CN110766048B - Image content recognition method, image content recognition device, computer equipment and storage medium - Google Patents
Image content recognition method, image content recognition device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN110766048B CN110766048B CN201910882184.1A CN201910882184A CN110766048B CN 110766048 B CN110766048 B CN 110766048B CN 201910882184 A CN201910882184 A CN 201910882184A CN 110766048 B CN110766048 B CN 110766048B
- Authority
- CN
- China
- Prior art keywords
- image
- preset
- coding model
- loss rate
- self
- 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
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 71
- 238000000605 extraction Methods 0.000 claims abstract description 39
- 238000012545 processing Methods 0.000 claims abstract description 20
- 238000004422 calculation algorithm Methods 0.000 claims description 74
- 238000011176 pooling Methods 0.000 claims description 44
- 230000006870 function Effects 0.000 claims description 33
- 238000004590 computer program Methods 0.000 claims description 25
- 238000012216 screening Methods 0.000 claims description 17
- 238000005096 rolling process Methods 0.000 claims description 10
- 230000010355 oscillation Effects 0.000 claims description 7
- 238000003709 image segmentation Methods 0.000 claims description 3
- 230000009514 concussion Effects 0.000 claims 1
- 238000007906 compression Methods 0.000 abstract description 4
- 238000001514 detection method Methods 0.000 abstract description 2
- 230000001537 neural effect Effects 0.000 abstract 1
- 238000013527 convolutional neural network Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000002441 reversible effect Effects 0.000 description 4
- 210000005036 nerve Anatomy 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The application relates to the technical field of image detection, and provides an image content identification method and device, computer equipment and a storage medium. The method comprises the following steps: based on a convolutional self-coding model obtained by training a historical image and an image loss rate, performing feature extraction processing on a to-be-identified image from the perspective of a neural layer to obtain a compressed image containing features of a preset dimension; the model is obtained based on image loss rate training; therefore, the compressed image output by the model is representative; the compressed image can be considered to represent an original image; and then decoding and reconstructing the compressed image to obtain a decoded image, and the model is trained based on a preset image loss rate, so that the loss rate of the output decoded image and the original image is low, the loss rate of the image in the image compression process can be reduced, and the accuracy of subsequent image recognition is improved.
Description
Technical Field
The present application relates to the field of image detection technologies, and in particular, to an image content identification method, an image content identification device, a computer device, and a storage medium.
Background
Image recognition technology is an important area of artificial intelligence. It refers to a technique of performing object recognition on an image to recognize objects and objects of various modes. At present, the image recognition technology is mature and widely applied to aspects of faces, animals, other objects and the like.
The scheme of image recognition often relates to a feature extraction scheme, but after an original image is compressed and feature extracted by using a traditional image recognition technology, the loss rate of a compressed and expanded picture and the original picture is higher, so that the accuracy rate of subsequent image recognition is lower and the efficiency is poorer.
Disclosure of Invention
Based on this, it is necessary to provide an efficient and accurate image content recognition method, apparatus, computer device and storage medium, aiming at the problem that the accuracy and efficiency of image recognition by the conventional image recognition technology are low.
An image content recognition method, the method comprising:
Acquiring an image to be identified;
Extracting the characteristics of the preset dimension of the image to be identified according to the trained convolution self-coding model to obtain a compressed image, wherein the trained convolution self-coding model is obtained by training based on the historical image and the preset image loss rate;
Inputting the compressed image into a trained convolution self-coding model for reconstruction to obtain a reconstructed decompressed image;
And performing similarity matching on the decompressed image and the historical images in the preset gallery according to a preset similarity matching algorithm, and screening out images similar to the decompressed image to obtain an image content identification result corresponding to the image to be identified.
In one embodiment, according to a preset similarity matching algorithm, performing similarity matching on the decompressed image and a history image in a preset gallery, and screening out an image similar to the decompressed image includes:
calculating the cosine value of an included angle between the decompressed image and each historical image in a preset gallery according to a cosine similarity matching algorithm;
Converting the cosine value of the included angle into an angle similarity coefficient, and screening out images similar to the decompressed images from a preset gallery according to the angle similarity coefficient.
In one embodiment, the feature extraction is performed on the image to be identified according to the coding layer of the preset convolution self-coding model, and before the image features of the preset dimension are extracted, the method further includes:
collecting an original image and constructing an image training set;
Establishing a convolution self-coding model of interleaving of a convolution layer and a pooling layer, wherein the convolution self-coding model is provided with an initial feature extraction number;
inputting an original image in an image training set into a convolution self-coding model for coding treatment, and extracting image features with preset dimensions;
decoding the image features with preset dimensions to obtain a reconstructed original image;
calculating and optimizing the loss rate of each reconstructed original image and each reconstructed original image to obtain the overall image loss rate of the image training set;
and when the integral image loss rate is larger than the preset image loss rate, adjusting the feature extraction number to obtain an updated integral image loss rate, and determining the feature extraction number until the integral image loss rate is smaller than or equal to the preset image loss rate to obtain the trained convolution self-coding model.
In one embodiment, calculating, and optimizing, the overall image loss rate of the training set of images includes:
Determining a loss function of the convolution self-coding model by combining Adadelta algorithm and random gradient descent algorithm;
and calculating and optimizing the overall image loss rate of the image training set based on the determined loss function.
In one embodiment, determining the loss function of the convolutional self-encoding model in combination with Adadelta algorithm and random gradient descent algorithm comprises:
according to Adadelta algorithm, adjusting the learning rate of the convolution self-coding model;
when the oscillation caused by the overlarge learning rate is detected, the optimal solution of the loss function is obtained by adopting a random gradient descent algorithm, and the loss function of the convolution self-coding model is determined.
In one embodiment, extracting features of a preset dimension of an image to be identified from a trained convolutional self-coding model includes:
rolling and pooling the image to be identified according to the coding layer of the trained convolution self-coding model, and extracting the characteristics of the preset dimension of the image to be identified;
inputting the compressed image into a decoding layer of a preset convolution self-coding model for reconstruction, and obtaining a reconstructed decompressed image comprises the following steps:
inputting the compressed image into a decoding layer of a preset convolution self-coding model, and carrying out deconvolution and reverse pooling on the compressed image according to the decoding layer of the preset convolution self-coding model to obtain a reconstructed decompressed image.
An image content recognition apparatus, the apparatus comprising:
the image acquisition module is used for acquiring an image to be identified;
The image coding module is used for extracting the characteristics of the preset dimension of the image to be identified according to the trained convolution self-coding model to obtain a compressed image, and the trained convolution self-coding model is obtained by training based on the historical image and the preset image loss rate;
the image decoding module inputs the compressed image to a decoding layer of a preset convolution self-coding model for reconstruction to obtain a reconstructed decompressed image;
The image content recognition module is used for carrying out similarity matching on the decompressed image and the historical images in the preset gallery according to a preset similarity matching algorithm, screening out similar images of the decompressed image, wherein the historical images are images containing features of preset dimensions.
In one embodiment, the apparatus further comprises:
The model training module is used for collecting original images, constructing an image training set, establishing a convolution self-coding model with staggered convolution layers and pooling layers, wherein the convolution self-coding model is provided with an initial feature extraction number, the original images in the image training set are input into the convolution self-coding model for coding, image features with preset dimensions are extracted, decoding is carried out on the image features with preset dimensions to obtain reconstructed original images, the loss rate of each reconstructed original image and each original image is calculated and optimized to obtain the integral image loss rate of the image training set, when the integral image loss rate is larger than the preset image loss rate, the feature extraction number is adjusted to obtain an updated integral image loss rate, and when the integral image loss rate is smaller than or equal to the preset image loss rate, the feature extraction number is determined to obtain the trained convolution self-coding model.
A computer device comprising a memory storing a computer program and a processor which when executing the computer program performs the steps of:
Acquiring an image to be identified;
Extracting the characteristics of the preset dimension of the image to be identified according to the trained convolution self-coding model to obtain a compressed image, wherein the trained convolution self-coding model is obtained by training based on the historical image and the preset image loss rate;
Inputting the compressed image into a trained convolution self-coding model for reconstruction to obtain a reconstructed decompressed image;
And performing similarity matching on the decompressed image and the historical images in the preset gallery according to a preset similarity matching algorithm, and screening out images similar to the decompressed image to obtain an image content identification result corresponding to the image to be identified.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
Acquiring an image to be identified;
Extracting the characteristics of the preset dimension of the image to be identified according to the trained convolution self-coding model to obtain a compressed image, wherein the trained convolution self-coding model is obtained by training based on the historical image and the preset image loss rate;
Inputting the compressed image into a trained convolution self-coding model for reconstruction to obtain a reconstructed decompressed image;
And performing similarity matching on the decompressed image and the historical images in the preset gallery according to a preset similarity matching algorithm, and screening out images similar to the decompressed image to obtain an image content identification result corresponding to the image to be identified.
According to the image content identification method, the device, the computer equipment and the storage medium, the convolutional self-coding model is obtained based on the historical image and the image loss rate training, the feature extraction processing is carried out on the image to be identified from the nerve layer angle, the compressed image containing the features of the preset dimension is obtained, the model is obtained based on the image loss rate training, so that the compressed image output by the model is representative and can be considered to represent an original image, then the decoding reconstruction processing is carried out on the compressed image, the decoded image is obtained, the model is based on the preset image loss rate training, the loss rate of the output decoded image and the original image is lower, the image loss rate in the image compression process can be reduced, and the accuracy rate of the subsequent image identification is improved.
Drawings
FIG. 1 is an application environment diagram of an image content recognition method in one embodiment;
FIG. 2 is a flow chart of a method for identifying image content in one embodiment;
FIG. 3 is a detailed flowchart of an image content recognition method according to another embodiment;
FIG. 4 is a flow diagram of a convolutional self-coding model training process in one embodiment;
FIG. 5 is a block diagram showing the structure of an image content recognition apparatus according to another embodiment;
FIG. 6 is a detailed block diagram of an image content recognition apparatus according to another embodiment;
Fig. 7 is an internal structural diagram of a computer device in one embodiment.
Detailed Description
The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
The image content identification method provided by the application can be applied to an application environment shown in fig. 1. The method comprises the steps that a user inputs an image to be identified into a terminal 102, the terminal uploads the image to be identified to a server 104 in real time, then the user sends an image identification instruction to the server 104 through the terminal 102, when the server 104 receives the image identification instruction, the image to be identified is obtained, feature extraction is carried out on the image to be identified according to a coding layer of a trained convolution self-coding model, the feature of a preset dimension of the image to be identified is extracted, a compressed image is obtained, then the compressed image is input into a decoding layer of the convolution self-coding model for reconstruction, a reconstructed decompressed image is obtained, finally, similarity matching is carried out on the decompressed image and a historical image in a preset gallery according to a preset similarity matching algorithm, and an image similar to the decompressed image is screened out, so that an image content identification result corresponding to the image to be identified is obtained. The trained convolution self-coding model is obtained through training based on historical images and a preset image loss rate. The terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices, and the server 104 may be implemented as a stand-alone server or a server cluster composed of a plurality of servers.
In one embodiment, as shown in fig. 2, there is provided an image content recognition method, which is described by taking the application of the method to the server in fig. 1 as an example, and includes the following steps:
step S200, an image to be identified is acquired.
Image recognition is an important area of artificial intelligence. The development of image recognition has gone through three phases: character recognition, digital image processing and recognition, and object recognition. Image recognition, as the name implies, is to perform various processing and analysis on the image, and finally identify the required target to be studied. After receiving an image recognition instruction sent by a user through a terminal, an image to be recognized uploaded by the terminal is obtained from a database, and in this embodiment, the image to be recognized may be a picture with a pixel size of 640×480×3, and is an image that is segmented based on an extreme point. Specifically, the image segmentation may be that gray value differential computation is performed on a pixel point row and a pixel point column of an original image, a target area is identified according to a differential computation result and a preset threshold value, extreme points of the original image are marked according to gray values of abrupt points of the target area, and when a certain extreme point simultaneously meets the extreme points of which the number is higher than the preset number on the left side and the extreme points of which the number is higher than the preset number on the right side, the extreme points are determined as first segmentation points; when a certain extreme point simultaneously meets the extreme points lower than the preset number of the left extreme point and the preset number of the right extreme point, dividing the K line image to be divided into extreme points of a plurality of sub-images according to the determined dividing points and a preset dividing algorithm, determining the extreme point as a second dividing point, wherein the first dividing point refers to the maximum dividing point, the second dividing point refers to the minimum dividing point, and dividing the original image into a plurality of sub-images according to the determined dividing points and the preset dividing algorithm to obtain the image to be recognized. In order to facilitate the subsequent image recognition, after the image to be recognized is obtained, the image to be recognized is further subjected to image preprocessing, wherein the image preprocessing mainly refers to denoising, smoothing, transformation and other operations in the image processing, so that important characteristics of the image are enhanced.
Step S400, extracting the features of the preset dimension of the image to be identified according to the trained convolution self-coding model to obtain a compressed image, wherein the trained convolution self-coding model is trained based on the historical image and the preset image loss rate.
The convolutional self-coding model is a model of a convolutional neural network, which is constructed based on a CNN (Convolutional Neural Network, convolutional neural network algorithm framework and AutoEncoder) method in deep learning, and comprises a coding layer and a decoding layer, and concretely comprises two parts of an encoder (used for compressing an input into a potential spatial representation) and a decoder (used for reconstructing the input from a hidden spatial representation), wherein the convolutional self-coding model is a neural network which adopts the convolutional layer to replace a full-connection layer, the principle is the same as the self-encoder, the input symbols are downsampled to provide a potential representation with smaller dimension, the CNN (Convolutional Neural Network, the convolutional neural network) is a feedforward neural network, an artificial neuron can respond to surrounding units, and large-scale image processing can be performed, the convolutional neural network comprises the convolutional layer and a pooling layer, autoEncoder is self-coding, is another important content of deep learning, the input and the output are more and more similar through a design of an end code and a decoding process, the convolutional self-coding model is an unsupervised learning process, the characteristics of the convolutional image can be well recognized by extracting the convolutional model from the input images to a training image, the training image is well-recognized by extracting the convolutional model with a preset degree, and the characteristic of the convolutional image is well-recognized by the training layer, and the characteristic is well recognized by the convolutional model, the image is well-recognized by the training layer, and the image is well-recognized by the training image, and the image is well-recognized by the training layer.
In one embodiment, as shown in FIG. 3, extracting features of a preset dimension of an image to be identified from a trained convolutional self-coding model includes: step S420, rolling and pooling the image to be identified according to the trained convolution self-coding model, and extracting the feature of the preset dimension of the image to be identified.
The coding layer is composed of a convolution layer and a pooling layer, the decoding layer is composed of a deconvolution layer and a pooling layer, and specifically, the convolution layer carries out local perception on the characteristics of an input original picture and then carries out comprehensive operation on the local at a higher level, so that global information of the original picture is obtained. The pooling layer is mainly used for carrying out feature dimension reduction on the picture, compressing the number of data and parameters and reducing the overfitting. For example, by pooling the layers, the picture originally comprising 200-dimensional features can be compressed into a picture comprising 60-dimensional features, and the feature dimension of the picture can be reduced. And inputting the image to be identified into a coding layer of the trained convolution self-coding model, and rolling and pooling the image to be identified, so that the 150-dimensional characteristics of the preset dimension of the image to be identified can be extracted. By rolling and pooling the images to be identified, the fault tolerance of the model can be improved.
And S600, inputting the compressed image into a preset convolution self-coding model for reconstruction, and obtaining a reconstructed decompressed image.
Specifically, after the convolution self-coding model finishes coding processing on an input image to be identified, obtaining a compressed image which can represent the image and contains 150-dimensional characteristics, inputting the compressed image into a decoding layer of the model, performing a series of decoding (i.e. deconvolution and deconvolution processing) on the compressed image through the decoding layer (staggered deconvolution layer and deconvolution layer) of the model, and reconstructing the compressed image, so that the size of the compressed image is restored to 640 x 480 x 3, and the decompressed image after being decoded and unfolded by the coded image to be identified is obtained, and the loss rate of the decompressed image and the coded image to be identified is less than 10%.
As shown in fig. 3, in one embodiment, inputting a compressed image to a decoding layer of a preset convolutional self-coding model for reconstruction, and obtaining a reconstructed decompressed image includes: and step S620, performing deconvolution and anti-pooling on the compressed image according to a preset convolution self-coding model to obtain a reconstructed decompressed image.
Deconvolution, also called transpose convolution, is not the complete inverse of forward convolution, and deconvolution is a special forward convolution, in which the size of an input image is first enlarged by supplementing 0 according to a certain proportion, then the convolution kernel is rotated, and then forward convolution is performed. Inverse pooling is the inverse of pooling, i.e., restoring a portion of the original data. The compressed image is input to a decoding layer of the convolution self-coding model, and is subjected to deconvolution and reverse pooling, so that the compressed image can be restored, a reconstructed decompressed image is obtained, and the loss rate of the image is reduced.
Step S800, according to a preset similarity matching algorithm, similarity matching is carried out on the decompressed image and the historical images in the preset gallery, images similar to the decompressed image are screened out, and an image content recognition result corresponding to the image to be recognized is obtained.
The preset similarity matching algorithm can be a distance algorithm, a cosine similarity algorithm and the like. After the reconstructed decompressed image is obtained, performing similarity matching on the decompressed image and all historical images in a preset image library through a similarity matching algorithm, namely calculating a similarity coefficient between two images, and screening a plurality of images similar to the decompressed image according to the similarity coefficient to obtain an image content identification result corresponding to the image to be identified. In this embodiment, the history image in the preset gallery is a picture including 150-dimensional image features after feature extraction processing of the convolutional self-coding model.
According to the image content identification method, the convolutional self-coding model obtained through training of the historical image and the image loss rate is based, the image to be identified is subjected to feature extraction processing from the perspective of the nerve layer, the compressed image containing the features of the preset dimension is obtained, and the model is obtained through training based on the image loss rate, so that the compressed image output by the model is representative and can be considered to represent an original image, then the compressed image is subjected to decoding reconstruction processing, a decoded image is obtained, and the model is trained based on the preset image loss rate, so that the loss rate of the output decoded image and the original image is lower, the loss rate of the image in the image compression process can be reduced, and the accuracy of the subsequent image identification is improved.
As shown in fig. 3, in one embodiment, according to a preset similarity matching algorithm, performing similarity matching on a decompressed image and a history image in a preset gallery, and screening out an image similar to the decompressed image includes: step S820, calculating the cosine value of the included angle of each historical image in the decompressed image and the preset gallery according to the cosine similarity matching algorithm, converting the cosine value of the included angle into an angle similarity coefficient, and screening the images similar to the decompressed image from the preset gallery according to the angle similarity coefficient.
In this embodiment, the similarity matching algorithm is a cosine similarity matching algorithm. Specifically, as described in the above embodiments, the loss rate of the compressed image and the original image converges to 10%, so that the compressed image can be considered to contain a certain representative feature, and can be used for image matching. Because the compressed data is still high-dimensional data, the method is not suitable for directly measuring the similarity by adopting the distance, and therefore, the similarity can be measured by directly adopting the angular similarity coefficient. Specifically, the cosine similarity algorithm may be:
through the formula, the cosine values among the pictures are obtained through calculation, the similarity of included angles is calculated because the cosine values have no distance meaning, and the compressed data are all larger than 0, so that the following conversion is carried out on the result of the formula:
Converting the cosine value of the included angle into an angle similarity coefficient to form a strict distance measurement, and screening a plurality of similar images different from the decompressed image from a preset image library according to the angle similarity coefficient. In this embodiment, by converting the cosine value into the angle similarity coefficient, when the angle similarity coefficient is taken as a difference coefficient (subtracting it from 1), the generated function is a strict distance measure, and the accuracy of image matching can be improved.
In one embodiment, before performing similarity matching on the decompressed image and the historical image in the preset gallery according to a preset similarity matching algorithm, the method further includes: and carrying out RGB (Red, green, blue) value processing on the decompressed image and all the pixels of the images in the preset gallery, namely subtracting the original RGB value from each pixel by 255 to obtain an RGB value as a new RGB value.
In the experiment of actually calculating the similarity between the images, the developer finds that the similarity of all the calculated images is very high, and in order to ensure the accuracy of similarity calculation, the developer performs adjustment similar to 'anti-color processing' on all the images, namely, the original RGB values of the pixel points in all the images are subtracted by 255, and the difference value obtained by subtraction is used as a new RGB value of the pixel points. Through the operation, RGB values of a large number of white pixels contained in the image are changed from 255 to 0, so that the influence of a large number of white color blocks on the calculation of the angular similarity coefficients between the images is directly eliminated, and the accuracy of image matching is improved.
As shown in fig. 4, in one embodiment, the feature extraction is performed on the image to be identified according to the coding layer of the preset convolution self-coding model, and before the image features of the preset dimension are extracted, the method further includes: step S100, collecting an original image and constructing an image training set; step S110, a convolutional neural network with staggered convolutional layers and pooling layers is established; step S120, inputting an original image in an image training set into a convolution self-coding model for coding processing, and extracting image features with preset dimensions; step S130, decoding the image features with preset dimensions to obtain a reconstructed original image; step S140, calculating and optimizing the loss rate of each reconstructed original image and each reconstructed original image to obtain the overall image loss rate of the image training set; and step S150, when the integral image loss rate is larger than the preset image loss rate, adjusting the feature extraction number to obtain an updated integral image loss rate, and when the integral image loss rate is smaller than or equal to the preset image loss rate, determining the feature extraction number to obtain a trained convolution self-coding model.
The construction and training process of the trained convolution self-coding model can be as follows: thousands of original images are collected, the pixel size of the collected images is uniformly set to 640 x 480 x 3, an image training set is constructed, a neural network comprising 10 layers of convolution layers and pooling layer interleaving (10 layers in total) is established as an encoding layer of an initial convolution self-encoding model, the 10 layers of deconvolution and anti-pooling interleaving (10 layers in total) neural network is established as a decoding layer of the initial convolution self-encoding model, and the model itself has initial feature extraction parameters. Compressing each original picture (640 x 480 x 3) to a certain dimension (also called feature extraction, regarding how many feature dimensions are most suitable to extract, obtaining by fitting and adjusting a model itself, decoding each compressed picture through a 10-layer deconvolution and anti-pooling staggered neural network, returning the compressed image to the original dimension 640 x 480 x 3, obtaining a decompressed image, calculating the loss rate of each original image input into the convolution self-coding model and the decompressed image output by convolution pooling, deconvolution and anti-pooling processing of the convolution self-coding model, taking the mean value of the loss rate as the integral loss rate of an image training set, determining the current feature extraction number as the feature extraction number of the initial convolution self-coding model when the integral loss rate of the image training set is smaller than or equal to the preset image loss rate, adjusting the initial feature extraction number according to preset step length when the integral image loss rate of the image training set is larger than the preset image loss rate, determining the convolution self-coding model, calculating the loss rate of the initial feature extraction algorithm is smaller than or equal to 150% when the convolution self-coding model is applied to the preset dimension of the actual loss rate, calculating the average value is smaller than 150, and calculating the actual loss rate of the image training model is smaller than the actual loss in the initial image is calculated as the initial loss rate is smaller than 150, the feature extraction number of the convolution self-coding model can be other values such as 151 dimension, 149 dimension and 140 dimension, and the effective loss rate threshold can be other values except 10%, such as 10.1%, according to practical situations. The model is trained according to the loss rate, so that the loss rate of the image can be reduced, and the quality of the decompressed image output by the model is ensured.
In one embodiment, calculating the overall image loss rate for the training set of images includes: and combining Adadelta algorithm and random gradient descent algorithm, determining a loss function of the convolution self-coding model, and calculating and optimizing the integral image loss rate of the image training set based on the determined loss function.
Since the deconvolution operation can only recover the size of the picture, but not recover each element value, and since only main information is reserved in the pooling process and part of information is removed, all original data cannot be recovered through the pooling result by the reverse pooling, and part of information is lost. Therefore, there is a certain loss rate in comparison with the picture after the deconvolution and the pooling processes and the original picture. Therefore, in this embodiment, the loss rate can be calculated and optimized by combining Adadelta algorithm and SGD (Stochastic GRADIENT DESCENT, random gradient descent method). The Adadelta algorithm is an extension algorithm of AdaGrad, which mainly solves the problem that the AdaGrad algorithm monotonically decreases the learning rate, and the Adadelta algorithm does not accumulate all previous square gradients, but limits the window of gradients before accumulation to a certain fixed size, and replaces accumulating all historical gradient squares by constraining historical gradient accumulation. Specifically, in the process of adaptively adjusting the learning rate in the early stage and the middle stage, the learning rate of the convolution self-coding model is adjusted according to Adadelta algorithm, when the occurrence of oscillation (namely, the fluctuation of the loss function is gradually reduced and the oscillation occurs) of the too large learning rate is detected, the method is switched to a random gradient descent algorithm to obtain the optimal solution of the loss function, the loss function of the convolution self-coding model is determined, and the integral image loss rate of the image training set is calculated and optimized based on the determined loss function. It will be appreciated that in other embodiments, the loss rate calculation optimization algorithm may also be a batch descent algorithm, an Adam algorithm, or the like. In this embodiment, the overall loss rate of the image training set is calculated and optimized by Adadelta algorithm and random gradient descent algorithm, without setting a default learning rate, and the problems of learning rate attenuation or gradient disappearance and the like can be prevented, and the problem that Adadelta itself may jump back and forth between locally optimal solutions in the later stage of gradient descent can be also optimized.
For clarity, the image content recognition method provided by the present application is described below with reference to a specific example: taking a K line graph as an example of an image to be identified, acquiring a K line graph (the K line graph is a segmented K line small graph) to be identified in a preset time period (such as a K line graph of 30 trading days), inputting the K line graph to be identified into a trained convolution self-coding model, extracting 150-dimension line characteristics of the K line graph to be identified by a coding layer of the model to obtain a compressed K line graph, inputting the compressed K line graph into a decoding layer of the trained convolution self-coding model for reconstruction (namely deconvolution and anti-pooling), obtaining a reconstructed decompressed K line graph (the image loss rate of the decompressed K line graph and an original image is less than 10%), and carrying out 'anti-color processing' on the decompressed K line graph and the image in a preset image library, namely subtracting the original RGB value from each pixel point in the image by 255 to obtain an RGB value as a new RGB value, thereby eliminating the influence of a large number of white blocks on image matching. And then, calculating an included angle cosine value of each historical K line graph (the time length of the historical K line graph is different and is an image after feature extraction is completed) in a decompressed K line graph and a preset K line graph library according to a cosine similarity matching algorithm, converting the included angle cosine value into an angle similarity coefficient, screening out an image similar to the decompressed K line graph from the preset line graph library according to the angle similarity coefficient, and obtaining a corresponding K line graph content identification result, and further comprising a prediction result of a K line future state in a preset time period according to the screened similar K line graph and preset similar historical data information.
According to the scheme, a plurality of historical K line pictures which are close in morphology and different from the K line pictures in the preset time period in time length can be screened out from the historical K line picture library according to the K line pictures in the preset time period, the output pictures do not limit the date length, the traditional K line picture screening is free from the limitation of fixed days, for example, the time window of the K line picture is 30 days, and the similar K line picture which is also found by utilizing the traditional similar K line matching is also 30 days. In addition, the scheme utilizes the convolution self-coding model to identify the K line shape of the K line graph from the nerve layer, can screen the K line graph which is closest to the K line graph in the form of the preset time period, and ensures that the loss rate of the picture is less than 10 percent.
It should be understood that, although the steps in the flowcharts of fig. 2-4 are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 2-4 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily occur sequentially, but may be performed alternately or alternately with at least a portion of the sub-steps or stages of other steps or steps.
In one embodiment, as shown in fig. 5, there is provided an image content recognition apparatus including: an image acquisition module 410, an image encoding module 420, an image decoding module 430, and an image content identification module 440, wherein:
An image acquisition module 410 is configured to acquire an image to be identified.
The image coding module 420 is configured to extract features of a preset dimension of an image to be identified according to a coding layer of a trained convolutional self-coding model, obtain a compressed image, and train the trained convolutional self-coding model based on a historical image, an image compression degree and a preset image loss rate, where the preset image loss rate is an image loss rate lower than a preset image loss rate threshold.
The image decoding module 430 inputs the compressed image to a decoding layer of a preset convolutional self-coding model for reconstruction, and obtains a reconstructed decompressed image.
The image content recognition module 440 is configured to perform similarity matching on the decompressed image and a history image in a preset gallery according to a preset similarity matching algorithm, and screen out similar images to the decompressed image, where the history image is an image including features of a preset dimension.
As shown in fig. 6, in one embodiment, the image content recognition device further includes a model training module 450, configured to collect original images, construct an image training set, establish a convolution self-coding model with staggered convolution layers and pooling layers, set an initial feature extraction number, input the original images in the image training set to the convolution self-coding model for coding, extract image features of a preset dimension, perform decoding processing on the image features of the preset dimension to obtain reconstructed original images, calculate and optimize loss rates of each reconstructed original image and the original image to obtain an overall image loss rate of the image training set, and adjust the feature extraction number when the overall image loss rate is greater than a preset image loss rate, obtain an updated overall image loss rate until the overall image loss rate is less than or equal to the preset image loss rate, determine the feature extraction number, and obtain the trained convolution self-coding model.
In one embodiment, the image content identification module 440 is further configured to calculate an angle cosine value of each of the decompressed images and each of the historical images in the preset gallery according to a cosine similarity matching algorithm, convert the angle cosine value into an angle similarity coefficient, and screen out images similar to the decompressed images from the preset gallery according to the angle similarity coefficient.
In one embodiment, the model training module 450 is further configured to determine a loss function of the convolutional self-coding model by combining Adadelta algorithm and random gradient descent algorithm, and calculate and optimize an overall image loss rate of the image training set based on the determined loss function.
In one embodiment, the model training module 450 is further configured to adjust a learning rate of the convolutional self-coding model according to the Adadelta algorithm, and when an oscillation occurs due to the learning rate being too high, calculate an optimal solution of the loss function by using a random gradient descent algorithm, and determine the loss function of the convolutional self-coding model.
In one embodiment, the image encoding module 420 is further configured to perform rolling and pooling on the image to be identified according to the trained convolutional self-encoding model, and extract features of a preset dimension of the image to be identified; the image decoding module 430 is further configured to input the compressed image to the preset convolutional self-coding model, and perform deconvolution and inverse pooling on the compressed image according to a decoding layer of the preset convolutional self-coding model, so as to obtain a reconstructed decompressed image.
For specific limitations of the image content recognition apparatus, reference may be made to the above limitations of the image content recognition method, and no further description is given here. The respective modules in the image content recognition apparatus described above may be implemented in whole or in part by software, hardware, and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a server, and the internal structure of which may be as shown in fig. 7. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is for storing image data. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of image content identification.
It will be appreciated by those skilled in the art that the structure shown in FIG. 7 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program: obtaining an image to be identified, extracting the characteristics of preset dimensions of the image to be identified according to a trained convolution self-coding model, obtaining a compressed image, training the trained convolution self-coding model based on a historical image and a preset image loss rate, inputting the compressed image into the trained convolution self-coding model for reconstruction, obtaining a reconstructed decompressed image, carrying out similarity matching on the decompressed image and the historical image in a preset gallery according to a preset similarity matching algorithm, screening out images similar to the decompressed image, and obtaining an image content identification result corresponding to the image to be identified.
In one embodiment, the processor when executing the computer program further performs the steps of: according to a cosine similarity matching algorithm, calculating an included angle cosine value of each historical image in the decompressed image and a preset image library, converting the included angle cosine value into an angle similarity coefficient, and screening images similar to the decompressed image from the preset image library according to the angle similarity coefficient.
In one embodiment, the processor when executing the computer program further performs the steps of: collecting original images, constructing an image training set, establishing a convolution self-coding model with staggered convolution layers and pooling layers, wherein the convolution self-coding model is provided with an initial feature extraction number, inputting the original images in the image training set into the convolution self-coding model for coding, extracting image features with preset dimensions, decoding the image features with preset dimensions to obtain reconstructed original images, calculating and optimizing the loss rate of each reconstructed original image and each original image to obtain the integral image loss rate of the image training set, adjusting the feature extraction number when the integral image loss rate is larger than the preset image loss rate to obtain an updated integral image loss rate, and determining the feature extraction number until the integral image loss rate is smaller than or equal to the preset image loss rate to obtain the trained convolution self-coding model.
In one embodiment, the processor when executing the computer program further performs the steps of: and combining Adadelta algorithm and random gradient descent algorithm, determining a loss function of the convolution self-coding model, and calculating and optimizing the integral image loss rate of the image training set based on the determined loss function.
In one embodiment, the processor when executing the computer program further performs the steps of: according to Adadelta algorithm, the learning rate of the convolution self-coding model is adjusted, when the occurrence of oscillation caused by the overlarge learning rate is detected, the optimal solution of the loss function is obtained by adopting a random gradient descent algorithm, and the loss function of the convolution self-coding model is determined.
In one embodiment, the processor when executing the computer program further performs the steps of: rolling and pooling the image to be identified according to the trained convolution self-coding model, and extracting the characteristics of the preset dimension of the image to be identified; inputting the compressed image into a preset convolution self-coding model, and carrying out deconvolution and anti-pooling on the compressed image according to a decoding layer of the preset convolution self-coding model to obtain a reconstructed decompressed image.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of: obtaining an image to be identified, extracting the characteristics of preset dimensions of the image to be identified according to a trained convolution self-coding model, obtaining a compressed image, training the trained convolution self-coding model based on a historical image and a preset image loss rate, inputting the compressed image into the trained convolution self-coding model for reconstruction, obtaining a reconstructed decompressed image, carrying out similarity matching on the decompressed image and the historical image in a preset gallery according to a preset similarity matching algorithm, screening out images similar to the decompressed image, and obtaining an image content identification result corresponding to the image to be identified.
In one embodiment, the computer program when executed by the processor further performs the steps of: according to a cosine similarity matching algorithm, calculating an included angle cosine value of each historical image in the decompressed image and a preset image library, converting the included angle cosine value into an angle similarity coefficient, and screening images similar to the decompressed image from the preset image library according to the angle similarity coefficient.
In one embodiment, the computer program when executed by the processor further performs the steps of: collecting original images, constructing an image training set, establishing a convolution self-coding model with staggered convolution layers and pooling layers, wherein the convolution self-coding model is provided with an initial feature extraction number, inputting the original images in the image training set into the convolution self-coding model for coding, extracting image features with preset dimensions, decoding the image features with preset dimensions to obtain reconstructed original images, calculating and optimizing the loss rate of each reconstructed original image and each original image to obtain the integral image loss rate of the image training set, adjusting the feature extraction number when the integral image loss rate is larger than the preset image loss rate to obtain an updated integral image loss rate, and determining the feature extraction number until the integral image loss rate is smaller than or equal to the preset image loss rate to obtain the trained convolution self-coding model.
In one embodiment, the computer program when executed by the processor further performs the steps of: and combining Adadelta algorithm and random gradient descent algorithm, determining a loss function of the convolution self-coding model, and calculating and optimizing the integral image loss rate of the image training set based on the determined loss function.
In one embodiment, the computer program when executed by the processor further performs the steps of: according to Adadelta algorithm, the learning rate of the convolution self-coding model is adjusted, when the occurrence of oscillation caused by the overlarge learning rate is detected, the optimal solution of the loss function is obtained by adopting a random gradient descent algorithm, and the loss function of the convolution self-coding model is determined.
In one embodiment, the computer program when executed by the processor further performs the steps of: rolling and pooling the image to be identified according to the trained convolution self-coding model, and extracting the characteristics of the preset dimension of the image to be identified; inputting the compressed image into a preset convolution self-coding model, and carrying out deconvolution and anti-pooling on the compressed image according to a decoding layer of the preset convolution self-coding model to obtain a reconstructed decompressed image.
In one embodiment, the computer program when executed by the processor further performs the steps of: rolling and pooling the image to be identified according to the coding layer of the trained convolution self-coding model, and extracting the characteristics of the preset dimension of the image to be identified; inputting the compressed image into a decoding layer of a preset convolution self-coding model, and carrying out deconvolution and reverse pooling on the compressed image according to the decoding layer of the preset convolution self-coding model to obtain a reconstructed decompressed image.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (SYNCHLINK) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The foregoing examples illustrate only a few embodiments of the application, which are described in detail and are not to be construed as limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of protection of the present application is to be determined by the appended claims.
Claims (10)
1. A method of image content identification, the method comprising:
Acquiring an image to be identified;
Collecting an original image, constructing an image training set, establishing a convolution self-coding model with staggered convolution layers and pooling layers, wherein the convolution self-coding model is provided with an initial feature extraction number, inputting the original image in the image training set into the convolution self-coding model for coding, extracting image features with preset dimensions, decoding the image features with preset dimensions to obtain a reconstructed original image, determining a loss function of the convolution self-coding model by combining Adadelta algorithm and random gradient descent algorithm, calculating and optimizing the overall image loss rate of the image training set based on the determined loss function, obtaining the overall image loss rate of the image training set, adjusting the feature extraction number when the overall image loss rate is larger than the preset image loss rate, obtaining an updated overall image loss rate, and determining the feature extraction number until the overall image loss rate is smaller than or equal to the preset image loss rate, so as to obtain a trained convolution self-coding model;
Extracting the characteristics of the preset dimension of the image to be identified according to the trained convolution self-coding model to obtain a compressed image, wherein the trained convolution self-coding model is obtained by training based on a historical image and a preset image loss rate;
Inputting the compressed image into the trained convolution self-coding model for reconstruction to obtain a reconstructed decompressed image;
RGB value processing is carried out on pixel points of all historical images in the decompressed image and the preset gallery to obtain updated decompressed images and updated historical images of the preset gallery, an included angle cosine value of each historical image in the decompressed images and the updated preset gallery is calculated according to a cosine similarity matching algorithm, the included angle cosine value is converted into an angle similarity coefficient, and images similar to the updated decompressed images are screened out from the updated preset gallery according to the angle similarity coefficient.
2. The image content identification method of claim 1, wherein the combining Adadelta algorithm with the random gradient descent algorithm to determine the loss function of the convolutional self-encoding model comprises:
according to Adadelta algorithm, adjusting the learning rate of the convolution self-coding model;
When the oscillation occurrence caused by the overlarge learning rate is detected, the optimal solution of the loss function is obtained by adopting the random gradient descent algorithm, and the loss function of the convolution self-coding model is determined.
3. The image content recognition method of claim 1, wherein the extracting the feature of the preset dimension of the image to be recognized according to the trained convolutional self-coding model comprises:
Rolling and pooling the image to be identified according to the trained convolution self-coding model, and extracting the characteristics of the preset dimension of the image to be identified;
Inputting the compressed image into the trained convolution self-coding model for reconstruction, and obtaining a reconstructed decompressed image comprises the following steps:
And carrying out deconvolution and anti-pooling on the compressed image according to the trained convolution self-coding model to obtain a reconstructed decompressed image.
4. The image content recognition method according to claim 1, wherein the image to be recognized is an image in which a pixel size is preset and image segmentation has been completed based on extreme points.
5. An image content recognition device, the device comprising:
the image acquisition module is used for acquiring an image to be identified;
The model training module is used for collecting original images, constructing an image training set, establishing a convolution self-coding model with staggered convolution layers and pooling layers, wherein the convolution self-coding model is provided with an initial feature extraction number, inputting the original images in the image training set into the convolution self-coding model for coding, extracting image features with preset dimensions, decoding the image features with preset dimensions to obtain a reconstructed original image, combining Adadelta algorithm and random gradient descent algorithm, determining a loss function of the convolution self-coding model, calculating and optimizing the overall image loss rate of the image training set based on the determined loss function, obtaining the overall image loss rate of the image training set, adjusting the feature extraction number when the overall image loss rate is larger than the preset image loss rate, obtaining an updated overall image loss rate, and determining the feature extraction number until the overall image loss rate is smaller than or equal to the preset image loss rate, so as to obtain a trained self-coding model;
the image coding module is used for extracting the characteristics of the preset dimension of the image to be identified according to the trained convolution self-coding model to obtain a compressed image, and the trained convolution self-coding model is obtained by training based on a historical image and a preset image loss rate;
The image decoding module is used for inputting the compressed image into the trained convolution self-coding model for reconstruction to obtain a reconstructed decompressed image;
The image content recognition module is used for carrying out similarity matching on the decompressed image and a history image in a preset gallery according to a preset similarity matching algorithm, screening out similar images of the decompressed image, and obtaining an image content recognition result corresponding to the image to be recognized;
the device is also used for carrying out RGB value processing on the pixel points of the decompressed image and all the historical images in the preset gallery to obtain updated decompressed images and updated historical images of the preset gallery;
The image content identification module is further configured to calculate an included angle cosine value of each historical image in the updated decompressed image and the updated preset gallery according to a cosine similarity matching algorithm, convert the included angle cosine value into an angle similarity coefficient, and screen out an image similar to the updated decompressed image from the updated preset gallery according to the angle similarity coefficient.
6. The image content recognition device according to claim 5, wherein the model training module is further configured to adjust a learning rate of the convolutional self-coding model according to the Adadelta algorithm, and when an occurrence of a concussion is detected due to an excessive learning rate, calculate an optimal solution of a loss function by using the random gradient descent algorithm, and determine the loss function of the convolutional self-coding model.
7. The image content recognition device of claim 5, wherein the model training module is further configured to perform rolling and pooling on the image to be recognized according to the trained convolutional self-coding model, and extract features of a preset dimension of the image to be recognized;
The image decoding module is also used for carrying out deconvolution and anti-pooling on the compressed image according to the trained convolution self-coding model to obtain a reconstructed decompressed image.
8. The image content recognition device according to claim 5, wherein the image to be recognized is an image in which a pixel size is preset and image segmentation has been completed based on extreme points.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 4 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910882184.1A CN110766048B (en) | 2019-09-18 | 2019-09-18 | Image content recognition method, image content recognition device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910882184.1A CN110766048B (en) | 2019-09-18 | 2019-09-18 | Image content recognition method, image content recognition device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110766048A CN110766048A (en) | 2020-02-07 |
| CN110766048B true CN110766048B (en) | 2024-08-02 |
Family
ID=69329725
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910882184.1A Active CN110766048B (en) | 2019-09-18 | 2019-09-18 | Image content recognition method, image content recognition device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110766048B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111401272B (en) * | 2020-03-19 | 2021-08-24 | 支付宝(杭州)信息技术有限公司 | Face feature extraction method, device and equipment |
| CN112861950A (en) * | 2021-01-29 | 2021-05-28 | 深圳前海微众银行股份有限公司 | Method and device for identifying copied image, electronic equipment and storage medium |
| CN117196996B (en) * | 2023-10-17 | 2024-06-04 | 山东鸿业信息科技有限公司 | Interface-free interaction management method and system for data resources |
| CN117576519B (en) * | 2024-01-15 | 2024-04-09 | 浙江航天润博测控技术有限公司 | Image recognition model training optimization method and device, electronic equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106682649A (en) * | 2017-01-24 | 2017-05-17 | 成都容豪电子信息科技有限公司 | Vehicle type recognition method based on deep learning |
| CN109259784A (en) * | 2018-08-27 | 2019-01-25 | 上海铱硙医疗科技有限公司 | AI prediction technique, device, equipment and the storage medium of cerebral infarction |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10417560B2 (en) * | 2016-12-01 | 2019-09-17 | Via Alliance Semiconductor Co., Ltd. | Neural network unit that performs efficient 3-dimensional convolutions |
-
2019
- 2019-09-18 CN CN201910882184.1A patent/CN110766048B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106682649A (en) * | 2017-01-24 | 2017-05-17 | 成都容豪电子信息科技有限公司 | Vehicle type recognition method based on deep learning |
| CN109259784A (en) * | 2018-08-27 | 2019-01-25 | 上海铱硙医疗科技有限公司 | AI prediction technique, device, equipment and the storage medium of cerebral infarction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110766048A (en) | 2020-02-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110766048B (en) | Image content recognition method, image content recognition device, computer equipment and storage medium | |
| US10834415B2 (en) | Devices for compression/decompression, system, chip, and electronic device | |
| CN111738244B (en) | Image detection method, image detection device, computer equipment and storage medium | |
| Manap et al. | Non-distortion-specific no-reference image quality assessment: A survey | |
| US11741581B2 (en) | Training method for image processing model, image processing method, network device, and storage medium | |
| CN109816615B (en) | Image restoration method, device, equipment and storage medium | |
| CN110929622A (en) | Video classification method, model training method, device, equipment and storage medium | |
| CN112001914A (en) | Depth image completion method and device | |
| CN106548159A (en) | Reticulate pattern facial image recognition method and device based on full convolutional neural networks | |
| CN111260055A (en) | Model training method based on three-dimensional image recognition, storage medium and equipment | |
| CN111079764A (en) | Low-illumination license plate image recognition method and device based on deep learning | |
| CN116681584A (en) | Multistage diffusion image super-resolution algorithm | |
| CN112602114A (en) | Image processing method and device, neural network and training method, and storage medium | |
| CN112634120A (en) | Image reversible watermarking method based on CNN prediction | |
| CN115345866A (en) | Method for extracting buildings from remote sensing images, electronic equipment and storage medium | |
| CN114641792A (en) | Image processing method, image processing apparatus, and readable storage medium | |
| CN114331913B (en) | Motion blurred image restoration method based on residual attention block | |
| CN117333409A (en) | Big data analysis method based on image | |
| CN113920147B (en) | Remote sensing image building extraction method and device based on deep learning | |
| CN110659641B (en) | Text recognition method and device and electronic equipment | |
| CN113518229B (en) | Method and device for training loop filter network, computer equipment and storage medium | |
| CN112258527B (en) | Image processing method, apparatus, electronic device, and computer-readable storage medium | |
| CN113222016B (en) | Change detection method and device based on cross enhancement of high-level and low-level features | |
| CN114926348A (en) | Device and method for removing low-illumination video noise | |
| CN116952954B (en) | Concave-convex detection method, device, equipment and storage medium based on stripe light |
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 |