CN112926479A - Cat face identification method and system, electronic device and storage medium - Google Patents

Abstract

The present application discloses a cat face recognition method, system, electronic device and storage medium. The method includes: acquiring a frontal face image of a cat collected at a predetermined location; extracting key points of the cat's face on the frontal image; Affine transformation to obtain a fixed point; input the fixed point into the pre-trained cat face recognition model to obtain the output result of the cat face recognition model, the output result contains at least one cat face label, and the cat face label of different cats is different; The behavior information and time information of the location are associated with the cat face tag and recorded. By performing cat face recognition on the front face image of the cat collected at the predetermined location, the label of the cat appearing at the predetermined location can be accurately identified. It can constitute what the cat does at the predetermined place and the time of doing it, which is the cat's living habits, so as to accurately record the living habits of each cat.

Description

Cat face recognition method, system, electronic device and storage medium

technical field

The present application relates to the technical field of cat face recognition, and in particular to a method, system, electronic device and storage medium for cat face recognition.

Background technique

Among the pets raised by people, cats account for a large proportion. When cats are sick, their living habits (such as eating, drinking, excretion time, etc.) will be abnormal, so it is necessary to record the living habits of cats.

However, some cat owners have a large number of cats, and the cat owners may not have enough energy to record the cats' living habits, so they cannot find out whether the cats' living habits are abnormal.

At present, there are some technologies that recognize cat faces through face recognition models. Although there are face recognition models that can perform face recognition, compared with human faces, cat faces have hair and other influences, which makes the face recognition model more sensitive to cat faces. Recognition is difficult, so using existing face recognition methods to recognize cat faces cannot accurately identify each cat, and further results in inability to accurately record each cat's living habits.

SUMMARY OF THE INVENTION

In view of this, the application provides a cat face recognition method, system, electronic device and storage medium, to solve the problem that the existing face recognition method to recognize the cat face cannot accurately identify each cat and record the living habits of each cat question.

A first aspect of the present application provides a method for cat face recognition, including: acquiring a frontal face image of a cat collected at a predetermined location; extracting key points of a cat face on the frontal face image, where the key points of the cat face include two parts of the cat's face. The key points of the eyes and nose; perform affine transformation on the key points of the cat face to obtain fixed points; input the fixed points into the pre-trained cat face recognition model to obtain the output result of the cat face recognition model, so The output result contains at least one cat face tag, and the cat face tags of different cats are different; the behavior information and time information of the cat at the predetermined location are obtained, associated with the cat face tag, and recorded and saved.

Among them, the training method of the cat face recognition model includes: establishing an original face recognition model with a feature database of tens of thousands, the original face recognition model including at least one fully connected layer and one feature layer; optimizing the original face recognition model; The cat face sample data is input into the original face recognition model for training, and the cat face recognition model is obtained. The cat face sample data at least includes the cat face features and cat face labels of cats.

Among them, the optimization of the original face recognition model includes: constructing a cross entropy loss function; obtaining the feature vector output by the feature layer and the weight output by the classification network; The distance between them is expressed as cosine similarity, which completes the optimization of the original face recognition model.

Among them, the optimization of the original face recognition model also includes: screening the difficult samples in the cat face sample data, and constructing a triple loss function; using the triple loss function to calculate the cosine similarity of the difficult samples to complete the original face Recognition model optimization.

Among them, the difficult samples include: N sample data with the same cat face label and the largest difference in cosine similarity, and N sample data with different cat face labels and the smallest difference in cosine similarity, where N is greater than or equal to two.

Wherein, optimizing the original face recognition model includes: inputting two sets of cat face sample data into the original face recognition model, and constructing a cross-entropy loss function and a triplet loss function in the feature layer; using the cross-entropy loss function to calculate Cross-entropy loss value of the first group of cat face sample data and cat face labels; use the original face recognition model to filter out the difficult samples in the second group of cat face sample data; use triplet loss function for difficult samples to calculate triples Loss value; the triple loss value and the cross entropy loss value are added and summed to obtain the final loss value of the original face recognition model, and the optimization of the original face recognition model is completed.

Among them, the data magnitude calculated by the cross-entropy loss function is ten times the data magnitude calculated by the triple loss function.

A second aspect of the present application provides a cat face recognition system, including: an image acquisition module for acquiring a frontal face image of a cat collected at a predetermined location; a cat face key point extraction module for performing key point analysis on the cat face from the frontal image Extraction, the key points of the cat face include the key points of the cat's two eyes and nose; the affine transformation module is used to perform affine transformation on the key points of the cat face extracted by the cat face key point extraction module to obtain fixed points; cat face recognition The module is used to integrate a cat face recognition model that can recognize cat faces by using cat face sample data in advance. The output of the cat face recognition model contains at least one cat face label, and different cats have different cat face labels; data interaction module, use The fixed point obtained by the affine transformation module is input into the cat face recognition module, and the output result of the cat face recognition model in the cat face recognition module is obtained; the associated recording module is used to obtain the behavior information and time information of the cat at the predetermined location, And associated with the cat face tag, and record keeping.

A third aspect of the present application provides an electronic device, comprising: a memory and a processor, wherein a program is stored on the memory, wherein the program is used to implement any one of the above when executed by the processor The cat face recognition method described in item.

A fourth aspect of the present application provides a readable storage medium on which a program is stored, wherein the program is configured to implement any one of the above methods for cat face recognition when executed by a processor.

The above-mentioned cat face recognition method of the present application, on the one hand, uses a pre-trained cat face recognition model to recognize cat faces, which can reduce the influence of hair on recognition, thereby reducing the difficulty of cat face recognition; Cat face recognition from the front face image of the cat collected at the location can accurately identify the label of the cat appearing at the predetermined location. The label of the cat is associated with the behavior information and time information of the cat at the predetermined location, and is recorded and saved, so the living habits of each cat can be accurately recorded.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present application more clearly, the following briefly introduces the drawings that are used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, other drawings can also be obtained from these drawings without creative effort.

1 is a schematic flowchart of a method for cat face recognition according to an embodiment of the present application;

FIG. 2 is a schematic flow chart of an optimization process for an original face recognition model of a cat face recognition method according to an embodiment of the present application;

3 is a schematic flow diagram of another optimization process for the original face recognition model of the cat face recognition method according to an embodiment of the present application;

4 is a schematic structural block diagram of a cat face recognition system according to an embodiment of the present application;

FIG. 5 is a schematic structural block diagram of an electronic device according to an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of this application. In the case of no conflict, the following various embodiments and their technical features can be combined with each other.

Referring to FIG. 1, a method for recognizing a cat face provided in an embodiment of the present application includes: S1, acquiring a frontal face image of a cat collected at a predetermined location; S2, extracting key points of the cat's face on the frontal face image; S3, Perform affine transformation on the key points of the cat face to obtain the fixed point; S4, input the fixed point into the pre-trained cat face recognition model, and obtain the output result of the cat face recognition model, the output result contains at least one cat face label, cats of different cats The face tags are different; S5, obtain the behavior information and time information of the cat at the predetermined place, associate with the cat face tag, and save it.

In this embodiment, the key points of the cat's face are the key points of the cat's two eyes and the tip of the nose. The predetermined location is the execution location of the cat's living habits, and each predetermined location matches the living habits of a cat; the predetermined location can be a cat litter box, a cat food basin, a cat water basin, a cat climbing frame, a cat litter, etc., corresponding to The cat’s living habits are excretion, eating, drinking, leisure, and sleeping, and the time when the frontal face image is collected is the time information. In this embodiment, the predetermined location is the cat litter box, and the corresponding cat’s living habits are the cat’s excretion, The time information is the time when the frontal face image was collected when the cat entered the litter box.

A camera device can be set next to the cat litter box to capture the cat's face image, and a closed cat litter box can also be used. , since the face is always facing inward, the frontal face image of the cat can be obtained in the corresponding direction.

After obtaining the cat's frontal face image, the cat's face is recognized. Since the cat's mouth and nose are connected, only one of the mouth and nose can be used as a key point. In this embodiment, the There are three key points, namely the nose and the key points of the two eyes.

After extracting the key points of the cat face, the key points of the cat face are affine transformed to fixed points. The pre-trained cat face recognition model can use the fixed points as input, and then output the result. The output result contains the cat face label, so it can According to the cat face label, it is determined which cat is excreting at the litter box, so that the cat's living habits can be recorded. It is convenient for cat owners to check which cat has excreted in the litter box, whether the excretion time is normal, or which cat has not gone to the litter box for excretion, and the cat's excretion habit is also a kind of living habit, so the cat owner It is possible to check whether the cat's excretion habits are abnormal through records, so as to timely determine whether the cat is sick.

In one embodiment, the training method of the cat face recognition model includes: establishing an original face recognition model with a feature library of tens of thousands, the original face recognition model including at least one fully connected layer and one feature layer; Carry out optimization; input the cat face sample data into the original face recognition model for training, and obtain a cat face recognition model capable of recognizing cat faces. The cat face sample data includes at least the cat face features and cat face labels of cats.

In this embodiment, the original face model used can use any relevant model in the prior art, such as residual network model, Mobilenetv3 and other network models, and the original face recognition model is a cat face recognition model after training , and the feature library of tens of thousands can ensure the ability of the cat face recognition model to perform feature extraction. By optimizing the original face model, the performance of the original face model can be improved, so that the cat face sample data can be used for subsequent training. , which can train a cat face recognition model with better performance.

In one embodiment, optimizing the original face recognition model includes: constructing a cross-entropy loss function; acquiring the feature vector output by the feature layer and the weight output by the classification network; performing L2 norm normalization on the feature vector and the weight, The optimization of the original face recognition model is completed by expressing the distance between feature vectors as cosine similarity.

The optimization direction of the feature vector is expressed as the cosine similarity between the vectors, which can express the loss value more clearly, which is more convenient for the use of the original face recognition model and enhances the consistency of the optimization direction after combining with other loss functions.

In this embodiment, the cross-entropy loss function first outputs a probability value P _j , which can be expressed by formula 1:

Among them, z _j represents the normalized value of the feature vector and the weight calculated by the original face recognition model, z _j =w*x, w represents the weight after the normalization of the L2 norm of the fully connected layer in the classification network, x is the eigenvector normalized by the L2 norm of the feature layer, K represents the number of samples in the sample set, j represents the jth sample in the sample set, z _k represents the real eigenvector of the sample in the sample set and the normalized value of the weight, z _k can be calculated using the true distribution model of the sample set.

Then the cross-entropy loss function outputs the metric loss value, which can be expressed by Equation 2:

表示样本集的真实分布模型，p表示概率值，K表示样本集的样本数量，

表示第j个样本在真实分布模型中的标签值，p_j表示第j个样本的概率值。in,

Represents the true distribution model of the sample set, p represents the probability value, K represents the number of samples in the sample set,

represents the label value of the jth sample in the true distribution model, and p _j represents the probability value of the jth sample.

In this embodiment, the cross-entropy directly measures the distribution difference between the two models, one of which is the original face recognition model and the other is the real distribution model of the sample set, and the cross-entropy loss function is used to explain the original face recognition model. The output of the identification model and the degree of interpretation of the sample set by the real distribution model of the sample set. The real distribution model of the sample set is a conventional model, which will not be described in this embodiment.

The above label value is 0 or 1. When the predicted output of the corresponding category is closer to the true label value, the value of the cross-entropy loss function is smaller, and the recognized result is closer to the sample, that is, the model recognition accuracy rate is higher. The more the predicted output deviates from the true label value, the larger the value of the cross-entropy loss function, and the farther the recognized result is from the sample, that is, the lower the model recognition accuracy rate is.

In one embodiment, optimizing the original face recognition model further includes: screening difficult samples in the cat face sample data, and constructing a triplet loss function; using the triplet loss function to calculate the cosine similarity of the difficult samples to complete Optimization of the original face recognition model.

After using the cross-entropy loss function, the triplet loss function can be used to process the data that the cross-entropy loss function cannot process on difficult samples, thereby improving the performance of the trained cat face recognition model. In this embodiment , using the triple loss function to calculate the cosine similarity of the difficult samples separately, that is, the difficult samples no longer use the cross entropy loss function to calculate the cosine similarity.

Referring to FIG. 2, in this embodiment, the cross-entropy loss function and the triplet loss function are used in turn to perform the calculation of the cosine similarity of ordinary samples and the calculation of cosine similarity of difficult samples, so as to optimize the original face recognition model, Ordinary samples are all samples contained in the training set. In this process, since the cross-entropy loss function does not need to normalize the weights and feature layer features of the fully-connected layer, when the cross-entropy loss function and cosine-based loss function are used later When using the triple loss function of similarity, it will lead to differences in the consistency of the optimization direction. In order to ensure the consistency of this optimization direction and allow the training of the original face recognition model to converge better, when using the cross entropy loss function, it is necessary to perform L2 normalization between the feature vector obtained by the feature layer and the weight of the fully connected layer. Finally, the distance between feature vectors is expressed as cosine similarity. After that, select the difficult samples in this batch of cat face sample data, and use the triple loss function to optimize the cosine similarity of the difficult samples alone, so that the difficult samples can also be learned by the original face recognition model, and finally make the training result. The cat face recognition model is able to adapt to cat face image recognition in more situations.

In one embodiment, the difficult samples include: N sample data with the same cat face label and the largest difference in cosine similarity, and N sample data with different cat face labels and the smallest difference in cosine similarity, where N is greater than or equal to two .

N sample data with the same cat face label and the largest difference in cosine similarity. Such sample data represent the same cat, but the cat face has subtle changes in different situations. For example, when the cat is sick, the facial features will have subtle changes. , when the cat meows, the nose and eyes will also change accordingly. Therefore, when the cat face changes, the cosine similarity will be greatly different from that of the normal cat face. Such a sample can be used as a positive Samples are used to let the original face recognition model learn, and the model's inclusiveness of the class spacing within the same cat face is enhanced.

When there are N sample data with different cat face labels and the smallest difference in cosine similarity, such sample data represents different cats. Such samples can be used as negative samples for the original face recognition model to learn, and the model can be enhanced for different cat faces. Discrimination of between-class differences.

Therefore, after the original face recognition model learns positive samples, the trained cat face recognition model can increase the probability of identifying the same cat face in the image, and after learning negative samples, it can increase the probability of filtering out images that are not the same cat face, so that This further increases the probability of correct recognition by the cat face recognition model.

In other embodiments, the optimization of the original face recognition model can also be achieved by the following steps: inputting two groups of cat face sample data into the original face recognition model, and constructing a cross-entropy loss function and a triplet loss function in the feature layer ;Use the cross-entropy loss function to calculate the cross-entropy loss value of the first set of cat face sample data and the cat face label; use the original face recognition model to filter out the difficult samples in the second set of cat face sample data; use ternary for difficult samples The group loss function calculates the triple loss value; the triple loss value and the cross entropy loss value are added and summed to obtain the final loss value of the original face recognition model, and the optimization of the original face recognition model is completed. In this embodiment, the data magnitude calculated by the cross-entropy loss function is ten times the data magnitude calculated by the triplet loss function. In addition, the definition and selection of difficult samples are the same as the above-mentioned embodiments. For details, please refer to the above-mentioned embodiments, which will not be repeated here.

Referring to FIG. 3, in this embodiment, the cross-entropy loss function and the triplet loss function are used simultaneously to calculate the cosine similarity of common samples and the cosine similarity of difficult samples, so as to optimize the original face recognition model, Ordinary samples are samples whose cross-entropy loss function can calculate cosine similarity. In this process, cross-entropy loss function and triplet loss function are used at the same time. Specifically, the cross-entropy loss function based on cosine similarity is the main loss function. The similarity between the cat face recognition model and the results trained on the independent cross-entropy loss function is maintained, and the triple loss function is used as the auxiliary loss function to learn the difficult samples, which optimizes the learning ability of the original face recognition model for difficult samples. , so that the trained cat face recognition model can be better applied to the scene of cat face recognition.

In any of the above-mentioned embodiments, the cat face recognition method may further include: recording the facial features of the cat by using the cat face recognition model, and recording the facial features into the feature database. In this way, after the cat enters the litter box, the image can be collected, and the key points of the cat's face can be extracted to identify the cat, and the cosine similarity with the recorded facial features can be calculated to identify the cat's label, name or code.

On the one hand, recognizing cat faces through a pre-trained cat face recognition model can reduce the impact of hair on recognition, thereby reducing the difficulty of cat face recognition; Cat face recognition can accurately identify the cat's label that appears at the predetermined location, and when applied to the scene where the predetermined location is where the cat eats, drinks, excretion, etc. Behavior information and time information are correlated and recorded, so the living habits of each cat can be accurately recorded.

Referring to FIG. 4 , a cat face recognition system provided in an embodiment of the present application includes: an image acquisition module 1, a cat face key point extraction module 2, an affine transformation module 3, a cat face recognition module 4, and a data interaction module 5 and associated recording module 6.

The image acquisition module 1 is used to acquire the frontal face image of the cat collected at a predetermined location;

The cat face key point extraction module 2 is used to extract the cat face key points from the frontal image, and the cat face key points include the key points of the cat's two eyes and nose; the affine transformation module 3 is used to extract the cat face key points. Module 2 The extracted cat face key points are subjected to affine transformation to obtain fixed points; the cat face recognition module 4 is used to integrate the pre-trained cat face recognition model capable of recognizing cat faces using cat face sample data, and the output result of the cat face recognition model contains at least A cat face label, different cats have different cat face labels; the data interaction module 5 is used to input the fixed point obtained by the affine transformation module 3 into the cat face recognition model module, and obtain the cat face recognition model in the cat face recognition model module. Output the result; the association recording module 6 is used to obtain the behavior information and time information of the cat at the predetermined location, associate it with the cat face tag, and record and save it.

In this embodiment, the key points of the cat's face are the key points of the cat's two eyes and the tip of the nose. The predetermined location is the execution location of the cat's living habits, and each predetermined location matches the living habits of a cat; the predetermined location can be a cat litter box, a cat food basin, a cat water basin, a cat climbing frame, a cat litter, etc., corresponding to The cat’s living habits are excretion, eating, drinking, leisure, and sleeping, and the time when the frontal face image is collected is the time information. In this embodiment, the predetermined location is the cat litter box, and the corresponding cat’s living habits are the cat’s excretion, The time information is the time when the frontal face image was collected when the cat entered the litter box.

In one embodiment, the cat face recognition module 4 includes: an original face recognition model establishment unit, an optimization unit and a training unit; the original face recognition model establishment unit is used to establish an original face recognition model of a ten-thousand-level feature library, the original face recognition model The face recognition model includes at least one fully connected layer and one feature layer; the optimization unit is used to optimize the original face recognition model; the training unit is used to input the cat face sample data into the original face recognition model for training, and get the ability to recognize cats The cat face recognition model of the face, the cat face sample data includes at least the cat face features and cat face labels of the cat.

In one embodiment, the optimization unit includes: a first function construction subunit, a data acquisition subunit, and a normalization subunit; the first function construction subunit is used to construct a cross-entropy loss function; the data acquisition subunit is used to acquire features The feature vector output by the layer, and the weight output by the classification network; the normalization subunit is used to normalize the feature vector and the weight with the L2 norm, so as to express the distance between the feature vectors as cosine similarity, and complete the analysis of primitive people. Optimization of face recognition models.

In one embodiment, the optimization unit further includes: a second function construction subunit and a difficult sample optimization unit, the second function construction subunit is used for screening difficult samples in the cat face sample data, and constructing a triplet loss function; The sample optimization unit is used to separately calculate and optimize the cosine similarity of difficult samples by using the triple loss function to complete the optimization of the original face recognition model.

In this embodiment, the difficult samples include: N sample data with the same cat face label and the largest difference in cosine similarity, and N sample data with different cat face labels and the smallest difference in cosine similarity, where N is greater than or equals two.

In other embodiments, the optimization unit may further include: a third function construction subunit, a cross entropy loss value calculation subunit, a difficult sample screening subunit, a triple loss value calculation subunit, and a summation subunit, the third function construction subunit The subunit is used to input two sets of cat face sample data to the original face recognition model, and constructs the cross-entropy loss function and triplet loss function in the feature layer; the cross-entropy loss value calculation sub-unit is used to use the cross-entropy loss function to calculate the first The cross entropy loss value of a set of cat face sample data and cat face labels; the difficult sample screening subunit is used to use the original face recognition model to screen out the difficult samples in the second set of cat face sample data; triple loss value calculator The unit is used to use the triple loss function to calculate the triple loss value for difficult samples; the summation subunit is used to add and sum the triple loss value and the cross entropy loss value to obtain the final loss of the original face recognition model. value to complete the optimization of the original face recognition model. In this embodiment, the definition of the difficult sample is the same as that in the above-mentioned embodiment. For details, please refer to the above-mentioned embodiment, which will not be repeated here.

In addition, in this embodiment, the data magnitude calculated by the cross-entropy loss function is ten times the data magnitude calculated by the triplet loss function.

The cat face recognition system provided in this embodiment, on the one hand, uses a pre-trained cat face recognition model to recognize cat faces, which can reduce the influence of hair on recognition, thereby reducing the difficulty of cat face recognition; Cat face recognition is performed on the front face image of the cat collected at the predetermined location, and the label of the cat that appears at the predetermined location can be accurately recognized. The cat's label is associated with the cat's behavior information and time information at the predetermined location, and is recorded and saved, so the living habits of each cat can be accurately recorded.

Referring to FIG. 5 , an embodiment of the present application provides an electronic device. The electronic device includes: a memory 601 and a processor 602 . The memory 601 stores a program that can be executed on the processor 602 , and the program is used to be executed by the processor 602 . , implement the cat face recognition method described above.

Further, the electronic device further includes: at least one input device 603 and at least one output device 604 .

The above-mentioned memory 601 , processor 602 , input device 603 and output device 604 are connected through a bus 605 .

The input device 603 may specifically be a camera, a touch panel, a physical button, a mouse, or the like. The output device 604 may specifically be a display screen.

The memory 601 may be a high-speed random access memory (RAM, Random Access Memory) memory, or may be a non-volatile memory (non-volatile memory), such as a disk memory. Memory 601 is used to store a set of executable program codes, and processor 602 is coupled to memory 601 .

Further, an embodiment of the present application further provides a readable storage medium. The readable storage medium may be provided in the electronic device in each of the foregoing embodiments, and the readable storage medium may be the aforementioned memory 601 . A program is stored on the readable storage medium, and when the program is executed by the processor 602, the method for recognizing the cat face described in the foregoing embodiments is implemented.

Further, the storable medium may also be a USB flash drive, a removable hard disk, a read-only memory 601 (ROM, Read-Only Memory), a RAM, a magnetic disk or an optical disk and other media that can store program codes.

In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or modules, and may be in electrical, mechanical or other forms.

The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, may be located in one place, or may be distributed to multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically alone, or two or more modules may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware, and can also be implemented in the form of software function modules.

If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may be stored in a readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the software product is stored in a storage medium, Several instructions are included to cause a device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

It should be noted that, for the convenience of description, the foregoing method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence. As in accordance with the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily all necessary to the present invention.

In the above-mentioned embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

In the above description, various details are set forth for the purpose of explanation. It is to be understood that one of ordinary skill in the art can realize that the present application may be practiced without the use of these specific details. In other instances, well-known structures and procedures have not been described in detail so as not to obscure the description of the present application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and features disclosed herein.

The above is a description of a cat face recognition method, system, electronic device and storage medium provided by the present invention. For those skilled in the art, based on the idea of the embodiments of the present invention, there will be Changes, in conclusion, the content of this specification should not be construed as a limitation to the present invention.

Claims (10)

1. A cat face identification method, comprising:

acquiring a front face image of a cat collected at a predetermined place;

extracting key points of a cat face from the front face image, wherein the key points of the cat face comprise key points of two eyes and a nose of a cat;

carrying out affine transformation on the key points of the cat face to obtain fixed points;

inputting the fixed point into a pre-trained cat face recognition model to obtain an output result of the cat face recognition model, wherein the output result at least comprises one cat face label, and the cat face labels of different cats are different;

and acquiring behavior information and time information of the cat in the preset place, associating the behavior information and the time information with the cat face label, and recording and storing the behavior information and the time information.

2. The cat face recognition method according to claim 1,

the training method of the cat face recognition model comprises the following steps:

establishing an original face recognition model of a ten-thousand-level feature library, wherein the original face recognition model at least comprises a full connection layer and a feature layer;

optimizing the original face recognition model;

and inputting cat face sample data into the original face recognition model for training to obtain a cat face recognition model, wherein the cat face sample data at least comprises cat face characteristics and cat face labels of cats.

3. The cat face recognition method according to claim 2,

the optimizing the original face recognition model comprises:

constructing a cross entropy loss function;

acquiring a feature vector output by the feature layer and the weight output by the classification network;

and performing L2 norm normalization on the feature vectors and the weights to express the distance between the feature vectors as cosine similarity, and completing the optimization of the original face recognition model.

4. The cat face recognition method according to claim 3,

the optimizing the original face recognition model further comprises:

screening difficult samples in the cat face sample data, and constructing a triple loss function;

and calculating the cosine similarity of the difficult samples by using the triple loss function, and finishing the optimization of the original face recognition model.

5. The cat face recognition method according to claim 4,

the difficult samples include: the method comprises the steps of providing N sample data with the same cat face label and the largest cosine similarity difference, providing N sample data with different cat face labels and the smallest cosine similarity difference, wherein N is larger than or equal to two.

6. The cat face recognition method according to claim 2,

the optimizing the original face recognition model comprises:

inputting two groups of cat face sample data into the original face recognition model, and constructing a cross entropy loss function and a triple loss function on the feature layer;

calculating a cross entropy loss value of the first group of cat face sample data and the cat face label by using the cross entropy loss function;

screening out difficult samples in the second group of cat face sample data by using the original face recognition model;

calculating a triplet loss value using the triplet loss function for the difficult sample;

and adding and summing the triple loss values and the cross entropy loss values to obtain a final loss value of the original face recognition model, and finishing the optimization of the original face recognition model.

7. The cat face recognition method according to claim 6,

the data magnitude calculated by the cross entropy loss function is ten times that calculated by the triplet loss function.

8. A cat face identification system comprising:

the image acquisition module is used for acquiring a front face image of the cat acquired at a preset place;

the face image processing module is used for processing the face image, and the face image processing module is used for processing the face image;

the affine transformation module is used for carrying out affine transformation on the cat face key points extracted by the cat face key point extraction module to obtain fixed points;

the cat face identification module is used for training a cat face identification model capable of identifying a cat face by integrating and using cat face sample data in advance, the output result of the cat face identification model at least comprises a cat face label, and the cat face labels of different cats are different;

the data interaction module is used for inputting the fixed point obtained by the affine transformation module into the cat face recognition module and obtaining an output result of a cat face recognition model in the cat face recognition module;

and the association recording module is used for acquiring behavior information and time information of the cat in the preset place, associating the behavior information and the time information with the cat face tag, and recording and storing the behavior information and the time information.

9. An electronic device, comprising: comprising a memory and a processor, said memory having stored thereon a program, characterized in that said program is adapted to carry out the method of any one of claims 1 to 7 when executed by said processor.

10. A readable storage medium on which a program is stored, the program being adapted to carry out the method of any one of claims 1 to 7 when executed by a processor.

Images