An image processing method and system based on artificial intelligence

Abstract

The application discloses an image processing method and system based on artificial intelligence, relates to the technical field of image processing, and comprises the following steps: performing preliminary extraction of image feature information, optimizing image quality, obtaining optimized image data, training a model and performing target detection, identifying a target region in the image and providing an identifier, performing local feature alignment on the target region, and classifying and labeling the target region, performing enhancement and repair processing, and customizing an output format according to an application scenario; and constructing a function module. Through the setting of adaptive image enhancement algorithms, target detection, feature extraction technology, and the generation of a generative adversarial network (GAN) for image repair, efficient positioning, optimization and recovery of the target region of the image are realized. Not only does the application have a significant advantage in the accuracy of noise removal and target detection, but also avoids distortion caused by excessive processing through dynamic adjustment of the enhancement strategy, thereby ensuring the outstanding performance and natural integration of the target region in the image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an image processing method and system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of artificial intelligence and deep learning technology, image processing technology has been widely applied in many fields, such as medical image analysis, automatic driving, security monitoring, industrial detection, etc. However, the existing image processing methods still have some shortcomings in the face of target detection, target region enhancement and repair, etc. in complex scenes.

[0003] Traditional image processing methods often rely on hand-designed feature extraction algorithms, and the processing process is difficult to balance accuracy and efficiency, and is easily affected by image noise, blur and uneven illumination, etc. In addition, the existing target detection technology has weak recognition ability in high noise, low resolution or complex background, which leads to inaccurate target positioning and affects the quality of subsequent processing. Although deep learning methods have made significant progress in image recognition, existing models still have limitations in processing the details of target regions, image repair and adaptive enhancement of target regions, etc., and cannot fully meet the real-time, efficient and high-quality requirements.

[0004] In the prior art, although deep learning models have achieved good performance in target detection, in the process of image enhancement and repair, they usually rely on fixed enhancement strategies and over-rely on generative adversarial network (GAN) repair models, which may cause some details to be over-processed or distorted, affecting the naturalness and authenticity of the image. SUMMARY

[0005] In view of the shortcomings of the prior art, the present application provides an image processing method and system based on artificial intelligence to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides an image processing method based on artificial intelligence, comprising the following steps:

[0008] S1, based on the image input signal, preliminary extraction of image feature information is performed;

[0009] S2, based on the extracted preliminary feature information, the image quality is optimized, the noise is removed and the image clarity is improved, and the optimized image data is obtained;

[0010] S3, using the optimized image data, the model is trained and target detection is performed, the target region in the image is identified and the identification is provided;

[0011] S4, after completing the target detection, local feature alignment is performed on the target region, and the target region is classified and labeled according to semantic information;

[0012] S5, the labeled target region is enhanced and repaired, image enhancement and image repair are performed, and processed image data is obtained;

[0013] S6, the processed image data is output according to the application scenario.

[0014] Further optimize the technical solution, in step S1, the image is preliminarily extracted by the deep learning model of convolutional neural network CNN, the low-level features in the image are recognized, including edge, texture and color distribution, and a preliminary image description is formed;

[0015] In the process of preliminary extraction, the local feature of the image is extracted using the convolution kernel, and the dimension reduction is performed by combining the pooling operation to extract the spatial hierarchical structure information of the image.

[0016] Further optimize the technical solution, in step S2, an adaptive filter is used for noise suppression, the adaptive filter dynamically adjusts the parameters of the filter according to the local features of the image, and different degrees of suppression are performed on different types of noise according to the noise characteristics of different regions in the image.

[0017] Further optimize the technical solution, in step S3, the model trained by the optimized image data is as follows:

[0018] ;

[0019] Among them,

[0020] : the total loss function of the training model;

[0021] : the number of training samples;

[0022] : the coordinate set of the real box and the predicted box;

[0023] : the intersection over union, which measures the overlap degree of the predicted box and the real box;

[0024] : the loss weight coefficient, used to balance the position loss and the category loss;

[0025] : the total number of categories;

[0026] : the sample In the category real labels on the images;

[0027] : sample predicted probabilities on the classes

[0028] Further optimize the technical solution, in the step S3, when the target detection is carried out, the multiscale target detection model is constructed, and the multiscale target detection model is as follows:

[0029] ;

[0030] Among them,

[0031] : the detected target set

[0032] : the overall inference function of the model

[0033] : the feature map set of the input image

[0034] : the weight parameter set of the model

[0035] : the multiscale layer set, representing the resolution level of different feature maps

[0036] : the target detection function of the first layer scale feature map

[0037] : the first layer feature map and the corresponding weight

[0038] Further optimize the technical solution, in the step S4, the local feature matching algorithm is used to align the local features of the target region, including:

[0039] Key point detection

[0040] Feature descriptor extraction

[0041] Feature matching

[0042] Geometric transformation estimation

[0043] Target alignment

[0044] Further optimize the technical solution, in the step S4, the deep semantic network is used to classify and label the target region according to semantic information, including:

[0045] Image description generation ​

[0046] semantic information embedding;

[0047] refined classification;

[0048] target labeling and relationship reasoning;

[0049] output labeling.

[0050] Further optimization of the technical solution, in the step S5, the image enhancement adopts adaptive enhancement algorithm, constructs adaptive enhancement loss function, carries out adaptive enhancement to the target area, and the adaptive enhancement loss function is as follows:

[0051] ;

[0052] wherein,

[0053] : adaptive enhancement total loss;

[0054] : target area quantity;

[0055] : original contrast value of target area ;

[0056] : enhanced contrast value of target area ;

[0057] : original brightness value of target area ;

[0058] : enhanced brightness value of target area ;

[0059] : sharpening degree measurement function for target area ;

[0060] : weight coefficient, used for balancing the optimization of contrast, brightness and sharpening degree;

[0061] : enhanced target area feature.

[0062] Further optimization of the technical solution, in the step S5, the image repair adopts the generation of the adversarial network GAN, and the GAN generates and optimizes the image through the discriminator and the generator, to ensure the naturalness and authenticity of the target area;

[0063] The loss function between the discriminator and the generator of the generation of the adversarial network is designed as follows:

[0064] ;

[0065] wherein,

[0066] GAN total loss;

[0067] discriminator judgment on real image ;

[0068] generated image generated by generator, based on noise vector ;

[0069] real data distribution;

[0070] noise data distribution.

[0071] An image processing system based on artificial intelligence is constructed based on the above-mentioned image processing method based on artificial intelligence, and the functional modules of the system include a preliminary feature extraction module, a data preprocessing module, a target region detection module, a target classification labeling module, an image enhancement repair module, and an image format conversion module.

[0072] In a second aspect, an embodiment of the present application provides a computer device, comprising a memory and a processor, and the memory stores a computer program, wherein the computer program instructions are executed by the processor to realize the steps of the image processing method and system based on artificial intelligence according to the first aspect of the present application.

[0073] In a third aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program instructions are executed by the processor to realize the steps of the image processing method and system based on artificial intelligence according to the first aspect of the present application.

[0074] Compared with the prior art, the present application provides an image processing method and system based on artificial intelligence, which has the following beneficial effects:

[0075] The artificial intelligence-based image processing method and system realize efficient positioning, optimization and recovery of the target region of the image by setting adaptive image enhancement algorithms, target detection, feature extraction technology and generating an adversarial network GAN for image restoration. Not only does it have a significant advantage in noise removal and target detection accuracy, but it can also dynamically adjust the enhancement strategy to avoid distortion caused by over-processing, ensuring the prominent performance and natural integration of the target region in the image. Compared with traditional image processing techniques, it effectively improves the quality and detail recovery capability of the image, has stronger robustness and adaptability, can handle more complex image scenes, and has a wide application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0076] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0077] Figure 1 A flowchart of an artificial intelligence-based image processing method according to the present application is shown.

[0078] Figure 2 A flowchart of a local feature matching algorithm in an artificial intelligence-based image processing method according to the present application is shown.

[0079] Figure 3 A flowchart of a deep semantic network in an artificial intelligence-based image processing method according to the present application is shown.

[0080] Figure 4 A functional module composition diagram of an artificial intelligence-based image processing system according to the present application is shown. DETAILED DESCRIPTION

[0081] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.

[0082] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from the description, and those skilled in the art can make similar generalizations without departing from the scope of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0083] Second, the "one embodiment" or "an embodiment" referred to herein can include a particular feature, structure, or characteristic. The various embodiments appearing at different places in this specification are not necessarily all cumulative or mutually exclusive of each other.

[0084] Embodiment one:

[0085] Referring to Figures 1-3 For the first embodiment of the present application, the embodiment provides an image processing method based on artificial intelligence, comprising the following steps:

[0086] S1, based on the image input signal, the preliminary extraction of image feature information is carried out.

[0087] In this embodiment, the deep learning model of convolutional neural network (CNN) is used to preliminarily extract the features of the image, identify the low-level features in the image, including edges, textures and color distribution, and form a preliminary image description; these low-level features will provide a basis for subsequent image enhancement, target detection and classification tasks.

[0088] In the process of preliminary extraction, the local feature extraction of the image is carried out using the convolution kernel, and the dimension reduction of the size is carried out combining with the pooling operation to extract the spatial hierarchical structure information of the image. The data set after feature extraction not only provides rough outline information for image understanding, but also provides effective input for subsequent reasoning and identification.

[0089] S2, based on the extracted preliminary feature information, the image quality is optimized, the noise is removed and the image clarity is improved, and the optimized image data is obtained.

[0090] In this embodiment, an adaptive filter is used for noise suppression. The adaptive filter dynamically adjusts the parameters of the filter according to the local features of the image, and suppresses different types of noise to different degrees according to the noise characteristics of different regions in the image.

[0091] Traditional fixed filter often has poor effect when dealing with complex scenes, while adaptive filter can optimize the filtering process according to the noise characteristics of different regions in the image. This method not only reduces the influence of noise on image quality, but also preserves important detail information in the image, avoids excessive blurring processing, and thus improves the accuracy of subsequent image analysis tasks.

[0092] S3, using the optimized image data, the model is trained and the target detection is carried out, the target region in the image is identified and the identification is provided.

[0093] In this embodiment, the model trained by using the optimized image data is as follows:

[0094] ;

[0095] wherein,

[0096] : total loss function of training model;

[0097] : number of training samples;

[0098] : coordinate set of real box and predicted box;

[0099] : intersection over union, which measures the overlap between predicted box and real box;

[0100] : loss weight coefficient, used to balance position loss and class loss;

[0101] : total number of classes;

[0102] : sample real label in class ;

[0103] : predicted probability of sample in class .

[0104] The model aims to jointly optimize target position detection and class classification. The first term calculates the error of target position, ensuring that the model can accurately predict the bounding box of the object; the second term is a classification loss based on cross-entropy, used to evaluate the accuracy of class prediction. During training, by optimizing minimization, gradually adjust the weights of the deep network, so that it can learn the significant features of the target in the optimized data (step S2)

[0105] Further, when detecting the target, a multi-scale target detection model is constructed, which is as follows:

[0106] ;

[0107] wherein,

[0108] : set of detected targets;

[0109] : overall inference function of model;

[0110] : a feature map set of an input image;

[0111] : a weight parameter set of the model;

[0112] : a multi-scale layer set, representing different resolution levels of feature maps;

[0113] : a target detection function for the i-th layer scale feature map;

[0114] : the i-th layer feature map and the corresponding weight.

[0115] The core of the model is multi-scale target detection, The function will detect targets in feature maps of different resolutions, and the detection results will be fused between different scales through joint operations to ensure accurate detection of small and large targets. Each layer detection function extracts regional features through convolution operations, and then combines non-maximum suppression (NMS) to filter duplicate detection boxes to generate the final target set .

[0116] The multi-scale design can effectively solve the problem of target scale difference. For example, small targets rely more on high-resolution feature maps for detection, while large targets are more captured by low-resolution feature maps. In actual use, the detection process is carried out in steps:

[0117] 1. Input the optimized image features .

[0118] 2. Process each scale feature map in turn .

[0119] 3. Merge the detection results of all scales to generate the target set .

[0120] First, the model is trained using the training model. The image data optimized in step S2 is input into the training framework, and the model weight is adjusted through the loss function, so that it learns the high-level features in the image.

[0121] Using the feature maps optimized in step S2, input into the multi-scale target detection model to generate multiple resolution feature levels (such as high-resolution to capture small targets, low-resolution to capture large targets), process feature maps of different scales layer by layer, and generate candidate targets.

[0122] S4, after completing the target detection, align the local features of the target region, and classify and label the target region according to the semantic information.

[0123] In this embodiment, a local feature matching algorithm (such as SIFT, SURF) is used to align the target region, including:

[0124] Key point detection;

[0125] First, the SIFT or SURF algorithm is used to detect the key points in the image. Key points are usually areas in the image with significant local features, such as edges, corners, and places with obvious texture changes.

[0126] Feature descriptor extraction;

[0127] Local feature descriptors are extracted for the area around each key point. SIFT and SURF generate descriptors (such as 128-dimensional vectors generated by SIFT and 64-dimensional vectors generated by SURF), which can effectively capture the texture, color and shape information of the local area, and have rotation and scale invariance.

[0128] Feature matching;

[0129] The nearest neighbor search algorithm (such as brute force matching or k-d tree) is used to match the key point descriptors in different images, and the most similar feature point pairs are found. These matching point pairs will help to register the images in the subsequent stage.

[0130] Geometric transformation estimation;

[0131] The geometric transformation (such as homography matrix) between images is estimated through the matching key points, and one of the images is transformed into the coordinate system of the other image. The commonly used method includes RANSAC (Random Sample Consensus) to eliminate false matching points.

[0132] Target alignment;

[0133] After geometric transformation, the target region in the image is accurately aligned. These aligned image regions can provide accurate input data for subsequent target recognition, classification and labeling.

[0134] Deep semantic network (such as BERT) is used to classify and label the target region according to semantic information, including:

[0135] Image description generation;

[0136] After key point matching and target alignment of the image, the system generates a preliminary image description based on the context information of the image (such as the background around the target, the spatial relationship between objects, etc.). This description can be a short sentence in natural language form, summarizing the main objects in the image and their relationships.

[0137] Semantic information embedding;

[0138] The image description is input into a pre-trained semantic network (such as BERT). The deep semantic network will convert the input text into a higher-level semantic representation through multiple layers of semantic understanding modules. Through the relationship of context, the network can understand the semantic connection between different objects.

[0139] Refined classification;

[0140] Based on the semantic representation of the deep semantic network, the system will refine the classification of the target in the image. That is, the model not only identifies the approximate category of the target (such as "car" or "person"), but also further describes the attributes of the target (such as "red sedan" or "person sitting on a chair"), improving the accuracy and details of the classification.

[0141] Target labeling and relationship reasoning;

[0142] After completing the refined classification, the deep semantic network will infer the relationship between the target and the target. For example, the model may infer that the relationship between a "person" and a "car" is "standing next to" or "driving", further improving the semantic understanding level of the image.

[0143] Output labeling;

[0144] Finally, the network will identify the target and its attributes (such as category, position, relationship) in the form of text labeling. These information can be used as the basis for further processing, display or storage.

[0145] S5, the enhanced and repaired target region is processed, the image enhancement and image repair are performed, and the processed image data is obtained.

[0146] In this embodiment, the image enhancement adopts an adaptive enhancement algorithm, an adaptive enhancement loss function is constructed, and the target region is adaptively enhanced. The adaptive enhancement loss function (abbreviated as AELF) is as follows:

[0147] ;

[0148] Wherein,

[0149] : adaptive enhancement total loss;

[0150] : target region number;

[0151] : original contrast value of target region ;

[0152] : target region enhanced contrast value of the target region;

[0153] : target region original brightness value of the target region;

[0154] : target region enhanced brightness value of the target region;

[0155] : sharpness measure function for the target region ;

[0156] : weight coefficient for balancing contrast, brightness, and sharpness optimization;

[0157] : enhanced target region features.

[0158] The AELF loss function mainly optimizes the enhancement effect through three parts:

[0159] Contrast optimization: first term Calculate the difference in target region contrast to ensure that the target region's enhanced contrast is improved without excessive distortion.

[0160] Brightness optimization: second term Aims to adjust the brightness of the target region, so that the target can maintain clarity and adapt to complex backgrounds.

[0161] Sharpness optimization: third term Used to enhance the sharpness of the target edge. Through sharpening operations such as Laplacian operator or sharpening convolution kernel in deep learning models, the details of the target region are enhanced, making the target more prominent.

[0162] During the enhancement process, the weight coefficient will be dynamically adjusted according to the specific content of the target region to avoid detail loss or information distortion caused by excessive processing.

[0163] Image restoration uses a generative adversarial network (GAN) to generate and optimize images through discriminators and generators, ensuring the naturalness and authenticity of the target region;

[0164] The loss function between the discriminator and the generator of the generative adversarial network is designed as follows:

[0165] ;

[0166] Where,

[0167] : total GAN loss;

[0168] : Discriminator's judgment on real images

[0169] : Image generated by Generator, based on noise vector

[0170] : Real data distribution

[0171] : Noise data distribution

[0172] Discriminator judges input images by training whether they are from real data distribution or fake data generated by Generator Generator then generates as realistic images as possible by accepting random noise vectors

[0173] Through the optimization of , the Generator can fill in missing details in the target area and perform detail repair and image quality improvement during the enhancement process.

[0174] In this step, based on the detection and calibration of step S4, the features of the target area in the image are extracted, including the boundary, brightness, contrast and sharpness of the target. These features will be used as input for the enhancement process.

[0175] Using adaptive enhancement loss function, the contrast, brightness and sharpness of the target area are optimized through AELF loss function. According to the local features of the target area, the enhancement strategy is adjusted adaptively to ensure that the target details are improved, while the background area is not affected too much.

[0176] After enhancement, GAN generative adversarial network is used to further repair missing parts or details in the target area. The Generator generates missing details according to the features of the target area, and the Discriminator verifies the authenticity of the image to ensure the naturalness and consistency of the supplemented content in the target area.

[0177] Finally, the enhanced target area is optimized by a deep learning model to ensure that the target can stand out more in a complex background, enhancing the visual effect of the image and the usability of the actual application.

[0178] S6, according to the application scenario, the output format of the processed image data is customized.

[0179] ​​​In this embodiment, the image is intelligently optimized and output according to the target application scenario. The specific requirements of the application scenario are combined, such as the image needs to be used in the fields of medical diagnosis, automatic driving, industrial detection, etc., and images of different formats and resolutions are output. The saving format of the image (such as JPG, PNG, TIFF, etc.) is considered, as well as the compatibility and processing efficiency of the subsequent image. According to the specific requirements of the application, the optimization algorithm can automatically select the best output parameters, and perform the final compression, format conversion and data storage on the image. The artificial intelligence is deeply combined with the specific application scenario to ensure that the processed image can maximize its effect in practical application.

[0180] Embodiment two:

[0181] Reference Figure 4 For the second embodiment of the present application, the embodiment provides an image processing system based on artificial intelligence, which is constructed based on the image processing method based on artificial intelligence described in embodiment one. The functional modules of the system include a preliminary feature extraction module, a data preprocessing module, a target region detection module, a target classification labeling module, an image enhancement repair module, and an image format conversion module.

[0182] The above modules are all constructed based on the image processing method based on artificial intelligence described in embodiment one to realize the flow of the method.

[0183] Embodiment three:

[0184] The embodiment also provides a computer device suitable for the image processing method and system based on artificial intelligence, which includes a memory and a processor. The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the image processing method and system based on artificial intelligence as proposed in the above embodiment.

[0185] The embodiment also provides a storage medium having a computer program stored thereon, which is executed by a processor to realize the image processing method and system based on artificial intelligence as proposed in the above embodiment.

[0186] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or can be a key, a trackball or a touchpad arranged on the shell of the computer device, or can be an external keyboard, a touchpad or a mouse, etc.

[0187] If the functions are implemented in the form of software function units and sold or used as independent products, the functions can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0188] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer readable medium for use by an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor, or other system that can fetch and execute instructions from the instruction execution system, apparatus or device, or in conjunction with these instructions. For the purpose of this specification, the "computer readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, apparatus or device, or in conjunction with these instructions.

[0189] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, as necessary, and stored in a computer memory.

[0190] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies, known in the art, or combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0191] It should be noted that the above examples are merely intended to illustrate the technical solutions of the present application and not to limit the same. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all such modifications or replacements should be included in the scope of the claims of the present application.

Claims

1. An artificial intelligence-based image processing method, characterized by, Comprising the following steps: S1, based on image input signal, preliminary extraction of image feature information is carried out; S2, based on the extracted preliminary feature information, the image quality is optimized, the noise is removed and the image definition is improved, and the optimized image data is obtained; S3, using the optimized image data, the model is trained and the target detection is carried out, the target area in the image is recognized and the identification is provided; The model trained by using the optimized image data is as follows: ； Wherein, : total loss function of the trained model; : number of training samples; : a set of coordinates of the real and predicted boxes; : intersection over union, measures the degree of overlap between a predicted box and a true box; : loss weight coefficient for balancing position loss and class loss; : total number of categories; : sample true label on a class ​ : sample in class prediction probability; When carrying out target detection, a multi-scale target detection model is constructed, and the multi-scale target detection model is as follows: ； Wherein, : detected target set; : overall inference function of the model; : a set of feature maps of the input image; : a set of weight parameters of the model; : a set of multi-scale layers representing different resolution levels of feature maps; : for the first layer scale feature map target detection function; : first layer feature map and the corresponding weight S4, after completing the target detection, the local feature alignment of the target area is carried out, and the target area is classified and labeled according to the semantic information; S5, the labeled target area is enhanced and repaired, the image enhancement and image repair are carried out, and the processed image data is obtained; S6, the processed image data is output according to the application scene.

2. The image processing method based on artificial intelligence according to claim 1, characterized in that, In the step S1, the deep learning model of convolutional neural network (CNN) is used to preliminarily extract the features of the image, to recognize the low-level features in the image, including edges, textures and color distribution, and to form a preliminary image description; In the process of preliminary extraction, the local feature extraction of the image is carried out by using convolution kernel, and the dimension reduction is carried out by combining the pooling operation, so as to extract the spatial hierarchical structure information of the image.

3. The image processing method based on artificial intelligence according to claim 1, characterized in that, In the step S2, an adaptive filter is used for noise suppression, and the adaptive filter dynamically adjusts the parameters of the filter according to the local features of the image, and different types of noise are suppressed to different degrees according to the noise characteristics of different regions in the image.

4. The image processing method based on artificial intelligence according to claim 1, characterized in that, In the step S4, a local feature matching algorithm is used to align the local features of the target area, including: Key point detection; Feature descriptor extraction; Feature matching; Geometric transformation estimation; Target alignment.

5. The image processing method based on artificial intelligence according to claim 1, characterized in that, In the step S4, a deep semantic network is used to classify and label the target area according to the semantic information, including: Image description generation; Semantic information embedding; Refined classification; Target labeling and relationship reasoning; Output labeling.

6. The image processing method based on artificial intelligence according to claim 1, characterized in that, In the step S5, the image enhancement adopts an adaptive enhancement algorithm, and an adaptive enhancement loss function is constructed to adaptively enhance the target area, and the adaptive enhancement loss function is as follows: ； Wherein, : adaptive enhancement total loss; : number of target regions; : target region original contrast value; : target region : enhanced contrast value; : target region original luminance value; : target region enhanced luminance value; : a sharpening measure function for the target region : a sharpening measure function for the target region : weight coefficients for balancing the optimization of contrast, brightness and sharpness; : enhanced target region features.

7. The artificial intelligence-based image processing method of claim 1, wherein, In the step S5, the image repair is carried out by using a generative adversarial network (GAN), and the GAN generates and optimizes the image through a discriminator and a generator to ensure the naturalness and authenticity of the target area; The loss function design between the discriminator and the generator of the generative adversarial network is as follows: ； Wherein, : GAN total loss; : the discriminator's judgment of real images ; : image generated by the generator, based on the noise vector ; : real data distribution; : Noise data distribution.

8. An artificial intelligence-based image processing system constructed based on the artificial intelligence-based image processing method according to any one of claims 1 to 7, characterized by, The function modules of the system include preliminary feature extraction module, data preprocessing module, target area detection module, target classification and labeling module, image enhancement and repair module and image format conversion module.

Images