Palmprint identity recognition method, system and device based on palmprint data contribution weight

Abstract

The application discloses a palmprint identity recognition method and system based on palmprint data contribution weight and equipment, and relates to the technical field of identity permission management. The method comprises the following steps: calculating the local anomaly factor value of each palmprint training sample data; establishing a mapping relationship from the local anomaly factor value to the contribution weight based on the Weibull distribution function; fusing the contribution weights corresponding to different scale neighborhoods based on the Bayesian framework to obtain the multi-scale fusion contribution weight of each palmprint training sample data; optimizing the palmprint recognition model parameters according to the multi-scale fusion contribution weight of each palmprint training sample data; and inputting the palmprint data to be identified into the optimized palmprint recognition model to perform identity determination on the palmprint data. The palmprint identity recognition method and system based on palmprint data contribution weight and equipment can distinguish the importance of different palmprint training samples through the contribution weight, and effectively improve the robustness of palmprint identity recognition without increasing the calculation amount.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of identity permission management, in particular to a palmprint identity recognition method, system and device based on palmprint data contribution weight. BACKGROUND

[0002] Palmprint is widely used in the field of identity permission management due to its good privacy, stability and friendliness. The core of palmprint identity recognition technology is to construct an identity recognition model based on registered palmprint sample data through machine learning methods such as deep learning. Among them, data is the key to computer knowledge formation as the basis of palmprint recognition technology.

[0003] Deep learning, as a key technology of artificial intelligence, its network parameters highly depend on large-scale training data. Studies have shown that not all palmprint data samples are equally suitable for identity feature learning, and deep neural networks are particularly evident in this regard. In fact, palmprint samples affected by noise often have a negative effect on identity recognition network parameter learning. Traditional learning methods often ignore the distinction between data contribution sizes, and noise data and non-noise data have the same impact on knowledge representation, making them vulnerable to training sample offset interference in existing methods. In order to reduce the influence of noise, the commonly used method is to increase the size of palmprint data to improve the generalization performance of the recognition model. However, large-scale palmprint data means more computing, which increases the cost of the model. SUMMARY

[0004] The purpose of the present application is to provide a palmprint identity recognition method, system and device based on palmprint data contribution weight, which can distinguish the importance of different palmprint training samples through contribution weight and effectively improve the robustness of palmprint identity recognition without increasing the amount of calculation.

[0005] To achieve the above purpose, the present application provides the following scheme:

[0006] A palmprint identity recognition method based on palmprint data contribution weight, comprising:

[0007] Obtaining palmprint training sample data and calculating the local anomaly factor value of each palmprint training sample data;

[0008] Based on the Weibull distribution function, a mapping relationship from the local anomaly factor value to the contribution weight is established, and the contribution weight of each palmprint training sample data is obtained;

[0009] Based on the Bayesian framework, the contribution weights corresponding to different scales of neighborhoods are fused to obtain the multi-scale fusion contribution weight of each palmprint training sample data;

[0010] The parameters of the palmprint recognition model are optimized by the multi-scale fusion contribution weight of each palmprint training sample data to obtain an optimized palmprint recognition model.

[0011] The palmprint data to be identified is input into the optimized palmprint recognition model for identity determination.

[0012] Optionally, calculating the local anomaly factor value for each palmprint training sample data specifically includes:

[0013] Using the formula reach_dist k (P,O)=max{d k Calculate the reachable distance reach_dist for each palmprint training sample data. k (P,O); where P is a palmprint training sample data; O is a palmprint training sample data within the k-neighborhood of the palmprint training sample data P; d k (·) represents the k-distance between the palmprint training sample data within the brackets; d(P,O) represents the Euclidean distance between the palmprint training sample data P and O.

[0014] Based on reach_dist k (P,O), using the formula Calculate the local reachability density of each palmprint training sample data P. Where, N k (P) represents the set of k-neighbor palmprint training sample data of palmprint training sample data P; |N k (P)| represents all points in the k-neighborhood of the palmprint training sample data P;

[0015] Based on locally accessible density Using formula Calculate the local anomaly factor value for each palmprint training sample data P. in, Let O be the local reachability density of the palmprint training sample data.

[0016] Optionally, the step of establishing a mapping relationship from the local outlier factor values ​​to contribution weights based on the Weiber distribution function to obtain the contribution weight of each palmprint training sample data specifically includes:

[0017] A mapping relationship from the local outlier factor values ​​to contribution weights is established based on the Weiber distribution function. Where x is the local anomaly factor value of each palmprint training sample data P. λ is the proportional control parameter; θ is the shape control parameter; The cumulative probability is within the range (0,1). The cumulative probability is within the range [1,∞).

[0018] according to a mapping relationship from the local anomaly factor value to the contribution weight obtain the contribution weight of each palmprint training sample data P

[0019] Optionally, the Bayesian framework is used to fuse the contribution weights corresponding to different scale neighborhoods to obtain a multi-scale fused contribution weight of each palmprint training sample data, and specifically includes:

[0020] The contribution weight ω corresponding to different scale neighborhoods k is calculated as the conditional probability p(x|k) after a given k;

[0021] The Bayesian framework is used to fuse the conditional probability p(x|k) after a given k to obtain a multi-scale fused contribution weight of each palmprint training sample data P Wherein, p(k) is the prior probability, and Φ is the set of k values.

[0022] Optionally, the palmprint recognition model parameters are optimized according to the multi-scale fused contribution weight of each palmprint training sample data to obtain an optimized palmprint recognition model, and specifically includes:

[0023] According to the multi-scale fused contribution weight of the i-th palmprint training sample data P i The weighted loss is adopted The palmprint recognition model parameter α is optimized to obtain the optimal model parameter α*; wherein, loss i (f(P i |α)) is the prediction loss of the palmprint recognition model f(·||α) on the palmprint training sample data P i ; α is the palmprint recognition model parameter; N is the number of palmprint training sample data;

[0024] The optimal model parameter α* is assigned to the palmprint recognition model to obtain an optimized palmprint recognition model f(·|α*).

[0025] Optionally, the palmprint data to be identified is input into the optimized palmprint recognition model for palmprint data identity determination, and specifically includes:

[0026] The palmprint data to be identified P* is input into the optimized palmprint recognition model f(·|α*) to determine the corresponding palmprint data identity f(P*|α*).

[0027] A palmprint identity recognition system based on palmprint data contribution weight, comprising:

[0028] A local anomaly factor value calculation module is configured to obtain palmprint training sample data and calculate a local anomaly factor value of each palmprint training sample data.​

[0029] The contribution weight calculation module is configured to establish a mapping relationship from the local anomaly factor value to the contribution weight based on a Weibull distribution function, and obtain the contribution weight of each palmprint training sample data.

[0030] The multi-scale fusion contribution weight calculation module is configured to fuse the contribution weights corresponding to different scale neighborhoods based on a Bayesian framework, and obtain the multi-scale fusion contribution weight of each palmprint training sample data.

[0031] The model parameter optimization module is configured to optimize the palmprint recognition model parameters according to the multi-scale fusion contribution weight of each palmprint training sample data, and obtain an optimized palmprint recognition model.

[0032] The palmprint identity determination module is configured to input the palmprint data to be identified into the optimized palmprint recognition model for identity determination of the palmprint data.

[0033] An electronic device includes a memory for storing a computer program and a processor for running the computer program to enable the electronic device to perform the palmprint identity recognition method based on the contribution weight of palmprint data.

[0034] Optionally, the memory is a computer-readable storage medium.

[0035] According to the embodiments of the present application, the following technical effects are provided.

[0036] The present application provides a palmprint identity recognition method, system and device based on the contribution weight of palmprint data, which calculates the local anomaly factor value of each palmprint training sample data, establishes a mapping relationship from the local anomaly factor value to the contribution weight, fuses the contribution weights corresponding to different scale neighborhoods, optimizes the palmprint recognition model parameters, and finally inputs the palmprint data to be identified into the optimized palmprint recognition model for identity determination of the palmprint data, so that the importance of different palmprint training samples can be distinguished through the contribution weight, and the robustness of palmprint identity recognition is effectively improved without increasing the amount of calculation. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0038] Figure 1 The flowchart of the palmprint identity recognition method based on the contribution weight of palmprint data provided by the present application;

[0039] Figure 2 The principle diagram of the palmprint identity recognition method based on palmprint data contribution weight provided by the present application is shown in the figure.

[0040] Figure 3 The Weibull distribution function corresponding to different parameters; wherein, Figure 3 (a) in the Weibull distribution function under the condition that θ is invariable and λ changes; Figure 3 (b) in the Weibull distribution function under the condition that λ is invariable and θ changes; Figure 3 (c) in the Weibull distribution function under the condition that θ and λ change. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0042] The purpose of the present application is to provide a palmprint identity recognition method, system and device based on palmprint data contribution weight, which can distinguish the importance of different palmprint training samples through contribution weight and effectively improve the robustness of palmprint identity recognition without increasing the amount of calculation. The present application takes the distribution density of palmprint training sample data as the basis to evaluate the importance of palmprint training sample data for palmprint recognition model learning.

[0043] In order to make the above-mentioned purposes, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0044] Figure 1 The flowchart of the palmprint identity recognition method based on palmprint data contribution weight provided by the present application is shown in the figure, Figure 2 The principle diagram of the palmprint identity recognition method based on palmprint data contribution weight provided by the present application is shown in the figure, Figure 1 and Figure 2 The palmprint identity recognition method based on palmprint data contribution weight provided by the present application comprises:

[0045] Step 1: Obtain palmprint training sample data and calculate the local anomaly factor value of each palmprint training sample data.

[0046] For the input palmprint training sample data (palmprint sample for short), N groups of different neighborhood values are selected, and the local anomaly factor value of each palmprint training sample data is calculated respectively. Among them, the local anomaly factor value defines the dispersion degree of each palmprint training sample data relative to other sample points in the field, which is mainly obtained based on the reachable distance and the local reachable density of each palmprint training sample data.

[0047] The step 1 specifically comprises:

[0048] Step 1.1: adopt the formula reach_dist k (P, O) = max{d k (O), d(P, O)} to calculate the reachable distance reach_dist k (P, O) of each palmprint training sample data; wherein P is a palmprint training sample data; O is a palmprint training sample data in the k neighborhood of the palmprint training sample data P; d k (·) is the k distance of the palmprint training sample data in the parentheses; d(P, O) is the Euclidean distance between the palmprint training sample data P and O.

[0049] When d k (·) is expressed as the k distance of the palmprint training sample data P, it is defined as follows:

[0050] Wherein, N k (P) is the k neighborhood training sample set of the palmprint training sample data P.

[0051] Step 1.2: according to the reachable distance reach_dist k (P, O), adopt the formula to calculate the local reachable density of each palmprint training sample data P k Wherein, N k (P) is the k neighborhood palmprint training sample data set of the palmprint training sample data P; |N k (P) | is all points in the k neighborhood of the palmprint training sample data P.

[0052] It can be seen that the local reachable density of the center point of the palmprint training sample P is the reciprocal of the average reachable distance of P in its k nearest neighborhood. Therefore, when the palmprint training samples are densely gathered, the local reachable density tends to produce a high value, and using it to describe the density of the palmprint training sample can maximize the influence of similar samples on the density evaluation.

[0053] Step 1.3: according to the local reachable density Adopt the formula to calculate the local anomaly factor value of each palmprint training sample data P wherein, is the local reachable density of the palmprint training sample data O.

[0054] The local anomaly factor value is used to replace the absolute density of the palmprint training sample point, and whether the center point of the palmprint training sample P is an abnormal point is determined by comparing the density of the center point with the densities of other sample points in the k-neighborhood.

[0055] If the density of the center point of the palmprint training sample is greater than the densities of other sample points in the k-neighborhood, then The value is less than 1. Obviously, when is closer to 1, it indicates that the distribution of the palmprint training sample data region is more uniform and more representative. When the value is much greater than 1, it indicates that the density of the center point of the palmprint training sample is much lower than the densities of other sample points in the k-neighborhood, so the palmprint training sample is more likely to be an abnormal value and is more affected by noise.

[0056] Generally, if the local anomaly factor value of the palmprint training data is close to 1, it means that it is sampled from a uniformly distributed region. That is, the palmprint training sample is likely to be a core sample. The local anomaly factor value less than 1 means that the corresponding palmprint training sample is located in the sample group and is an internal sample. Otherwise, the palmprint training sample is located in an isolated region and is an outlier sample. In addition, in terms of the contribution of various palmprint training samples, the core sample, the internal sample and the outlier sample occupy the first, the second and the third places, respectively.

[0057] Step 2: establishing a mapping relationship from the local anomaly factor value to the contribution weight based on the Weibull distribution function to obtain the contribution weight of each palmprint training sample data.

[0058] The density calculation method based on the local anomaly factor value depends on the selection of the data neighborhood value. When the neighborhood value is too small, the density value is easily affected by the normal data deviation; when the neighborhood value is too large, the density value is too small to reflect the noise. Therefore, based on the local anomaly factor value, the present application calculates the mapping relationship from the local anomaly factor value to the contribution weight probability based on the Weibull distribution function, generates the contribution weight between 0 and 1 for each palmprint training sample data; the greater the weight, the greater the contribution of the palmprint training sample data to the palmprint recognition model learning, and vice versa.

[0059] The step 2 specifically comprises:

[0060] Step 2.1: establishing a mapping relationship from the local anomaly factor value to the contribution weight based on the Weibull distribution function wherein, x is the local anomaly factor value of each palmprint training sample data P λ is a proportional control parameter; θ is a shape control parameter. dx is a cumulative probability in the range of (0, 1); dx is a cumulative probability in the range of (0, 1).

[0061] Specifically, the mapping is based on a Weibull distribution function, defined as follows:

[0062]

[0063] Figure 3 Figures of Weibull distribution functions corresponding to different parameters are shown in Figure 3 As shown in Figure 3 (a), when θ is fixed, the position of the maximum value increases with the increase of λ. As shown in Figure 3 (b), when λ is fixed, the position of the maximum value increases with the increase of θ.

[0064] Since the Weibull distribution function has only one extreme value and the required mapping needs to reach the maximum value at input x = 1, the value of λ is derived from θ, and the specific relationship is derived as follows:

[0065]

[0066] As shown in Figure 3 (c), when λ and θ satisfy , the maximum value position of all functions is 1. The smoothness of each function decreases with the increase of θ. Among them, lower smoothness means that the Weibull distribution function has a sharper peak near position 1, that is, the palmprint training sample with a local anomaly factor value far from 1 is assigned a small contribution weight close to 0, and the palmprint training sample near 1 is assigned a large contribution weight. However, the contribution weight is an evaluation index in the range of 0 to 1, which cannot be guaranteed by the Weibull distribution function.

[0067] Therefore, the mapping of the local anomaly factor value to the contribution weight of the present application does not directly use the probability density, but uses the cumulative probability of the segmented interval as the final contribution weight of each palmprint training sample, which is defined as follows:

[0068]

[0069] wherein, dx is a cumulative probability in the range of (0, 1); dx is a cumulative probability in the range of (0, 1). - (P + ) is finally normalized.

[0070] Step 2.2: obtaining the contribution weight of each palmprint training sample data P according to the mapping relationship from the local anomaly factor value to the contribution weight obtaining the contribution weight of each palmprint training sample data P

[0071] Step 3: fusing the contribution weights corresponding to different scale neighborhoods based on the Bayesian framework to obtain the multi-scale fusion contribution weight of each palmprint training sample data.

[0072] Since the contribution degree is based on the local reachable density, and the local reachable density is greatly affected by the k-neighborhood value of the palmprint training sample; too small k value will cause the palmprint training sample density anomaly caused by oversampling; on the contrary, too large k value will make the palmprint training sample density value too small to reflect the noise. As can be seen, the k nearest neighborhood value is a key factor affecting the contribution weight. Therefore, the present application combines the Bayesian probability fusion strategy, and comprehensively considers the characteristics of multi-scale neighborhood to form the final palmprint sample contribution weight, and fuses the data contribution weights corresponding to different scale neighborhoods to avoid the adverse effects caused by improper selection of neighborhood value on the distribution density.

[0073] The step 3 specifically includes:

[0074] Step 3.1: calculating the contribution weight ω corresponding to different scale neighborhoods k as the conditional probability p(x|k) after a given k.

[0075] Step 3.2: fusing the conditional probability p(x|k) after a given k based on the Bayesian framework to obtain the multi-scale fusion contribution weight of each palmprint training sample data P Wherein, p(k) is the prior probability, and Φ is the k value set.

[0076] In order to make full use of the multi-scale contribution weight, a Bayesian probability-based fusion strategy is proposed combined with the probability based on different k values:

[0077]

[0078] Wherein, x represents the local anomaly factor value, p(x,k) represents the joint probability of x and k, p(x|k) represents the conditional probability after a given k, p(k) represents the prior probability, and Φ is the set of possible k values. In the present application, Φ∈{3,5,7,9,11}.

[0079] In order to obtain the prior probability p(k), the following definition is used:

[0080]

[0081] Wherein, |Φ| is the number of elements in the set Φ, and γ is a constant. The definition gives all candidate values of the prior probability under the ranking manner When 0 < γ < 1, the candidate values are arranged in descending order. When γ = 0, all candidate values are arranged in descending order. When γ > 1, the candidate values are arranged in ascending order. In the present application, γ = 0.2 is set.

[0082] In order to generate a mapping corresponding to the k value, the conditional probability p(x|k) is aligned with these prior probabilities according to the average entropy m(k):

[0083]

[0084] where N is the number of palmprint training samples, x i represents the local anomaly factor value of the i-th palmprint training sample data P i . The average entropy m(k) represents the effectiveness of the contribution weight at a fixed k, and the larger the value means that all palmprint samples have greater diversity.

[0085] Since the core palmprint samples, internal palmprint samples and outlier palmprint samples have different contribution degrees to model learning, the k value that produces a larger average entropy m(k) should have a larger prior probability.

[0086] Let represent the conditional probability p(x|k) corresponding to the j-th largest average entropy m(k) among all k values k∈Φ, and when 0 < γ < 1, the corresponding prior probability value is Therefore, the ordered pair of conditional probability and prior probability is obtained Therefore, the multi-scale fusion contribution weight can be calculated as:

[0087]

[0088] By aligning the ordered conditional probability with the ordered prior probability, the k nearest neighbors under multi-scale are considered hierarchically, the influence of the k value on the palmprint sample contribution weight is reduced to the greatest extent, and the multi-scale fusion contribution weight of each palmprint training sample data is obtained.

[0089] Step 4: optimizing the palmprint recognition model parameters according to the multi-scale fusion contribution weight of each palmprint training sample data, to obtain an optimized palmprint recognition model.

[0090] In the learning process of the recognition model, the learning loss is adjusted according to the data contribution weight, and the palmprint recognition model parameters are optimized in combination with the palmprint training sample data contribution weight and the prediction loss.

[0091] The step 4 specifically includes:

[0092] Step 4.1: according to the multi-scale fusion contribution weight of the i-th palmprint training sample data P i ​ adopting a weighted loss optimizing a palmprint recognition model parameter a to obtain an optimal model parameter a*; wherein, loss i (f(P i |a)) is a prediction loss of the palmprint recognition model f(·||a) on palmprint training sample data P i ; a is a palmprint recognition model parameter; and N is a quantity of palmprint training sample data.

[0093] Step 4.2: assigning the optimal model parameter a* to the palmprint recognition model to obtain an optimized palmprint recognition model f(·|a*).

[0094] After obtaining the multi-scale evaluation weight of each palmprint training sample, the palmprint recognition model is finally optimized by using the following weighted loss:

[0095]

[0096] wherein, loss i (f(P i |a)) is a prediction loss of the palmprint recognition model f(·||a) on palmprint training sample data P i ; a is a palmprint recognition model parameter; and N is a quantity of palmprint training sample data.

[0097] The optimal recognition model parameter a* is selected to minimize the above loss function, and a corresponding recognition model f(·|a*) is obtained, which can be used for identity determination of palmprint data.

[0098] Step 5: inputting the palmprint data to be identified into the optimized palmprint recognition model for identity determination of the palmprint data.

[0099] Specifically, inputting the palmprint data to be identified P* into the optimized palmprint recognition model f(·|a*) to determine the corresponding palmprint data identity f(P*|a*).

[0100] Next, taking the CASIA palmprint database as an example, the effectiveness of the method is verified. The loss single neighborhood scale weighted loss (ω i i is the contribution weight of the i-th palmprint training sample data P i and the multi-scale neighborhood weighted loss are considered, and the accuracy, precision, sensitivity and F1 index are used as evaluation (the larger the value is, the better), and the results of the three methods are compared, as shown in Table 1, and it can be found that the present application is better than the comparison method in the four indexes.

[0101] Table 1 comparison results of different optimization strategies

[0102]

[0103] Based on the method provided by the application, the application also provides a palmprint identity recognition system based on palmprint data contribution weight, comprising:

[0104] A local anomaly factor value calculation module is configured to acquire palmprint training sample data and calculate a local anomaly factor value of each palmprint training sample data.

[0105] A contribution weight calculation module is configured to establish a mapping relationship from the local anomaly factor value to the contribution weight based on a Weibull distribution function, and obtain the contribution weight of each palmprint training sample data.

[0106] A multi-scale fusion contribution weight calculation module is configured to fuse the contribution weights corresponding to different scales of neighborhoods based on a Bayesian framework, and obtain a multi-scale fusion contribution weight of each palmprint training sample data.

[0107] A model parameter optimization module is configured to optimize palmprint recognition model parameters according to the multi-scale fusion contribution weight of each palmprint training sample data, and obtain an optimized palmprint recognition model.

[0108] A palmprint identity determination module is configured to input palmprint data to be identified into the optimized palmprint recognition model for palmprint data identity determination.

[0109] Further, the application also provides an electronic device, which can include a processor, a communication interface, a memory and a communication bus. Wherein the processor, the communication interface and the memory complete mutual communication through the communication bus. The processor can call the computer program in the memory to execute the palmprint identity recognition method based on palmprint data contribution weight.

[0110] In addition, the computer program in the memory described above is implemented in the form of a software functional unit and sold or used as an independent product. It can be stored in a non-transitory computer readable storage medium. Based on such understanding, the technical solutions of the application or the part of the technical solutions that make essential contributions to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, and various media that can store program codes.

[0111] The application provides a sample contribution weight generation method based on data density, which includes sample density generation based on a local anomaly factor, contribution weight calculation based on a Weibull distribution, and multi-neighbor scale fusion weight generation based on a Bayesian framework.

[0112] The various embodiments are described in a progressive manner in the specification, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0113] The principles and implementation manners of the application are described by using specific examples in the specification, and the above embodiment description is only used to help understand the method and core idea of the application; meanwhile, for the general technical personnel in the field, the specific implementation manner and application range can be changed according to the idea of the application. In conclusion, the content of the specification should not be understood as the limitation of the application.

Claims

1. A palmprint identity recognition method based on palmprint data contribution weight, characterized in that, The method comprises the following steps: Obtain palmprint training sample data and calculate a local anomaly factor value of each palmprint training sample data; The calculation of the local anomaly factor value for each palmprint training sample data specifically includes: using the formula Calculate the reachable distance for each palmprint training sample data ;in, This is a palmprint training sample data; Palmprint training sample data A palmprint training sample data within the neighborhood; The palmprint training sample data is in parentheses. distance; Palmprint training sample data and Euclidean distance between them; based on reachability Using formula Calculate training sample data for each palm print Locally achievable density ;in, Palmprint training sample data of Neighborhood palmprint training sample data set; For palmprint training sample data of All points within the neighborhood; based on local reachability density Using formula Calculate training sample data for each palm print Local anomaly factor values ;in, Palmprint training sample data Locally achievable density; mapping relationship from the local anomaly factor value to the contribution weight is established based on the Weibull distribution function to obtain the contribution weight of each palmprint training sample data, and specifically includes: mapping relationship from the local anomaly factor value to the contribution weight is established based on the Weibull distribution function ; wherein x is the local anomaly factor value of each palmprint training sample data ; ; is a proportional control parameter; is a shape control parameter; is a cumulative probability in the range of [0, 1]; is a cumulative probability in the range of [0, 1]; is a cumulative probability in the range of [0, 1]; is a cumulative probability in the range of [0, 1]; obtain the contribution weight of each palmprint training sample data ; ; Based on a Bayesian framework, the contribution weights corresponding to different scale neighborhoods are fused to obtain a multi-scale fusion contribution weight of each palmprint training sample data; Optimize the palmprint recognition model parameters according to the multi-scale fusion contribution weight of each palmprint training sample data to obtain an optimized palmprint recognition model; Input the palmprint data to be identified into the optimized palmprint recognition model to determine the identity of the palmprint data.

2. The palmprint identification method based on palmprint data contribution weight according to claim 1, characterized in that, The method for fusing the contribution weights corresponding to different scale neighborhoods based on the Bayesian framework to obtain the multi-scale fusion contribution weight of each palmprint training sample data specifically comprises the following steps: Computing different scale neighborhoods Corresponding contribution weights As given Posterior conditional probabilities ; Fusing the conditional probability after the given based on the Bayesian framework , the multi-scale fusion contribution weight of each palm print training sample data is obtained ; wherein, is the prior probability, is the value set.

3. The palmprint identification method based on palmprint data contribution weight according to claim 2, characterized in that, The method for optimizing the palmprint recognition model parameters according to the multi-scale fusion contribution weight of each palmprint training sample data to obtain the optimized palmprint recognition model specifically comprises the following steps: According to the first i palmprint training sample data multi-scale fusion contribution weight , the weighted loss palmprint recognition model parameters optimization, get the optimal model parameters ; wherein, palmprint recognition model palmprint training sample data prediction loss; palmprint recognition model parameters; the number of palmprint training sample data; assigning the optimal model parameters to the palmprint recognition model to obtain an optimized palmprint recognition model assigning the optimal model parameters to the palmprint recognition model to obtain an optimized palmprint recognition model .

4. The palmprint identification method based on palmprint data contribution weight according to claim 3, characterized in that, The method for inputting the palmprint data to be identified into the optimized palmprint recognition model to determine the identity of the palmprint data specifically comprises the following steps: The palmprint data to be identified Input the optimized palmprint recognition model In the middle, the corresponding palmprint data identity is determined .

5. A palmprint identification system based on palmprint data contribution weight, characterized in that, The method comprises the following steps: A local anomaly factor value calculation module is configured to obtain palmprint training sample data and calculate a local anomaly factor value of each palmprint training sample data; The calculation of the local anomaly factor value for each palmprint training sample data specifically includes: using the formula Calculate the reachable distance of each palmprint training sample data ;in, This is a palmprint training sample data; Palmprint training sample data A palmprint training sample data within the neighborhood; The palmprint training sample data is in parentheses. distance; Palmprint training sample data and Euclidean distance between them; based on reachable distance Using formula Calculate training sample data for each palm print Locally achievable density ;in, Palmprint training sample data of Neighborhood palmprint training sample data set; For palmprint training sample data of All points within the neighborhood; based on local reachability density Using formula Calculate training sample data for each palm print Local anomaly factor values ;in, Palmprint training sample data Locally achievable density; The contribution weight calculation module is used to establish a mapping relationship from the local outlier factor values ​​to the contribution weights based on the Weiber distribution function, thereby obtaining the contribution weight of each palmprint training sample data. Specifically, it includes: establishing a mapping relationship from the local outlier factor values ​​to the contribution weights based on the Weiber distribution function. Where x represents each palmprint training sample data. Local anomaly factor values ; This is a proportional control parameter; These are shape control parameters; for Cumulative probability within the range; for The cumulative probability within the range; based on the mapping relationship from the local anomaly factor value to the contribution weight. Obtain training sample data for each palm print. Contribution weight ; A multi-scale fusion contribution weight calculation module is configured to fuse the contribution weights corresponding to different scale neighborhoods based on a Bayesian framework to obtain a multi-scale fusion contribution weight of each palmprint training sample data; A model parameter optimization module is configured to optimize the palmprint recognition model parameters according to the multi-scale fusion contribution weight of each palmprint training sample data to obtain an optimized palmprint recognition model; A palmprint identity determination module is configured to input the palmprint data to be identified into the optimized palmprint recognition model to determine the identity of the palmprint data.

6. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the palmprint identity recognition method based on the palmprint data contribution weight.

7. The electronic device of claim 6, wherein, The memory is a computer readable storage medium.

Images