WO2019227616A1 - 动物身份的识别方法、装置、计算机设备和存储介质 - Google Patents

动物身份的识别方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2019227616A1
WO2019227616A1 PCT/CN2018/095667 CN2018095667W WO2019227616A1 WO 2019227616 A1 WO2019227616 A1 WO 2019227616A1 CN 2018095667 W CN2018095667 W CN 2018095667W WO 2019227616 A1 WO2019227616 A1 WO 2019227616A1
Authority
WO
WIPO (PCT)
Prior art keywords
animal
feature
facial
facial features
identity
Prior art date
Application number
PCT/CN2018/095667
Other languages
English (en)
French (fr)
Inventor
马进
王健宗
肖京
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2019227616A1 publication Critical patent/WO2019227616A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Definitions

  • the present application relates to the field of computer technology, and in particular, to a method, a device, a computer device, and a storage medium for identifying an animal identity.
  • identity recognition usually recognizes a person's face to identify a person.
  • Animal identification is rare.
  • industries such as smart farms, animals such as cattle and horses need to be identified for animal management; in the insurance industry, animals such as cows and horses need to be identified for animal insurance.
  • the main purpose of this application is to provide an animal identification method, device, computer equipment, and storage medium, which overcomes the defects that animal identification cannot be performed simply and effectively at present.
  • the present application provides a method for identifying an animal identity, including the following steps:
  • the feature database stores the animal facial features of the preset animals and their corresponding identity information in advance;
  • identity information corresponding to the facial features of the animal that are matched is obtained to identify the identity of the first animal.
  • the application also provides an animal identification device, including:
  • a first obtaining unit configured to obtain a face image of a first animal to be identified
  • a first extraction unit configured to extract facial features of the first animal face image through an animal identification model trained based on a FaceNet network model
  • a matching unit configured to match an animal face feature corresponding to a facial feature of the first animal face image in a feature database; the feature face of the preset animal and its correspondence are stored in the feature database Identity information
  • the second obtaining unit is configured to obtain identity information corresponding to the matched facial features of the animal if the matching is successful, so as to identify the identity of the first animal.
  • the present application further provides a computer device including a memory and a processor, where the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, implements the steps of any of the foregoing methods.
  • the present application also provides a computer non-volatile readable storage medium having computer-readable instructions stored thereon, which are executed by a processor to implement the steps of the method according to any one of the foregoing.
  • An animal identity recognition method, device, computer equipment and storage medium provided in this application, obtain a face image of a first animal to be identified, and extract the first animal face through an animal identity recognition model trained based on a FaceNet network model The facial features of the animal image; matching the facial features of the animal corresponding to the facial features of the first animal facial image in the feature database; if the matching is successful, obtaining the corresponding facial features of the animal Identity information to identify the identity of the first animal. It overcomes the defect that animal identification cannot be performed simply and effectively at present, which facilitates the management of animals.
  • FIG. 1 is a schematic diagram of steps of an animal identification method according to an embodiment of the present application.
  • FIG. 2 is a schematic diagram of steps of an animal identification method according to another embodiment of the present application.
  • FIG. 3 is a structural block diagram of an animal identification device according to an embodiment of the present application.
  • FIG. 4 is a structural block diagram of an animal identification device according to another embodiment of the present application.
  • FIG. 5 is a structural block diagram of an animal identification device according to another embodiment of the present application.
  • FIG. 6 is a schematic block diagram of a structure of a computer device according to an embodiment of the present application.
  • an embodiment of the present application provides a method for identifying an animal identity, including the following steps:
  • Step S1 Obtain a facial image of a first animal to be identified.
  • the first animal described above is an animal that needs to be identified, and the animal includes cows, horses, etc. In this embodiment, a cow is used as an example for illustration.
  • the face image of the first animal is first obtained.
  • the animal is identified by the features of the face image of the animal.
  • the above step S1 specifically includes: acquiring an image of the first animal, and detecting a face image of the first animal from the image of the first animal by using an image detection algorithm.
  • the first animal is a cow whose identity is to be identified
  • the above step S1 is specifically: obtaining an image of the cow to be identified, and performing cow face detection on the image by using a cow face detection algorithm to detect An image of a cow's face.
  • the aforementioned bull face detection algorithm is an algorithm for detecting bull face contour information, which includes a supervised gradient descent algorithm (SDM), a local binary algorithm (LBF), and the like.
  • step S2 the facial features of the first animal face image are extracted through an animal identification model trained based on a FaceNet network model.
  • the animal identification model is an animal identification model trained based on a FaceNet network model.
  • the FaceNet network model is generally used for face recognition.
  • the FaceNet network model used for face recognition is migrated to animal face recognition.
  • the FaceNet network model is pre-trained on the LFW (Labeled Faces in the Wild, face recognition public test set) data set, and then the parameters of the network model obtained after training are used as initialization parameters of the animal identity recognition model, where the initialization
  • the network in the animal identification model and the loss function are the same as the FaceNet network model pre-trained on the public face recognition test set described above.
  • the animal identification model in step S2 can be obtained.
  • the facial feature is a feature vector
  • the facial feature of the first animal facial image is extracted through the animal identification model.
  • the face image of the first animal is input into the animal identification model.
  • the last layer of the animal identification model is a loss function layer.
  • the input data of the loss function layer is the above facial features, and the output result is the Euclidean distance between different facial features. Therefore, only the output vector of the layer before the loss function layer needs to be extracted to obtain the facial features of the first animal face image.
  • Step S3 matching the animal facial features corresponding to the facial features of the first animal facial image in the feature database.
  • a plurality of preset animal face features (also a feature vector) and identity information corresponding to each animal face feature are pre-stored in the feature database.
  • the identity information corresponding to the above 100 cows and the facial features of each cow can be pre-stored in the feature database.
  • the facial features of the first animal face image are obtained, and they are matched with the facial features of the animals stored in the feature database. If the matching facial features of the animals are matched, it indicates that The first animal is the identity of the animal corresponding to the facial features of the matched animal. If the matching animal facial features cannot be matched, the first animal cannot be identified. In the scenario of a smart farm, it indicates that the first animal is not an animal raised in the farm.
  • step S4 if the matching is successful, the identity information corresponding to the matched facial features of the animal is acquired to identify the identity of the first animal.
  • the above-mentioned feature database stores identity information corresponding to each animal's facial features one by one.
  • identity information corresponding to the facial feature of the animal can be obtained, and the identity information is the identity information of the first animal.
  • the identity of the above-mentioned first animal is simply and effectively recognized, and it is matched and identified through facial features with high accuracy.
  • the step S3 of matching the facial features of the animal corresponding to the facial features of the first animal facial image in the feature database includes:
  • Step S31 Calculate the Euclidean distance between the facial features of the animal in the feature library and the facial features of the first animal facial image through the animal identification model;
  • Step S32 selecting an animal facial feature having the smallest Euclidean distance to the facial feature of the first animal and less than a preset value, as the animal face corresponding to the facial feature of the first animal ⁇ traits.
  • the facial features of the first animal face image are input into the animal identification model, meanwhile, the animal facial features in the feature library are sequentially input, and the features of the animal identification model are sequentially passed through.
  • the loss function layer calculates the Euclidean distance between the facial features of the first animal face image and the animal facial features in the feature library. In this animal identification model, the smaller the Euclidean distance, the more the two facial features are explained. The more similar.
  • An animal facial feature having the smallest Euclidean distance from the facial features of the first animal is selected, and when the Euclidean distance is less than a preset value, the animal's facial features and the first animal's face can be considered If the facial features are the same, the identity information corresponding to the facial features of the animal may be used as the identity information of the first animal.
  • the FaceNet network model is a network model for face recognition. It extracts features during training to make the Euclidean distance between the faces of the same person closer, and the Euclidean distance between different face images farther. Specifically, when training the FaceNet network model, three images are input at the same time each time, one is a picture A of a face to be predicted with labels, and the other two are auxiliary pictures B0, B1. B0 is a picture different from the face in picture A, and B1 is a different picture from the same person in picture A.
  • step S1 of obtaining a face image of a first animal to be identified the method includes:
  • Step S11 Extract animal facial features of the preset animal through the animal identification model trained based on the FaceNet network model;
  • step S12 animal facial features of the preset animal and corresponding identity information are stored in the feature database.
  • the preset animals in this embodiment are all animals that may need to be identified, such as all animals insured by the user, and all animals raised in the smart farm.
  • the cow face features of all the cows in the smart farm to be managed are extracted, each cow has its corresponding cow identity, and the cow identity of each cow and its corresponding cow face Features are stored in the feature library to form a bull face feature library.
  • the process of extracting the facial features of the animal in the above step S11 is the same as the process of extracting the facial features of the first animal's face image in the above step S2, and the difference is that the targeted facial images are different, and details are not described herein again.
  • the cow's face features are extracted and matched with the cow face features stored in the aforementioned cow face feature database to obtain the identity of the cow to be identified.
  • step S11 of extracting animal facial features of a preset animal by using the animal identification model obtained based on the FaceNet network model training the method includes:
  • Step S101 input the sample pictures in the training set into a FaceNet network model for training to obtain the animal identity recognition model; wherein the FaceNet network model is a face recognition network model trained based on the face recognition public test set training .
  • the cattle picture is used as the training set for training.
  • 16 sets of pictures in the training set are iteratively input to the FaceNet network model for training, where each group of pictures contains three bull face pictures, two of which are It comes from the same kind of cattle, and the other one is a different one. If two pictures come from the same cow, the label is 1, which means similar; if they are from different cattle, the label is 0, which means dissimilar.
  • the learning rate of the aforementioned FaceNet network model is set to 0.01, and the loss function is triplet loss.
  • the feature vectors of the three pictures in the set of pictures are extracted, and after calculation of the loss function layer, the two features corresponding to the two pictures with the label 1 are obtained.
  • the Euclidean distance between the vectors decreases (reduced to less than the preset value), and the Euclidean distance between the two feature vectors corresponding to the two pictures with the label 0 becomes larger (larger than the preset value) until After the Euclidean distance no longer changes, the training is completed, the training parameters are obtained, and the FaceNet network model from which the training parameters are obtained is used as the animal identification model for animal identification in the above embodiment.
  • step S101 of inputting the sample pictures in the training set into a FaceNet network model for training to obtain the animal identification model the method includes:
  • Step S102 input the sample pictures in the test set into the animal identification model for verification.
  • the ratio of the number of sample pictures in the training set to the test set is 4: 1.
  • the process of making a training set and a test set with a large number of cattle pictures in an intelligent farm includes:
  • the picture data of 120 cows is divided into 120 categories, each category has about 100 pictures, and the cow face detection algorithm is used to detect the cow faces of all pictures to generate a cow face database of 120 cows.
  • Two pictures are randomly selected from the same cow, and one picture is selected from the other cow.
  • the three pictures form a set of photos in the training and test sets. Randomly selected pictures are 10,000 groups. Among them, if the two pictures are from the same cow, the label is 1, which means they are similar. If they are from different cattle, the label is 0, which means they are not similar. Finally, the above 10,000 groups of pictures are divided into a training set and a test set according to a ratio of 4: 1, that is, there are 8000 groups of photos in the training set and 2000 groups of images in the test set.
  • step S4 of obtaining the identity information corresponding to the facial features of the matched animal to identify the identity of the first animal the method includes:
  • Step S5a Match an insurance policy corresponding to the identity of the first animal in a policy database according to the identity information of the first animal.
  • the first animal is a cow.
  • the claimant uses the animal identification method to identify the cow; then, According to S5a, the policy corresponding to the identity of the first animal is matched in the policy database, and information such as a claim amount corresponding to the identity of the cow is obtained from the above policy to facilitate claim settlement.
  • the method includes:
  • step S5b according to the identity information of the first animal, a farming database corresponding to the identity information of the first animal is queried in the farm breeding database.
  • the first animal is a cow raised in a smart farm.
  • it is necessary to identify the identity of the cow when tracking the status of the cow for example, collecting images of cows , And use the above method to identify the identity of the cow, and query the breeding information corresponding to the identity of the cow in the farm breeding database, such as: breeding start time, number of days of breeding, weight change, etc. Weight difference determines the weight change); or you can also query the historical photos of the cow in the farm breeding database, so that managers can view the changes and so on.
  • the animal identity recognition method obtains a face image of a first animal to be recognized, and extracts the first animal face through an animal identity recognition model trained based on a FaceNet network model.
  • the facial features of the animal image matching the facial features of the animal corresponding to the facial features of the first animal facial image in the feature database; obtaining the identity information corresponding to the matched facial features of the animal to Identifying the identity of the first animal. It overcomes the defect that animal identification cannot be performed simply and effectively at present, which facilitates the management of animals.
  • an embodiment of the present application further provides an animal identity recognition device, including:
  • the first obtaining unit 10 is configured to obtain a face image of a first animal to be recognized.
  • the first animal described above is an animal that needs to be identified, and the animal includes cows, horses, etc. In this embodiment, a cow is used as an example for illustration.
  • the face image of the first animal is first obtained.
  • the animal is identified by the features of the face image of the animal.
  • the first obtaining unit 10 is specifically configured to obtain an image of a first animal, and detect an image of a face of the first animal from the image of the first animal by using an image detection algorithm.
  • the first animal is a cow whose identity is to be identified
  • the above step S1 is specifically: obtaining an image of the cow to be identified, and performing cow face detection on the image by using a cow face detection algorithm to detect An image of a cow's face.
  • the aforementioned bull face detection algorithm is an algorithm for detecting bull face contour information, which includes a supervised gradient descent algorithm (SDM), a local binary algorithm (LBF), and the like.
  • the first extraction unit 20 is configured to extract facial features of the first animal face image through an animal identity recognition model trained based on a FaceNet network model.
  • the animal identification model is an animal identification model trained based on a FaceNet network model.
  • the FaceNet network model is generally used for face recognition.
  • the FaceNet network model used for face recognition is migrated to animal face recognition.
  • the FaceNet network model is pre-trained on the LFW (Labeled Faces in the Wild, face recognition public test set) data set, and then the parameters of the network model obtained after training are used as initialization parameters of the animal identity recognition model, where the initialization
  • the network in the animal identification model and the loss function are the same as the FaceNet network model pre-trained on the public face recognition test set described above.
  • the animal identification used in the first extraction unit 20 can be obtained. model.
  • the facial feature is a feature vector
  • the first extraction unit 20 uses the animal identification model to extract the facial feature of the first animal face image.
  • the face image of the first animal is input into the animal identification model.
  • the last layer of the animal identification model is a loss function layer.
  • the input data of the loss function layer is the above facial features, and the output result is the Euclidean distance between different facial features. Therefore, only the output vector of the layer before the loss function layer needs to be extracted to obtain the facial features of the first animal face image.
  • the matching unit 30 is configured to match an animal facial feature corresponding to a facial feature of the first animal facial image in a feature database.
  • the feature library stores in advance the animal facial features of the preset animals and their corresponding identity information.
  • a plurality of preset animal face features (also a feature vector) and identity information corresponding to each animal face feature are pre-stored in the feature database.
  • the identity information corresponding to the above 100 cows and the facial features of each cow can be pre-stored in the feature database.
  • the facial features of the above-mentioned first animal facial image are obtained, and the matching unit 30 matches them with the facial features of the animals stored in the feature database. If the matching facial features of the animals are matched, It indicates that the first animal is the identity of the animal corresponding to the facial features of the matched animal. If the matching animal facial features cannot be matched, the first animal cannot be identified. In the scenario of a smart farm, it indicates that the first animal is not an animal raised in the farm.
  • the second obtaining unit 40 is configured to obtain identity information corresponding to the matched facial features of the animal if the matching is successful, so as to identify the identity of the first animal.
  • the above-mentioned feature database stores identity information corresponding to each animal's facial features one by one.
  • the second obtaining unit 40 can obtain the identity information corresponding to the facial feature of the animal, and the identity information is the first Animal identity information.
  • the identity of the above-mentioned first animal is simply and effectively recognized, and it is matched and identified through facial features with high accuracy.
  • the matching unit 30 includes:
  • a calculation module configured to calculate an Euclidean distance between an animal face feature in the feature library and a face feature of the first animal face image through the animal identification model;
  • a selection module configured to select an animal facial feature having the smallest Euclidean distance from the facial feature of the first animal and smaller than a preset value, as the facial feature corresponding to the facial feature of the first animal Animal facial features.
  • the calculation module inputs the facial features of the first animal face image into the animal identification model, and simultaneously inputs the animal facial features in the feature library in order, and sequentially recognizes them through the animal identity.
  • the loss function layer of the model calculates the Euclidean distance between the facial features of the first animal face image and the animal facial features in the feature library. In this animal identification model, the smaller the Euclidean distance, the two faces are explained The more similar the features are.
  • the selection module selects an animal facial feature with the smallest Euclidean distance from the facial features of the first animal, and when the Euclidean distance is less than a preset value, the animal's facial features and the first animal can be considered And the facial features of the same animal feature, the identity information corresponding to the facial features of the animal can be used as the identity information of the first animal.
  • the FaceNet network model is a network model for face recognition. It extracts features during training to make the Euclidean distance between the faces of the same person closer, and the Euclidean distance between different face images farther. Specifically, when training the FaceNet network model, three images are input at the same time each time, one is a picture A of a face to be predicted with labels, and the other two are auxiliary pictures B0, B1. B0 is a picture different from the face in picture A, and B1 is a different picture from the same person in picture A.
  • the above-mentioned animal identification device further includes:
  • a second extraction unit 11 configured to extract animal facial features of a preset animal by using an animal identification model trained based on a FaceNet network model
  • the storage unit 12 is configured to store animal facial features of the preset animal and corresponding identity information in the feature database.
  • the preset animals in this embodiment are all animals that may need to be identified, such as all animals insured by the user, and all animals raised in the smart farm.
  • the second extraction unit 11 extracts the cow face features of all the cows in the smart farm that need to be managed, and each cow has its corresponding cow identity.
  • the storage unit 12 stores the bull identity of each cow and its corresponding bull face feature in a feature database to form a bull face feature database.
  • the process of extracting the facial features of the animal by the second extraction unit 11 is the same as the process of extracting the facial features of the first animal face image by the first extraction unit 20, and the difference is only that the targeted facial images are different and will not be performed here To repeat.
  • the cow's face features are extracted and matched with the cow face features stored in the aforementioned cow face feature database to obtain the identity of the cow to be identified.
  • the above-mentioned animal identification device further includes:
  • a training unit 101 is configured to input sample pictures in a training set into a FaceNet network model for training to obtain the animal identity recognition model; wherein the FaceNet network model is a human face that has been trained based on a face recognition public test set. Identify the network model.
  • the training unit 101 iteratively inputs 16 groups of pictures in the training set into the FaceNet network model for training, where each group of pictures includes three cattle face pictures. Two of them come from the same cow, and the other is a different cow. If the two pictures are from the same cow, the label is 1, which indicates similarity; if they are from different cattle, the label is 0, which indicates dissimilarity.
  • the learning rate of the aforementioned FaceNet network model is set to 0.01, and the loss function is triplet loss.
  • the feature vectors of the three pictures in the set of pictures are extracted, and after calculation of the loss function layer, the two features corresponding to the two pictures with the label 1 are obtained.
  • the Euclidean distance between the vectors decreases (reduced to less than the preset value), and the Euclidean distance between the two feature vectors corresponding to the two pictures with the label 0 becomes larger (larger than the preset value) until After the Euclidean distance no longer changes, the training is completed, the training parameters are obtained, and the FaceNet network model from which the training parameters are obtained is used as the animal identification model for animal identification in the above embodiment.
  • the above-mentioned animal identification device further includes:
  • the testing unit 102 is configured to input the sample pictures in the test set into the animal identification model for verification.
  • the test unit 102 calculates by inputting the sample pictures in the test set into the animal identification model each time, and determines the two feature vectors corresponding to the two pictures with the label of 1. Whether the Euclidean distance between them is less than a preset value; at the same time, determine whether the Euclidean distance between two feature vectors corresponding to two pictures with a label of 0 is greater than another preset value.
  • the ratio of the number of sample pictures in the training set to the test set is 4: 1.
  • the process of making a training set and a test set with a large number of cattle pictures in an intelligent farm includes:
  • the picture data of 120 cows is divided into 120 categories, each category has about 100 pictures, and the cow face detection algorithm is used to detect the cow faces of all pictures to generate a cow face database of 120 cows.
  • Two pictures are randomly selected from the same cow, and one picture is selected from the other cow.
  • the three pictures form a set of photos in the training and test sets. Randomly selected pictures are 10,000 groups. Among them, if the two pictures are from the same cow, the label is 1, which means they are similar. If they are from different cattle, the label is 0, which means they are not similar. Finally, the above 10,000 groups of pictures are divided into a training set and a test set according to a ratio of 4: 1, that is, there are 8000 groups of photos in the training set and 2000 groups of images in the test set.
  • the above-mentioned animal identification device further includes:
  • a claim unit is configured to match an insurance policy corresponding to the identity of the first animal in a policy database according to the identity information of the first animal.
  • the first animal is a cow
  • the claimant uses the animal identification method to identify the cow
  • a policy corresponding to the identity of the first animal is matched in the policy database, and information such as a claim amount corresponding to the identity of the cow is obtained from the above policy to facilitate claim settlement.
  • the above-mentioned animal identification device further includes:
  • the management unit is configured to query, based on the identity information of the first animal, a breeding information corresponding to the identity information of the first animal in a farm breeding database.
  • the first animal is a cow raised in a smart farm.
  • the management unit queries the farming information corresponding to the identity of the cow in the farm breeding database, such as: breeding start time, breeding days, weight changes, etc. Weight, the weight change is determined according to the weight difference); or the historical photos of the cow can also be queried in the farm breeding database, so that managers can view the changes and so on.
  • the animal identity recognition device obtains a face image of a first animal to be recognized, and extracts the first animal face image by using an animal identity recognition model trained based on a FaceNet network model. Matching the facial features of the animal corresponding to the facial features of the first animal facial image in the feature database; obtaining the identity information corresponding to the matched facial features of the animal to identify the Describe the identity of the first animal. It overcomes the defect that animal identification cannot be performed simply and effectively at present, which facilitates the management of animals.
  • an embodiment of the present application further provides a computer device.
  • the computer device may be a server, and its internal structure may be as shown in FIG.
  • the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the computer design processor is used 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-readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer-readable instructions in a non-volatile storage medium.
  • the computer equipment database is used to store data such as the FaceNet network model.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by a processor to implement a method for identifying an animal identity.
  • the processor executes the steps of the animal identification method:
  • the feature database stores the animal facial features of the preset animals and their corresponding identity information in advance;
  • identity information corresponding to the facial features of the animal that are matched is obtained to identify the identity of the first animal.
  • the step of matching, by the processor, an animal facial feature corresponding to a facial feature of the first animal facial image in the feature database includes:
  • An animal facial feature having the smallest Euclidean distance to a facial feature of the first animal and smaller than a preset value is selected as the animal facial feature corresponding to the facial feature of the first animal.
  • the method before the step of obtaining the facial image of the first animal to be identified by the processor, the method includes:
  • the animal facial features of the preset animals and their corresponding identity information are stored in the feature database.
  • the processor before the step of extracting the animal facial features of the preset animal by the processor through the animal identification model trained based on the FaceNet network model, the processor includes:
  • the sample pictures in the training set are input to the FaceNet network model for training to obtain the animal identification model; wherein the FaceNet network model is a face recognition network model trained based on the face recognition public test set training.
  • the step of inputting the sample pictures in the training set into a FaceNet network model for training to obtain the animal identification model includes:
  • the sample pictures in the test set are input into the animal identification model for verification.
  • the step of acquiring, by the processor, the identity information corresponding to the facial features of the animal that are matched to identify the identity of the first animal includes:
  • a policy corresponding to the identity of the first animal is matched in the policy database.
  • the step of acquiring, by the processor, the identity information corresponding to the facial features of the animal that are matched to identify the identity of the first animal includes:
  • the farming database corresponding to the identity information of the first animal is queried in the farm breeding database.
  • FIG. 6 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • An embodiment of the present application further provides a computer non-volatile readable storage medium, which stores computer readable instructions, and the computer readable instructions implement a method for identifying an animal identity when executed by a processor, specifically:
  • the feature database stores the animal facial features of the preset animals and their corresponding identity information in advance;
  • identity information corresponding to the facial features of the animal that are matched is obtained to identify the identity of the first animal.
  • the step of matching, by the processor, an animal facial feature corresponding to a facial feature of the first animal facial image in the feature database includes:
  • An animal facial feature having the smallest Euclidean distance to a facial feature of the first animal and smaller than a preset value is selected as the animal facial feature corresponding to the facial feature of the first animal.
  • the method before the step of obtaining the facial image of the first animal to be identified by the processor, the method includes:
  • the animal facial features of the preset animals and their corresponding identity information are stored in the feature database.
  • the processor before the step of extracting the animal facial features of the preset animal by the processor through the animal identification model trained based on the FaceNet network model, the processor includes:
  • the sample pictures in the training set are input to the FaceNet network model for training to obtain the animal identification model; wherein the FaceNet network model is a face recognition network model trained based on the face recognition public test set training.
  • the step of inputting the sample pictures in the training set into a FaceNet network model for training to obtain the animal identification model includes:
  • the sample pictures in the test set are input into the animal identification model for verification.
  • the step of acquiring, by the processor, the identity information corresponding to the facial features of the animal that are matched to identify the identity of the first animal includes:
  • a policy corresponding to the identity of the first animal is matched in the policy database.
  • the step of acquiring, by the processor, the identity information corresponding to the facial features of the animal that are matched to identify the identity of the first animal includes:
  • the farming database corresponding to the identity information of the first animal is queried in the farm breeding database.
  • the animal identification method, device, computer equipment, and storage medium provided in the embodiments of the present application are used to obtain the facial image of the first animal to be identified, and the animal identification obtained through training based on the FaceNet network model
  • the model extracts the facial features of the first animal face image; matches the animal facial features corresponding to the facial features of the first animal face image in a feature database; and obtains the matched animal face Identity information corresponding to the feature to identify the identity of the first animal. It overcomes the defect that animal identification cannot be performed simply and effectively at present, which facilitates the management of animals.
  • Non-volatile memory may 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.
  • RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), two-speed data rate SDRAM (SSRSDRAM), 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), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)
  • Collating Specific Patterns (AREA)

Abstract

一种动物身份的识别方法、装置、计算机设备和存储介质,包括:通过基于FaceNet网络模型训练得到的动物身份识别模型提取待识别的第一动物脸部图像的脸部特征;在特征库中匹配与脸部特征相对应的动物脸部特征及其所对应的身份信息,以识别所述第一动物的身份。克服了目前无法简单有效进行动物身份识别的缺陷。

Description

动物身份的识别方法、装置、计算机设备和存储介质
本申请要求于2018年6月1日提交中国专利局、申请号为2018105560506,发明名称为“动物身份的识别方法、装置、计算机设备和存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域,特别涉及一种动物身份的识别方法、装置、计算机设备和存储介质。
背景技术
当前,身份识别通常是对人脸进行识别,从而识别人物身份。对动物进行身份识别却很少见。而目前,智能农场等行业中,需要对牛、马等动物进行身份识别,以便对动物进行管理;在保险行业,需要对牛、马等动物进行身份识别,以便对动物进行投保等。
目前还无法简单有效地进行牛、马等动物身份的识别,不便于对牛、马等动物进行管理。
技术问题
本申请的主要目的为提供一种动物身份的识别方法、装置、计算机设备和存储介质,克服目前无法简单有效进行动物身份识别的缺陷。
技术解决方案
为实现上述目的,本申请提供一种动物身份的识别方法,包括以下步骤:
获取待识别的第一动物的脸部图像;
通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
本申请还提供了一种动物身份的识别装置,包括:
第一获取单元,用于获取待识别的第一动物的脸部图像;
第一提取单元,用于通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
匹配单元,用于在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
第二获取单元,用于若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
本申请还提供一种计算机设备,包括存储器和处理器,所述存储器存储有计算机可读指令,所述处理器执行所述计算机可读指令时实现上述任一项所述方法的步骤。
本申请还提供一种计算机非易失性可读存储介质,其上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现上述任一项所述的方法的步骤。
有益效果
本申请中提供的动物身份的识别方法、装置、计算机设备和存储介质,获取待识别的第一动物的脸部图像,通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。克服了目前无法简单有效进行动物身份识别的缺陷,便于对动物进行管理。
附图说明
图1 是本申请一实施例中动物身份的识别方法步骤示意图;
图2 是本申请另一实施例中动物身份的识别方法步骤示意图;
图3 是本申请一实施例中动物身份的识别装置结构框图;
图4 是本申请另一实施例中动物身份的识别装置结构框图;
图5 是本申请又一实施例中动物身份的识别装置结构框图;
图6 为本申请一实施例的计算机设备的结构示意框图。
本发明的最佳实施方式
参照图1,本申请实施例中提供了一种动物身份的识别方法,包括以下步骤:
步骤S1,获取待识别的第一动物的脸部图像。
在本实施例中,在保险行业、智能农场行业,通常需要对动物进行身份识别。上述第一动物为需要识别身份的动物,该动物包括牛、马等,本实施例中主要以牛为例进行阐述。在对第一动物进行身份识别时,首先获取该第一动物的脸部图像,本实施例中是以动物的脸部图像特征来识别动物。
上述步骤S1具体包括:获取第一动物的图像,并通过图像检测算法从所述第一动物的图像中检测出所述第一动物的脸部图像。在一具体实施例中,上述第一动物为待识别身份的牛,则上述步骤S1具体为:获取待识别的牛的图像,并通过牛脸检测算法对所述图像进行牛脸检测,以检测出牛的脸部图像。上述牛脸检测算法为一种检测牛脸轮廓信息的算法,其包括监督式梯度下降算法(SDM)、局部二值算法(LBF)等。
步骤S2,通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征。
本实施例中,上述动物身份识别模型是基于FaceNet网络模型训练得到的动物身份识别模型,FaceNet网络模型通常用于人脸识别。本实施例中,将用于人脸识别的FaceNet网络模型迁移到动物脸部识别中。具体地,在LFW(Labeled Faces in the Wild,人脸识别公开测试集)数据集上预先训练FaceNet网络模型,然后将训练完成得到的网络模型的参数作为动物身份识别模型的初始化参数,其中,初始化的动物身份识别模型中的网络以及损失函数均与上述在人脸识别公开测试集上预先训练过的FaceNet网络模型相同。
本实施例中,将上述FaceNet网络模型迁移到初始化的动物身份识别模型中之后,再使用动物的训练集训练上述初始化的动物身份识别模型,则可以得到本步骤S2中的动物身份识别模型。
本实施例中,上述脸部特征为一特征向量,通过上述动物身份识别模型去提取第一动物脸部图像的脸部特征。具体地,将上述第一动物的脸部图像输入至上述动物身份识别模型中。上述动物身份识别模型的最后一层为损失函数层,该损失函数层的输入数据即为上述脸部特征,其输出结果为不同的脸部特征之间的欧式距离。因此,只需要提取出上述损失函数层之前一层的输出向量,则可以得到上述第一动物脸部图像的脸部特征。
步骤S3,在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征。
本实施例中,在特征库中预存有多个预设动物的动物脸部特征(也是一个特征向量)以及与每个动物脸部特征一一所对应的身份信息。例如,在某个智能农场中,饲养了100头牛,则可以在该特征库中预存有上述100头牛对应的身份信息以及每一头牛的脸部特征。
在识别第一动物的身份时,得到上述第一动物脸部图像的脸部特征,将其与特征库中存储的动物脸部特征进行匹配,若匹配到一致的动物脸部特征,则表明该第一动物即为该匹配出的动物脸部特征对应的动物的身份。若匹配不到一致的动物脸部特征,则无法识别出第一动物。在智能农场的场景中,则表明该第一动物不是该农场中饲养的动物。
步骤S4,若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
本实施例中,上述特征库中存储有与每个动物脸部特征一一所对应的身份信息。当匹配到与第一动物脸部图像的脸部特征一致的动物脸部特征,则可以获取到该动物脸部特征所对应的身份信息,而该身份信息即为上述第一动物的身份信息。综上,则简单有效地识别出上述第一动物的身份,且其是通过脸部特征进行匹配识别,准确率高。
在上述步骤S3的匹配过程中,其匹配的是两个特征向量之间的相似性。具体地,在一个实施例中,上述在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征的步骤S3,包括:
步骤S31,通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
步骤S32,选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
在本实施例中,将上述第一动物脸部图像的脸部特征输入至上述动物身份识别模型中,同时,依次输入上述特征库中的动物脸部特征,并依次通过上述动物身份识别模型的损失函数层计算第一动物脸部图像的脸部特征与特征库中的动物脸部特征之间的欧式距离,在该动物身份识别模型中,欧式距离越小,则说明两个脸部特征之间越相似。选择出一个与所述第一动物的脸部特征之间的欧式距离最小的动物脸部特征,且当该欧式距离小于预设值时,则可以认为该动物脸部特征与第一动物的脸部特征为同一个特征,则可以将该动物脸部特征所对应的身份信息作为上述第一动物的身份信息。
为了便于对上述动物身份识别模型计算欧式距离判断两个特征是否相似的过程进行理解,现对FaceNet网络模型进行阐述。
FaceNet网络模型为对人脸进行识别的网络模型,其在训练时对图像进行特征提取,让同一个人脸图片之间的欧式距离更近,不同人脸图片之间的欧式距离更远。具体地,在训练FaceNet网络模型时,每次都同时输入三张图片,一张为有标签的需要预测的人脸的图片A,另外两张为辅助图片B0,B1。其中B0是不同于图片A中的人脸的图片,B1是与图片A中同一个人的不同图片。提取上述三张图片的特征向量,并计算A与B0、B1之间的欧式距离,同时使得A与B0之间的欧式距离变大,而A与B1之间的欧式距离减少,以训练上述FaceNet网络模型。使用人脸识别公开测试集中的数据集训练上述FaceNet网络模型之后,将训练完成得到的FaceNet网络模型的参数作为动物身份识别模型的初始化参数,其中,初始化的动物身份识别模型中的网络以及损失函数均与上述在人脸识别公开测试集上预先训练过的FaceNet网络模型相同。最后,使用动物的训练集训练上述初始化的动物身份识别模型,则可以得到上述步骤S2中的动物身份识别模型。
参照图2,在一实施例中,上述获取待识别的第一动物的脸部图像的步骤S1之前,包括:
步骤S11,通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征;
步骤S12,将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
在保险行业、智能农场等行业,若要对某一只动物进行准确识别,以识别出其身份信息,则需要预先在特征库中存储有该动物所对应动物脸部特征以及其对应的身份信息。本实施例中的预设动物为所有可能需要用到身份识别的动物,例如用户投保的所有动物,智能农场中饲养的所有动物。
具体一实施例中,如上述步骤S11所述,提取智能农场内所有需要管理的牛的牛脸特征,每一只牛具有其相应的牛身份,每一只牛的牛身份与其对应的牛脸特征存储在特征库中,形成牛脸特征库。上述步骤S11中提取动物脸部特征的过程与上述步骤S2中提取第一动物脸部图像的脸部特征的过程相同,区别仅在于针对的脸部图像不同,在此不再进行赘述。当需要识别智能农场中牛身份的时候,则提取牛的牛脸特征与上述牛脸特征库中预存的牛脸特征进行匹配,便可以获取出需要识别的牛的身份。
在一实施例中,上述通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征的步骤S11之前,包括:
步骤S101,将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型;其中,所述FaceNet网络模型为基于人脸识别公开测试集训练完成的人脸识别网络模型。
在本实施例中,以牛图片作为训练集进行训练为例,每次将训练集中的16组图片迭代输入至FaceNet网络模型中进行训练,其中每组图片包含三张牛脸图片,其中两张来于同一种牛,另一张为不同的牛,若两张图片来自同一头牛,标签为1,代表相似;若来自不同的牛,标签为0,代表不相似。上述FaceNet网络模型的学习率设置为0.01,损失函数为triplet loss。具体地,将训练集中图片输入至FaceNet网络模型中进行训练时,提取该组图片中的三张图片的特征向量,经过损失函数层计算之后,使得标签为1的两张图片对应的两个特征向量之间的欧式距离减小(减小到小于预设值),同时标签为0的两张图片对应的两个特征向量之间的欧式距离变大(变大到大于预设值),直到欧式距离不再变化之后,则训练完成,得到训练参数,将得到训练参数的FaceNet网络模型作为上述实施例中进行动物身份识别的动物身份识别模型。
在一实施例中,上述将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型的步骤S101之后,包括:
步骤S102,将测试集中的样本图片输入至所述动物身份识别模型中进行验证。
为了验证上述训练的动物身份识别模型的检测准确率,每次将测试集中的样本图片输入至所述动物身份识别模型中计算,判断标签为1的两张图片对应的两个特征向量之间的欧式距离是否小于预设值;同时,判断标签为0的两张图片对应的两个特征向量之间的欧式距离大于另一个预设值。
在一实施例中,上述训练集与测试集中的样本图片数量比例为4:1。在一个具体实施例中,应用于智能农场中以大量牛图片制作训练集以及测试集的过程包括:
将120头牛的图片数据分为120类,每一类有约100张图片,使用牛脸检测算法检测所有图片的牛脸,生成120头牛的牛脸数据库。从同一只牛中随机选出两张图片,另一只牛中选择一张图片,三张图片组成训练集中和测试集中的一组照片,随机选取图片为10000组。其中,若两张图片来自同一头牛,标签为1,代表相似。若来自不同的牛,标签为0,代表不相似。最后,按照4:1的比例将上述10000组图片分为训练集和测试集,即训练集中有8000组照片,测试集中有2000组图片。
在一实施例中,上述获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤S4之后,包括:
步骤S5a,根据所述第一动物的身份信息,在保单数据库匹配与所述第一动物的身份相对应的保单。
具体地,在保险理赔场景中,上述第一动物为牛,用户需要对其投保的牛进行理赔时,上传牛的图像,理赔方采用上述动物身份的识别方法进行牛的身份识别;然后如步骤S5a所述,在保单数据库中匹配与所述第一动物的身份相对应的保单,并从上述保单中获取到对应于该牛身份的理赔金额等信息,方便进行理赔。
在另一实施例中,上述获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤S4之后,包括:
步骤S5b,根据所述第一动物的身份信息,在农场养殖数据库中查询出与所述第一动物的身份信息相对应的养殖信息。
具体地,在智能农场应用场景中,上述第一动物为智能农场中饲养的牛,为了对每一只牛进行管理,在对牛进行状态跟踪时,需要识别牛的身份;如采集牛的图像,并使用上述方法识别牛的身份,在农场养殖数据库中查询与该牛身份对应的养殖信息,如:养殖开始时间、养殖天数、体重变化等(可以在采集牛图像时采集牛的体重,根据体重差确定体重变化);或者还可以在农场养殖数据库中查询到该牛的历史照片,便于管理人员查看变化等。
综上所述,为本申请实施例中提供的动物身份的识别方法,获取待识别的第一动物的脸部图像,通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。克服了目前无法简单有效进行动物身份识别的缺陷,便于对动物进行管理。
参照图3,本申请一实施例中还提供了一种动物身份的识别装置,包括:
第一获取单元10,用于获取待识别的第一动物的脸部图像。
在本实施例中,在保险行业、智能农场行业,通常需要对动物进行身份识别。上述第一动物为需要识别身份的动物,该动物包括牛、马等,本实施例中主要以牛为例进行阐述。在对第一动物进行身份识别时,首先获取该第一动物的脸部图像,本实施例中是以动物的脸部图像特征来识别动物。
上述第一获取单元10具体用于获取第一动物的图像,并通过图像检测算法从所述第一动物的图像中检测出所述第一动物的脸部图像。在一具体实施例中,上述第一动物为待识别身份的牛,则上述步骤S1具体为:获取待识别的牛的图像,并通过牛脸检测算法对所述图像进行牛脸检测,以检测出牛的脸部图像。上述牛脸检测算法为一种检测牛脸轮廓信息的算法,其包括监督式梯度下降算法(SDM)、局部二值算法(LBF)等。
第一提取单元20,用于通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征。
本实施例中,上述动物身份识别模型是基于FaceNet网络模型训练得到的动物身份识别模型,FaceNet网络模型通常用于人脸识别。本实施例中,将用于人脸识别的FaceNet网络模型迁移到动物脸部识别中。具体地,在LFW(Labeled Faces in the Wild,人脸识别公开测试集)数据集上预先训练FaceNet网络模型,然后将训练完成得到的网络模型的参数作为动物身份识别模型的初始化参数,其中,初始化的动物身份识别模型中的网络以及损失函数均与上述在人脸识别公开测试集上预先训练过的FaceNet网络模型相同。
本实施例中,将上述FaceNet网络模型迁移到初始化的动物身份识别模型中之后,再使用动物的训练集训练上述初始化的动物身份识别模型,则可以得到第一提取单元20中使用的动物身份识别模型。
本实施例中,上述脸部特征为一特征向量,第一提取单元20通过上述动物身份识别模型去提取第一动物脸部图像的脸部特征。具体地,将上述第一动物的脸部图像输入至上述动物身份识别模型中。上述动物身份识别模型的最后一层为损失函数层,该损失函数层的输入数据即为上述脸部特征,其输出结果为不同的脸部特征之间的欧式距离。因此,只需要提取出上述损失函数层之前一层的输出向量,则可以得到上述第一动物脸部图像的脸部特征。
匹配单元30,用于在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征。所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息。
本实施例中,在特征库中预存有多个预设动物的动物脸部特征(也是一个特征向量)以及与每个动物脸部特征一一所对应的身份信息。例如,在某个智能农场中,饲养了100头牛,则可以在该特征库中预存有上述100头牛对应的身份信息以及每一头牛的脸部特征。
在识别第一动物的身份时,得到上述第一动物脸部图像的脸部特征,匹配单元30将其与特征库中存储的动物脸部特征进行匹配,若匹配到一致的动物脸部特征,则表明该第一动物即为该匹配出的动物脸部特征对应的动物的身份。若匹配不到一致的动物脸部特征,则无法识别出第一动物。在智能农场的场景中,则表明该第一动物不是该农场中饲养的动物。
第二获取单元40,用于若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
本实施例中,上述特征库中存储有与每个动物脸部特征一一所对应的身份信息。当匹配到与第一动物脸部图像的脸部特征一致的动物脸部特征,第二获取单元40则可以获取到该动物脸部特征所对应的身份信息,而该身份信息即为上述第一动物的身份信息。综上,则简单有效地识别出上述第一动物的身份,且其是通过脸部特征进行匹配识别,准确率高。
在一实施例中,上述匹配单元30包括:
计算模块,用于通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
选择模块,用于选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
在本实施例中,计算模块将上述第一动物脸部图像的脸部特征输入至上述动物身份识别模型中,同时,依次输入上述特征库中的动物脸部特征,并依次通过上述动物身份识别模型的损失函数层计算第一动物脸部图像的脸部特征与特征库中的动物脸部特征之间的欧式距离,在该动物身份识别模型中,欧式距离越小,则说明两个脸部特征之间越相似。选择模块选择出一个与所述第一动物的脸部特征之间的欧式距离最小的动物脸部特征,且当该欧式距离小于预设值时,则可以认为该动物脸部特征与第一动物的脸部特征为同一个特征,则可以将该动物脸部特征所对应的身份信息作为上述第一动物的身份信息。
为了便于对上述动物身份识别模型计算欧式距离判断两个特征是否相似的过程进行理解,现对FaceNet网络模型进行阐述。
FaceNet网络模型为对人脸进行识别的网络模型,其在训练时对图像进行特征提取,让同一个人脸图片之间的欧式距离更近,不同人脸图片之间的欧式距离更远。具体地,在训练FaceNet网络模型时,每次都同时输入三张图片,一张为有标签的需要预测的人脸的图片A,另外两张为辅助图片B0,B1。其中B0是不同于图片A中的人脸的图片,B1是与图片A中同一个人的不同图片。提取上述三张图片的特征向量,并计算A与B0、B1之间的欧式距离,同时使得A与B0之间的欧式距离变大,而A与B1之间的欧式距离减少,以训练上述FaceNet网络模型。使用人脸识别公开测试集中的数据集训练上述FaceNet网络模型之后,将训练完成得到的FaceNet网络模型的参数作为动物身份识别模型的初始化参数,其中,初始化的动物身份识别模型中的网络以及损失函数均与上述在人脸识别公开测试集上预先训练过的FaceNet网络模型相同。最后,使用动物的训练集训练上述初始化的动物身份识别模型,则可以得到上述第一提取单元20中使用的动物身份识别模型。
参照图4,在一实施例中,上述动物身份的识别装置还包括:
第二提取单元11,用于通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征;
存储单元12,用于将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
在保险行业、智能农场等行业,若要对某一只动物进行准确识别,以识别出其身份信息,则需要预先在特征库中存储有该动物所对应动物脸部特征以及其对应的身份信息。本实施例中的预设动物为所有可能需要用到身份识别的动物,例如用户投保的所有动物,智能农场中饲养的所有动物。
具体一实施例中,如上述第二提取单元11,提取智能农场内所有需要管理的牛的牛脸特征,每一只牛具有其相应的牛身份。存储单元12将每一只牛的牛身份与其对应的牛脸特征存储在特征库中,形成牛脸特征库。上述第二提取单元11提取动物脸部特征的过程与上述第一提取单元20提取第一动物脸部图像的脸部特征的过程相同,区别仅在于针对的脸部图像不同,在此不再进行赘述。当需要识别智能农场中牛身份的时候,则提取牛的牛脸特征与上述牛脸特征库中预存的牛脸特征进行匹配,便可以获取出需要识别的牛的身份。
参照图5,在一实施例中,上述动物身份的识别装置还包括:
训练单元101,用于将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型;其中,所述FaceNet网络模型为基于人脸识别公开测试集训练完成的人脸识别网络模型。
在本实施例中,以牛图片作为训练集进行训练为例,训练单元101每次将训练集中的16组图片迭代输入至FaceNet网络模型中进行训练,其中每组图片包含三张牛脸图片,其中两张来于同一种牛,另一张为不同的牛,若两张图片来自同一头牛,标签为1,代表相似;若来自不同的牛,标签为0,代表不相似。上述FaceNet网络模型的学习率设置为0.01,损失函数为triplet loss。具体地,将训练集中图片输入至FaceNet网络模型中进行训练时,提取该组图片中的三张图片的特征向量,经过损失函数层计算之后,使得标签为1的两张图片对应的两个特征向量之间的欧式距离减小(减小到小于预设值),同时标签为0的两张图片对应的两个特征向量之间的欧式距离变大(变大到大于预设值),直到欧式距离不再变化之后,则训练完成,得到训练参数,将得到训练参数的FaceNet网络模型作为上述实施例中进行动物身份识别的动物身份识别模型。
在一实施例中,上述动物身份的识别装置还包括:
测试单元102,用于将测试集中的样本图片输入至所述动物身份识别模型中进行验证。
为了验证上述训练的动物身份识别模型的检测准确率,测试单元102每次将测试集中的样本图片输入至所述动物身份识别模型中计算,判断标签为1的两张图片对应的两个特征向量之间的欧式距离是否小于预设值;同时,判断标签为0的两张图片对应的两个特征向量之间的欧式距离大于另一个预设值。
在一实施例中,上述训练集与测试集中的样本图片数量比例为4:1。在一个具体实施例中,应用于智能农场中以大量牛图片制作训练集以及测试集的过程包括:
将120头牛的图片数据分为120类,每一类有约100张图片,使用牛脸检测算法检测所有图片的牛脸,生成120头牛的牛脸数据库。从同一只牛中随机选出两张图片,另一只牛中选择一张图片,三张图片组成训练集中和测试集中的一组照片,随机选取图片为10000组。其中,若两张图片来自同一头牛,标签为1,代表相似。若来自不同的牛,标签为0,代表不相似。最后,按照4:1的比例将上述10000组图片分为训练集和测试集,即训练集中有8000组照片,测试集中有2000组图片。
在一实施例中,上述动物身份的识别装置还包括:
理赔单元,用于根据所述第一动物的身份信息,在保单数据库匹配与所述第一动物的身份相对应的保单。
具体地,在保险理赔场景中,上述第一动物为牛,用户需要对其投保的牛进行理赔时,上传牛的图像,理赔方采用上述动物身份的识别方法进行牛的身份识别;然后如理赔单元所述,在保单数据库中匹配与所述第一动物的身份相对应的保单,并从上述保单中获取到对应于该牛身份的理赔金额等信息,方便进行理赔。
在一实施例中,上述动物身份的识别装置还包括:
管理单元,用于根据所述第一动物的身份信息,在农场养殖数据库中查询出与所述第一动物的身份信息相对应的养殖信息。
具体地,在智能农场应用场景中,上述第一动物为智能农场中饲养的牛,为了对每一只牛进行管理,在对牛进行状态跟踪时,需要识别牛的身份;如采集牛的图像,并使用上述方法识别牛的身份;然后,管理单元在农场养殖数据库中查询与该牛身份对应的养殖信息,如:养殖开始时间、养殖天数、体重变化等(可以在采集牛图像时采集牛的体重,根据体重差确定体重变化);或者还可以在农场养殖数据库中查询到该牛的历史照片,便于管理人员查看变化等。
综上所述,为本申请中提供的动物身份的识别装置,获取待识别的第一动物的脸部图像,通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。克服了目前无法简单有效进行动物身份识别的缺陷,便于对动物进行管理。
参照图6,本申请实施例中还提供一种计算机设备,该计算机设备可以是服务器,其内部结构可以如图6所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中,该计算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储FaceNet网络模型等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种动物身份的识别方法。
上述处理器执行上述动物身份的识别方法的步骤:
获取待识别的第一动物的脸部图像;
通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
在一实施例中,上述处理器在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征的步骤,包括:
通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
在一实施例中,上述处理器获取待识别的第一动物的脸部图像的步骤之前,包括:
通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征;
将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
在一实施例中,上述处理器通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征的步骤之前,包括:
将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型;其中,所述FaceNet网络模型为基于人脸识别公开测试集训练完成的人脸识别网络模型。
在一实施例中,上述处理器将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型的步骤之后,包括:
将测试集中的样本图片输入至所述动物身份识别模型中进行验证。
在一实施例中,上述处理器获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤之后,包括:
根据所述第一动物的身份信息,在保单数据库匹配与所述第一动物的身份相对应的保单。
在一实施例中,上述处理器获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤之后,包括:
根据所述第一动物的身份信息,在农场养殖数据库中查询出与所述第一动物的身份信息相对应的养殖信息。
本领域技术人员可以理解,图6中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备的限定。
本申请一实施例还提供一种计算机非易失性可读存储介质,其上存储有计算机可读指令,计算机可读指令被处理器执行时实现一种动物身份的识别方法,具体为:
获取待识别的第一动物的脸部图像;
通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
在一实施例中,上述处理器在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征的步骤,包括:
通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
在一实施例中,上述处理器获取待识别的第一动物的脸部图像的步骤之前,包括:
通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征;
将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
在一实施例中,上述处理器通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征的步骤之前,包括:
将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型;其中,所述FaceNet网络模型为基于人脸识别公开测试集训练完成的人脸识别网络模型。
在一实施例中,上述处理器将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型的步骤之后,包括:
将测试集中的样本图片输入至所述动物身份识别模型中进行验证。
在一实施例中,上述处理器获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤之后,包括:
根据所述第一动物的身份信息,在保单数据库匹配与所述第一动物的身份相对应的保单。
在一实施例中,上述处理器获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤之后,包括:
根据所述第一动物的身份信息,在农场养殖数据库中查询出与所述第一动物的身份信息相对应的养殖信息。
综上所述,为本申请实施例中提供的动物身份的识别方法、装置、计算机设备和存储介质,获取待识别的第一动物的脸部图像,通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。克服了目前无法简单有效进行动物身份识别的缺陷,便于对动物进行管理。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机可读指令来指令相关的硬件来完成,所述的计算机可读指令可存储与一非易失性计算机可读取存储介质中,该计算机可读指令在执行时,可包括如上述各方法的实施例的流程。其中,本申请所提供的和实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和/或易失性存储器。非易失性存储器可以包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限,RAM通过多种形式可得,诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双速据率SDRAM(SSRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其它变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素,而且还包括没有明确列出的其它要素,或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
以上所述仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其它相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种动物身份的识别方法,其特征在于,包括以下步骤:
    获取待识别的第一动物的脸部图像;
    通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
    在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
    若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
  2. 根据权利要求1所述的动物身份的识别方法,其特征在于,所述在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征的步骤,包括:
    通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
    选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
  3. 根据权利要求1所述的动物身份的识别方法,其特征在于,所述获取待识别的第一动物的脸部图像的步骤之前,包括:
    通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述预设动物的动物脸部特征;
    将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
  4. 根据权利要求3所述的动物身份的识别方法,其特征在于,所述通过基于FaceNet网络模型训练得到的动物身份识别模型提取预设动物的动物脸部特征的步骤之前,包括:
    将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型;其中,所述FaceNet网络模型为基于人脸识别公开测试集训练完成的人脸识别网络模型。
  5. 根据权利要求4所述的动物身份的识别方法,其特征在于,所述将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型的步骤之后,包括:
    将测试集中的样本图片输入至所述动物身份识别模型中进行验证。
  6. 根据权利要求1所述的动物身份的识别方法,其特征在于,所述获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤之后,包括:
    根据所述第一动物的身份信息,在保单数据库匹配与所述第一动物的身份相对应的保单。
  7. 根据权利要求1所述的动物身份的识别方法,其特征在于,所述获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份的步骤之后,包括:
    根据所述第一动物的身份信息,在农场养殖数据库中查询出与所述第一动物的身份信息相对应的养殖信息。
  8. 一种动物身份的识别装置,其特征在于,包括:
    第一获取单元,用于获取待识别的第一动物的脸部图像;
    第一提取单元,用于通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
    匹配单元,用于在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
    第二获取单元,用于若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
  9. 根据权利要求8所述的动物身份的识别装置,其特征在于,所述匹配单元包括:
    计算模块,用于通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
    选择模块,用于选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
  10. 根据权利要求8所述的动物身份的识别装置,其特征在于,还包括:
    第二提取单元,用于通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述预设动物的动物脸部特征;
    存储单元,用于将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
  11. 根据权利要求10所述的动物身份的识别装置,其特征在于,还包括:
    训练单元,用于将训练集中的样本图片输入至FaceNet网络模型中进行训练,以得到所述动物身份识别模型;其中,所述FaceNet网络模型为基于人脸识别公开测试集训练完成的人脸识别网络模型。
  12. 根据权利要求11所述的动物身份的识别装置,其特征在于,还包括:
    测试单元,用于将测试集中的样本图片输入至所述动物身份识别模型中进行验证。
  13. 根据权利要求8所述的动物身份的识别装置,其特征在于,还包括:
    理赔单元,用于根据所述第一动物的身份信息,在保单数据库匹配与所述第一动物的身份相对应的保单。
  14. 根据权利要求8所述的动物身份的识别装置,其特征在于,还包括:
    管理单元,用于根据所述第一动物的身份信息,在农场养殖数据库中查询出与所述第一动物的身份信息相对应的养殖信息。
  15. 一种计算机设备,包括存储器和处理器,所述存储器中存储有计算机可读指令,其特征在于,所述处理器执行所述计算机可读指令时实现动物身份的识别方法,所述方法包括:
    获取待识别的第一动物的脸部图像;
    通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
    在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
    若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
  16. 根据权利要求15所述的计算机设备,其特征在于,所述处理器在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征的步骤,包括:
    通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
    选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
  17. 根据权利要求15所述的计算机设备,其特征在于,所述处理器获取待识别的第一动物的脸部图像的步骤之前,包括:
    通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述预设动物的动物脸部特征;
    将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
  18. 一种计算机非易失性可读存储介质,其上存储有计算机可读指令,其特征在于,所述计算机可读指令被处理器执行时实现动物身份的识别,所述方法包括:
    获取待识别的第一动物的脸部图像;
    通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述第一动物脸部图像的脸部特征;
    在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征;所述特征库中预先存储有预设动物的动物脸部特征及其对应的身份信息;
    若匹配成功,获取匹配出的所述动物脸部特征所对应的身份信息,以识别所述第一动物的身份。
  19. 根据权利要求18所述的计算机非易失性可读存储介质,其特征在于,所述处理器在特征库中匹配与所述第一动物脸部图像的脸部特征相对应的动物脸部特征的步骤,包括:
    通过所述动物身份识别模型计算所述特征库中的动物脸部特征与所述第一动物脸部图像的脸部特征之间的欧式距离;
    选择出与所述第一动物的脸部特征之间的欧式距离最小且小于预设值的动物脸部特征,作为与所述第一动物的脸部特征相对应的所述动物脸部特征。
  20. 根据权利要求18所述的计算机非易失性可读存储介质,其特征在于,所述处理器获取待识别的第一动物的脸部图像的步骤之前,包括:
    通过基于FaceNet网络模型训练得到的动物身份识别模型提取所述预设动物的动物脸部特征;
    将所述预设动物的动物脸部特征及其对应的身份信息存储于所述特征库中。
PCT/CN2018/095667 2018-06-01 2018-07-13 动物身份的识别方法、装置、计算机设备和存储介质 WO2019227616A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810556050.6A CN108921026A (zh) 2018-06-01 2018-06-01 动物身份的识别方法、装置、计算机设备和存储介质
CN201810556050.6 2018-06-01

Publications (1)

Publication Number Publication Date
WO2019227616A1 true WO2019227616A1 (zh) 2019-12-05

Family

ID=64418078

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/095667 WO2019227616A1 (zh) 2018-06-01 2018-07-13 动物身份的识别方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN108921026A (zh)
WO (1) WO2019227616A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111310703A (zh) * 2020-02-26 2020-06-19 深圳市巨星网络技术有限公司 基于卷积神经网络的身份识别方法、装置、设备及介质

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325059A (zh) * 2018-12-14 2020-06-23 技嘉科技股份有限公司 脸部识别方法、装置及计算机可读取媒体
CN109829381A (zh) * 2018-12-28 2019-05-31 北京旷视科技有限公司 一种犬只识别管理方法、装置、系统及存储介质
CN109886145B (zh) * 2019-01-29 2021-09-03 浙江泽曦科技有限公司 宠物脸部识别方法及系统
CN109948458A (zh) * 2019-02-25 2019-06-28 广东智媒云图科技股份有限公司 基于鼻纹的宠物身份识别方法、装置、设备及存储介质
CN110222725B (zh) * 2019-05-15 2022-01-04 广州影子科技有限公司 猪只核验方法、猪只核验装置与猪只核验系统
CN110334593A (zh) * 2019-05-28 2019-10-15 浙江泽曦科技有限公司 宠物脸部识别方法及系统
CN110610125A (zh) * 2019-07-31 2019-12-24 平安科技(深圳)有限公司 基于神经网络的牛脸识别方法、装置、设备及存储介质
CN110765849B (zh) * 2019-09-09 2024-04-09 中国平安财产保险股份有限公司 基于微表情的身份信息采集方法、装置和计算机设备
CN110929099B (zh) * 2019-11-28 2023-07-21 杭州小影创新科技股份有限公司 一种基于多任务学习的短视频帧语义提取方法及系统
CN110956149A (zh) * 2019-12-06 2020-04-03 中国平安财产保险股份有限公司 宠物身份核验方法、装置、设备及计算机可读存储介质
CN111523479A (zh) * 2020-04-24 2020-08-11 中国农业科学院农业信息研究所 动物的生物特征识别方法、装置、计算机设备及存储介质
CN111985477B (zh) * 2020-08-27 2024-06-28 平安科技(深圳)有限公司 基于单目摄像头的动物体在线核赔方法、装置及存储介质
CN112132026B (zh) * 2020-09-22 2024-07-05 深圳赛安特技术服务有限公司 动物识别方法及装置
CN113962336B (zh) * 2021-08-06 2023-11-24 杨甦 可实时的牛只正脸id编码方法
CN114399787B (zh) * 2021-12-09 2022-10-21 慧之安信息技术股份有限公司 一种基于ai的牛脸识别方法
CN115457593A (zh) * 2022-07-26 2022-12-09 南京清湛人工智能研究院有限公司 一种牛脸识别方法、系统、存储介质和电子设备
CN116704264B (zh) * 2023-07-12 2024-01-30 北京万里红科技有限公司 动物分类方法、分类模型训练方法、存储介质及电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090060274A1 (en) * 2007-08-31 2009-03-05 Casio Computer Co., Ltd. Image pick-up apparatus having a function of recognizing a face and method of controlling the apparatus
CN103218610A (zh) * 2013-04-28 2013-07-24 宁波江丰生物信息技术有限公司 狗脸检测器的形成方法和狗脸检测方法
CN103858156A (zh) * 2011-10-18 2014-06-11 本田技研工业株式会社 车辆周围监测装置
CN107292298A (zh) * 2017-08-09 2017-10-24 北方民族大学 基于卷积神经网络和分类器模型的牛脸识别方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101751551B (zh) * 2008-12-05 2013-03-20 比亚迪股份有限公司 一种基于图像的人脸识别方法、装置、系统及设备
CN106056064B (zh) * 2016-05-26 2019-10-11 汉王科技股份有限公司 一种人脸识别方法及人脸识别装置
CN106203533B (zh) * 2016-07-26 2019-09-20 厦门大学 基于混合训练的深度学习人脸验证方法
CN107229947B (zh) * 2017-05-15 2020-11-06 翔创科技(北京)有限公司 一种基于动物识别的金融保险方法及系统
CN107766807A (zh) * 2017-09-30 2018-03-06 平安科技(深圳)有限公司 电子装置、投保牲畜识别方法和计算机可读存储介质
CN108090433B (zh) * 2017-12-12 2021-02-19 厦门集微科技有限公司 人脸识别方法及装置、存储介质、处理器
CN107895160A (zh) * 2017-12-21 2018-04-10 曙光信息产业(北京)有限公司 人脸检测与识别装置及方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090060274A1 (en) * 2007-08-31 2009-03-05 Casio Computer Co., Ltd. Image pick-up apparatus having a function of recognizing a face and method of controlling the apparatus
CN103858156A (zh) * 2011-10-18 2014-06-11 本田技研工业株式会社 车辆周围监测装置
CN103218610A (zh) * 2013-04-28 2013-07-24 宁波江丰生物信息技术有限公司 狗脸检测器的形成方法和狗脸检测方法
CN107292298A (zh) * 2017-08-09 2017-10-24 北方民族大学 基于卷积神经网络和分类器模型的牛脸识别方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111310703A (zh) * 2020-02-26 2020-06-19 深圳市巨星网络技术有限公司 基于卷积神经网络的身份识别方法、装置、设备及介质
CN111310703B (zh) * 2020-02-26 2022-06-03 深圳市巨星网络技术有限公司 基于卷积神经网络的身份识别方法、装置、设备及介质

Also Published As

Publication number Publication date
CN108921026A (zh) 2018-11-30

Similar Documents

Publication Publication Date Title
WO2019227616A1 (zh) 动物身份的识别方法、装置、计算机设备和存储介质
WO2020140665A1 (zh) 双录视频质量检测方法、装置、计算机设备和存储介质
CN109740620B (zh) 人群画像分类模型的建立方法、装置、设备和存储介质
WO2019120115A1 (zh) 人脸识别的方法、装置及计算机装置
WO2017113680A1 (zh) 声纹认证处理方法及装置
WO2019227617A1 (zh) 动物种类的分类方法、装置、计算机设备和存储介质
CN110222791B (zh) 样本标注信息的审核方法及装置
WO2021043085A1 (zh) 命名实体识别方法、装置、计算机设备及存储介质
CN111275207A (zh) 基于半监督的横向联邦学习优化方法、设备及存储介质
WO2021121129A1 (zh) 雷同病例检测方法、装置、设备及存储介质
CN112232117A (zh) 一种人脸识别方法、装置及存储介质
CN108170759B (zh) 投诉案件处理的方法、装置、计算机设备及存储介质
CN109711874A (zh) 用户画像生成方法、装置、计算机设备和存储介质
WO2019232893A1 (zh) 文本的情感分析方法、装置、计算机设备和存储介质
CN110321870B (zh) 一种基于lstm的掌静脉识别方法
CN111832581B (zh) 肺部特征识别方法、装置、计算机设备及存储介质
WO2020038138A1 (zh) 样本标注方法及装置、损伤类别的识别方法及装置
WO2021017316A1 (zh) 基于残差网络的信息识别方法、装置和计算机设备
CN112734202B (zh) 基于电子病历的医疗能力评价方法、装置、设备及介质
CN111914842A (zh) 车牌信息的识别方法、装置、计算机设备及存储介质
CN110728182B (zh) 基于ai面试系统的面试方法、装置和计算机设备
WO2019056503A1 (zh) 门店监控评价方法、装置及存储介质
JP2019087250A5 (zh)
CN110427992A (zh) 数据匹配方法、装置、计算机设备及存储介质
CN113705685B (zh) 疾病特征识别模型训练、疾病特征识别方法、装置及设备

Legal Events

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

Ref document number: 18920816

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

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

Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 26.03.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18920816

Country of ref document: EP

Kind code of ref document: A1