classified subject fusion image in the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set; Specifically, the specific implementation method of the image feature fusion algorithm is as follows: The classification fusion image feature set is expressed as ,in, is the classification fusion image feature set, For the Zhang classified subject fusion image, For the The feature set of the fusion image of the classification subject, For the The first fusion image of the classified subject Features, is the number of fused images of the classified subject, For the The number of features in the fused image of the classified subject.

[0055] Based on the classification fusion image feature set through the image feature fusion algorithm The first Classification subject fusion image The image features are fused, where For the The image feature fusion value of the classified subject fusion image, For the The number of image features of the fused image of the classified subject, For the The first fusion image of the classified subject The fusion of image features affects the weight, For the The first fusion image of the classified subject image features, is the fusion parameter for the classified subject fused image.

[0056] The image feature set is fused by the image feature fusion algorithm. The image features of all the classified subject fusion images are fused to obtain the image feature fusion value of the classified subject fusion image Set, according to the image feature fusion value of the classification fusion image set and the classification subject fusion image The collection generates a fusion set of classification image features.

[0057] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. An artificial intelligence image recognition data processing method, characterized in that: include: Step S1: The cloud platform collects target images according to the application field of image recognition and the recognition target, and preprocesses the collected target images through image preprocessing technology, and generates a target image set based on the preprocessed target images; Step S2: The cloud platform performs image quality detection and restoration on the target image set through an image quality detection and restoration algorithm to generate a high-quality image set; The image quality detection and restoration algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and to restore the target image whose image detection quality value does not reach the preset image quality threshold; obtain the image recognition subject according to the recognition target of the image recognition and highlight the image recognition subject through the image enhancement technology to generate a high-quality highlighted image set; Step S3, the cloud platform generates a classification standard according to the recognition target of image recognition, and classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; the target image in the classified highlighted image subset in the classified highlighted image set is subjected to image recognition subject fusion through an image subject structure texture fusion algorithm to generate a classified subject fusion image and construct a classified fusion image set; Step S4: The cloud platform extracts the image features of the classified main fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses the features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set.

2. The artificial intelligence image recognition data processing method according to claim 1, characterized in that: The cloud platform collects target images according to the application field and recognition target of image recognition and preprocesses the collected target images through image preprocessing technology. Generating a target image set based on the preprocessed target images includes the following sub-steps: Step S11: The cloud platform obtains an image acquisition object of a target image according to the application field of image recognition and the recognition target; Step S12: The cloud platform selects an image acquisition method according to the image acquisition object of the target image and acquires the target image; Step S13: The cloud platform performs image preprocessing on the collected target images through image preprocessing technology to construct a target image set.

3. The artificial intelligence image recognition data processing method according to claim 1, characterized in that: The cloud platform performs image quality detection and repair on the target image set through an image quality detection and repair algorithm to generate a high-quality image set; the image quality detection and repair algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and repairs the target image whose image detection quality value does not reach the preset image quality threshold; the image recognition subject is obtained according to the recognition target of the image recognition and the image recognition subject is highlighted through the image enhancement technology, and the generation of a high-quality highlighted image set includes the following sub-steps: Step S21: The cloud platform performs image quality detection and restoration on the target image set through an image quality detection and restoration algorithm to generate a high-quality image set; Step S22: The cloud platform obtains the image recognition subject of the target image according to the recognition target of the image recognition, and highlights the image recognition subject of the target image in the high-quality image set through image enhancement technology to generate a high-quality highlighted image set.

4. The artificial intelligence image recognition data processing method according to claim 1, characterized in that: The cloud platform generates a classification standard according to the recognition target of image recognition, and classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; the target image in the classified highlighted image subset in the classified highlighted image set is subjected to image recognition subject fusion to generate a classified subject fusion image and construct a classified fusion image set through an image subject structure texture fusion algorithm, which includes the following sub-steps: Step S31: The cloud platform generates a classification standard according to the recognition target of the image recognition, classifies the high-quality highlighted image set according to the classification standard, and generates a classified highlighted image set; Step S32: The cloud platform uses an image main structure texture fusion algorithm to perform image main fusion on the target image in the classified highlighted image set, generates a classified main fusion image, and constructs a classified fusion image set.

5. The artificial intelligence image recognition data processing method according to claim 1, characterized in that: The cloud platform extracts the image features of the classified subject fusion image in the classified fusion image set through the convolutional neural network to generate a classified fusion image feature set. The features of the classified fusion image feature set are fused through the image feature fusion algorithm to generate a classified image feature fusion set, which includes the following sub-steps: Step S41: The cloud platform extracts image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set; Step S42: The cloud platform fuses the image features of the classified subject fusion image in the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set.

6. An artificial intelligence image recognition data processing system, characterized in that: include: An image acquisition and processing module, which acquires target images according to the application field of image recognition and the recognition target, pre-processes the acquired target images through image pre-processing technology, and generates a target image set based on the pre-processed target images; The image detection and restoration highlighting module performs image quality detection and restoration on the target image set through the image quality detection and restoration algorithm to generate a high-quality image set; The image quality detection and restoration algorithm is used to detect whether the image detection quality value of the target image in the target image set reaches a preset image quality threshold, and to restore the target image whose image detection quality value does not reach the preset image quality threshold; obtain the image recognition subject according to the recognition target of the image recognition and highlight the image recognition subject through the image enhancement technology to generate a high-quality highlighted image set; An image classification and fusion module generates a classification standard according to the recognition target of image recognition, and classifies the high-quality highlighted image set according to the classification standard to generate a classified highlighted image set; the classified highlighted image set includes a classified highlighted image subset and a classification number; the target image in the classified highlighted image subset in the classified highlighted image set is subjected to image recognition subject fusion through an image subject structure texture fusion algorithm to generate a classified subject fusion image and construct a classified fusion image set; The image feature extraction and fusion module extracts the image features of the classified main fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set, and fuses the features of the classified fusion image feature set through an image feature fusion algorithm to generate a classified image feature fusion set.

7. An artificial intelligence image recognition data processing system as claimed in claim 6, characterized in that: Image acquisition and processing module, specifically including: An image acquisition object acquisition submodule obtains an image acquisition object of a target image according to the application field of image recognition and the recognition target; An image acquisition submodule selects an image acquisition method according to an image acquisition object of a target image and acquires the target image; The image preprocessing submodule performs image preprocessing on the collected target images through image preprocessing technology to construct a target image set.

8. An artificial intelligence image recognition data processing system as claimed in claim 6, characterized in that: Image detection, repair and highlighting module, including: The image detection and restoration submodule performs image quality detection and restoration on the target image set through the image quality detection and restoration algorithm to generate a high-quality image set; The image highlighting submodule obtains the image recognition subject of the target image according to the recognition target of image recognition, highlights the image recognition subject of the target image in the high-quality image set through image enhancement technology, and generates a high-quality highlighted image set.

9. An artificial intelligence image recognition data processing system as claimed in claim 6, characterized in that: Image classification fusion module, including: The image classification submodule generates a classification standard according to the recognition target of the image recognition, classifies the high-quality highlighted image set according to the classification standard, and generates a classified highlighted image set; The image fusion submodule uses the image main structure texture fusion algorithm to fuse the target image in the classified highlight image set, generate a classified main fusion image and construct a classified fusion image set.

10. An artificial intelligence image recognition data processing system as claimed in claim 6, characterized in that: Image feature extraction and fusion module, specifically including: The image feature extraction submodule extracts the image features of the classified subject fusion image in the classified fusion image set through a convolutional neural network to generate a classified fusion image feature set; The image feature fusion submodule fuses the image features of the classified subject fusion image in the classified fusion image feature set through the image feature fusion algorithm to generate a classified image feature fusion set.

Images