CN117746517A

CN117746517A - Human body biological information recognition model training method, system, equipment and storage medium

Info

Publication number: CN117746517A
Application number: CN202311641034.4A
Authority: CN
Inventors: 段兴; 陈晨; 林威宇; 吴陈涛; 兰兴增; 汪博; 朱力; 吕方璐
Original assignee: Chongqing Guangjian Aoshen Technology Co ltd; Zhuhai Hengqin Guangjian Technology Co ltd; Shenzhen Guangjian Technology Co Ltd
Current assignee: Chongqing Guangjian Aoshen Technology Co ltd; Zhuhai Hengqin Guangjian Technology Co ltd; Shenzhen Guangjian Technology Co Ltd
Priority date: 2023-12-04
Filing date: 2023-12-04
Publication date: 2024-03-22

Abstract

A human body biological information recognition model training method comprises the following steps: step S1: acquiring a human body biological information training set, and training a first training set according to part of samples in the human body biological information training set to acquire an initial model; step S2: identifying a second training set by using the initial model, and taking the sample pair with the error of at least one image larger than a first preset value as a new first training set; step S3: training according to the new first training set to obtain a second model; step S4: labeling a human body biological information area on at least one image of a positive sample in the first training set; step S5: and repeating the steps S2-S4 until the second model converges. The human body biological information recognition model training and the model training process are synchronously carried out, so that the training time is greatly saved, the training speed is increased, and the human body biological information recognition model training method has the advantages of being high in training speed, good in human body biological information recognition model training effect and low in calculation force requirement.

Description

Human body biological information recognition model training method, system, equipment and storage medium

Technical Field

The invention relates to the technical field of human body biological feature recognition, in particular to a human body biological information recognition model training method, a system, equipment and a storage medium.

Background

Training a human face model human body biological information recognition model refers to finding out challenging human face image sample pairs which are difficult to recognize from a large amount of training data in a human face recognition system through a certain method and technology. These difficult samples are generally of high inter-class and intra-class variability, and are of great significance in improving the performance of face recognition systems.

The main purpose of human body biological information recognition model training is:

1) The generalization capability of the face recognition system is improved: through learning the difficult sample pairs, the model can be better adapted to various complex environments and scenes, and the accuracy and the robustness of the system in practical application are improved.

2) Training process of optimization model: the training of the human body biological information recognition model can help us to better understand the weaknesses and the shortages of the model, so that the training strategy is adjusted, and the training efficiency is improved.

3) The false recognition rate is reduced: through intensive research on difficult sample pairs, some reasons for false recognition can be found, so that corresponding measures are taken to reduce the false recognition rate.

The human body biological information recognition model training method mainly comprises the following steps:

(1) A threshold-based method: by setting a certain threshold value, the sample pair with the wrong classification is used as a difficult sample pair. This method is simple and easy to implement, but may not adequately take into account the complex relationships between pairs of samples.

(2) Distance-based method: by calculating the distance between the sample pairs, the sample pair with the longer distance is taken as a difficult sample pair. The method can better reflect the similarity and the difference between the sample pairs, but has higher calculation complexity.

(3) Clustering-based methods: similar pairs of samples are clustered together by a clustering algorithm, and then pairs of samples at the cluster center or boundary are treated as difficult pairs of samples. This approach can better handle cases where the inter-class variation is large, but may not adequately account for the diversity within the class.

(4) Based on the method of generating a countermeasure network (GAN): the generalization ability of the model is improved by generating pairs of difficult samples that are challenging to combat network generation. This approach can generate high quality pairs of difficult samples, but requires large computational resources and time.

In the prior art, special improvement is often carried out on difficult sample pairs, more time is required, and the excavation effect and the time have positive correlation, so that the training cost of the human body biological information identification model is high.

The foregoing background is only for the purpose of providing an understanding of the inventive concepts and technical aspects of the present invention and is not necessarily prior art to the present application and is not intended to be used to evaluate the novelty and creativity of the present application in the event that no clear evidence indicates that such is already disclosed at the filing date of the present application.

Disclosure of Invention

Therefore, the invention obtains the initial model through the first training set formed by partial sample pairs, then identifies the second training set, identifies the sample pair with larger error, updates the first training set, marks the human body biological information area, and then trains again until the model converges, and synchronously carries out the training of the human body biological information identification model and the training process of the model through the rapid training of partial sample pairs, thereby greatly saving the training time, accelerating the training speed and having the advantages of high training speed, good training effect of the human body biological information identification model and low calculation force requirement.

In a first aspect, the present invention provides a training method for a human body biological information recognition model, which is characterized by comprising:

step S1: acquiring a human body biological information training set, and training a first training set according to part of samples in the human body biological information training set to acquire an initial model; the human body biological information training set comprises a plurality of groups of sample pairs, wherein the sample pairs comprise at least two images, and the images are obtained by shooting with a binocular or multi-view camera;

Step S2: identifying a second training set by using the initial model, and taking the sample pair with the error of at least one image larger than a first preset value as a new first training set; wherein the second training set is formed by a sample pair outside the first training set in the human biological information training set;

step S3: training according to the new first training set to obtain a second model;

step S4: labeling a human body biological information area on at least one image of a positive sample in the first training set;

step S5: and repeating the steps S2-S4 until the second model converges.

Optionally, the training method of the human body biological information recognition model is characterized in that step S2 includes:

step S21: removing the sample pair of the first training set from the human body biological information training set to obtain a second training set;

step S22: predicting each sample pair in the second training set using the initial model;

step S23: calculating an error for each image;

step S24: and selecting a sample pair with the error of at least one image in the human body biological information training set larger than a first preset value as a new first training set.

Optionally, the human body biological information recognition model training method is characterized by further comprising the following steps:

step S25: and detecting the new first training set by adopting a third model to obtain a second probability, and changing labels of part of sample pairs in the new first training set according to the second probability.

step S26: calculating the error of each image, and further obtaining the comprehensive error of the sample pair;

step S27: and selecting a sample pair with the comprehensive error larger than a fourth preset value in the human body biological information training set as a new first training set.

Optionally, the training method of the human body biological information recognition model is characterized in that step S4 includes:

step S41: labeling a human body biological information area for each positive sample pair of the first training set;

Step S42: uniformly adjusting the human body biological information area into a minimum rectangular area containing human body biological information;

step S43: and amplifying the positive sample pair and keeping the human body characteristic information marked.

Optionally, the training method of the human body biological information recognition model is characterized in that sub-category labels are added to the positive sample pair and the amplified image so as to improve learning effect.

Optionally, the training method of the human body biological information recognition model is characterized in that step S5 includes:

step S51: repeating the step S2-the step S4, and judging whether convergence is achieved or not according to the parameters of the second model;

step S52: if the number of times of circularly executing the step S2 to the step S4 reaches the second preset value and is not converged yet, executing the step S53;

step S53: forming a third training set by using the sample pairs which are formed in the step S2 and are common in the first training set, and forming a fourth training set by using other sample pairs except the third sample pair in the human biological information training set;

step S54: identifying the fourth training set by using the second model, and judging that the second model converges if the accuracy is higher than a third preset value; if the accuracy is not higher than the third preset value, executing step S55;

Step S55: and learning the fourth training set until the parameters of the second model are converged.

In a second aspect, the present invention provides a training system for a human body biological information recognition model, for implementing the training method for a human body biological information recognition model according to any one of the preceding claims, which is characterized by comprising:

the acquisition module is used for acquiring a human body biological information training set and training a first training set formed according to part of samples in the human body biological information training set to obtain an initial model; the human body biological information training set comprises a plurality of groups of sample pairs, wherein the sample pairs comprise at least two images, and the images are obtained by shooting with a binocular or multi-view camera;

the identification module is used for identifying the second training set by using the initial model, and taking the sample pair with the error of at least one image larger than a first preset value as a new first training set; wherein the second training set is formed by a sample pair outside the first training set in the human biological information training set;

the second model module is used for training according to the new first training set to obtain a second model;

the labeling module is used for labeling the human body biological information area for at least one image of the positive sample in the first training set;

And the circulation module is used for repeatedly executing the identification module, the second model module and the labeling module until the second model converges.

In a third aspect, the present invention provides a training apparatus for a human body biological information recognition model, comprising:

a processor;

a memory having stored therein executable instructions of the processor;

wherein the processor is configured to perform the steps of the human biometric model training method of any of the preceding claims via execution of the executable instructions.

In a fourth aspect, the present invention provides a computer readable storage medium storing a program, wherein the program when executed implements the steps of the human biological information identification model training method according to any one of the preceding claims.

Compared with the prior art, the invention has the following beneficial effects:

according to the invention, the training of the identification model and the training of the human body biological information identification model are fused into the same process, and the training work of the human body biological information identification model is not required to be specially carried out, so that the flow is greatly simplified, and the efficiency is improved.

The invention accelerates the training speed of the model through training the small sample pairs, saves a great amount of time, and makes the model more robust through dynamically adjusting the sample pairs of the first training set, and simultaneously can save the training cost.

When the invention is applied to larger images such as depth images, a lightweight model can be obtained quickly, the requirement on calculation force is low, and the invention can run on a carrier with smaller calculation force, so that the terminal equipment can locally perform the human body feature recognition model.

The invention marks the human body biological information area on the first training set, can accelerate the training speed and improves the learning effect on difficult sample pairs.

According to the invention, training is carried out according to images shot by the binocular or multi-view cameras, so that the model can learn together for images with higher correlation, and especially for different types of binocular images, the learning effect can be greatly improved, and better learning effect than that of a single type of image can be obtained by utilizing different types of characteristic cross learning.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art. Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings in which:

FIG. 1 is a flowchart showing steps of a training method for a human body biological information recognition model according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating steps for updating a first training set according to an embodiment of the present invention;

FIG. 3 is a flowchart illustrating another step of updating the first training set according to an embodiment of the present invention;

FIG. 4 is a flowchart illustrating another step of updating the first training set according to an embodiment of the present invention;

FIG. 5 is a flowchart illustrating steps for labeling a human body bioinformation area according to an embodiment of the present invention;

FIG. 6 is a flowchart illustrating steps of a loop until the second model converges in an embodiment of the present invention;

FIG. 7 is a schematic diagram of a training system for identifying human biological information according to an embodiment of the present invention;

FIG. 8 is a schematic structural diagram of a training device for a human body biological information recognition model according to an embodiment of the present invention; and

fig. 9 is a schematic diagram of a computer-readable storage medium according to an embodiment of the present invention.

Detailed Description

The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the present invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications could be made by those skilled in the art without departing from the inventive concept. These are all within the scope of the present invention.

The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented, for example, in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

The embodiment of the invention provides a method for training a depth module by using a human body biological information recognition model, which aims to solve the problems in the prior art.

The following describes the technical scheme of the present invention and how the technical scheme of the present application solves the above technical problems in detail with specific embodiments. The following embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. Embodiments of the present invention will be described below with reference to the accompanying drawings.

According to the invention, an initial model is obtained through a first training set formed by partial sample pairs, then the second training set is identified, the sample pair with larger error is identified, the first training set is updated, the human body biological information area is marked, and then the model is trained again until the model converges, and the human body biological information identification model training and the model training process are synchronously carried out through the rapid training of partial sample pairs, so that the training time is greatly saved, the training speed is accelerated, and the method has the advantages of high training speed, good human body biological information identification model training effect and low calculation force requirement.

Fig. 1 is a flowchart illustrating steps of a training method for a human body biological information recognition model according to an embodiment of the present invention. As shown in fig. 1, the method for training the human body biological information recognition model according to the embodiment of the invention includes the following steps:

step S1: and acquiring a human body biological information training set, and training the first training set according to part of samples in the human body biological information training set to acquire an initial model.

In this step, the human biological information training set includes a plurality of sample pairs, the sample pairs including at least two images, the images being captured by a binocular or multi-view camera. The human biological information training set comprises a positive sample pair and a negative sample pair. The positive sample pair is an image set containing human body biological characteristics, such as a human face image set, a palm image set, a pupil image set and the like. Each sample pair has a corresponding label to indicate whether it is a positive or negative sample pair. For the human body biological information training set, the positive sample pair is determined according to the expected obtained model, for example, if the expected obtained face recognition model is obtained, the image containing the face is marked as the positive sample pair, and the image not containing the face is marked as the negative sample pair. The ratio of the number of positive to negative pairs of samples ranges from [0.2,5]. The first training set is formed by equally proportionally selecting positive sample pairs and negative sample pairs, namely, randomly selecting part of sample pairs in the positive sample pairs and the negative sample pairs in the same proportion. The number of the first training sets is not less than 2 kilopairs and may be any integer greater than 1 kilopair.

In some embodiments, the step includes:

step S11: a human bioinformation dataset is obtained from a reliable data source.

Step S12: preprocessing the human body biological information data set, including data cleaning, missing value processing, abnormal value processing and the like.

Step S13: dividing the preprocessed human body biological information data set into a training set and a testing set.

Step S14: a portion of the sample pairs from the training set is randomly selected as a first training set.

Step S15: and training a machine learning algorithm by using the first training set to obtain an initial model.

Step S2: and identifying a second training set by using the initial model, and taking the sample pair with the error of at least one image larger than a first preset value as a new first training set.

In this step, the second training set is formed by pairs of samples in the human biological information training set other than the first training set. Operationally, the sample pairs of the first training set are removed from the human body biological information training set to obtain a second training set, each sample pair in the second training set is predicted by using the initial model, and the error of the image in each sample pair is calculated. And selecting a sample pair of the image with the error larger than the first preset value as a new first training set. In the step, a first training set is updated by adopting a sample pair with the error larger than a first preset value in the human body biological information training set, so that the first training set is always a difficult sample pair for the current model, namely a sample pair with larger error. Since the selection of the first training set is determined in relation to the error and the first preset value, the number of the first training sets is not fixed. The method judges the sample pair according to the maximum error of the images in the sample pair, so that the model can judge the relatively difficult images with the aid of the relatively easy images, and the learning effect of the difficult samples is improved. For example, for a binocular camera composed of an RGB sensor and an infrared sensor, the obtained RGB image and the obtained infrared image are different in information, and even a large difference in image quality is possible, so that a situation that the RGB image and the infrared image are easy to identify and difficult to identify easily occurs. The model can be better learned for difficult samples through the easily identified images.

Step S3: and training according to the new first training set to obtain a second model.

In this step, the new pairs of samples in the first training set are input into a machine learning algorithm for training, and model parameters are adjusted to minimize errors. The parameters of the second model are different from the first model. The second model is an optimization of the first model.

Step S4: and labeling the human body biological information area on at least one image of the positive sample in the first training set.

In this step, unlike the labeling in the prior art, this step also labels the human bioinformation region in the positive sample pair. For example, when training a face recognition model, the region where the face is located is labeled. The human body biological information region marked in this step is not a strict human body biological information region, but a region containing human body biological information and part of non-human body biological information. For example, for an image containing a human face, the human body biological information area refers to a rectangular frame, a circular frame or other closed area of any shape including a human face, and contains both human body biological information and background information. The step can be used for marking one image in the sample pair, and can also be used for marking both images. If a manual labeling mode is adopted, only one image in the sample pair can be labeled, so that time is saved. Taking a binocular image as an example, because of the corresponding relation of the sample pairs, when only one label exists, a certain corresponding relation exists between the binocular images, and the label range can be expanded and then the images are corresponding to the other image, so that the automatic label of the other image is realized.

After each model training, labeling the updated first training set. If the images in the first training set are already marked, the images do not need to be marked again, and only the unmarked images need to be marked.

The method aims at the image in the first training set, not only learns the labeling information by adopting positive and negative samples, but also provides the human body biological information area of the positive sample pair and trains the second model at the same time, so that the convergence speed and the learning effect of the second model can be accelerated.

Step S5: and repeating the steps S2-S3 until the second model converges.

In the step, the steps S2-S3 are repeated, and the first training set and the second model are continuously updated until the final second model converges, namely the model obtained by final training. And when the step S2 is circularly executed, a second model is used for identifying a second training set, and a sample pair with the error larger than a first preset value in the human body biological information training set is used as a new first training set. And when the step S3 is circularly executed, training is carried out according to the new first training set, and an updated second model is obtained. And S4, judging the parameter change of the second model after the adjacent circulation, and judging whether the second model converges or not according to the parameter change.

The pairs of samples in the first training set are dynamically changing. For example, 1000 ten thousand pairs of images are arranged in a human body biological information training set, 10 ten thousand pairs of images are randomly selected to form a first training set, and an initial model is obtained through training. And then, the initial model is adopted to identify the rest 990 ten thousand pairs of images, and 3 ten thousand pairs of images with errors larger than 0.9 are obtained. In the first training set, 4 ten thousand pairs of images with errors less than 0.1 are removed, and 6 ten thousand pairs of images remain. The 3 ten thousand pairs of images and the 6 ten thousand pairs of images are combined into a new first training set. And continuing training by the new first training set to obtain a second model.

Fig. 2 is a flowchart illustrating steps for updating the first training set according to an embodiment of the present invention. As shown in fig. 2, a step of updating a first training set in an embodiment of the present invention includes:

step S21: and removing the sample pair of the first training set from the human body biological information training set to obtain a second training set.

In this step, the human body biological information training set is divided into two parts, one part is the first training set, and the rest is the second training set.

Step S22: each sample pair in the second training set is predicted using the initial model.

In this step, each sample pair in the second training set is input into the initial model for prediction, and the probability a that each sample pair is true is obtained.

Step S23: the error for each image is calculated.

In this step, the error of each image is calculated from the probability a. For positive samples, the error is (1-a). For negative samples, the error is a.

In this step, the first training set is a difficult pair of samples for the current model, at least one image of a pair of samples being difficult, and the second training set is a simple pair of samples for the current model. The new first training set includes both partial pairs of samples in the original first training set and partial pairs of samples in the original second training set.

Fig. 3 is a flowchart illustrating another step of updating the first training set according to an embodiment of the present invention. As shown in fig. 3, compared to the foregoing embodiment, another step of updating the first training set in the embodiment of the present invention further includes:

In the step, a new first training set is input into a third model for detection, and a prediction result and a corresponding second probability of each sample pair are obtained. The third model is a mature generic model, different from the model to be trained in this embodiment. The universal model is generally relatively more resource consuming and relatively slow in response, and real-time arrangement is difficult to realize in the intelligent terminal. However, the generic model is often mature, with relatively high accuracy, and especially with relatively high confidence for sample pairs with relatively high confirmations. The step utilizes a general model, namely a third model to recheck a third training set, comprises a positive sample pair and a negative sample pair with higher confidence, and modifies a sample pair with wrong labeling, so that the correctness of the labeling of the sample pair in the first training set is ensured.

Fig. 4 is a flowchart illustrating another step of updating the first training set according to an embodiment of the present invention. As shown in fig. 4, another step of updating the first training set in the embodiment of the present invention includes:

Step S26: and calculating the error of each image, and further obtaining the comprehensive error of the sample pair.

In this step, the calculation of the error for each image refers to the foregoing embodiment. The integrated error for a sample pair is an average of the errors of the images in the sample pair. The comprehensive error is used for evaluating the error level of the sample pair, is used for evaluating the whole sample pair, can identify the most difficult sample, the more difficult sample and the easy sample according to the error condition of the whole sample pair, forms a gradient learning state and improves the learning effect.

In this step, the fourth preset value is a preset value for the error level of the screening sample pair. Depending on the fourth preset value, a portion can be selected from the most difficult sample, the more difficult sample, and the easy sample for learning.

According to the method, the sample pair is evaluated by the comprehensive error of the training sample pair, the sample pair with the comprehensive error larger than the fourth preset value is selected as the first training set, the overall learning effect of the training sample pair can be improved, the most difficult sample learning times are guaranteed, and the learning effect is improved.

Fig. 5 is a flowchart illustrating steps for labeling a biological information area of a human body according to an embodiment of the present invention. As shown in fig. 5, the steps of labeling a biological information area of a human body in an embodiment of the present invention include:

step S41: and labeling the human body biological information area for each positive sample pair of the first training set.

In this step, the operated sample pair of the positive sample pair is noted. The labeling can be performed manually or automatically. Automatic labeling requires recognition by means of an additional model so that a human body birth control information region can be obtained. For example, a mature face recognition model is adopted to recognize and obtain the region where the face is located.

Step S42: and uniformly adjusting the human body biological information area to a minimum rectangular area containing the human body biological information.

In this step, the human body biological information area may be in various forms, which is unfavorable for the learning of the second model. This step adjusts all the regions to rectangular regions. And during adjustment, the minimum rectangular area corresponding to the marked area is taken as the target. Rectangular areas on different sample pairs are different in size.

In this step, the positive sample pair is amplified. And when the image is amplified, the human body characteristic information is kept in a marked state, and the minimum rectangular area is kept unchanged relative to the pixel points. For example, changing the brightness, contrast, etc. of an image, keeping the minimum rectangular area unchanged; changing the size or the size, etc., the minimum rectangular area is kept to be the same as the image, so that the content in the minimum rectangular area is kept unchanged.

The embodiment amplifies the difficult sample pairs, so that the number of the difficult sample pairs is more, and the second model is more stable and has better learning effect.

In some embodiments, when the augmentation is performed, an area other than the human body biological information area is at least partially replaced, so as to strengthen the learning effect of the human body biological information area. In execution, for each positive sample pair labeled with a minimum rectangular region, a non-human biological information region within a certain range around it is determined. One area is randomly selected from the non-human body biological information areas as an area to be replaced. And at least partially replacing the area to be replaced according to a preset replacement strategy. The replacement policy may include the following:

Replacing the corresponding region in the other sample pair;

performing substitution by using the randomly generated area;

and (v) replacing the region with a region similar to the region to be replaced.

In the augmentation process, the position and the size of the minimum rectangular area are ensured to be unchanged, so that the labeling of the human body characteristic information is maintained. The above operations are repeated until all positive sample pairs have been or cannot be subjected to further amplification operations.

In some embodiments, sub-category labels are added to the positive sample pair and the amplified image to improve learning effect. When executing, for each positive sample pair marked with the smallest rectangular area, determining the corresponding sub-category label. And distributing one or more sub-category labels to each positive sample according to the preset sub-category label set. For example, only one face is assigned a subcategory label; there are two faces assigned two subcategory labels. And for the amplified image, determining the corresponding sub-category label according to the position and the corresponding relation of the sub-category label in the original image. And adding the sub-category labels into the positive sample pairs and the label information of the amplified images. Repeating the above operation until all positive sample pairs and the amplified images are added with corresponding sub-category labels or no more adding operation can be performed.

FIG. 6 is a flowchart illustrating steps of a loop until the second model converges according to an embodiment of the present invention. As shown in fig. 6, the steps of cycling until the second model converges in the embodiment of the present invention include:

step S51: and repeating the steps S2-S4, and judging whether convergence is achieved or not according to the parameters of the second model.

In this step, steps S2 to S4 are cyclically executed, and whether the model converges is determined according to the parameter change of the second model.

Step S52: if the number of times of circularly executing the steps S2-S4 reaches the second preset value, the step S53 is executed.

In this step, the second preset value is a larger value preset for evaluating the number of cycles. When the number of the cyclic training times reaches the second preset value, it is determined that the current model cannot automatically achieve convergence, and step S53 is executed.

Step S53: and (2) forming a third training set by the sample pairs which are formed in the step (S2) and are common in the first training set, and forming a fourth training set by the other sample pairs except the third sample pair in the human biological information training set.

In this step, when steps S2 to S4 are cyclically performed, the pairs of samples in the first training set are updated each time, pairs of samples common to the first training set that are different a plurality of times are recorded, and the pairs of samples are formed into a third training set. The third training set is a set of pairs of samples that the model cannot recognize. And removing the third training set from the human body biological information training set, and forming a fourth training set by the rest sample pairs. The fourth training set is a set of sample pairs that the model can identify.

Step S54: identifying the fourth training set by using the second model, and judging that the second model converges if the accuracy is higher than a third preset value; if the accuracy is not higher than the third preset value, step S55 is performed.

In the step, the fourth training set is identified, and if the accuracy is higher than a third preset value, the second model is judged to have converged. Otherwise, step S55 is performed.

In the step, the second model is trained by adopting the fourth training set, and the second model is directly converged.

According to the embodiment, the problem that the small sample cannot be converged in the training process is considered, a more flexible scheme is adopted for training the second model, the possible problems of the sample pair are eliminated, and the second model can be converged finally.

Fig. 7 is a schematic structural diagram of a training system for a human body biological information recognition model according to an embodiment of the present invention. As shown in fig. 7, a training system for a human body biological information recognition model according to an embodiment of the present invention includes:

According to the method, the initial model is obtained through the first training set formed by the partial sample pairs, the second training set is identified, the sample pair with larger error is identified, the first training set is updated, the human body biological information area is marked, and then the model is trained again until the model converges, the human body biological information identification model is trained synchronously with the training process of the model through the rapid training of the partial sample pairs, so that the training time is greatly saved, the training speed is accelerated, and the method has the advantages of being high in training speed, good in human body biological information identification model training effect and low in calculation force requirement.

The embodiment of the invention also provides human body biological information recognition model training equipment which comprises a processor. A memory having stored therein executable instructions of a processor. Wherein the processor is configured to perform the steps of a human biometric model training method via execution of executable instructions.

As described above, in this embodiment, an initial model is obtained through a first training set formed by partial sample pairs, and then the second training set is identified, and a sample pair with a larger error is identified, and then the first training set is updated and labeled with a human body biological information area, and then the model is trained again until the model converges.

Those skilled in the art will appreciate that the various aspects of the invention may be implemented as a system, method, or program product. Accordingly, aspects of the invention may be embodied in the following forms, namely: an entirely hardware embodiment, an entirely software embodiment (including firmware, micro-code, etc.) or an embodiment combining hardware and software aspects may be referred to herein as a "circuit," module "or" platform.

Fig. 8 is a schematic structural diagram of a training device for a human body biological information recognition model in an embodiment of the present invention. An electronic device 600 according to this embodiment of the invention is described below with reference to fig. 8. The electronic device 600 shown in fig. 8 is merely an example, and should not be construed as limiting the functionality and scope of use of embodiments of the present invention.

As shown in fig. 8, the electronic device 600 is in the form of a general purpose computing device. Components of electronic device 600 may include, but are not limited to: at least one processing unit 610, at least one memory unit 620, a bus 630 connecting the different platform components (including memory unit 620 and processing unit 610), a display unit 640, etc.

Wherein the storage unit stores program codes that can be executed by the processing unit 610, so that the processing unit 610 performs the steps according to various exemplary embodiments of the present invention described in the above-described one of the human body biological information recognition model training method sections of the present specification. For example, the processing unit 610 may perform the steps as shown in fig. 1.

The storage unit 620 may include readable media in the form of volatile storage units, such as Random Access Memory (RAM) 6201 and/or cache memory unit 6202, and may further include Read Only Memory (ROM) 6203.

The storage unit 620 may also include a program/utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment.

Bus 630 may be a local bus representing one or more of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or using any of a variety of bus architectures.

The electronic device 600 may also communicate with one or more external devices 700 (e.g., keyboard, pointing device, bluetooth device, etc.), one or more devices that enable a user to interact with the electronic device 600, and/or any device (e.g., router, modem, etc.) that enables the electronic device 600 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/O) interface 650. Also, electronic device 600 may communicate with one or more networks such as a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet, through network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 over the bus 630. It should be appreciated that although not shown in fig. 8, other hardware and/or software modules may be used in connection with electronic device 600, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage platforms, and the like.

The embodiment of the invention also provides a computer readable storage medium for storing a program, and the method is realized when the program is executed. In some possible embodiments, the aspects of the present invention may also be implemented in the form of a program product comprising program code for causing a terminal device to carry out the steps according to the various exemplary embodiments of the invention as described in the above-mentioned one of the human biological information identification model training methods section of the present specification, when the program product is run on the terminal device.

As shown above, in this embodiment, an initial model is obtained through a first training set formed by partial sample pairs, and then the second training set is identified, and a sample pair with a larger error is identified, and then the first training set is updated and labeled with a human body biological information area, and then the model is trained again until the model converges.

Fig. 9 is a schematic structural view of a computer-readable storage medium in an embodiment of the present invention. Referring to fig. 9, a program product 800 for implementing the above-described method according to an embodiment of the present invention is described, which may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable storage medium may also be any readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing describes specific embodiments of the present invention. It is to be understood that the invention is not limited to the particular embodiments described above, and that various changes and modifications may be made by one skilled in the art within the scope of the claims without affecting the spirit of the invention.

Claims

1. The human body biological information recognition model training method is characterized by comprising the following steps of:

Step S5: and repeating the steps S2-S4 until the second model converges.

2. The human biological information recognition model training method according to claim 1, wherein step S2 comprises:

step S23: calculating an error for each image;

3. The human biological information recognition model training method of claim 2, further comprising:

4. The human biological information recognition model training method according to claim 1, wherein step S2 comprises:

5. The human biological information identification model training method according to claim 1, wherein step S4 comprises:

6. The training method of human biological information recognition model according to claim 4, wherein sub-category labels are added to the positive sample pair and the amplified image to improve learning effect.

7. The human biological information identification model training method according to claim 1, wherein step S5 comprises:

8. A human body biological information recognition model training system for implementing the human body biological information recognition model training method according to any one of claims 1 to 7, characterized by comprising:

9. A human biological information recognition model training apparatus, comprising:

a processor;

a memory having stored therein executable instructions of the processor;

wherein the processor is configured to perform the steps of the human bioinformation recognition model training method of any one of claims 1 to 7 via execution of the executable instructions.

10. A computer-readable storage medium storing a program, wherein the program when executed implements the steps of the human biological information identification model training method according to any one of claims 1 to 7.