A visual perception recognition method, device and medium based on deep learning

Abstract

The present invention discloses a visual perception and recognition method, device, and medium based on deep learning, which belongs to the field of computer vision technology and is used to solve the technical problems that existing computer vision systems have poor accuracy and robustness in visual perception and recognition, and also need to rely on manual processing and analysis, which easily consumes a large amount of computing resources and time. The method includes: performing data enhancement processing on large-scale data in the computer vision system to obtain a computer vision data set; based on a deep learning network, automatically matching feature points of the computer vision data set and constructing a deep learning network architecture; according to the deep learning network architecture, performing parameter training on the computer vision data set to generate a deep learning model; performing performance generalization processing on the deep learning model to determine a deep learning perception and recognition model; and applying and deploying the deep learning perception and recognition model to complete the perception and recognition of computer vision data.

Description

Technical Field

[0001] The present application relates to the field of computer vision, and in particular to a method, device, and medium for visual perception and recognition based on deep learning. Background Art

[0002] In the era of big data, visual perception and recognition systems can solve complex problems in computer vision, enabling computers to understand and process image and video data more accurately and efficiently. Furthermore, systems based on deep learning can achieve real-time performance through efficient parallel computing and optimized algorithms, resulting in improved response speed and real-time processing capabilities in practical applications.

[0003] Traditional image processing and pattern recognition methods have certain limitations when dealing with complex scenes and large-scale data. Traditional computer vision tasks require manual processing and analysis to complete visual tasks such as object detection, image classification, and face recognition.

[0004] Existing computer vision systems suffer from poor accuracy and robustness in visual perception and recognition, making it difficult for them to accurately identify and understand complex image backgrounds, lighting changes, occlusions, and other conditions. Furthermore, traditional computer vision algorithms typically consume significant computing resources and time. Summary of the Invention

[0005] The embodiments of the present application provide a deep learning-based visual perception and recognition method, device, and medium for solving the following technical problems: existing computer vision systems have poor accuracy and robustness in visual perception and recognition, and also need to rely on manual processing and analysis, which easily consumes a large amount of computing resources and time.

[0006] The embodiments of this application adopt the following technical solutions:

[0007] On the one hand, an embodiment of the present application provides a visual perception and recognition method based on deep learning, including: performing data enhancement processing on large-scale data in a computer vision system to obtain a computer vision data set; based on a deep learning network, automatically matching feature points of the computer vision data set and constructing a deep learning network architecture; according to the deep learning network architecture, performing parameter training on the computer vision data set to generate a deep learning model; performing performance generalization processing on the deep learning model to determine a deep learning perception and recognition model; and applying and deploying the deep learning perception and recognition model to complete the perception and recognition of computer vision data.

[0008] The visual perception and recognition systems after deep learning in the embodiments of the present application have broad application potential and application scalability in various fields. For example, in the field of intelligent security, these systems can be used for tasks such as pedestrian detection and abnormal behavior recognition. In medical diagnosis, they can be used for tumor detection, disease diagnosis, etc. The widespread application of these systems has brought convenience to people's lives and work. By continuously improving network architecture, training algorithms and large-scale data sets, it provides an important driving force for the development of computer vision technology. At the same time, it improves the accuracy and robustness of computer vision systems in visual perception and recognition, and can also learn richer and more abstract feature representations from big data, saving a lot of computing resources and time, and having better response speed and real-time processing capabilities.

[0009] In a feasible implementation, data enhancement processing is performed on large-scale data in a computer vision system to obtain a computer vision dataset, specifically including: extracting large-scale data from the computer vision system; wherein the large-scale data includes at least: object detection data, image classification data, face recognition data and medical diagnosis data; clearing the noise data and abnormal data of the large-scale data to obtain clean data; performing data scaling processing on the clean data within a range scale through a preset feature normalization technology to obtain normalized data; performing data enhancement processing on the normalized data, and performing set processing on the enhanced normalized data to obtain the computer vision dataset; wherein the data enhancement processing includes at least: data translation expansion processing, data rotation expansion processing and data flip expansion processing.

[0010] In a feasible implementation, the computer vision dataset includes an image dataset and a video dataset; both the image dataset and the video dataset include feature annotation data; and the dataset features under different computer vision datasets are determined through the feature annotation data.

[0011] In a feasible embodiment, based on a deep learning network, the computer vision dataset is automatically matched with feature points, and a deep learning network architecture is constructed, specifically comprising: performing feature recognition on the computer vision dataset through a deep learning convolutional neural network in the deep learning network and based on a feature extractor to extract dataset features; and performing parameter calculation and output of the fully connected layer on the identified dataset features through a convolution window in the deep learning convolutional neural network to obtain convolutional neural network parameters; performing time series data expansion on the computer vision dataset through a recurrent neural network in the deep learning network, and performing multi-layer feedforward recurrent connection on the expanded computer vision dataset so that the time series data in the computer vision dataset is copied and connected under the same neural network structure to obtain position sharing parameters under different time series; the deep learning network includes the deep learning convolutional neural network and the recurrent neural network; configuring a deep learning network environment based on the convolutional neural network parameters and the position sharing parameters; and automatically matching feature points between the dataset features and preset target task features through the deep learning network environment to determine the deep learning network architecture; wherein the automatic matching includes: mean square error algorithm matching and cross entropy loss function algorithm matching.

[0012] In a feasible implementation, according to the deep learning network architecture, the computer vision dataset is parameter trained to generate a deep learning model, specifically including: classifying and dividing the computer vision dataset into datasets to obtain visual type datasets; wherein the visual type datasets include at least: intelligent security datasets, intelligent traffic datasets, medical diagnosis datasets, driving navigation datasets, and entertainment life datasets; the visual type datasets are input into the deep learning network architecture, and according to the data characteristics of each type of data in the visual type datasets, the network parameters of the deep learning network architecture are adjusted and trained to adapt the network parameters to the data characteristics to obtain the deep learning model.

[0013] In a feasible implementation, the deep learning model is subjected to performance generalization processing to determine a deep learning perception and recognition model, specifically including: adjusting the learning rate of the deep learning model through a preset stochastic gradient descent algorithm to determine the learning rate optimization parameter; adjusting the step size of the learning rate optimization parameter in multiple component directions through the Adagrad algorithm to obtain the iterative cycle amount of the deep learning model; wherein the iterative cycle amount includes: the maximum number of iterations, the total number of samples, and the iterative gradient norm; based on the learning rate optimization parameter and the iterative cycle amount, the deep learning model is subjected to batch descent optimization to obtain the deep learning perception and recognition model.

[0014] In a feasible embodiment, before the deep learning perception and recognition model is applied and deployed to complete the perception and recognition of computer vision data, the method also includes: performing calculation and evaluation on the visual accuracy of the deep learning perception and recognition model through an inherent computer vision test data set to obtain an accuracy index; performing calculation and evaluation on the visual recall rate of the deep learning perception and recognition model through an inherent computer vision test data set to obtain a recall rate index; if the accuracy rate index is greater than a first preset threshold and the recall rate index is greater than a second preset threshold, the test judgment result of the deep learning perception and recognition model is a qualified test; otherwise, the test judgment result of the deep learning perception and recognition model is a failed test, and the deep learning perception and recognition model continues to be subjected to performance generalization processing until the test judgment result is a qualified test.

[0015] In a feasible implementation, the deep learning perception and recognition model is applied and deployed to complete the perception and recognition of computer vision data, specifically including: configuring and associating the deep learning perception and recognition model with a preset recognition application system to make the deep learning perception and recognition model adaptable to the recognition application system platform; through the recognition application system platform, the multi-scene application system and the deep learning perception and recognition model are scene-embedded and deployed; wherein the multi-scene application system includes at least: a camera acquisition scene system, a traffic driving scene system, a security monitoring scene system, and a medical diagnosis scene system; based on the embedded and deployed multi-scene application system, the computer vision data to be processed is perceived and recognized.

[0016] In a second aspect, an embodiment of the present application also provides a deep learning-based visual perception and recognition device, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute a deep learning-based visual perception and recognition method described in any of the above embodiments.

[0017] In a third aspect, an embodiment of the present application further provides a non-volatile computer storage medium, characterized in that the storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each of which includes instructions, and when the instructions are executed by the terminal, the terminal executes a deep learning-based visual perception and recognition method described in any of the above embodiments.

[0018] This application provides a deep learning-based visual perception and recognition method, device, and medium. Compared with the existing technology, the embodiments of this application have the following beneficial technical effects:

[0019] The visual perception and recognition systems after deep learning in the embodiments of the present application have broad application potential and application scalability in various fields. For example, in the field of intelligent security, these systems can be used for tasks such as pedestrian detection and abnormal behavior recognition. In medical diagnosis, they can be used for tumor detection, disease diagnosis, etc. The widespread application of these systems has brought convenience to people's lives and work. By continuously improving network architecture, training algorithms and large-scale data sets, it provides an important driving force for the development of computer vision technology. At the same time, it improves the accuracy and robustness of computer vision systems in visual perception and recognition, and can also learn richer and more abstract feature representations from big data, saving a lot of computing resources and time, and having better response speed and real-time processing capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0021] Figure 1 A flowchart of a deep learning-based visual perception and recognition method provided in an embodiment of the present application;

[0022] Figure 2 A simplified diagram of the deep learning model optimization and training process provided in an embodiment of the present application;

[0023] Figure 3 A schematic diagram of the structure of a deep learning-based visual perception and recognition device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0025] The present application embodiment provides a visual perception recognition method based on deep learning, such as Figure 1As shown, the visual perception recognition method based on deep learning specifically includes steps S101-S105:

[0026] S101. Perform data enhancement processing on large-scale data in a computer vision system to obtain a computer vision dataset.

[0027] Specifically, before building a deep learning network architecture, it is necessary to first extract large-scale data from the computer vision system. This large-scale data includes at least object detection data, image classification data, face recognition data, and medical diagnosis data.

[0028] Furthermore, the noise data and abnormal data are cleared from the large-scale data to obtain clean data.

[0029] Furthermore, the clean data is scaled within a range of scales using a preset feature normalization technique to obtain normalized data.

[0030] Furthermore, data augmentation processing is performed on the normalized data, and the normalized data after the augmentation processing is aggregated to obtain a computer vision dataset. The data augmentation processing includes at least: data translation augmentation processing, data rotation augmentation processing, and data flip augmentation processing.

[0031] Computer vision datasets include image datasets and video datasets. Both image and video datasets contain feature annotation data. This feature annotation data is used to determine the dataset characteristics of different computer vision datasets.

[0032] In one embodiment, preprocessing large-scale data primarily involves operations such as data cleaning, data normalization, and data augmentation. Data cleaning removes noise and abnormal data. Data normalization scales the data to a similar scale range, improving training stability. Data augmentation expands the dataset through operations such as translation, rotation, and flipping, increasing data diversity and improving the model's generalization capabilities.

[0033] S102. Based on the deep learning network, automatically match the feature points of the computer vision dataset and build a deep learning network architecture.

[0034] Specifically, the deep learning convolutional neural network in the deep learning network, based on the feature extractor, performs feature recognition on the computer vision dataset and extracts the dataset features. The convolution window in the deep learning convolutional neural network is then used to calculate and output the parameters of the fully connected layer based on the identified dataset features, thus obtaining the convolutional neural network parameters.

[0035] Furthermore, the recurrent neural network in the deep learning network is used to expand the time series data of the computer vision dataset. The expanded computer vision dataset is then connected through multiple layers of feedforward loops. This allows the time series data in the computer vision dataset to be replicated and connected under the same neural network structure, thereby obtaining positional shared parameters at different time series. The deep learning network includes deep learning convolutional neural networks and recurrent neural networks.

[0036] Furthermore, a deep learning network environment is configured based on the convolutional neural network parameters and the position sharing parameters.

[0037] Furthermore, through the deep learning network environment, the dataset features are automatically matched with the preset target task features to determine the deep learning network architecture. The automatic matching includes: mean square error algorithm matching and cross entropy loss function algorithm matching.

[0038] In one embodiment, the deep learning network architecture is based on convolutional neural networks (CNN) and recurrent neural networks (RNN). The network design and customized environment configuration are performed according to the convolutional neural network parameters and the position sharing parameters, and automatic matching is performed based on function algorithms such as mean square error loss and cross entropy loss according to preset target task characteristics and data set characteristics, so as to determine the deep learning network architecture for subsequent construction and use of deep learning models.

[0039] S103. Perform parameter training on the computer vision dataset according to the deep learning network architecture to generate a deep learning model.

[0040] Specifically, the computer vision dataset is classified and divided into visual type datasets, wherein the visual type datasets include at least: intelligent security datasets, intelligent transportation datasets, medical diagnosis datasets, driving navigation datasets, and entertainment life datasets.

[0041] Furthermore, the visual type data set is input into the deep learning network architecture, and the network parameters of the deep learning network architecture are adjusted and trained according to the data characteristics of each type of data in the visual type data set, so that the network parameters are adapted to the data characteristics to obtain a deep learning model.

[0042] In one embodiment, Figure 2 A simplified diagram of the deep learning model optimization and training process provided in the embodiment of this application is as follows: Figure 2As shown, the image and video datasets used to train and test the deep learning model are used to guide the deep learning model to perform model optimization, and data training is performed on the deep learning network architecture built based on the convolutional neural network and the recurrent neural network. That is, by adjusting multiple parameters in the deep learning model to be trained, it can accurately recognize and understand the content in the images and videos in the computer vision dataset, thereby training the deep learning model.

[0043] S104: Generalize the performance of the deep learning model to determine a deep learning perception and recognition model.

[0044] Specifically, the learning rate of the deep learning model is adjusted through a preset stochastic gradient descent algorithm to determine the learning rate optimization parameters.

[0045] The Adagrad algorithm is used to adjust the step size of the learning rate optimization parameter in multiple directions to obtain the iteration cycle of the deep learning model. The iteration cycle includes: the maximum number of iterations, the total number of samples, and the iteration gradient norm.

[0046] According to the learning rate optimization parameters and the number of iterative cycles, the deep learning model is optimized by batch reduction to obtain a deep learning perception and recognition model.

[0047] As a feasible implementation method, in the process of optimizing the deep learning model, stochastic gradient descent (SGD), Adagrad, Adam optimization algorithms, etc. can also be used to adjust the learning rate, regularize, enhance data, etc., to improve the performance and generalization ability of the deep learning model, and finally optimize the deep learning perception and recognition model.

[0048] S105. Apply and deploy the deep learning perception and recognition model to complete the perception and recognition of computer vision data.

[0049] Specifically, through the inherent computer vision test data set, the deep learning perception and recognition model is evaluated on the visual accuracy to obtain the accuracy index.

[0050] Through the inherent computer vision test dataset, the deep learning perception and recognition model is evaluated on the visual recall rate to obtain the recall rate indicator.

[0051] If the accuracy index is greater than the first preset threshold and the recall index is greater than the second preset threshold, the test judgment result of the deep learning perception recognition model is that the test is qualified.

[0052] Otherwise, the test result of the deep learning perception and recognition model is determined to be unqualified, and the deep learning perception and recognition model is further subjected to performance generalization processing until the test result is determined to be qualified.

[0053] As a feasible implementation method, an independent and inherent computer vision test dataset is used to evaluate and test the trained deep learning perception and recognition model, that is, the accuracy, recall rate and other evaluation indicators of the model are calculated, and then the performance of the deep learning perception and recognition model is further evaluated based on the determined test results, so as to ensure the visual perception and recognition effect of the deep learning perception and recognition model.

[0054] Furthermore, the deep learning perception recognition model is configured and associated with a preset recognition application system so that the deep learning perception recognition model is adapted to the recognition application system platform.

[0055] Furthermore, through the identification application system platform, the multi-scenario application system and the deep learning perception recognition model are scene-embedded and deployed; wherein, the multi-scenario application system includes at least: a camera acquisition scene system, a traffic driving scene system, a security monitoring scene system and a medical diagnosis scene system.

[0056] Furthermore, based on the integrated and deployed multi-scenario application system, the computer vision data to be processed is perceived and recognized. That is, the qualified deep learning perception and recognition model is developed and adapted to the recognition application system platform, and then applied to the actual multi-scenario application system, such as embedded in smart cameras, self-driving cars, security monitoring systems, etc., thereby realizing real-time visual perception and recognition functions of the computer vision data to be processed, saving a lot of computing resources and time, and having better response speed and real-time processing capabilities. At the same time, the use of deep learning perception and recognition models deployed in the multi-scenario application system also improves the accuracy and robustness of the computer vision system in visual perception and recognition.

[0057] In addition, the present application also provides a visual perception recognition device based on deep learning, such as Figure 3 As shown, the deep learning-based visual perception recognition device 300 specifically includes:

[0058] At least one processor 301. And a memory 302 in communication with the at least one processor 301. The memory 302 stores instructions that can be executed by the at least one processor 301, so that the at least one processor 301 can execute:

[0059] Perform data enhancement processing on large-scale data in computer vision systems to obtain computer vision datasets;

[0060] Based on the deep learning network, the computer vision dataset is automatically matched with feature points, and a deep learning network architecture is constructed;

[0061] According to the deep learning network architecture, the parameters of the computer vision dataset are trained to generate a deep learning model;

[0062] Perform performance generalization processing on the deep learning model to determine the deep learning perception and recognition model;

[0063] Apply and deploy deep learning perception and recognition models to complete the perception and recognition of computer vision data.

[0064] The present application provides a method, device and medium for visual perception and recognition based on deep learning. The visual perception and recognition system after deep learning has broad application potential and application scalability in various fields. For example, in the field of intelligent security, these systems can be used for tasks such as pedestrian detection and abnormal behavior recognition. In medical diagnosis, they can be used for tumor detection, disease diagnosis and other aspects. The wide application of these systems has brought convenience to people's lives and work. By continuously improving network architecture, training algorithms and large-scale data sets, it provides an important driving force for the development of computer vision technology. At the same time, it improves the accuracy and robustness of computer vision systems in visual perception and recognition, and can also learn richer and more abstract feature representations from big data, saving a lot of computing resources and time, and having better response speed and real-time processing capabilities.

[0065] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device and medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0066] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0067] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0068] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0069] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0071] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0072] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0073] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0074] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0075] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A visual perception recognition method based on deep learning, characterized in that: The method comprises: Perform data enhancement processing on large-scale data in computer vision systems to obtain computer vision datasets; Based on a deep learning network, the computer vision dataset is automatically matched with feature points, and a deep learning network architecture is constructed, specifically including: Performing feature recognition on the computer vision dataset based on a feature extractor and extracting dataset features through a deep learning convolutional neural network in the deep learning network; and performing parameter calculation and output of a fully connected layer on the identified dataset features through a convolution window in the deep learning convolutional neural network to obtain convolutional neural network parameters; Expanding the time series data of the computer vision dataset through a recurrent neural network in the deep learning network, and performing multi-layer feedforward loop connections on the expanded computer vision dataset, so that the time series data in the computer vision dataset are copied and connected under the same neural network structure, and position sharing parameters under different time series are obtained; The deep learning network includes the deep learning convolutional neural network and the recurrent neural network; Configuring a deep learning network environment based on the convolutional neural network parameters and the position sharing parameters; Automatically matching feature points between the dataset features and preset target task features in the deep learning network environment to determine the deep learning network architecture; wherein the automatic matching includes: mean square error algorithm matching and cross entropy loss function algorithm matching; Performing parameter training on the computer vision dataset according to the deep learning network architecture to generate a deep learning model; Performing performance generalization processing on the deep learning model to determine a deep learning perception and recognition model; The deep learning perception and recognition model is applied and deployed to complete the perception and recognition of computer vision data.

2. The method for visual perception and recognition based on deep learning according to claim 1, characterized in that: Perform data enhancement processing on large-scale data in computer vision systems to obtain computer vision datasets, including: Extracting large-scale data from the computer vision system; wherein the large-scale data includes at least object detection data, image classification data, face recognition data, and medical diagnosis data; Clearing noise data and abnormal data from the large-scale data to obtain clean data; The clean data is scaled within a range using a preset feature normalization technique to obtain normalized data; The normalized data is subjected to data enhancement processing, and the enhanced normalized data is subjected to set processing to obtain the computer vision data set; wherein the data enhancement processing at least includes: data translation expansion processing, data rotation expansion processing, and data flip expansion processing.

3. The method for visual perception and recognition based on deep learning according to claim 2, characterized in that: The computer vision dataset includes an image dataset and a video dataset; The image dataset and the video dataset both contain feature annotation data; and the dataset features under different computer vision datasets are determined through the feature annotation data.

4. The method for visual perception and recognition based on deep learning according to claim 1, characterized in that: According to the deep learning network architecture, parameter training is performed on the computer vision dataset to generate a deep learning model, specifically including: Classify the computer vision dataset into datasets to obtain visual type datasets; wherein the visual type datasets include at least: intelligent security datasets, intelligent traffic datasets, medical diagnosis datasets, driving navigation datasets, and entertainment life datasets; The visual type data set is input into the deep learning network architecture, and the network parameters of the deep learning network architecture are adjusted and trained according to the data characteristics of each type of data in the visual type data set, so that the network parameters are adapted to the data characteristics to obtain the deep learning model.

5. The method of visual perception recognition based on deep learning according to claim 1, characterized in that: The deep learning model is subjected to performance generalization processing to determine a deep learning perception and recognition model, specifically including: Adjusting the learning rate of the deep learning model through a preset stochastic gradient descent algorithm to determine a learning rate optimization parameter; Using the Adagrad algorithm, the learning rate optimization parameter is adjusted in multiple component directions to obtain the iterative cycle amount of the deep learning model; wherein the iterative cycle amount includes: the maximum number of iterations, the total number of samples, and the iterative gradient norm; According to the learning rate optimization parameter and the iterative cycle amount, the deep learning model is subjected to batch reduction optimization to obtain the deep learning perception recognition model.

6. The method of visual perception recognition based on deep learning according to claim 1, characterized in that: Before applying and deploying the deep learning perception and recognition model to complete perception and recognition of computer vision data, the method further includes: Using an inherent computer vision test dataset, the deep learning perception and recognition model is evaluated for visual accuracy to obtain an accuracy index; Using an inherent computer vision test dataset, the deep learning perception and recognition model is evaluated on the visual recall rate to obtain a recall rate indicator; If the accuracy index is greater than a first preset threshold and the recall index is greater than a second preset threshold, the test result of the deep learning perception recognition model is determined to be qualified; Otherwise, the test result of the deep learning perception and recognition model is determined to be a test failure, and the deep learning perception and recognition model is continued to be subjected to performance generalization processing until the test result is determined to be a test pass.

7. The method of visual perception recognition based on deep learning according to claim 1, characterized in that: Apply and deploy the deep learning perception and recognition model to complete the perception and recognition of computer vision data, specifically including: Associating the deep learning perception recognition model with a preset recognition application system so that the deep learning perception recognition model is adapted to the recognition application system platform; Through the recognition application system platform, the multi-scenario application system and the deep learning perception recognition model are integrated and deployed; wherein the multi-scenario application system includes at least: a camera acquisition scene system, a traffic driving scene system, a security monitoring scene system, and a medical diagnosis scene system; Based on the integrated and deployed multi-scenario application system, the computer vision data to be processed is perceived and recognized.

8. A visual perception and recognition device based on deep learning, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the deep learning-based visual perception recognition method according to any one of claims 1-7.

9. A non-volatile computer storage medium, characterized in that The storage medium is a non-volatile computer-readable storage medium, which stores at least one program. Each of the programs includes instructions, and when the instructions are executed by the terminal, the terminal executes a deep learning-based visual perception recognition method according to any one of claims 1 to 7.

Images