WO2019223154A1 - 单页高负载图像识别方法、装置、计算机设备及存储介质 - Google Patents

单页高负载图像识别方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2019223154A1
WO2019223154A1 PCT/CN2018/102471 CN2018102471W WO2019223154A1 WO 2019223154 A1 WO2019223154 A1 WO 2019223154A1 CN 2018102471 W CN2018102471 W CN 2018102471W WO 2019223154 A1 WO2019223154 A1 WO 2019223154A1
Authority
WO
WIPO (PCT)
Prior art keywords
neural network
convolutional neural
local
network model
picture
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2018/102471
Other languages
English (en)
French (fr)
Inventor
黄度新
方小伟
李双灵
张川
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2019223154A1 publication Critical patent/WO2019223154A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the present application relates to the field of image recognition technology, and in particular, to a single-page high-load image recognition method, device, computer equipment, and storage medium.
  • mainstream image recognition applications need to upload pictures to the enterprise back-end server for image recognition, such as Baidu picture search. It cannot use neural networks to identify pictures on the client. If the network speed is poor and the upload speed is slow, it will affect the speed of image recognition result feedback.
  • This application provides a single-page high-load image recognition method, device, computer equipment, and storage medium, which are aimed at solving the mainstream image recognition applications in the prior art, and all need to upload pictures to the enterprise background server for image recognition. If the network speed is poor and the upload speed is slow, it will affect the speed of image recognition result feedback.
  • the present application provides a single-page high-load image recognition method, which includes:
  • the present application provides a single-page high-load image recognition device, which includes:
  • a model acquisition unit configured to acquire a plurality of convolutional neural network models stored in a local single-page application at a user end for image recognition
  • a local picture recognition unit is configured to obtain a picture to be recognized, and use the picture to be recognized as an input of multiple convolutional neural network models in the local single-page application to obtain a one-to-one correspondence with each convolutional neural network model. Pending results
  • a display unit configured to display a keyword corresponding to the local picture recognition result if the local picture recognition result is successful
  • the background recognition unit is configured to upload the to-be-recognized picture to the background server if the local picture recognition result is recognition failure, and receive the background recognition result fed back by the background server.
  • the present application further provides a computer device including a memory, a processor, and a computer program stored on the memory and executable on the processor.
  • the processor is implemented when the computer program is executed.
  • the present application further provides a storage medium, wherein the storage medium stores a computer program, where the computer program includes program instructions, and the program instructions, when executed by a processor, cause the processor to execute the application A single-page high-load image recognition method according to any one of the provided.
  • the present application provides a single-page high-load image recognition method, device, computer equipment, and storage medium.
  • the method realizes the recognition of pictures in a local single-page application on the user side by using a convolutional neural network model, and quickly obtains the recognition results. It is not necessary to directly upload pictures to the background server for recognition, which improves the recognition efficiency.
  • FIG. 1 is a schematic flowchart of a single-page high-load image recognition method according to an embodiment of the present application
  • FIG. 2 is another schematic flowchart of a single-page high-load image recognition method according to an embodiment of the present application
  • FIG. 3 is a schematic diagram of a sub-flow of a single-page high-load image recognition method according to an embodiment of the present application
  • FIG. 4 is a schematic diagram of another sub-flow of a single-page high-load image recognition method according to an embodiment of the present application.
  • FIG. 5 is a schematic diagram of another sub-flow of a single-page high-load image recognition method according to an embodiment of the present application.
  • FIG. 6 is a schematic block diagram of a single-page high-load image recognition device according to an embodiment of the present application.
  • FIG. 7 is another schematic block diagram of a single-page high-load image recognition device according to an embodiment of the present application.
  • FIG. 8 is a schematic block diagram of a subunit of a single-page high-load image recognition device according to an embodiment of the present application.
  • FIG. 9 is a schematic block diagram of another subunit of a single-page high-load image recognition device according to an embodiment of the present application.
  • FIG. 10 is a schematic block diagram of another subunit of a single-page high-load image recognition device according to an embodiment of the present application.
  • FIG. 11 is a schematic block diagram of a computer device according to an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of a single-page high-load image recognition method according to an embodiment of the present application. This method is applied to terminals such as desktop computers, laptop computers, and tablet computers. As shown in FIG. 1, the method includes steps S101 to S104.
  • the convolutional neural network model stored in the local single-page application is a convolutional neural model that can be used for image recognition when the installation package of the single-page application is downloaded.
  • the convolutional neural model It is stored in a local single-page application built with AngularJS (a JavaScript-based front-end language). After the installation package is installed on the client, image recognition can be performed through the convolutional neural network model stored in the local single-page application.
  • AngularJS a JavaScript-based front-end language
  • AngularJS only needs HTML (that is, Hypertext Markup Language), CSS (that is, Cascading Style Sheets), and JavaScript (which is a translatable scripting language) to create single-page applications on the client. Its goal is to make development and testing more Easy to enhance the performance of MVC web applications.
  • the so-called single-page application refers to the integration of multiple functions on one page, even the entire system has only one page, and all business functions are its submodules, which are linked to the main interface in a specific way.
  • the stand-alone web page has a simple operation interface, such as an operation area for adding pictures to be identified and a display area for outputting recognition results.
  • the convolutional neural network model stored in the local single-page application is obtained by inputting pre-selected imported image data sets (such as the image data set) into the convolutional neural network on the enterprise back-end server and training it.
  • ImageNet is the name of a computer vision system recognition project. It is the largest database for image recognition in the world. It is an image recognition database created by a computer scientist at Stanford in the United States that simulates a human recognition system.
  • the ImageNet data set (that is, the image data set) has more than 14 million pictures, covering more than 20,000 categories; more than one million of these pictures have clear category annotations and object location annotations in the image.
  • the ImageNet dataset The size is about 1TB.
  • the method before step S101, the method further includes:
  • S1002 Decompress the installation package of the local single-page application to obtain a local single-page application including multiple convolutional neural network models.
  • the convolutional neural network model included in the installation package of the local single-page application (the installation package stores a convolutional neural network model for image recognition) is to perform a large number of images on the background server of the enterprise Recognition training. That is, a convolutional neural network is trained on the background server of the enterprise by using the image data set as input.
  • the obtained convolutional neural network model includes multiple types of models. For example, there is a convolutional neural network model that recognizes vehicles. Animal convolutional neural network models, etc. After the above-mentioned types of convolutional neural network models are trained, they can directly import installation packages (such as apk installation packages or iOs installation packages) for single-page applications. After the user downloads the installation package of the single-page application, decompress and install the installation package of the single-page application on the local terminal to obtain a single-page application with the identification image, without uploading the image to be identified to the enterprise background server. Identify.
  • the step S1001 includes:
  • the above training process may be performed on the enterprise server or the user.
  • you can select the type of image to be identified such as vehicles, animals, plants, etc.
  • the type of image to be identified such as vehicles, animals, plants, etc.
  • the trained convolutional neural network model is packaged into an installation package of a local single-page application for use by the terminal itself, Or for other downloading the installation package uploaded by the client (the installation package is the installation package of the local single page application including the convolutional neural network model).
  • the more types of convolutional neural network models imported into the installation package of a single-page application the better, but generally the size of the installation package of the entire single-page application is controlled to about 200M That is, it can meet the basic picture recognition requirements, that is, 10-20 kinds of convolutional neural network models can be imported (for example, it can meet the categories of images of vehicles, animals, and plants).
  • the step S10011 includes:
  • S10011a Convolve pictures of the image data set according to a filter to obtain underlying features
  • S10011b Sampling the underlying features according to the sampling window to obtain pooled features
  • S10011c Connect the underlying features and the pooled features through a fully connected layer to obtain a convolutional neural network model.
  • the convolutional neural network is composed of a convolutional layer, a pooling layer, and a fully connected layer.
  • the convolutional layer and the pooling layer cooperate to form multiple convolutional groups, extract features layer by layer, and finally complete classification through several fully connected layers.
  • a convolutional network is essentially an input-to-output mapping. It can learn a large number of mappings between inputs and outputs without the need for any precise mathematical expressions between the inputs and outputs.
  • the model trains the convolutional network, and the network has the mapping capability between input and output pairs.
  • the convolutional network performs mentor training, so its sample set is composed of vector pairs of the form: (input vector, ideal output vector). All these vector pairs should be derived from the actual "running" results of the system that the network is about to simulate. They can be collected from actual operating systems. Before starting training, all weights should be initialized with some different small random numbers.
  • each convolution kernel represents an image mode. If an image block has a large convolution value with this convolution kernel, the image block is considered to be Very close to this convolution kernel. If you design 6 convolution kernels, you can understand that there are 6 underlying texture modes on this image, that is, an image can be drawn using 6 basic modes.
  • the original picture is 20x20, it is downsampled with a sampling window of 10x10, and finally it is downsampled into a 2x2 feature map.
  • the reason for the above pooling is that even after the convolution is done, the image is still large (because the convolution kernel is relatively small), so in order to reduce the data dimension, downsampling is performed. In the pooling process, even if a lot of data is reduced, the statistical properties of the features can still describe the image, and because the data dimension is reduced, overfitting is effectively avoided. In practical applications, pooling is divided into Max-Pooling and Mean-Pooling according to the down-sampling method.
  • the fully connected layer mainly re-fits the features to reduce the loss of feature information. For example, after fully connecting the underlying features with the pooled features, a convolutional neural network model is obtained.
  • S102 Obtain a to-be-recognized picture, and use the to-be-recognized picture as an input of a plurality of convolutional neural network models in the local single-page application, and obtain a to-be-processed result corresponding to each of the convolutional neural network models.
  • the step S102 includes:
  • a picture of a cat is input into a convolutional neural network model, and after calculating through the convolutional neural network model, an actual output value is 1 (1 represents a cat in a large animal category).
  • subsequent operations such as averaging, variance, or standard deviation are required to reduce the actual output value error.
  • a picture recognition result that is the same as the picture recognition result is obtained according to the picture recognition result. For example, you can also set the picture recognition result to 2 to indicate that the recognition is successful and the dogs in the animal category, and set the actual output value to be greater than 2 to indicate that the recognition has failed.
  • a plurality of to-be-processed results may be statistically calculated (such as an average operation, a variance operation, or an Standard deviation calculation) to obtain a representative local picture recognition result.
  • the to-be-processed results are counted to obtain a local picture recognition result.
  • a specific implementation manner is to perform an average operation, a variance calculation, or a standard deviation operation on the to-be-processed results. To get the local picture recognition result.
  • the keywords corresponding to the picture recognition result are displayed directly in the result display area in the local single-page application.
  • a picture of a cat is input into a convolutional neural network model, and after calculating through the convolutional neural network model, the picture recognition result is 1 (1 represents the picture recognition result of cats in the animal category), and then according to the picture recognition result, Get keywords (such as cats) that correspond to the picture recognition results.
  • the method further includes:
  • the convolutional neural network model carried in the local single-page application cannot accurately identify the picture to be identified
  • the picture to be identified can be uploaded to the enterprise back-end server, and a richer variety of convolutions can be performed.
  • the neural network model is used to recognize on the enterprise back-end server, and the correct recognition result is fed back after the recognition is completed. This method with both local recognition and background recognition ensures the efficiency and accuracy of picture recognition.
  • this method realizes the recognition of the picture through the convolutional neural network model in the local single-page application on the user side, and quickly obtains the recognition result. It is not necessary to directly upload the picture to the background server for recognition, which improves the recognition efficiency.
  • the embodiment of the present application further provides a single-page high-load image recognition device, and the single-page high-load image recognition device is configured to execute any one of the foregoing single-page high-load image recognition methods.
  • FIG. 6, is a schematic block diagram of a single-page high-load image recognition device according to an embodiment of the present application.
  • the single-page high-load image recognition device 100 may be configured in a terminal such as a desktop computer, a tablet computer, a laptop computer, or the like.
  • the single-page high-load image recognition device 100 includes a model acquisition unit 101, a local picture recognition unit 102, a statistics unit 103, and a display unit 104.
  • the model acquiring unit 101 is configured to acquire a plurality of convolutional neural network models stored in a local single-page application at a user end for image recognition.
  • the single-page high-load image recognition device 100 further includes:
  • An installation package obtaining unit 1001 is configured to obtain an installation package of a local single-page application including a convolutional neural network model
  • the decompression installation unit 1002 is configured to decompress the installation package of the local single-page application to obtain a local single-page application including multiple convolutional neural network models.
  • the installation package obtaining unit 1001 includes:
  • a model training unit 10011 is configured to train the convolutional neural network model by using a pre-imported image data set as training data of the convolutional neural network model to obtain a trained convolutional neural network model;
  • a model loading unit 10012 is configured to package the trained convolutional neural network model into an installation package of a local single-page application.
  • the model training unit 10011 includes:
  • a convolution layer unit 10011a configured to convolve pictures of the image data set according to a filter to obtain underlying features
  • a pooling unit 10011b configured to sample the underlying features according to the sampling window to obtain pooled features
  • a fully connected unit 10011c is used to connect the underlying features and pooled features through a fully connected layer to obtain a convolutional neural network model.
  • a local picture recognition unit 102 is configured to obtain a picture to be recognized, and use the picture to be recognized as an input of multiple convolutional neural network models in the local single-page application, and obtain a one-to-one correspondence with each convolutional neural network model. Pending results.
  • the local picture recognition unit 102 includes:
  • a picture input unit 1021 configured to input a picture to be identified into each of the convolutional neural network models included in the multiple convolutional neural network models;
  • the computing unit 1022 is configured to calculate an actual output value of each convolutional neural network model as a to-be-processed result corresponding to each of the convolutional neural network models.
  • the statistics unit 103 is configured to collect statistics on the results to be processed to obtain a local picture recognition result.
  • the to-be-processed results are counted to obtain a local picture recognition result.
  • a specific implementation manner is to perform an average operation, a variance calculation, or a standard deviation operation on the to-be-processed results. To get the local picture recognition result.
  • the display unit 104 is configured to display a keyword corresponding to the local picture recognition result if the local picture recognition result is successful.
  • the single-page high-load image recognition device 100 further includes:
  • the background recognition unit 105 is configured to upload the to-be-recognized picture to the background server if the local picture recognition result is recognition failure, and receive the background recognition result fed back by the background server.
  • the device realizes the recognition of the picture through the convolutional neural network model in the local single-page application on the user side, and quickly obtains the recognition result, without directly uploading the picture to the background server for recognition, which improves the recognition efficiency.
  • the above-mentioned single-page high-load image recognition device can be implemented in the form of a computer program, which can be run on a computer device as shown in FIG. 11.
  • the computer device 500 may be a terminal.
  • the terminal may be an electronic device such as a tablet computer, a notebook computer, a desktop computer, a personal digital assistant, and the like.
  • the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501.
  • the memory may include a non-volatile storage medium 503 and an internal memory 504.
  • the non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032.
  • the computer program 5032 includes program instructions. When the program instructions are executed, the processor 502 can execute a single-page high-load image recognition method.
  • the processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.
  • the internal memory 504 provides an environment for running a computer program 5032 in the non-volatile storage medium 503.
  • the processor 502 can execute a single-page high-load image recognition method.
  • the network interface 505 is used for network communication, such as sending assigned tasks.
  • the structure shown in FIG. 11 is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer equipment 500 to which the scheme of the present application is applied.
  • the specific computer equipment 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.
  • the processor 502 is configured to run a computer program 5032 stored in a memory to implement the following functions: obtaining a plurality of convolutional neural network models for image recognition stored in a local single-page application at a user terminal; Identify the picture and use the to-be-recognized picture as an input to a plurality of convolutional neural network models in the local single-page application to obtain a to-be-processed result corresponding to each of the convolutional neural network models; The results are counted to obtain a local picture recognition result; if the local picture recognition result is successful, the keywords corresponding to the local picture recognition result are displayed.
  • the processor 502 further performs the following operations: obtaining an installation package of a local single-page application including a convolutional neural network model; and decompressing the installation package of the local single-page application to obtain a plurality of convolutional nerves. Native single page application of the network model.
  • the processor 502 further performs the following operation: training the convolutional neural network model by using a pre-imported image data set as training data of the convolutional neural network model to obtain a trained convolutional neural network model ; Packaging the trained convolutional neural network model into an installation package of a local single-page application;
  • the processor 502 further performs the following operations: convolving the pictures of the image data set according to the filter to obtain the underlying features; sampling the underlying features according to the sampling window to obtain the pooled features; and the underlying features It is connected to the pooling feature through a fully connected layer to obtain a convolutional neural network model.
  • the processor 502 further performs the following operations: inputting the picture to be identified into each of the convolutional neural network models included in the plurality of convolutional neural network models; and calculating the actual output value of each of the convolutional neural network models. As a to-be-processed result corresponding to each convolutional neural network model.
  • the processor 502 further performs the following operations: performing an average operation, a variance operation, or a standard deviation operation on the results to be processed to obtain a local picture recognition result.
  • the processor 502 further performs the following operation: if the local picture recognition result is recognition failure, uploading the picture to be recognized to the background server, and receiving the background recognition result fed back by the background server.
  • the embodiment of the computer device shown in FIG. 11 does not constitute a limitation on the specific configuration of the computer device.
  • the computer device may include more or fewer components than shown in the figure. Either some parts are combined or different parts are arranged.
  • the computer device may include only a memory and a processor. In such an embodiment, the structure and function of the memory and the processor are consistent with the embodiment shown in FIG. 11, and details are not described herein again.
  • the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSPs), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor, or the processor may be any conventional processor.
  • a storage medium may be a computer-readable storage medium.
  • the storage medium stores a computer program, where the computer program includes program instructions.
  • the program instructions are executed by a processor, the single-page high-load image recognition method in the embodiment of the present application is implemented.
  • the storage medium may be an internal storage unit of the foregoing device, such as a hard disk or a memory of the device.
  • the storage medium may also be an external storage device of the device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, and a flash memory card provided on the device. (Flash Card), etc.
  • the storage medium may further include both an internal storage unit of the device and an external storage device.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

一种单页高负载图像识别方法、装置、计算机设备及存储介质。该方法包括：获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型（S101）；获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果（S102）；将所述待处理结果进行统计，得到本地的图片识别结果（S103）；若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示（S104）；若本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果（S105）。该方法实现在用户端的本地单页应用中通过卷积神经网络模型对图片进行识别，快速的得到识别结果，无需直接将图片上传至后台服务器进行识别，提高了识别效率。

Description

单页高负载图像识别方法、装置、计算机设备及存储介质
本申请要求于2018年5月25日提交中国专利局、申请号为201810513879.8、申请名称为“单页高负载图像识别方法、装置、计算机设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像识别技术领域，尤其涉及一种单页高负载图像识别方法、装置、计算机设备及存储介质。
背景技术
目前，主流的图像识别应用，都需要将图片上传到企业后台服务器才可以进行图像识别，比如百度图片搜索。其不能在客户端应用神经网络识别图片，如果网速较差，上传图片速度较慢，就会影响图像识别结果反馈的速度。
发明内容
本申请提供了一种单页高负载图像识别方法、装置、计算机设备及存储介质，旨在解决现有技术中主流的图像识别应用，都需要将图片上传到企业后台服务器才可以进行图像识别，如果网速较差，上传图片速度较慢，就会影响图像识别结果反馈的速度的问题。
第一方面，本申请提供了一种单页高负载图像识别方法，其包括：
获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型；
获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果；
若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示；
若本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果。
第二方面，本申请提供了一种单页高负载图像识别装置，其包括：
模型获取单元，用于获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型；
本地图片识别单元，用于获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果；
显示单元，用于若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示；
后台识别单元，用于若本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果。
第三方面，本申请又提供了一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现本申请提供的任一项所述的单页高负载图像识别方法。
第四方面，本申请还提供了一种存储介质，其中所述存储介质存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被处理器执行时使所述处理器执行本申请提供的任一项所述的单页高负载图像识别方法。
本申请提供一种单页高负载图像识别方法、装置、计算机设备及存储介质。该方法实现在用户端的本地单页应用中通过卷积神经网络模型对图片进行识别，快速的得到识别结果，无需直接将图片上传至后台服务器进行识别，提高了识别效率。
附图说明
为了更清楚地说明本申请实施例技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的一种单页高负载图像识别方法的示意流程图；
图2为本申请实施例提供的一种单页高负载图像识别方法的另一示意流程图；
图3是本申请实施例提供的一种单页高负载图像识别方法的子流程示意图；
图4为本申请实施例提供的一种单页高负载图像识别方法的另一子流程示意图；
图5为本申请实施例提供的一种单页高负载图像识别方法的另一子流程示意图；
图6为本申请实施例提供的一种单页高负载图像识别装置的示意性框图；
图7为本申请实施例提供的一种单页高负载图像识别装置的另一示意性框图；
图8为本申请实施例提供的一种单页高负载图像识别装置的子单元示意性框图；
图9为本申请实施例提供的一种单页高负载图像识别装置的另一子单元示意性框图；
图10为本申请实施例提供的一种单页高负载图像识别装置的另一子单元示意性框图；
图11为本申请实施例提供的一种计算机设备的示意性框图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
请参阅图1，图1是本申请实施例提供的一种单页高负载图像识别方法的示意流程图。该方法应用于台式电脑、手提电脑、平板电脑等终端中。如图1所示，该方法包括步骤S101～S104。
S101、获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型。
在本实施例中，本地单页应用中存储的卷积神经网络模型是在下载单页应用的安装包时，安装包中自带的可用于图像识别的卷积神经模型，该卷积神经模型是存储在通过AngularJS(是基于javascript的前端语言)所构建的本地单页应用中。用户端安装了该安装包后，即可通过本地单页应用中存储的卷积神经网络模型进行图像识别。
其中，AngularJS仅需HTML(即超文本标记语言)，CSS(即层叠样式表)和JavaScript(是一种直译式脚本语言)就可在客户端创建单页应用，其目标是使得开发和测试更容易，增强MVC Web应用的性能。
所谓单页应用，指的是在一个页面上集成多种功能，甚至整个系统就只有一个页面，所有的业务功能都是它的子模块，通过特定的方式挂接到主界面上。本申请中用户端的本地单页应用点击打开后，类似于进入一个单机网页页面，该单机网页页面上具有简单的操作界面，如添加待识别图片的操作区，及输出识别结果的显示区。
本地单页应用中存储的卷积神经网络模型是在企业后台服务器上通过将预选导入的图像数据集(例如所述图像数据集)的图片输入卷积神经网络，并进行训练而得到。其中，ImageNet是一个计算机视觉系统识别项目名称，其是目前世界上图像识别最大的数据库，它是美国斯坦福的计算机科学家，模拟人类的识别系统建立的图像识别数据库。ImageNet数据集(也即所述图像数据集)有1400多万幅图片，涵盖2万多个类别；其中有超过百万的图片有明确的类别标注和图像中物体位置的标注，该ImageNet数据集的大小约为1TB。
在一实施例中，如图2所示，在所述步骤S101之前还包括：
S1001、获取包括卷积神经网络模型的本地单页应用的安装包；
S1002、将所述本地单页应用的安装包进行解压，得到包括多个卷积神经网络模型的本地单页应用。
在本实施例中，本地单页应用的安装包(该安装包内存储有用于图像识别的卷积神经网络模型)中所包括的卷积神经网络模型，是在企业后台服务器上进行大量的图像识别训练而得到。即在企业后台服务器上通过所述图像数据集作为输入训练卷积神经网络，所得到的卷积神经网络模型是包含多个种类模型的，例如有识别交通工具的卷积神经网络模型，有识别动物的卷积神经网络模型等，上述多个种类的卷积神经网络模型训练完成后，可直接导入单页应用的安装包(如apk安装包或iOs安装包)。当用户下载了单页应用的安装包后，在本地终端上解压安装该单页应用的安装包，即可得到一个具有识别图像的单页应用，无需将待识别图片上传至企业后台服务器上进行识别。
在一实施例中，如图3所示，所述步骤S1001包括：
S10011、采用预先导入的图像数据集作为所述卷积神经网络模型的训练数 据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型；
S10012、将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中。
在本实施例中，上述训练过程既可选择在企业后台服务器端进行，也可在用户端进行。例如在用户端进行训练时，可以选择待识别图像的种类(如交通工具，动物，植物等)，直接根据所选择的待识别图像的种类，对应获取Image在数据集，并在用户端的本地中作为卷积神经网络的训练数据，在用户端训练得到卷积神经网络模型，并将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中，以供本终端自身使用，或者是供其他下载了该用户端上传的安装包(该安装包即包括卷积神经网络模型的本地单页应用的安装包)。
其中，为了提高对图片识别的正确率，导入到单页应用的安装包内的卷积神经网络模型的种类是越多越好，但是一般将整个单页应用的安装包的大小控制在200M左右即可满足基本的图片识别要求，也即导入10-20种的卷积神经网络模型即可(例如满足了识别交通工具，动物，植物等类别的图像)。
在一实施例中，如图4所示，所述步骤S10011包括：
S10011a、将所述图像数据集的图片根据过滤器进行卷积，得到底层特征；
S10011b、根据采样窗口对底层特征进行采样，得到池化特征；
S10011c、将底层特征与池化特征通过全连接层进行连接，得到卷积神经网络模型。
在本实施例中，卷积神经网络由卷积层、池化层、全连接层组成。其中卷积层与池化层配合，组成多个卷积组，逐层提取特征，最终通过若干个全连接层完成分类。
卷积网络在本质上是一种输入到输出的映射，它能够学习大量的输入与输出之间的映射关系，而不需要任何输入和输出之间的精确的数学表达式，只要用已知的模式对卷积网络加以训练，网络就具有输入输出对之间的映射能力。卷积网络执行的是有导师训练，所以其样本集是由形如：(输入向量，理想输出向量)的向量对构成的。所有这些向量对，都应该是来源于网络即将模拟的系统的实际“运行”结果。它们可以是从实际运行系统中采集来的。在开始训练前，所有的权都应该用一些不同的小随机数进行初始化。“小随机数”用来保证网络不会因权值过大而进入饱和状态，从而导致训练失败；“不同”用来 保证网络可以正常地学习。实际上，如果用相同的数去初始化权矩阵，则网络无能力学习。
例如，有一个5x5的图像(其原始大小为20x20)，用一个3x3的过滤器(卷积核)对图像进行卷积，得到了3x3的底层特征；这个过程可以理解为使用一个过滤器(卷积核)来过滤图像的各个小区域，从而得到这些小区域的特征值。在实际训练过程中，卷积核的值是在学习过程中学到的。在具体应用中，往往有多个卷积核，可以认为，每个卷积核代表了一种图像模式，如果某个图像块与此卷积核卷积出的值大，则认为此图像块十分接近于此卷积核。如果设计了6个卷积核，可以理解：这个图像上有6种底层纹理模式，也就是用6种基础模式就能描绘出一副图像。
由于原始图片是20x20的，对其进行下采样，采样窗口为10x10，最终将其下采样成为一个2x2大小的特征图。之所以进行上述池化，是因为即使做完了卷积，图像仍然很大(因为卷积核比较小)，所以为了降低数据维度，就进行下采样。在池化的过程中，即使减少了许多数据，特征的统计属性仍能够描述图像，而且由于降低了数据维度，有效地避免了过拟合。在实际应用中，池化根据下采样的方法，分为最大值下采样(Max-Pooling)与平均值下采样(Mean-Pooling)。
全连接层主要对特征进行重新拟合，减少特征信息的丢失，如将底层特征与池化特征进行全连接后，得到卷积神经网络模型。
S102、获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果。
在一实施例中，如图5所示，所述步骤S102包括：
S1021、将待识别图片输入多个卷积神经网络模型中所包括的每一卷积神经网络模型；
S1022、计算每一卷积神经网络模型的实际输出值以作为与每一卷积神经网络模型一一对应的待处理结果。
在本实施例中，例如将一猫的图片输入卷积神经网络模型，通过卷积神经网络模型计算后，得到实际输出值为1(1表示动物大类中的猫类)。由于通过一个卷积神经网络模型得到的实际输出值可能存在误差，后续需对每一实际输 出值进行求平均值、求方差、或求标准差等运算以降低实际输出值的误差。例如将所述待处理结果进行统计，得到本地的图片识别结果后，则根据图片识别结果，获取与图片识别结果的图片识别结果。例如，还可以设置图片识别结果为2时表示识别成功且表示动物大类中的狗类，设置实际输出值大于2时表示识别失败。
由于是直接调本地的单页应用中的卷积神经网络模型，也不用将图片上传到后台服务器，调用后台服务器的模型来进行识别，无需考虑上传图片至后台服务器网速慢而降低识别效率。
S103、将所述待处理结果进行统计，得到本地的图片识别结果。
在本实施例中，当获取了与每一卷积神经网络模型一一对应的待处理结果，则可对多个待处理结果进行统计(如进行求平均值运算、或求方差运算、或求标准差运算)，得到一个具有代表性的本地的图片识别结果。通过这种多重识别求统计值的方式，能有效避免单次运算的误差。
在一实施例中，将所述待处理结果进行统计，得到本地的图片识别结果的具体实现方式是，对将所述待处理结果进行求平均值运算、或求方差运算、或求标准差运算，得到本地的图片识别结果。
S104、若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示。
在本实施例中，当本地的图片识别结果为识别成功时，直接在本地的单页应用中的结果显示区域显示图片识别结果对应的关键词。例如将一猫的图片输入卷积神经网络模型，通过卷积神经网络模型计算后，得到图片识别结果为1(1表示动物大类中的猫类的图片识别结果)，则根据图片识别结果，获取与图片识别结果对应的关键词(如猫)。通过在本地的单页应用中直接显示识别结果，方便用户查看。
在一实施例中，如图1所示，步骤S103之后还包括：
S105、若本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果。
在本实施例中，当本地的单页应用中所搭载的卷积神经网络模型不能准确识别待识别图片时，则可将该待识别图片上传至企业后台服务器，在具有更丰富种类的卷积神经网络模型的企业后台服务器上进行识别，并在识别完成后反 馈正确的识别结果，这种兼具本地识别和后台识别的方式，确保了图片识别的效率和准确率。
可见，该方法实现在用户端的本地单页应用中通过卷积神经网络模型对图片进行识别，快速的得到识别结果，无需直接将图片上传至后台服务器进行识别，提高了识别效率。
本申请实施例还提供一种单页高负载图像识别装置，该单页高负载图像识别装置用于执行前述单页高负载图像识别方法的任一实施例。具体地，请参阅图6，图6是本申请实施例提供的一种单页高负载图像识别装置的示意性框图。单页高负载图像识别装置100可以配置于台式电脑、平板电脑、手提电脑、等终端中。
如图6所示，单页高负载图像识别装置100包括模型获取单元101、本地图片识别单元102、统计单元103、及显示单元104。
模型获取单元101，用于获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型。
在一实施例中，如图7所示，所述单页高负载图像识别装置100还包括：
安装包获取单元1001，用于获取包括卷积神经网络模型的本地单页应用的安装包；
解压安装单元1002，用于将所述本地单页应用的安装包进行解压，得到包括多个卷积神经网络模型的本地单页应用。
在一实施例中，如图8所示，所述安装包获取单元1001包括：
模型训练单元10011，用于采用预先导入的图像数据集作为所述卷积神经网络模型的训练数据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型；
模型加载单元10012，用于将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中。
在一实施例中，如图9所示，所述模型训练单元10011包括：
卷积层单元10011a，用于将所述图像数据集的图片根据过滤器进行卷积，得到底层特征；
池化单元10011b，用于根据采样窗口对底层特征进行采样，得到池化特征；
全连接单元10011c，用于将底层特征与池化特征通过全连接层进行连接， 得到卷积神经网络模型。
本地图片识别单元102，用于获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果。
在一实施例中，如图10所示，所述本地图片识别单元102包括：
图片输入单元1021，用于将待识别图片输入多个卷积神经网络模型中所包括的每一卷积神经网络模型；
计算单元1022，用于计算每一卷积神经网络模型的实际输出值以作为与每一卷积神经网络模型一一对应的待处理结果。
统计单元103，用于将所述待处理结果进行统计，得到本地的图片识别结果。
在一实施例中，将所述待处理结果进行统计，得到本地的图片识别结果的具体实现方式是，对将所述待处理结果进行求平均值运算、或求方差运算、或求标准差运算，得到本地的图片识别结果。
显示单元104，用于若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示。
在一实施例中，如图6所示，所述单页高负载图像识别装置100还包括：
后台识别单元105，用于若本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果。
可见，该装置实现在用户端的本地单页应用中通过卷积神经网络模型对图片进行识别，快速的得到识别结果，无需直接将图片上传至后台服务器进行识别，提高了识别效率。
上述单页高负载图像识别装置可以实现为一种计算机程序的形式，该计算机程序可以在如图11所示的计算机设备上运行。
请参阅图11，图11是本申请实施例提供的一种计算机设备的示意性框图。该计算机设备500设备可以是终端。该终端可以是平板电脑、笔记本电脑、台式电脑、个人数字助理等电子设备。
参阅图11，该计算机设备500包括通过系统总线501连接的处理器502、存储器和网络接口505，其中，存储器可以包括非易失性存储介质503和内存储器504。
该非易失性存储介质503可存储操作系统5031和计算机程序5032。该计算 机程序5032包括程序指令，该程序指令被执行时，可使得处理器502执行一种单页高负载图像识别方法。
该处理器502用于提供计算和控制能力，支撑整个计算机设备500的运行。
该内存储器504为非易失性存储介质503中的计算机程序5032的运行提供环境，该计算机程序5032被处理器502执行时，可使得处理器502执行一种单页高负载图像识别方法。
该网络接口505用于进行网络通信，如发送分配的任务等。本领域技术人员可以理解，图11中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备500的限定，具体的计算机设备500可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
其中，所述处理器502用于运行存储在存储器中的计算机程序5032，以实现如下功能：获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型；获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果；将所述待处理结果进行统计，得到本地的图片识别结果；若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示。
在一实施例中，处理器502还执行如下操作：获取包括卷积神经网络模型的本地单页应用的安装包；将所述本地单页应用的安装包进行解压，得到包括多个卷积神经网络模型的本地单页应用。
在一实施例中，处理器502还执行如下操作：采用预先导入的图像数据集作为所述卷积神经网络模型的训练数据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型；将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中。
在一实施例中，处理器502还执行如下操作：将所述图像数据集的图片根据过滤器进行卷积，得到底层特征；根据采样窗口对底层特征进行采样，得到池化特征；将底层特征与池化特征通过全连接层进行连接，得到卷积神经网络模型。
在一实施例中，处理器502还执行如下操作：将待识别图片输入多个卷积 神经网络模型中所包括的每一卷积神经网络模型；计算每一卷积神经网络模型的实际输出值以作为与每一卷积神经网络模型一一对应的待处理结果。
在一实施例中，处理器502还执行如下操作：将所述待处理结果进行求平均值运算、或求方差运算、或求标准差运算，得到本地的图片识别结果。
在一实施例中，处理器502还执行如下操作：若所述本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果。
本领域技术人员可以理解，图11中示出的计算机设备的实施例并不构成对计算机设备具体构成的限定，在其他实施例中，计算机设备可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。例如，在一些实施例中，计算机设备可以仅包括存储器及处理器，在这样的实施例中，存储器及处理器的结构及功能与图11所示实施例一致，在此不再赘述。
应当理解，在本申请实施例中，处理器502可以是中央处理单元(Central Processing Unit，CPU)，该处理器502还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中，通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
在本申请的另一实施例中提供一种存储介质。该存储介质可以为计算机可读存储介质。该存储介质存储有计算机程序，其中计算机程序包括程序指令。该程序指令被处理器执行时实现本申请实施例的单页高负载图像识别方法。
所述存储介质可以是前述设备的内部存储单元，例如设备的硬盘或内存。所述存储介质也可以是所述设备的外部存储设备，例如所述设备上配备的插接式硬盘，智能存储卡(Smart Media Card，SMC)，安全数字(Secure Digital，SD)卡，闪存卡(Flash Card)等。进一步地，所述存储介质还可以既包括所述设备的内部存储单元也包括外部存储设备。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的设备、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
以上所述，仅为本发明的具体实施方式，但本发明的保护范围并不局限于 此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本发明的保护范围之内。因此，本发明的保护范围应以权利要求的保护范围为准。

Claims (20)

  1. 一种单页高负载图像识别方法，其特征在于，包括：
    获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型；
    获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果；
    将所述待处理结果进行统计，得到本地的图片识别结果；
    若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示。
  2. 根据权利要求1所述的单页高负载图像识别方法，其特征在于，所述获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型之前，包括：
    获取包括卷积神经网络模型的本地单页应用的安装包；
    将所述本地单页应用的安装包进行解压，得到包括多个卷积神经网络模型的本地单页应用。
  3. 根据权利要求2所述的单页高负载图像识别方法，其特征在于，所述获取包括卷积神经网络模型的本地单页应用的安装包，包括：
    采用预先导入的图像数据集作为所述卷积神经网络模型的训练数据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型；
    将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中。
  4. 根据权利要求3所述的单页高负载图像识别方法，其特征在于，所述采用预先导入的图像数据集作为所述卷积神经网络模型的训练数据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型，包括：
    将所述图像数据集的图片根据过滤器进行卷积，得到底层特征；
    根据采样窗口对底层特征进行采样，得到池化特征；
    将底层特征与池化特征通过全连接层进行连接，得到卷积神经网络模型。
  5. 根据权利要求1所述的单页高负载图像识别方法，其特征在于，所述将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果，包括：
    将待识别图片输入多个卷积神经网络模型中所包括的每一卷积神经网络模型；
    计算每一卷积神经网络模型的实际输出值以作为与每一卷积神经网络模型一一对应的待处理结果。
  6. 根据权利要求5所述的单页高负载图像识别方法，其特征在于，所述将所述待处理结果进行统计，得到本地的图片识别结果，包括：
    将所述待处理结果进行求平均值运算、或求方差运算、或求标准差运算，得到本地的图片识别结果。
    根据实际输出值，获取与实际输出值对应的图片识别结果。
  7. 根据权利要求1所述的单页高负载图像识别方法，其特征在于，所述将所述待识别图片作为本地单页应用中卷积神经网络模型的输入，得到本地的图片识别结果之后，还包括：
    若所述本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果。
  8. 一种单页高负载图像识别装置，其特征在于，包括：
    模型获取单元，用于获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型；
    本地图片识别单元，用于获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果；
    统计单元，用于将所述待处理结果进行统计，得到本地的图片识别结果；
    显示单元，用于若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示。
  9. 根据权利要求8所述的单页高负载图像识别装置，其特征在于，还包括：
    安装包获取单元，用于获取包括卷积神经网络模型的本地单页应用的安装包；
    解压安装单元，用于将所述本地单页应用的安装包进行解压，得到包括多个卷积神经网络模型的本地单页应用。
  10. 根据权利要求9所述的单页高负载图像识别装置，其特征在于，所述安装包获取单元包括：
    模型训练单元，用于采用预先导入的图像数据集作为所述卷积神经网络模型的训练数据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型；
    模型加载单元，用于将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中。
  11. 根据权利要求10所述的单页高负载图像识别装置，其特征在于，所述模型训练单元包括：
    卷积层单元，用于将所述图像数据集的图片根据过滤器进行卷积，得到底层特征；
    池化单元，用于根据采样窗口对底层特征进行采样，得到池化特征；
    全连接单元，用于将底层特征与池化特征通过全连接层进行连接，得到卷积神经网络模型。
  12. 根据权利要求8所述的单页高负载图像识别装置，其特征在于，所述本地图片识别单元包括：
    图片输入单元，用于将待识别图片输入多个卷积神经网络模型中所包括的每一卷积神经网络模型；
    计算单元，用于计算每一卷积神经网络模型的实际输出值以作为与每一卷积神经网络模型一一对应的待处理结果。
  13. 根据权利要求12所述的单页高负载图像识别装置，其特征在于，所述统计单元中对将所述待处理结果进行求平均值运算、或求方差运算、或求标准差运算，得到本地的图片识别结果。
  14. 根据权利要求8所述的单页高负载图像识别装置，其特征在于，还包括：
    后台识别单元，用于若本地的图片识别结果为识别失败，将待识别图片上传至后台服务器，接收由后台服务器反馈的后台识别结果。
  15. 一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现以下步骤：
    获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型；
    获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果；
    将所述待处理结果进行统计，得到本地的图片识别结果；
    若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示。
  16. 根据权利要求15所述的计算机设备，其特征在于，所述获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型之前，包括：
    获取包括卷积神经网络模型的本地单页应用的安装包；
    将所述本地单页应用的安装包进行解压，得到包括多个卷积神经网络模型的本地单页应用。
  17. 根据权利要求16所述的计算机设备，其特征在于，所述获取包括卷积神经网络模型的本地单页应用的安装包，包括：
    采用预先导入的图像数据集作为所述卷积神经网络模型的训练数据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型；
    将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中。
  18. 一种存储介质，其特征在于，所述存储介质存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被处理器执行时使所述处理器执行以下操作：
    获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型；
    获取待识别图片，并将所述待识别图片作为所述本地单页应用中多个卷积神经网络模型的输入，得到与每一卷积神经网络模型一一对应的待处理结果；
    将所述待处理结果进行统计，得到本地的图片识别结果；
    若所述本地的图片识别结果为识别成功，将所述本地的图片识别结果对应的关键词进行显示。
  19. 根据权利要求18所述的存储介质，其特征在于，所述获取用户端的本地单页应用中存储的多个用于进行图像识别的卷积神经网络模型之前，包括：
    获取包括卷积神经网络模型的本地单页应用的安装包；
    将所述本地单页应用的安装包进行解压，得到包括多个卷积神经网络模型的本地单页应用。
  20. 根据权利要求19所述的存储介质，其特征在于，所述获取包括卷积神经网络模型的本地单页应用的安装包，包括：
    采用预先导入的图像数据集作为所述卷积神经网络模型的训练数据训练所述卷积神经网络模型，得到训练后的卷积神经网络模型；
    将所述训练后的卷积神经网络模型打包至本地单页应用的安装包中。
PCT/CN2018/102471 2018-05-25 2018-08-27 单页高负载图像识别方法、装置、计算机设备及存储介质 Ceased WO2019223154A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810513879.8A CN108846419A (zh) 2018-05-25 2018-05-25 单页高负载图像识别方法、装置、计算机设备及存储介质
CN201810513879.8 2018-05-25

Publications (1)

Publication Number Publication Date
WO2019223154A1 true WO2019223154A1 (zh) 2019-11-28

Family

ID=64213520

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/102471 Ceased WO2019223154A1 (zh) 2018-05-25 2018-08-27 单页高负载图像识别方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN108846419A (zh)
WO (1) WO2019223154A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110969143A (zh) * 2019-12-19 2020-04-07 深圳壹账通智能科技有限公司 基于图像识别的取证方法、系统、计算机设备及存储介质
CN111598226A (zh) * 2020-05-18 2020-08-28 济南浪潮高新科技投资发展有限公司 一种用于图像识别的卷积网络可视化方法及装置
CN111652262A (zh) * 2020-03-19 2020-09-11 深圳市彬讯科技有限公司 图像物体识别方法、装置、计算机设备及存储介质
CN111680775A (zh) * 2020-06-09 2020-09-18 河北科技师范学院 一种基于深度神经网络的图像识别模组
CN111797921A (zh) * 2020-07-02 2020-10-20 中国工商银行股份有限公司 一种图像数据对比方法及装置
CN112330816A (zh) * 2020-10-19 2021-02-05 杭州易现先进科技有限公司 Ar识别的处理方法、装置和电子装置
CN113537448A (zh) * 2020-04-22 2021-10-22 杭州智芯科微电子科技有限公司 流式数据处理的方法、装置、半导体芯片和计算机设备

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112241648A (zh) * 2019-07-16 2021-01-19 杭州海康威视数字技术股份有限公司 图像处理系统及图像设备
CN111917837A (zh) * 2020-07-13 2020-11-10 西安即刻易用网络科技有限公司 一种Web微应用程序发布系统及其实现方法
CN111913757B (zh) * 2020-07-13 2025-04-22 西安即刻易用网络科技有限公司 一种单页面应用程序加载方法及系统
CN111767902A (zh) * 2020-07-31 2020-10-13 上海中通吉网络技术有限公司 安检机危险品的识别方法、装置、设备及存储介质
CN114627459A (zh) * 2020-12-14 2022-06-14 菜鸟智能物流控股有限公司 Ocr识别方法、识别装置和识别系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8897577B2 (en) * 2011-06-09 2014-11-25 Electronics & Telecommunications Research Institute Image recognition device and method of recognizing image thereof
CN105938557A (zh) * 2015-03-06 2016-09-14 松下知识产权经营株式会社 图像识别方法、图像识别装置
CN106293972A (zh) * 2016-08-12 2017-01-04 金蝶软件（中国）有限公司 单页应用页面通讯方法和装置
CN106375378A (zh) * 2016-08-25 2017-02-01 北京每刻风物科技有限公司 一种基于局域网客户端服务器结构的应用部署方法及系统

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105512676A (zh) * 2015-11-30 2016-04-20 华南理工大学 一种智能终端上的食物识别方法
CN105825214B (zh) * 2016-03-14 2019-02-05 东南大学 一种基于tesseract引擎的文字识别方法及装置
CN107491726B (zh) * 2017-07-04 2020-08-04 重庆邮电大学 一种基于多通道并行卷积神经网络的实时表情识别方法
CN107657281A (zh) * 2017-09-28 2018-02-02 辽宁工程技术大学 一种基于改进的卷积神经网络的图像识别方法
CN107704853A (zh) * 2017-11-24 2018-02-16 重庆邮电大学 一种基于多分类器的交通信号灯的识别方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8897577B2 (en) * 2011-06-09 2014-11-25 Electronics & Telecommunications Research Institute Image recognition device and method of recognizing image thereof
CN105938557A (zh) * 2015-03-06 2016-09-14 松下知识产权经营株式会社 图像识别方法、图像识别装置
CN106293972A (zh) * 2016-08-12 2017-01-04 金蝶软件（中国）有限公司 单页应用页面通讯方法和装置
CN106375378A (zh) * 2016-08-25 2017-02-01 北京每刻风物科技有限公司 一种基于局域网客户端服务器结构的应用部署方法及系统

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110969143A (zh) * 2019-12-19 2020-04-07 深圳壹账通智能科技有限公司 基于图像识别的取证方法、系统、计算机设备及存储介质
CN111652262A (zh) * 2020-03-19 2020-09-11 深圳市彬讯科技有限公司 图像物体识别方法、装置、计算机设备及存储介质
CN113537448A (zh) * 2020-04-22 2021-10-22 杭州智芯科微电子科技有限公司 流式数据处理的方法、装置、半导体芯片和计算机设备
CN111598226A (zh) * 2020-05-18 2020-08-28 济南浪潮高新科技投资发展有限公司 一种用于图像识别的卷积网络可视化方法及装置
CN111598226B (zh) * 2020-05-18 2023-07-25 山东浪潮科学研究院有限公司 一种用于图像识别的卷积网络可视化方法及装置
CN111680775A (zh) * 2020-06-09 2020-09-18 河北科技师范学院 一种基于深度神经网络的图像识别模组
CN111680775B (zh) * 2020-06-09 2022-02-01 河北科技师范学院 一种基于深度神经网络的图像识别模组
CN111797921A (zh) * 2020-07-02 2020-10-20 中国工商银行股份有限公司 一种图像数据对比方法及装置
CN112330816A (zh) * 2020-10-19 2021-02-05 杭州易现先进科技有限公司 Ar识别的处理方法、装置和电子装置
CN112330816B (zh) * 2020-10-19 2024-03-26 杭州易现先进科技有限公司 Ar识别的处理方法、装置和电子装置

Also Published As

Publication number Publication date
CN108846419A (zh) 2018-11-20

Similar Documents

Publication Publication Date Title
WO2019223154A1 (zh) 单页高负载图像识别方法、装置、计算机设备及存储介质
EP4006909B1 (en) Method, apparatus and device for quality control and storage medium
CN108595583B (zh) 动态图表类页面数据爬取方法、装置、终端及存储介质
US10496653B1 (en) Data workflow analysis service
CN105512723B (zh) 一种用于稀疏连接的人工神经网络计算装置和方法
CN107239318A (zh) 应用转换方法、装置及设备
CN113568965A (zh) 一种结构化信息的提取方法、装置、电子设备及存储介质
WO2018059016A1 (zh) 针对机器学习的特征处理方法及特征处理系统
CN113887447A (zh) 对象分类模型的训练方法、对象分类预测方法及装置
CN112085056A (zh) 目标检测模型生成方法、装置、设备及存储介质
US11620496B2 (en) Convolutional neural network, and processing method, processing device, processing system and medium for the same
CN113688796A (zh) 基于边缘计算的对象识别方法、装置、设备及介质
WO2024001806A1 (zh) 一种基于联邦学习的数据价值评估方法及其相关设备
CN115674272A (zh) 机器人故障诊断方法、装置、设备以及存储介质
WO2023040147A1 (zh) 神经网络的训练方法及装置、存储介质和计算机程序
CN110827208A (zh) 卷积神经网络的通用池化增强方法、装置、设备及介质
CN112115266A (zh) 恶意网址的分类方法、装置、计算机设备和可读存储介质
WO2024114659A1 (zh) 一种摘要生成方法及其相关设备
CN113050921A (zh) 一种网页转换方法、装置、存储介质和计算机设备
CN114821191A (zh) 图像分类模型的训练方法、图像分类方法及装置
CN116309226A (zh) 一种图像处理方法及其相关设备
WO2025016352A1 (zh) 一种视频评估方法及其相关设备
WO2024179485A1 (zh) 一种图像处理方法及其相关设备
US10817313B2 (en) Machine learning techniques for adaptation of initial workspace
WO2024199404A1 (zh) 一种消费预测方法及其相关设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18920150

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 19.03.2021 )

122 Ep: pct application non-entry in european phase

Ref document number: 18920150

Country of ref document: EP

Kind code of ref document: A1