CN111488891B - Image identification processing method, device, equipment and computer readable storage medium - Google Patents
Image identification processing method, device, equipment and computer readable storage medium Download PDFInfo
- Publication number
- CN111488891B CN111488891B CN201910073615.XA CN201910073615A CN111488891B CN 111488891 B CN111488891 B CN 111488891B CN 201910073615 A CN201910073615 A CN 201910073615A CN 111488891 B CN111488891 B CN 111488891B
- Authority
- CN
- China
- Prior art keywords
- image
- identification
- images
- target image
- neural network
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides an image identifier processing method, an image identifier processing device, image identifier processing equipment and a computer readable storage medium, wherein the image identifier processing method comprises the following steps: acquiring a plurality of similar images of a target image in a preset image set in the preset image set; according to the identification information of the similar images, the identification of the target image is updated; the similar images are images with the matching degree with the target image reaching a preset threshold, and the number of the similar images is at least three. According to the scheme, the automatic correction of the image identification can be realized, and manual characteristics do not need to be manually extracted according to the image characteristics, so that the labor cost and the time cost are reduced, the working efficiency is improved, and the method is suitable for various image data; the problems that image identification processing schemes in the prior art all need manual participation, and are high in cost, low in efficiency and not universal are well solved.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to an image identifier processing method, an image identifier processing apparatus, an image identifier processing device, and a computer-readable storage medium.
Background
In recent years, with the development of internet technology, a deep convolutional neural network has been widely applied to the field of image classification, but in some fields requiring professional knowledge for image labeling, such as the medical field, the problems of unclear image classification boundary, low labeling personnel level and the like often exist, which directly results in low accuracy of image labeling, and thus the accuracy of a deep convolutional neural network model is seriously affected.
At present, although some optimized image classification methods based on the convolutional neural network exist, most of the image classification methods are based on network models and parameters, and the problem of inaccurate image labeling is not solved. The existing method for correcting image data annotation usually needs a plurality of annotation personnel to annotate the same image at the same time, which needs a large amount of labor cost and also reduces the efficiency of data set optimization and model establishment. Meanwhile, the accuracy of the trained classification network model is not high.
In addition, although there is no technical solution for directly correcting image annotation at present, there are relatively close solutions for annotating images, such as: firstly, manually extracting low-level fusion characteristic vectors of an image to be marked, such as a telescopic color characteristic, a homogeneous texture characteristic, an edge histogram characteristic, a grid color moment characteristic, a wavelet moment characteristic and the like; then, self-coding is carried out on the fusion characteristic vectors of all the images in the image library by using a depth automatic coding machine to obtain corresponding self-coding characteristics; and then, the image to be labeled finds N neighbor training images according to self-coding characteristics, and the automatic labeling of the image to be labeled is completed by using the labels of the group of neighbor training images.
However, the existing image annotation correction method or automatic annotation image scheme has the following disadvantages:
the first disadvantage is that:
the method for manually correcting the image annotation by multiple people to label the same batch of images or to find out the depth of label requires a large amount of labor cost and time cost, wastes time and labor, has low working efficiency, and often needs to pay expensive labeling cost for the field requiring professional knowledge to label.
The second defect:
at present, the existing automatic image labeling scheme usually adopts manual characteristics to represent an image, if the image is to be completely represented, a plurality of characteristic vectors are required to be adopted and screened, and the vectors can not necessarily well represent the image characteristics; the solution also has a problem of non-universality, i.e. for another batch of image data with different forms, the previously selected manual features may not be applicable, and different manual features need to be selected for different image characteristics.
As can be seen from the above, the existing image identification processing schemes all need manual participation, and have the problems of high cost, low efficiency and non-universality.
Disclosure of Invention
The invention aims to provide an image identification processing method, an image identification processing device, image identification processing equipment and a computer readable storage medium, and solves the problems that image identification processing schemes in the prior art all need manual participation, and are high in cost, low in efficiency and not universal.
In order to solve the above technical problem, an embodiment of the present invention provides an image identifier processing method, including:
acquiring a plurality of similar images of a target image in a preset image set in the preset image set;
according to the identification information of the similar images, the identification of the target image is updated;
the similar images are images with the matching degree with the target image reaching a preset threshold, and the number of the similar images is at least three.
Optionally, the obtaining of multiple similar images of the target image in the preset image set includes:
acquiring a feature vector of each image in a preset image set by using a convolutional neural network;
acquiring Euclidean distances between the target image and other images according to the characteristic vectors of the target image in the preset image set and the characteristic vectors of other images except the target image;
and acquiring a plurality of similar images of the target image from other images according to the Euclidean distance.
Optionally, the obtaining a feature vector of each image in the preset image set by using the convolutional neural network includes:
training a convolutional neural network by using all images in a preset image set;
and acquiring a characteristic vector obtained after each image in a preset image set passes through a penultimate full-connected layer of the convolutional neural network in the training process, wherein the characteristic vector is used as a characteristic vector corresponding to each image.
Optionally, after training the convolutional neural network by using all images in the preset image set, the method further includes:
obtaining a loss value of the convolutional neural network trained at this time;
judging whether the loss value meets a stable condition;
and if not, continuing to train the convolutional neural network according to the target image with the updated identifier, and returning to the operation of acquiring a plurality of similar images of the target image in the preset image set.
Optionally, the training the convolutional neural network continuously according to the target image updated by the identifier includes:
judging whether each image in the preset image set is respectively used as a target image to be subjected to identification updating;
if yes, forming an image set with updated identification according to the target image with updated identification;
and training the convolutional neural network again by using all the images in the image set after the identification is updated.
Optionally, the determining whether the loss value meets a stability condition includes:
and judging whether the loss value meets a stable condition or not according to the change rate of the loss value.
Optionally, the stabilizing conditions are:
therein, loss i Represents the loss value, loss k Representing the penalty value that would require a summation operation,
representing the rate of change of said loss value.
Optionally, after determining whether the loss value meets the stability condition, the method further includes:
and if the loss value accords with a stable condition, storing the parameter information of the convolutional neural network.
Optionally, after storing the parameter information of the convolutional neural network, the method further includes:
and identifying the image to be identified by utilizing a convolutional neural network according to the parameter information.
Optionally, the updating the identification of the target image according to the identification information of the multiple similar images includes:
and according to the identification information of the plurality of similar images, utilizing a voting method to update the identification of the target image.
An embodiment of the present invention further provides an image identifier processing apparatus, including:
the first acquisition module is used for acquiring a plurality of similar images of a target image in a preset image set in the preset image set;
the first processing module is used for carrying out identification updating on the target image according to the identification information of the similar images;
the similar images are images with the matching degree with the target image reaching a preset threshold, and the number of the similar images is at least three.
Optionally, the first obtaining module includes:
the first acquisition sub-module is used for acquiring a feature vector of each image in a preset image set by using a convolutional neural network;
the second obtaining submodule is used for obtaining Euclidean distances between the target image and other images according to the characteristic vectors of the target image in the preset image set and the characteristic vectors of other images except the target image;
and the third acquisition sub-module is used for acquiring a plurality of similar images of the target image from other images according to the Euclidean distance.
Optionally, the first obtaining sub-module includes:
the first processing unit is used for training a convolutional neural network by utilizing all images in a preset image set;
and the second processing unit is used for acquiring a feature vector obtained by each image in the preset image set after passing through a penultimate full-link layer of the convolutional neural network in the training process, and taking the feature vector as a feature vector corresponding to each image.
Optionally, the method further includes:
the second acquisition module is used for acquiring a loss value of the convolutional neural network trained at this time after the convolutional neural network is trained by using all images in a preset image set;
the first judgment module is used for judging whether the loss value meets a stable condition or not;
and if the target images do not meet the preset image set, continuing to train the convolutional neural network according to the target images after the identification update, and returning to the operation of acquiring a plurality of similar images of the target images in the preset image set.
Optionally, the second processing module includes:
the first judgment sub-module is used for judging whether each image in the preset image set is respectively used as a target image to be subjected to identification updating;
the first processing submodule is used for forming an image set after the identification is updated according to the target image after the identification is updated if the target image is the target image;
and the first updating submodule is used for retraining the convolutional neural network by utilizing all the images in the image set after the identification is updated.
Optionally, the first determining module includes:
and the second judgment submodule is used for judging whether the loss value meets the stability condition or not according to the change rate of the loss value.
Optionally, the stabilizing conditions are:
/>
therein, loss i Representing said loss value, loss k Representing the loss value of the required summation operation,
representing the rate of change of said loss value.
Optionally, the method further includes:
and the first storage module is used for storing the parameter information of the convolutional neural network if the loss value meets the stable condition after judging whether the loss value meets the stable condition.
Optionally, the method further includes:
and the first identification module is used for identifying the image to be identified by using the convolutional neural network according to the parameter information after the parameter information of the convolutional neural network is stored.
Optionally, the first processing module includes:
and the second processing submodule is used for carrying out identification updating on the target image by using a voting method according to the identification information of the plurality of similar images.
The embodiment of the invention also provides image identification processing equipment, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor; the processor implements the image identification processing method described above when executing the program.
Embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps in the image identification processing method.
The technical scheme of the invention has the following beneficial effects:
in the above scheme, the image identifier processing method obtains a plurality of similar images of a target image in a preset image set in the preset image set; according to the identification information of the similar images, the identification of the target image is updated; the similar images are images with the matching degree with the target image reaching a preset threshold value, and the number of the similar images is at least three; the automatic correction of the image identification can be realized, manual characteristics do not need to be manually extracted according to the image characteristics, so that the labor cost and the time cost are reduced, the working efficiency is improved, and the method is suitable for various image data; the problems that image identification processing schemes in the prior art all need manual participation, and are high in cost, low in efficiency and not universal are well solved.
Drawings
FIG. 1 is a schematic flow chart of an image identifier processing method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a specific application flow of the image identifier processing method according to the embodiment of the present invention;
fig. 3 is a schematic structural diagram of an image identifier processing apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments.
Aiming at the problems of high cost, low efficiency and non-universality of image identification processing schemes in the prior art which need manual participation, the invention provides an image identification processing method, as shown in figure 1, which comprises the following steps:
step 11: acquiring a plurality of similar images of a target image in a preset image set in the preset image set;
step 12: according to the identification information of the similar images, the identification of the target image is updated;
the similar images are images with the matching degree with the target image reaching a preset threshold, and the number of the similar images is at least three.
The preset image set can be selected and set according to the actual requirement, for example, the image set consisting of only images of cats and dogs is selected.
The identification information may be tagging information for a subject in the image, such as a cat, a dog, or the like, or may be other tagging information for the image, such as weather information, scene information, or the like, which is not limited herein.
The image identification processing method provided by the embodiment of the invention obtains a plurality of similar images of a target image in a preset image set in the preset image set; according to the identification information of the similar images, the identification of the target image is updated; the similar images are images with the matching degree with the target image reaching a preset threshold value, and the number of the similar images is at least three; the automatic correction of the image identification can be realized, manual characteristics do not need to be manually extracted according to the image characteristics, so that the labor cost and the time cost are reduced, the working efficiency is improved, and the method is suitable for various image data; the problems that image identification processing schemes in the prior art all need manual participation, and are high in cost, low in efficiency and not universal are well solved.
The acquiring of multiple similar images of a target image in a preset image set in the preset image set includes: acquiring a feature vector of each image in a preset image set by using a convolutional neural network; acquiring Euclidean distances between the target image and each other image according to the characteristic vectors of the target image in the preset image set and the characteristic vectors of each other image except the target image; and acquiring a plurality of similar images of the target image from other images according to the Euclidean distance.
Specifically, the obtaining of multiple similar images of the target image from other images according to the euclidean distance may include: and sequencing the obtained Euclidean distances in an ascending order, and taking other images corresponding to the Euclidean distances of the first n bits as similar images of the target image, wherein n is more than or equal to 3 and less than or equal to the total number of other images. After sorting according to the Euclidean distance in an ascending order, the more front the image is, the smaller the corresponding Euclidean distance is, the higher the matching degree with the target image is, and the closer the image is to the target image, the more similar the image is.
In the embodiment of the present invention, the obtaining of the feature vector of each image in the preset image set by using the convolutional neural network includes: training a convolutional neural network by using all images in a preset image set; and in the training process, acquiring a feature vector obtained after each image in a preset image set passes through a penultimate full-link layer of the convolutional neural network, and taking the feature vector as a feature vector corresponding to each image.
Therefore, the image coding is carried out by utilizing the full connection layer of the convolutional neural network, the relevant features of the image can be automatically extracted and coded, the complex feature extraction process can be avoided, the method is suitable for various image data, and manual features do not need to be manually extracted according to the image features.
Further, after training the convolutional neural network with all the images in the preset image set, the method further includes: obtaining a loss value of the convolutional neural network trained at this time; judging whether the loss value meets a stable condition; and if not, continuing to train the convolutional neural network according to the target image with the updated identifier, and returning to the operation of acquiring a plurality of similar images of the target image in the preset image set.
Therefore, based on the convolutional neural network, the image identification correction can be performed through loop iteration in the training process, so that the labor cost and the time cost are further reduced, and the working efficiency is improved.
Wherein the continuing to train the convolutional neural network according to the target image updated by the identifier comprises: judging whether each image in the preset image set is respectively used as a target image for identification updating (namely judging whether the target image traverses the preset image set); if yes, forming an image set with updated identification according to the target image with updated identification; and training the convolutional neural network again by using all the images in the image set after the identification is updated.
Therefore, the accuracy of the trained convolutional neural network (image classification model) can be improved.
Specifically, the determining whether the loss value meets the stability condition includes: and judging whether the loss value meets a stable condition or not according to the change rate of the loss value.
More specifically, the stabilizing conditions are:
among them, loss i Representing said loss value, loss k Represents a loss value that requires a summation operation, and->
Representing the rate of change of said loss value.
Here, k = i-2 · · i, i.e. the stability of the loss value is evaluated from the loss values of the previous two times and the loss value of the current time; of course, k = i-3 · i may also be used, that is, the stability of the loss value is evaluated according to the loss value of the previous three times and the loss value of the current time; k = i-4 · i, that is, stability of the loss value is judged according to the loss values of the previous four times and the loss value of the current time, or k = i-5 · i, and the like, which is not limited herein.
Further, after determining whether the loss value meets a stable condition, the method further includes: and if the loss value meets a stable condition, storing the parameter information of the convolutional neural network.
Further, after storing the parameter information of the convolutional neural network, the method further includes: and identifying the image to be identified by utilizing a convolutional neural network according to the parameter information.
The trained convolutional neural network (image classification model) is adopted for identification, and the identification accuracy can be improved.
In this embodiment of the present invention, the updating the identifier of the target image according to the identifier information of the multiple similar images includes: and according to the identification information of the plurality of similar images, utilizing a voting method to update the identification of the target image.
Therefore, the labor cost can be reduced, and the automatic correction of the image identification can be realized.
Specifically, the updating the identification of the target image by using a voting method according to the identification information of the plurality of similar images includes: acquiring the number of similar images with consistent identification information in a plurality of similar images; and according to the identification information of the similar images with the largest number, performing identification updating on the target image (updating the identification information of the target image into the acquired identification information).
The image identifier processing method provided by the embodiment of the present invention is further described below, wherein the preset image set takes an image set M including M images as an example.
To solve the above technical problem, an embodiment of the present invention provides an image identifier processing method, which may be specifically as shown in fig. 2:
the image classification training is carried out on a convolutional neural network by utilizing an image set M, the convolutional neural network comprises a convolutional layer, a pooling layer, a fully-connected layer (at least two layers) and the like, the feature vector of each image after passing through the second last fully-connected layer is saved as the feature vector of the image in the training process (namely class binary Hash coding, in the invention, the feature vector is a continuous value and does not need to be processed into values of 0 and 1), and the loss value (loss value) when the training stops (after passing through the last fully-connected layer and stopping) is recorded and recorded as the loss value i Where i represents the ith iteration.
Then, any one image A is selected from the image set M, euclidean distances between the feature vector of the A and feature vectors of other images left in the image set M are calculated, the Euclidean distances are sorted in an ascending order, and other images corresponding to the first n Euclidean distances are obtained. Wherein n is not less than 3 and not more than m, and n =5 is taken as an example for explanation: the 5 images B1, \ 8230;, B5 corresponding to the euclidean distances of the top 5 rows were acquired as the 5 images most similar to a. Then, the identification information of the image A is corrected and updated by voting according to the identification information of the 5 images B1, \8230;, B5 (which identification information is updated according to the number of the identification information and which identification information is used more). According to the method, A is taken until A traverses the image set M (namely each image in M images is taken as an image A to carry out the operation of updating the identification information), and the image set M after updating the label is obtained.
Then, based on the updated and labeled image set M, the image classification training is carried out on the convolutional neural network, and the process is continuously repeated until the loss value of the convolutional neural network is stable, for example, the loss value can be made to be stable i Satisfy the requirement of
When this happens, the iterative update (of the image identification and of the convolutional neural network) ends. And if the loss value of the convolutional neural network is unstable, continuously updating the identification information of the image and the parameter information of the convolutional neural network. Therein, loss k Representing the penalty value that needs to be summed. In the embodiment of the invention, whether the loss value is stable or not is actually determined according to the change rate of the loss value; wherein k may also be k = i-3 · i; or k = i-4 · i, etc.
After the iterative update is finished, the image set M after multiple iterative corrections can be obtained, and the accuracy of the image classification model (convolutional neural network) trained in the process is improved. And when the convolutional neural network is used for updating the marking information, the parameter information of the convolutional neural network is correspondingly updated.
In this way, in the embodiment of the present invention, each training has M images, and a loss value is obtained through the convolutional neural network. The M images are determined according to a training purpose, for example, if cat or dog is classified, the images are images of cat or dog, but not limited thereto.
As can be seen from the above, the embodiment of the present invention specifically provides an image classification and image identification information processing method based on a convolutional neural network, which mainly involves: (1) Utilizing a convolutional neural network to correct the image label by adopting a voting method; and (2) a process of iteratively correcting the image identification information. And (3) correcting the image identification information, and obtaining a trained convolutional neural network (image classification model).
Specifically, the scheme provided by the embodiment of the invention is as follows:
(1) The image classification method based on the convolutional neural network can correct the image identification information through loop iteration in the training process, so that the labor cost and the time cost can be reduced, and the working efficiency can be improved;
(2) The full-connection layer of the convolutional neural network is utilized to encode the image, the process can automatically extract the relevant features of the image and encode the image, the complex feature extraction process can be avoided, and the method is suitable for various image data and does not need to manually extract manual features according to the image features;
(3) Meanwhile, the accuracy of the trained image classification model can be improved.
In conclusion, the scheme provided by the embodiment of the invention can reduce the labor cost and realize the automatic correction of the image identification information; the method has no complicated manual feature extraction process, is suitable for various image data, and does not need to manually extract manual features according to the image features; meanwhile, an image classification method based on the convolutional neural network is formed, and the accuracy of image classification can be improved.
An embodiment of the present invention further provides an image identifier processing apparatus, as shown in fig. 3, including:
the first obtaining module 31 is configured to obtain multiple similar images of a target image in a preset image set in the preset image set;
the first processing module 32 is configured to perform identifier update on the target image according to the identifier information of the multiple similar images;
the similar images are images with the matching degree with the target image reaching a preset threshold, and the number of the similar images is at least three.
The image identification processing device provided by the embodiment of the invention acquires a plurality of similar images of a target image in a preset image set in the preset image set; according to the identification information of the similar images, the identification of the target image is updated; the similar images are images with the matching degree with the target image reaching a preset threshold value, and the number of the similar images is at least three; the automatic correction of the image identification can be realized, manual characteristics do not need to be manually extracted according to the image characteristics, so that the labor cost and the time cost are reduced, the working efficiency is improved, and the method is suitable for various image data; the problems that image identification processing schemes in the prior art all need manual participation, and are high in cost, low in efficiency and not universal are well solved.
Wherein, the first obtaining module comprises: the first acquisition sub-module is used for acquiring a feature vector of each image in a preset image set by using a convolutional neural network; the second obtaining submodule is used for obtaining Euclidean distances between the target image and other images according to the characteristic vectors of the target image in the preset image set and the characteristic vectors of other images except the target image; and the third acquisition sub-module is used for acquiring a plurality of similar images of the target image from other images according to the Euclidean distance.
In an embodiment of the present invention, the first obtaining sub-module includes: the first processing unit is used for training a convolutional neural network by utilizing all images in a preset image set; and the second processing unit is used for acquiring a feature vector obtained by each image in the preset image set after passing through a penultimate full-link layer of the convolutional neural network in the training process, and taking the feature vector as a feature vector corresponding to each image.
Further, the image identification processing apparatus further includes: the second acquisition module is used for acquiring a loss value of the convolutional neural network trained at this time after the convolutional neural network is trained by using all images in a preset image set; the first judgment module is used for judging whether the loss value meets a stable condition or not; and if the target images do not meet the preset image set, continuing to train the convolutional neural network according to the target images after the identification update, and returning to the operation of acquiring a plurality of similar images of the target images in the preset image set.
Wherein the second processing module comprises: the first judgment submodule is used for judging whether each image in the preset image set is respectively used as a target image to be subjected to identification updating; the first processing submodule is used for forming an image set after the identification is updated according to the target image after the identification is updated if the target image is the target image; and the first updating submodule is used for retraining the convolutional neural network by utilizing all the images in the image set after the identification is updated.
Specifically, the first determining module includes: and the second judgment submodule is used for judging whether the loss value meets a stable condition or not according to the change rate of the loss value.
More specifically, the stabilizing conditions are:
therein, loss i Represents the loss value, loss k Represents a loss value that requires a summation operation, and->
Representing the rate of change of said loss value.
Further, the image identification processing apparatus further includes: and the first storage module is used for storing the parameter information of the convolutional neural network if the loss value accords with the stable condition after judging whether the loss value accords with the stable condition.
Still further, the image identification processing apparatus further includes: and the first identification module is used for identifying the image to be identified by using the convolutional neural network according to the parameter information after the parameter information of the convolutional neural network is stored.
In an embodiment of the present invention, the first processing module includes: and the second processing submodule is used for carrying out identification updating on the target image by using a voting method according to the identification information of the similar images.
The implementation embodiments of the image identifier processing method are all applicable to the embodiment of the image identifier processing device, and the same technical effects can be achieved.
The embodiment of the invention also provides image identification processing equipment, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor; the processor implements the image identification processing method as described above when executing the program.
The implementation embodiments of the image identifier processing method are all applicable to the embodiment of the image identifier processing device, and the same technical effects can be achieved.
Embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps in the image identification processing method.
The implementation embodiments of the image identifier processing method are all applicable to the embodiment of the computer-readable storage medium, and the same technical effects can be achieved.
It should be noted that many of the functional components described in this specification are referred to as modules/sub-modules in order to more particularly emphasize their implementation independence.
In embodiments of the invention, the modules/sub-modules may be implemented in software for execution by various types of processors. An identified module of executable code may, for instance, comprise one or more physical or logical blocks of computer instructions which may, for instance, be constructed as an object, procedure, or function. Nevertheless, the executables of an identified module need not be physically located together, but may comprise disparate instructions stored in different bits which, when joined logically together, comprise the module and achieve the stated purpose for the module.
Indeed, a module of executable code may be a single instruction, or many instructions, and may even be distributed over several different code segments, among different programs, and across several memory devices. Likewise, operational data may be identified within the modules and may be embodied in any suitable form and organized within any suitable type of data structure. The operational data may be collected as a single data set, or may be distributed over different locations including over different storage devices, and may exist, at least partially, merely as electronic signals on a system or network.
When a module can be implemented by software, considering the level of existing hardware technology, a module implemented by software may build a corresponding hardware circuit to implement a corresponding function, without considering cost, and the hardware circuit may include a conventional Very Large Scale Integration (VLSI) circuit or a gate array and an existing semiconductor such as a logic chip, a transistor, or other discrete components. A module may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices or the like.
While the foregoing is directed to the preferred embodiment of the present invention, it will be appreciated by those skilled in the art that various changes and modifications may be made therein without departing from the principles of the invention as set forth in the appended claims.
Claims (18)
1. An image identifier processing method, comprising:
acquiring a plurality of similar images of a target image in a preset image set in the preset image set;
according to the identification information of the similar images, the identification of the target image is updated;
the similar images are images with the matching degree with the target image reaching a preset threshold value, and the number of the similar images is at least three;
the image identification processing method further comprises the following steps:
obtaining a loss value of the training convolutional neural network;
judging whether the loss value meets a stable condition or not;
if not, continuing to train the convolutional neural network according to the target image with the updated identifier, and returning to the operation of acquiring a plurality of similar images of the target image in the preset image set;
wherein the stable conditions are as follows:
therein, loss i Representing said loss value, loss k Representing the penalty value that would require a summation operation,
representing the rate of change of said loss value.
2. The image identifier processing method according to claim 1, wherein the acquiring multiple similar images of the target image in the preset image set comprises:
acquiring a feature vector of each image in a preset image set by using a convolutional neural network;
acquiring Euclidean distances between the target image and other images according to the characteristic vectors of the target image in the preset image set and the characteristic vectors of other images except the target image;
and acquiring a plurality of similar images of the target image from other images according to the Euclidean distance.
3. The method according to claim 2, wherein the obtaining the feature vector of each image in the preset image set by using the convolutional neural network comprises:
training a convolutional neural network by using all images in a preset image set;
and in the training process, acquiring a feature vector obtained after each image in a preset image set passes through a penultimate full-link layer of the convolutional neural network, and taking the feature vector as a feature vector corresponding to each image.
4. The image tag processing method of claim 1, wherein the continuing training of the convolutional neural network according to the target image after tag update comprises:
judging whether each image in the preset image set is respectively used as a target image to be subjected to identification updating;
if yes, forming an image set with an updated identifier according to the target image with the updated identifier;
and training the convolutional neural network again by using all the images in the image set after the identification is updated.
5. The method according to claim 1, wherein the determining whether the loss value meets a stability condition includes:
and judging whether the loss value meets a stable condition or not according to the change rate of the loss value.
6. The image identifier processing method according to claim 1, further comprising, after determining whether the loss value satisfies a stable condition:
and if the loss value meets a stable condition, storing the parameter information of the convolutional neural network.
7. The image identifier processing method according to claim 6, further comprising, after storing the parameter information of the convolutional neural network:
and identifying the image to be identified by utilizing a convolutional neural network according to the parameter information.
8. The image identifier processing method according to claim 1, wherein the identifier updating of the target image according to the identifier information of the plurality of similar images includes:
and according to the identification information of the plurality of similar images, utilizing a voting method to update the identification of the target image.
9. An image tag processing apparatus, comprising:
the first acquisition module is used for acquiring a plurality of similar images of a target image in a preset image set in the preset image set;
the first processing module is used for carrying out identification updating on the target image according to the identification information of the similar images;
the similar images are images with the matching degree with the target image reaching a preset threshold value, and the number of the similar images is at least three;
the image identification processing device further comprises:
the second acquisition module is used for acquiring the loss value of the training convolutional neural network;
the first judgment module is used for judging whether the loss value meets a stable condition or not;
the second processing module is used for continuing to train the convolutional neural network according to the target image after the identification update if the target image does not conform to the identification update, and returning to the operation of acquiring a plurality of similar images of the target image in the preset image set;
wherein the stable conditions are as follows:
therein, loss i Representing said loss value, loss k Representing the loss value of the required summation operation,
representing the rate of change of said loss value.
10. The image tag processing apparatus according to claim 9, wherein the first obtaining module comprises:
the first obtaining submodule is used for obtaining a characteristic vector of each image in a preset image set by using a convolutional neural network;
the second obtaining submodule is used for obtaining Euclidean distances between the target image and other images according to the characteristic vectors of the target image in the preset image set and the characteristic vectors of other images except the target image;
and the third acquisition submodule is used for acquiring a plurality of similar images of the target image from other images according to the Euclidean distance.
11. The image identifier processing apparatus according to claim 10, wherein the first acquisition sub-module includes:
the first processing unit is used for training a convolutional neural network by utilizing all images in a preset image set;
and the second processing unit is used for acquiring a feature vector obtained by each image in the preset image set after passing through a penultimate full-link layer of the convolutional neural network in the training process, and taking the feature vector as a feature vector corresponding to each image.
12. The image tag processing apparatus according to claim 9, wherein the second processing module comprises:
the first judgment submodule is used for judging whether each image in the preset image set is respectively used as a target image to be subjected to identification updating;
the first processing submodule is used for forming an image set after the identification is updated according to the target image after the identification is updated if the target image is the target image;
and the first updating submodule is used for retraining the convolutional neural network again by utilizing all the images in the image set after the identification is updated.
13. The image tag processing apparatus according to claim 9, wherein the first determining means comprises:
and the second judgment submodule is used for judging whether the loss value meets a stable condition or not according to the change rate of the loss value.
14. The image tag processing apparatus according to claim 9, further comprising:
and the first storage module is used for storing the parameter information of the convolutional neural network if the loss value meets the stable condition after judging whether the loss value meets the stable condition.
15. The image tag processing apparatus according to claim 14, further comprising:
and the first identification module is used for identifying the image to be identified by using the convolutional neural network according to the parameter information after the parameter information of the convolutional neural network is stored.
16. The image tag processing apparatus according to claim 9, wherein the first processing module comprises:
and the second processing submodule is used for carrying out identification updating on the target image by using a voting method according to the identification information of the similar images.
17. An image identification processing apparatus comprising a memory, a processor and a computer program stored on said memory and executable on said processor; characterized in that the processor implements the image identification processing method according to any one of claims 1 to 8 when executing the program.
18. A computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, carries out the steps of the image identification processing method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910073615.XA CN111488891B (en) | 2019-01-25 | 2019-01-25 | Image identification processing method, device, equipment and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910073615.XA CN111488891B (en) | 2019-01-25 | 2019-01-25 | Image identification processing method, device, equipment and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111488891A CN111488891A (en) | 2020-08-04 |
| CN111488891B true CN111488891B (en) | 2023-04-07 |
Family
ID=71810686
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910073615.XA Active CN111488891B (en) | 2019-01-25 | 2019-01-25 | Image identification processing method, device, equipment and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111488891B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112307238A (en) * | 2020-10-30 | 2021-02-02 | 维沃移动通信有限公司 | Image management method and device and electronic equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101395614A (en) * | 2006-03-03 | 2009-03-25 | 皇家飞利浦电子股份有限公司 | Identifying a set of image characteristics for assessing similarity of images |
| CN104915643A (en) * | 2015-05-26 | 2015-09-16 | 中山大学 | Deep-learning-based pedestrian re-identification method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9349152B2 (en) * | 2012-10-02 | 2016-05-24 | Simetricka Systems Ltd. | Image identifiers and methods and systems of presenting image identifiers |
-
2019
- 2019-01-25 CN CN201910073615.XA patent/CN111488891B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101395614A (en) * | 2006-03-03 | 2009-03-25 | 皇家飞利浦电子股份有限公司 | Identifying a set of image characteristics for assessing similarity of images |
| CN104915643A (en) * | 2015-05-26 | 2015-09-16 | 中山大学 | Deep-learning-based pedestrian re-identification method |
Non-Patent Citations (2)
| Title |
|---|
| 基于卷积神经网络的图像分类技术研究与实现;王超;《电脑知识与技术》(第35期);全文 * |
| 基于索引和相关反馈的图像检索研究;李迎新等;《现代计算机》(第02期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111488891A (en) | 2020-08-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109255364B (en) | Scene recognition method for generating countermeasure network based on deep convolution | |
| CN109859171B (en) | Automatic floor defect detection method based on computer vision and deep learning | |
| CN106971152B (en) | Method for detecting bird nest in power transmission line based on aerial images | |
| JP6935377B2 (en) | Systems and methods for automatic inference of changes in spatiotemporal images | |
| CN110992365A (en) | Loss function based on image semantic segmentation and design method thereof | |
| CN110705516B (en) | Sole pattern image clustering method based on collaborative network structure | |
| CN111931581A (en) | Agricultural pest identification method based on convolutional neural network, terminal and readable storage medium | |
| CN111898613B (en) | Semi-supervised semantic segmentation model training method, recognition method and device | |
| CN110738166B (en) | Fishing administration monitoring system infrared target identification method based on PCNN and PCANet and storage medium | |
| CN111488891B (en) | Image identification processing method, device, equipment and computer readable storage medium | |
| CN112883795A (en) | Rapid and automatic table extraction method based on deep neural network | |
| CN116071560A (en) | Fruit identification method based on convolutional neural network | |
| CN116612435B (en) | Corn high-yield cultivation method | |
| CN112396042A (en) | Real-time updated target detection method and system, and computer-readable storage medium | |
| CN116307266A (en) | Crop growth prediction method, device, electronic equipment and storage medium | |
| CN116206208A (en) | Forestry plant diseases and insect pests rapid analysis system based on artificial intelligence | |
| CN115690570A (en) | Fish shoal feeding intensity prediction method based on ST-GCN | |
| CN113192108B (en) | Man-in-loop training method and related device for vision tracking model | |
| CN115497021A (en) | Method for identifying fine granularity of sow lactation behavior based on computer vision | |
| CN114663791A (en) | Branch recognition method for pruning robot in unstructured environment | |
| CN111126513B (en) | Universal object real-time learning and recognition system and learning and recognition method thereof | |
| CN114743170A (en) | Automatic driving scene labeling method based on AI algorithm | |
| CN113850166A (en) | Ship image identification method and system based on convolutional neural network | |
| CN116883275B (en) | Image rain removing method, system, device and medium based on boundary guidance | |
| CN113658148B (en) | Central flower identification and positioning method and system based on regional attribute extraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |