Artificial intelligence-based image processing method, system and training method
By adding an attention enhancement module to the CSP module of the Yolov5 model, combined with convolutional neural networks and batch normalization processing, the problem of low intelligence caused by redundant backgrounds in construction scenes is solved, and various detection functions are improved.
Patent Information
- Application Number
- CN202211705855.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-29
- Publication Date
- 2026-02-06
- Estimated Expiration
- 2042-12-29
AI Technical Summary
The cluttered backgrounds in construction images result in poor dataset quality, leading to a low level of intelligence in construction engineering.
An image processing method based on the Yolov5 model is adopted. By adding an attention enhancement module to the first and second CSP modules, the network's ability to extract global features of the image is improved. Combined with convolutional neural networks and batch normalization processing, image processing of building construction scenes is realized.
It improves the level of intelligence in building construction scenarios, realizing functions such as edge detection, safety helmet detection, mask detection, crack detection, rebar count detection, and semantic segmentation of indoor scenes, thereby enhancing the accuracy and precision of image processing.
Smart Images
Figure CN115830331B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of artificial intelligence, and in particular to an image processing method, system and training method based on artificial intelligence. BACKGROUND
[0002] With the development of artificial intelligence technology, deep learning technology is applied to various fields, such as face recognition, target detection, image segmentation, etc., providing new methods and new ideas for solving problems in various industries. However, in the field of civil engineering, the image background under the construction scene is very miscellaneous, resulting in poor quality of related data sets, which causes the current intelligent level of construction in the field of civil engineering to be low. SUMMARY
[0003] The present application provides an image processing method, system and training method based on artificial intelligence to solve the technical problem of low intelligent level of construction in the field of civil engineering.
[0004] In a first aspect, the present application provides an image processing method based on artificial intelligence, comprising:
[0005] obtaining a to-be-processed image;
[0006] performing feature extraction on the to-be-processed image to obtain a plurality of feature maps;
[0007] The plurality of feature maps are processed in turn by a first CBL module and a first CSP module;
[0008] When the first CSP module processes the to-be-processed feature map, the to-be-processed feature map is processed in turn by a third CBL module, a plurality of residual components and a first convolutional neural network to obtain a first feature map, the to-be-processed feature map is processed by a second convolutional neural network to obtain a second feature map, the first feature map and the second feature map are merged, and then batch normalization processing and activation processing are performed;
[0009] The feature map processed by the first CBL module and the first CSP module is processed by a second CBL module and a second CSP module;
[0010] When the second CSP module processes the to-be-processed feature map, the to-be-processed feature map is processed in turn by a fourth CBL module, a plurality of residual components and a third convolutional neural network to obtain a third feature map, the to-be-processed feature map is processed by a fourth convolutional neural network to obtain a fourth feature map, the third feature map and the fourth feature map are merged, and then batch normalization processing and activation processing are performed;
[0011] obtaining a processing result of the to-be-processed image.
[0012] In a possible implementation, in the first CSP module, the feature map obtained through the residual component processing is subjected to attention promotion processing to obtain a fifth feature map, the fifth feature map is combined with the feature map obtained through the residual component processing, and then the combined feature map is subjected to feature extraction based on the first convolutional neural network to obtain a first feature map.
[0013] In the second CSP module, the feature map obtained through the residual component processing is subjected to attention promotion processing to obtain a sixth feature map, the sixth feature map is combined with the feature map obtained through the residual component processing, and then the combined feature map is subjected to feature extraction based on the third convolutional neural network to obtain a third feature map.
[0014] In a possible implementation, the attention promotion processing on the feature map obtained through the residual component processing includes the following steps:
[0015] Step S11. The feature map obtained through the residual component processing is subjected to convolution processing.
[0016] Step S12. The feature map after the convolution processing is subjected to activation processing to obtain a seventh feature map.
[0017] Step S13. The mean value of each channel is calculated along the channel of the feature map.
[0018] Step S14. The calculated mean value is subjected to 1*1 convolution processing for channel fusion.
[0019] Step S15. The fused mean value is subjected to batch normalization processing.
[0020] Step S16. The mean value after the batch normalization processing is subjected to activation processing to obtain a first mean value.
[0021] Step S17. The seventh feature map is multiplied by the first mean value to obtain a feature map after the attention promotion processing.
[0022] In a possible implementation, in step S11, the feature map obtained through the residual component processing is subjected to 3*3 convolution processing, in step S12, the Leaky ReLU function is adopted for nonlinear activation of the feature map after the convolution processing, and in step S16, the sigmoid function is adopted for nonlinear activation of the mean value after the batch normalization processing.
[0023] In a possible implementation, the first CBL module, the second CBL module, and the third CBL module include a convolutional neural network, batch normalization, and an activation function.
[0024] In a second aspect, the embodiments of the present application provide an image processing system based on artificial intelligence, which is used to implement the image processing method based on artificial intelligence in the first aspect, and includes a basic device layer, an application service layer and an interface access layer. The basic device layer is configured to provide storage devices and computing devices for deploying artificial intelligence algorithms and image processing. The application service layer includes an artificial intelligence algorithm deployment environment and at least one artificial intelligence algorithm. At least one of the artificial intelligence algorithms adopts the image processing method based on artificial intelligence in the first aspect. The interface access layer is configured to receive images to be processed and return processing results.
[0025] In a possible implementation, the artificial intelligence algorithms in the application service layer include an edge detection algorithm, a safety helmet detection algorithm, a mask detection algorithm, a crack detection algorithm, a number of steel bars detection algorithm, an indoor scene semantic segmentation algorithm and a dangerous area detection algorithm.
[0026] In a third aspect, the embodiments of the present application provide a training method of an image processing system based on artificial intelligence, which is used to train the image processing system based on artificial intelligence in the second aspect, and includes the following steps.
[0027] A data set for training is collected, and a weight is assigned to an open source data set for pre-training. A weight of attention promotion processing is set as a random value, and migration learning is performed.
[0028] In a fourth aspect, the embodiments of the present application provide an electronic device, which includes a processor and a memory.
[0029] The memory is configured to store a computer program. The processor is configured to implement the image processing method based on artificial intelligence in the first aspect by executing the computer program.
[0030] In a fifth aspect, the embodiments of the present application provide a readable storage medium, which stores computer execution instructions. When the processor executes the computer execution instructions, the image processing method based on artificial intelligence in the first aspect is implemented.
[0031] The application is implemented, and the following beneficial effects are obtained: the application provides an image processing method, system and training method based on artificial intelligence, the image processing method based on artificial intelligence comprises the following steps: acquiring an image to be processed; performing feature extraction on the image to be processed based on a slicing operation and a convolutional neural network to obtain a plurality of feature maps; the plurality of feature maps are processed in sequence through a first CBL module, a first CSP module, a second CBL module and a second CSP module; when the CSP module processes a feature map to be processed, the feature map to be processed is processed in sequence through a CBL module, a plurality of residual components and a convolutional neural network to obtain a first feature map, the feature map to be processed is processed through a second convolutional neural network to obtain a second feature map, the first feature map and the second feature map are merged, and then batch normalization processing and activation processing are performed; a processing result of the image to be processed is obtained; the image in a construction scene can be processed to realize the functions of edge detection, safety helmet detection, mask detection, crack detection, steel bar number detection, indoor scene semantic segmentation and dangerous area detection. The application can greatly improve the intelligent level in a construction scene. BRIEF DESCRIPTION OF DRAWINGS
[0032] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. 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.
[0033] Figure 1 is a flowchart of the image processing method based on artificial intelligence provided by the embodiments of the application;
[0034] Figure 2 is a flowchart of attention enhancement processing provided by the embodiments of the application;
[0035] Figure 3 is a structural schematic diagram of the image processing system based on artificial intelligence provided by the embodiments of the application;
[0036] Figure 4 is a structural schematic diagram of an electronic device provided by the embodiments of the application. DETAILED DESCRIPTION
[0037] The technical solutions in the embodiments of the application will be described clearly and completely below. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0038] For the image background in the construction scene is very miscellaneous, thus leading to the quality of related data set is poor, resulting in the current intelligent degree of building construction in the field of civil engineering is low, the embodiment of the application increases the attention promotion module in the first CSP module and the second CPS module based on the Yolov5 model, which is used to improve the ability of the network to extract global features of the image, has better target detection capability for the image in the miscellaneous construction scene, and realizes improving the intelligent degree of the construction scene.
[0039] The embodiment of the application provides an image processing method based on artificial intelligence, specifically relates to computer vision technology, that is, machine vision using a computer to identify, track and measure a target and further performing image processing.
[0040] Figure 1 The embodiment of the application provides an image processing method based on artificial intelligence, specifically relates to computer vision technology, that is, machine vision using a computer to identify, track and measure a target and further performing image processing. Figure 1 As shown in the figure, the method comprises the following steps:
[0041] S1. Obtain an image to be processed:
[0042] The image to be processed can be a picture, a video or a URL address of a public network picture. Receive a picture, a video or a URL address of a public network picture uploaded by a user; wherein, for the video uploaded by the user, read the video frames in sequence by using opencv, each video frame is equivalent to a picture, and a series of images to be processed are obtained; and for the URL address of the public network picture uploaded by the user, download the picture by parsing the URL, and then obtain the image to be processed.
[0043] S2. Obtain a feature map by using a Focus module:
[0044] The Focus module can slice the input image, and then perform convolution operation, and finally obtain the feature map processed by the Focus module. For example, the input image is 640*640*3, which is changed into a feature map of 320*320*12 through slicing operation, and then is changed into a feature map of 320*320*32 through convolution operation of 32 convolution kernels.
[0045] S3. Processed by a first CBL module:
[0046] The first CBL module comprises a convolutional neural network module Conv, a batch normalization module BN and a Leaky ReLU activation function.
[0047] S4. Processed by a first CSP module:
[0048] The first CSP module comprises two branches, as shown in the figure. Figure 1As shown, the feature map processed by the first CBL module enters the first CSP module:
[0049] First branch: the feature map processed by the first CBL module is sequentially processed by the third CBL module, a plurality of residual components, and then input into the first convolutional neural network to obtain the first feature map; here, the first convolutional neural network is preferably a Depthwise convolutional layer, in which a convolutional kernel is configured for each channel, that is, convolution operation can be independently performed on each channel.
[0050] Second branch: the feature map processed by the first CBL module is directly input into the second convolutional neural network for processing to obtain the second feature map; here, the first convolutional neural network is preferably a Depthwise convolutional layer, in which a convolutional kernel is configured for each channel, that is, convolution operation can be independently performed on each channel.
[0051] Then, the first feature map and the second feature map are merged, input into the batch normalization module BN for processing and activated by the activation function to obtain the feature map processed by the first CSP module.
[0052] S5. Processed by the second CBL module:
[0053] The structure of the second CBL module is the same as that of the first CBL module, and also includes the convolutional neural network module Conv, the batch normalization module BN, and the Leaky ReLU activation function.
[0054] S6. Processed by the second CSP module:
[0055] The structure of the second CSP module is the same as that of the first CSP module, and also includes two branches; as shown in Figure 1 As shown, the feature map processed by the second CBL module enters the second CSP module:
[0056] First branch: the feature map processed by the second CBL module is sequentially processed by the fourth CBL module, a plurality of residual components, and then input into the third convolutional neural network to obtain the third feature map; here, the third convolutional neural network is preferably a Depthwise convolutional layer, in which a convolutional kernel is configured for each channel, that is, convolution operation can be independently performed on each channel.
[0057] The feature map processed by the second CBL module is directly input into a fourth convolutional neural network for processing to obtain a fourth feature map. In this case, the fourth convolutional neural network is preferably a Depthwise convolutional layer, in which a convolution kernel is configured for each channel, that is, each channel can be independently subjected to convolution operation.
[0058] Then, the third feature map and the fourth feature map are merged and input into a batch normalization module BN for processing and activated by an activation function to obtain a feature map processed by the second CSP module.
[0059] S7. Obtain the processing result.
[0060] In some embodiments, in order to improve the ability of the network to extract global features of an image, an attention fuse module (AFM module) is added in the first CSP module and the second CSP module, respectively, as shown in Figure 1 In the first branch of the first CSP module and the second CSP module, the feature map processed by the plurality of residual components is input into a first convolutional neural network through two branches:
[0061] The first branch: the feature map processed by the plurality of residual components is processed by the attention fuse module to obtain a fifth feature map and a sixth feature map. As shown in Figure 2 The processing flow of the attention fuse module is as follows:
[0062] S11. Perform convolution processing on the feature map processed by the plurality of residual components; in this case, for example, a 3*3 convolutional layer is used to maintain the size and number of channels of the feature map unchanged.
[0063] S12. Activate through an activation function layer to obtain a seventh feature map; in this case, for example, the activation function is Leaky ReLU, thereby providing nonlinear activation. It should be understood that the embodiments of the present application do not limit the above-mentioned activation function, and other activation functions can also be selected according to actual needs.
[0064] S13. Calculate the mean value of each channel along the channel of the seventh feature map.
[0065] S14. Perform 1*1 convolution processing on the obtained mean value to fuse the channels.
[0066] S15. Perform batch normalization processing on the fused mean value.
[0067] S16. The mean value after batch normalization processing is activated, and in this example, nonlinear activation is performed by a sigmoid activation function to obtain a first mean value. It should be understood that the embodiments of the present application do not limit the activation function described above, and other activation functions can also be selected according to actual needs.
[0068] S17. The seventh feature map obtained by step S12 is multiplied by the first mean value obtained by step S16, that is, the output of the AFM module, that is, the feature map after attention enhancement processing.
[0069] The second branch: the feature map processed by the plurality of residual components is not processed.
[0070] Then the feature map processed by the first branch of the AFM module is combined with the original feature map processed by the second branch, and then input into the first convolutional neural network and the third convolutional neural network.
[0071] Through such a configuration, the weight of the AFM can be set and modified, and through transfer learning, the ability to extract global features of an image is improved, continuously improving the accuracy and precision of image processing.
[0072] The present application also provides an image processing system based on artificial intelligence, which can be used to execute the above-mentioned image processing method based on artificial intelligence. As shown in Figure 3 The application also provides an image processing system based on artificial intelligence, which can be used to execute the above-mentioned image processing method based on artificial intelligence. As shown in
[0073] The base device layer is used to provide storage devices, computing devices and network devices for artificial intelligence algorithm deployment and image processing. The storage devices include cloud servers and local storage devices, and the number of cloud servers and the number of local storage devices can be configured arbitrarily according to the processing needs. The computing device includes a GPU with high TFLOPs and large memory, which is used to provide image processing capability, especially to meet the demand of multiple users using image processing function concurrently.
[0074] The application service layer includes a Web framework, an OSS Python SDK, and an algorithm deployment environment built on the infrastructure layer and algorithms for image processing such as target detection and semantic segmentation.
[0075] The Web framework is used to meet the interaction needs between the application service layer and the user and the object storage container of the cloud server. At the same time, in order to ensure the interaction between the application service layer and the object storage container of the cloud server, the OSS Python SDK and the log management module are also configured.
[0076] The application service layer deploys a deep learning environment of python+pytorch+cuda, and deploys multiple artificial intelligence-based target detection and semantic segmentation algorithms in python. The deployed algorithms are mainly for construction site scenarios, including edge detection, crack detection, indoor scene semantic segmentation, safety helmet detection, mask detection, steel bar number detection, and dangerous area detection of fixed camera view. The objects that can be processed include user-uploaded pictures and videos, and public picture url addresses.
[0077] It should be understood that the embodiments of the present application do not limit the above-mentioned algorithm language, which is not limited to Python, and other languages can also be selected according to actual needs.
[0078] Illustratively, the edge detection algorithm based on the canny algorithm of opencv can identify the edge features of objects in the image, and can strengthen the user's visual judgment and analysis ability for the edge.
[0079] Illustratively, the crack detection algorithm based on DeepCrack can identify cracks in the picture and return a binary image, in which the background is black and the cracks are white.
[0080] Illustratively, the indoor scene semantic segmentation algorithm based on BiSeNet can segment the door, window, wall, column, ceiling, floor and other background in the picture and superimpose them on the original picture with different colors to show the segmentation effect.
[0081] Illustratively, the safety helmet detection algorithm, the mask detection algorithm, the steel bar number detection algorithm and the dangerous area detection algorithm of fixed camera view are written based on Yolov5. Among them, the safety helmet detection algorithm and the mask detection algorithm can identify whether a person in the picture wears a safety helmet or a mask in a standard manner. If a person does not wear a safety helmet or a mask in a standard manner, an alarm information is returned, which is a prompt of "please wear a safety helmet (or a mask) in a standard manner" in the upper left corner of the UI layer. The steel bar number detection algorithm can identify the steel bars in the picture and return the number of steel bars, which is printed in the upper left corner of the UI layer. The dangerous area detection algorithm can identify the person in the picture and receive the dangerous area labeled by the user in the interface access layer. The pnpoly algorithm is used to judge whether the center point of the identified human body is in the labeled dangerous area, so as to judge whether the person enters the dangerous area. If it is detected that a person is in the dangerous area, a prompt of "please leave the dangerous area" is printed in the upper left corner of the UI layer.
[0082] The interface access layer is an access layer for user to upload images, videos and public picture urls, and returns the images and videos processed by the user. It also includes user registration module, user login module and user permission management module.
[0083] For the artificial intelligence-based image processing system, the objects it can process include user-uploaded pictures and videos and public network picture URL addresses. The processing flow of user-uploaded images is as follows:
[0084] When the user-uploaded object to be processed is a picture, the picture uploaded by the user from the interface access layer is uploaded to the object storage container of the cloud server, and after the uploading is completed, the interface access layer sends a message field with the storage address in the object storage container to the application service layer, and then the application service layer starts to download the picture uploaded by the user in the object storage container, applies the deployed artificial intelligence algorithm to process the image data and returns the processed picture to the object storage container, and after the return is completed, sends a message field with the picture storage address in the object storage container to the interface access layer, and the user can view or download the processed picture on the web page.
[0085] When the user-uploaded object to be processed is a video, the video uploaded by the user from the interface access layer is uploaded to the object storage container of the cloud server, and after the uploading is completed, the interface access layer sends a message field with the storage address in the object storage container to the application service layer, and then the application service layer starts to download the picture uploaded by the user in the object storage container, and after the picture is downloaded to the local hard disk, the application service layer applies the deployed artificial intelligence algorithm to process each frame of image data and writes the processed video frame into the hard disk to form a webm format video. After the complete video file is processed, the processed video file is returned to the object storage container of the cloud server, and after the return is completed, a message field with the picture storage address in the object storage container is sent to the interface access layer, and the user can view or download the processed video on the web page.
[0086] When the user-uploaded object to be processed is a public network picture URL, the user sends the URL address directly to the application service layer from the interface access layer, the application service layer parses the URL and downloads the picture into the memory, applies the deployed artificial intelligence algorithm to process the image data in the memory and returns the processed picture to the object storage container, and after the return is completed, a message field with the picture storage address in the object storage container is sent to the interface access layer, and the user can view or download the processed picture on the web page.
[0087] The application also provides a training method embodiment of an artificial intelligence-based image processing system, which is used to train the artificial intelligence-based image processing system provided by the application, comprising:
[0088] Collecting a data set for training, and assigning a weight to an open source data set for pre-training; setting the weight of attention promotion processing to a random value for transfer learning.
[0089] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device may include a processor 301 and a memory 302. Figure 4 The image shown is of an electronic device using a processor as an example.
[0090] The memory 302 is used to store programs. Specifically, the program may include program code, which includes computer operation instructions.
[0091] Memory 302 may include high-speed RAM memory, and may also include non-volatile memory, such as multiple disk drives.
[0092] The processor 301 is used to execute computer execution instructions stored in the memory 302 to implement the artificial intelligence-based image processing method and training method provided in the above method embodiments.
[0093] The processor 301 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.
[0094] Optionally, in specific implementations, if the communication interface, memory 302, and processor 301 are implemented independently, they can be interconnected via a bus to complete communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.
[0095] Optionally, in a specific implementation, if the communication interface, memory 302, and processor 301 are integrated on a single chip, then the communication interface, memory 302, and processor 301 can communicate through an internal interface.
[0096] The embodiment of the present application further provides a chip comprising a processor and an interface. The interface is used to input or output data or instructions processed by the processor. The processor is used to execute the artificial intelligence-based image processing method and the training method provided in the above method embodiments.
[0097] The present application further provides a computer readable storage medium, which can include a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various storage program codes. Specifically, the computer readable storage medium stores program information, which is used for the artificial intelligence-based image processing method and the training method in the above method embodiments.
[0098] The present application further provides a computer program product comprising a computer program, which is executed by a processor to implement the artificial intelligence-based image processing method and the training method in the above method embodiments.
[0099] The present application further provides a computer program, which enables a computer to execute the artificial intelligence-based image processing method and the training method in the above method embodiments.
[0100] In the above embodiments, all or part of the embodiments can be realized by software, hardware, firmware or any combination thereof. When realized by software, all or part of the embodiments can be realized in the form of a computer program product. The computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the flow or function according to the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available media sets. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)) and the like.
[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.
Claims
1. An artificial intelligence-based image processing method, comprising: obtaining an image to be processed; performing feature extraction on the image to be processed to obtain a plurality of feature maps; the plurality of feature maps are processed in turn via a first CBL module and a first CSP module; wherein when the first CSP module processes the feature map to be processed, the feature map to be processed is processed in turn via a third CBL module, a plurality of residual components, and a first convolutional neural network to obtain a first feature map, the feature map to be processed is processed via a second convolutional neural network to obtain a second feature map, the first feature map and the second feature map are merged, and then batch normalization processing and activation processing are performed; the feature map processed via the first CBL module and the first CSP module is processed in turn via a second CBL module and a second CSP module; wherein when the second CSP module processes the feature map to be processed, the feature map to be processed is processed in turn via a fourth CBL module, a plurality of residual components, and a third convolutional neural network to obtain a third feature map, the feature map to be processed is processed via a fourth convolutional neural network to obtain a fourth feature map, the third feature map and the fourth feature map are merged, and then batch normalization processing and activation processing are performed; obtaining a processing result of the image to be processed; in the first CSP module, performing attention enhancement processing on the feature map processed by the residual component to obtain a fifth feature map, merging the fifth feature map and the feature map processed by the residual component, and then performing feature extraction on the merged feature map based on the first convolutional neural network to obtain the first feature map; in the second CSP module, performing attention enhancement processing on the feature map processed by the residual component to obtain a sixth feature map, merging the sixth feature map and the feature map processed by the residual component, and then performing feature extraction on the merged feature map based on the third convolutional neural network to obtain the third feature map; the attention enhancement processing on the feature map processed by the residual component comprises the following steps: step S11. performing convolution processing on the feature map processed by the residual component; step S12. performing activation processing on the feature map processed by the convolution to obtain a seventh feature map; step S13. calculating the mean value of each channel along the channel of the seventh feature map; step S14. performing 1*1 convolution processing on the calculated mean value to perform channel fusion; step S15. performing batch normalization processing on the fused mean value; step S16. performing activation processing on the batch-normalized mean value to obtain a first mean value; step S17. multiplying the seventh feature map and the first mean value to obtain the feature map processed by the attention enhancement; in step S11, 3*3 convolution processing is performed on the feature map processed by the residual component, in step S12, LeakyReLU function is used for nonlinear activation of the feature map processed by the convolution, and in step S16, sigmoid function is used for nonlinear activation of the batch-normalized mean value.
2. The image processing method of claim 1, wherein, The first CBL module, the second CBL module, and the third CBL module comprise a convolutional neural network, batch normalization, and an activation function.
3. An artificial intelligence-based image processing system comprising a base device layer, an application service layer, and an interface access layer, wherein, The base device layer is configured to provide storage devices and computing devices for artificial intelligence algorithm deployment and image processing, the application service layer includes an artificial intelligence algorithm deployment environment and at least one artificial intelligence algorithm, at least one of the artificial intelligence algorithms adopts the artificial intelligence-based image processing method according to claim 1, and the interface access layer is configured to receive images to be processed and return processing results.
4. The image processing system of claim 3, wherein, The artificial intelligence algorithms of the application service layer include an edge detection algorithm, a safety helmet detection algorithm, a mask detection algorithm, a crack detection algorithm, a steel bar number detection algorithm, an indoor scene semantic segmentation algorithm, and a dangerous area detection algorithm.
5. An electronic device comprising: A processor and a memory; The memory is configured to store a computer program; The processor is configured to implement the artificial intelligence-based image processing method according to any one of claims 1-2 by executing the computer program. 6.A computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and when the processor executes the computer execution instructions, the artificial intelligence-based image processing method according to any one of claims 1-2 is implemented.