Intelligent security system for hydropower station and signature authentication method thereof

Abstract

The application belongs to the technical field of water power station security authentication, and specifically discloses a water power station intelligent security system and a signature authentication method thereof, the signature authentication method comprising trajectory reconstruction on the entry signature of a user to obtain a reconstruction speed and a reconstruction trajectory; constructing a continuous feature sequence according to the reconstruction speed, the reconstruction trajectory and signature parameters of the entry signature, and carrying out noise reduction on the continuous feature sequence to obtain a noise-reduced feature sequence of the entry signature; inputting the noise-reduced feature sequence of the entry signature and the noise-reduced feature sequence of the registered signature into a twin network model to obtain the similarity between the entry signature and the registered signature, and determining whether the user is a legal user according to the similarity. The application performs identity authentication based on the signature behavior characteristics of the user, and is strong in attack resistance because the behavior characteristics are difficult to imitate. Meanwhile, the entry signature is noise-reduced after trajectory reconstruction, so that external noise interference can be effectively filtered, and the authentication accuracy can be improved.

Description

Technical Field

[0001] This invention belongs to the field of hydropower station security authentication technology, and discloses a smart security system for hydropower stations and its signature authentication method. Background Technology

[0002] During the construction and operation management of water conservancy projects, personnel management in key areas such as hydropower station dams is crucial for ensuring the safe and stable operation of the power station and the safety of public property. Current personnel control methods for key areas mainly involve staff on duty and registration; however, this method suffers from high labor costs and low management efficiency. Therefore, to improve personnel management efficiency, existing control methods primarily utilize facial recognition and video surveillance. However, both of these methods have limitations in authentication when users wear face coverings. Furthermore, traditional facial recognition relies on deep learning models and requires a large number of positive and negative samples for training, while real-world scenarios often lack such a massive amount of data. Training is time-consuming and costly, and authentication accuracy decreases when training data is scarce. Summary of the Invention

[0003] The purpose of this invention is to provide a smart security system for hydropower stations and its signature authentication method, so as to solve the technical problems of existing hydropower station security management and control relying on facial recognition for authentication, which has authentication limitations and requires deep learning models and a large amount of data for training, resulting in long training time and high cost.

[0004] The first aspect of this invention provides a signature authentication method for a smart security system in a hydropower station, comprising:

[0005] The user's entry signature is reconstructed to obtain the reconstruction speed and reconstruction trajectory.

[0006] Based on the reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature, a continuous feature sequence is constructed, and the continuous feature sequence is denoised to obtain the denoised feature sequence of the entry signature.

[0007] The denoised feature sequences of the entry signature and the registration signature are both input into the Siamese network model to obtain the similarity between the entry signature and the registration signature, and the user is determined to be a legitimate user based on the similarity.

[0008] Preferably, a continuous feature sequence is constructed based on the reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature. Specifically, this includes: obtaining the changes in the reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature; constructing a feature sequence from the reconstruction speed, reconstruction trajectory, signature parameters of the entry signature, and the changes; and determining the transition feature values ​​between adjacent features in the feature sequence to obtain a continuous feature sequence.

[0009] Preferably, the twin network model includes two identical subnetworks and a common network connected to the outputs of the two subnetworks.

[0010] Preferably, each of the sub-networks includes a first long short-term memory network layer, a second long short-term memory network layer, and a first fully connected layer connected in sequence; the output of the first fully connected layer is connected to the common network.

[0011] Preferably, the public network includes a similarity calculation layer, a third long short-term memory network layer, a flattening layer, and a second fully connected layer connected in sequence; the similarity calculation layer determines the similarity between the entry signature and the registration signature based on the L1 norm.

[0012] Preferably, the entry signature is reconstructed to obtain the reconstruction speed and reconstruction trajectory. Specifically, this involves: acquiring the features of each stroke of the entry signature, and determining the speed and angular position of each stroke based on its features; and determining the speed and angular position of each stroke based on its speed and angular position. , Reconstruction speed in the direction; according to , The reconstruction speed in a given direction determines the reconstruction trajectory in that direction.

[0013] Preferably, each stroke is characterized by amplitude, start time, logarithmic delay, logarithmic response time, start angle, and end angle.

[0014] Preferably, the signature parameters include timestamp, status, azimuth angle, and... At least one of shaft angle and pressure.

[0015] Preferably, the continuous feature sequence is denoised by using the FISTA algorithm.

[0016] A second aspect of the present invention provides a smart security system for hydropower stations, which uses the signature authentication method of the aforementioned smart security system for hydropower stations for authentication.

[0017] The intelligent security system for hydropower stations and its signature authentication method of the present invention have the following advantages compared with the prior art:

[0018] This invention employs an online signature authentication method, authenticating identity based on user signature behavior characteristics. Because these behavioral characteristics are difficult to imitate, when an unauthorized intruder attempts to forge a signature, they can only mimic the outline of the genuine user's signature; features such as pen speed during signing are difficult to imitate, thus providing strong resistance to attacks. Furthermore, compared to authentication methods based on physiological characteristics, this invention can compensate for authentication failures caused by changes in physiological characteristics (such as sweaty fingers, peeling skin, or wearing face coverings). It is highly versatile, has a high authentication accuracy rate, and can be applied to intelligent security systems in hydropower stations, demonstrating broad application prospects.

[0019] This invention addresses noise interference from two aspects: first, it employs a trajectory reconstruction method, and then it applies the FISTA algorithm to the reconstructed signature. When environmental interference is significant at the power plant entrance, this method can effectively filter out interference caused by external noise. By reconstructing the user's trajectory, deep-seated intrinsic features of the user's signature can be extracted; simultaneously, the FISTA algorithm, through multiple iterations, allows the noisy signal to continuously approximate the original signal, thereby achieving a high authentication accuracy.

[0020] The twin network model used in this invention performs authentication based on a set of similarity metrics of signatures, rather than learning features from training samples. Therefore, it is not affected by the amount of signature data or the number of negative signature samples during training. The model can be trained based on a small number of samples, which solves the problem of not being able to collect a large amount of data on-site and the imbalance of positive and negative samples. It can be extended to the classification of signatures with few negative samples. Attached Figure Description

[0021] Figure 1 This is a simplified flowchart of the signature authentication method for the intelligent security system of a hydropower station according to an embodiment of the present invention.

[0022] Figure 2 This is a detailed flowchart of the signature authentication method for the intelligent security system of a hydropower station according to an embodiment of the present invention.

[0023] Figure 3 This is a structural diagram of the twin network model in an embodiment of the present invention. Detailed Implementation

[0024] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0025] The first aspect of this invention provides a signature authentication method for a smart security system in a hydropower station, such as... Figures 1 to 3 As shown, it includes:

[0026] Step 1: Reconstruct the trajectory of the entry signature to obtain the reconstruction speed and reconstruction trajectory, specifically:

[0027] Step 1.1: Obtain the characteristics of each stroke in the entrance signature, and determine the speed and angular position of each stroke based on these characteristics. The characteristics of each stroke include amplitude, start time, logarithmic delay, logarithmic response time, start angle, and end angle.

[0028] For example, the user's entry signature is first collected to obtain the complete signature. .

[0029] Secondly, the obtained complete signature Divided into Each stroke is Get each stroke after division The characteristics, namely It can be represented as , .in, For the first The amplitude of each stroke; For the first The start time of each stroke; For logarithmic delay, that is, writing the first... The time delay of the neuromuscular system during each stroke is expressed on a logarithmic time scale. For logarithmic response time, that is, writing the first... The response time of the neuromuscular system during each stroke is expressed on a logarithmic time scale. For the first The starting angle of each stroke For the first The ending angle of each stroke.

[0030] Finally, the speed and angular position of each stroke are determined based on the characteristics of each stroke.

[0031] Specifically, using Determine the first The speed of each stroke , As shown in formula (1):

[0032] (1)

[0033] In the formula, This refers to the current moment.

[0034] Furthermore, utilizing and Determine the first The corner position of each stroke , As shown in formula (2):

[0035] (2)

[0036] In the formula, Error function .

[0037] Step 1.2: Determine the speed and angular position of each stroke separately. , Reconstruction speed in the direction.

[0038] This invention can determine a complete signature after obtaining the speed of each stroke. Reconstruction speed As shown in formula (3):

[0039] (3)

[0040] In this embodiment of the invention, after obtaining the speed of each stroke, the complete signature is calculated based on Cartesian space. exist Reconstruction speed in direction and full signature exist Reconstruction speed in direction :

[0041] (4)

[0042] (5)

[0043] Step 1.3, according to , The reconstruction velocity in a given direction determines the reconstruction trajectory in that direction, where The reconstructed trajectory in the direction is shown in formula (6). The reconstructed trajectory in the direction is shown in formula (7):

[0044] (6)

[0045] (7)

[0046] Step 2: Based on the reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature, construct a continuous feature sequence, and denoise the continuous feature sequence to obtain the denoised feature sequence of the entry signature, including:

[0047] Step 2.1: Based on the reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature, construct a continuous feature sequence, specifically including:

[0048] Step 2.1.1: Obtain the changes in reconstruction speed, reconstruction trajectory, and entry signature parameters respectively.

[0049] In this embodiment of the invention, the signature parameters for entry signature include a timestamp. ,state Azimuth ,and Axis angle and pressure At least one of them.

[0050] The specific changes in the above-mentioned reconstruction speed, reconstruction trajectory, and entry signature parameters are shown in formula (8):

[0051] (8)

[0052] In the formula, for The amount of change in the reconstructed trajectory; for The amount of change in the reconstructed trajectory; This represents the change in reconstruction speed; The change in timestamp; This represents the change in azimuth angle; To and The change in the included angle of the axis; This refers to the change in pressure. for time The trajectory of directional reconstruction; for time The trajectory of directional reconstruction; for time The trajectory of directional reconstruction; for time The trajectory of directional reconstruction; for Complete signature at all times Reconstruction speed; for Complete signature at all times Reconstruction speed; for Time stamp; for Time stamp; for Azimuth angle at time; for Azimuth angle at time; for Time and Angle between axes; for Time and Angle between axes; for Constant pressure; for Constant pressure.

[0053] Step 2.1.2: Construct a feature sequence from the reconstruction speed, reconstruction trajectory, signature parameters of the entry signature, and changes.

[0054] The reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature in this embodiment of the invention can be constructed. Time series data in 8 dimensions at a given time: ;

[0055] Furthermore, in this embodiment of the invention, the time-series data is augmented using the aforementioned changes to obtain a feature sequence consisting of 15 features:

[0056]

[0057] Step 2.1.3: Determine the transitional feature values ​​between adjacent features in the feature sequence to obtain a continuous feature sequence.

[0058] In this embodiment of the invention, the end value of the preceding feature in two adjacent features is set as... The initial value of the latter feature is The length of the previous feature is The length of the latter feature is Then the transition eigenvalue and the number of intermediate interpolation points The calculation formula is as follows:

[0059] (9)

[0060] (10)

[0061] To reduce data dimensionality while ensuring data continuity and stability, this invention provides a feature sequence containing 15 features. The transition sequence, composed of the 14 transition eigenvalues ​​determined by formulas (9) and (10), is sequentially combined into the following continuous eigense sequence. :

[0062] .

[0063] Step 2.2: Denoise the continuous feature sequence to obtain the denoised feature sequence of the entry signature.

[0064] This invention specifically utilizes the FISTA algorithm to denoise continuous feature sequences, obtaining the denoised feature sequence of the entry signature as follows: Let .

[0065] a) Initialize the initial signal Regularization parameters Lipschitz constant ,momentum The number of iterations is The maximum number of iterations is , 500, order ;

[0066] b) Through gradient For signal Update the gradient The calculation formula is as follows:

[0067] (11)

[0068] in The calculation formula is as follows:

[0069] (12)

[0070] c) Gradient-based Lipschitz constant and regularization parameters Soft thresholding update signal The updated formula is as follows:

[0071] (13)

[0072] d) Update momentum ,but Through momentum and updating the original signal The updated formula is as follows:

[0073] (14)

[0074] e) Judgment If true, obtain the denoised feature sequence of the entry signature. And proceed to step 3; otherwise, let Then, proceed with steps (b) through (e).

[0075] Step 3: Input the denoised feature sequence of the entry signature and the denoised feature sequence of the registration signature into the trained Siamese network model to obtain the similarity between the entry signature and the registration signature, and determine whether the user is a legitimate user based on the similarity.

[0076] In this embodiment of the invention, the noise reduction feature sequence of the registration signature is determined with reference to the method for determining the noise reduction feature sequence of the entry signature described above, and will not be repeated here.

[0077] The architecture of the twin network model in this embodiment of the invention is as follows: Figure 3 As shown, the network comprises two identical subnetworks and a common network connected to the outputs of the two subnetworks, with the two subnetworks sharing weights. Each subnetwork consists of a first Long Short-Term Memory (LSTM) network layer, a second LSM network layer, and a first fully connected layer, connected in sequence; the output of the first fully connected layer is connected to the common network. The common network consists of a similarity calculation layer, a third LSM network layer, a flattening layer, and a second fully connected layer, connected in sequence; the similarity calculation layer determines the similarity between the entry signature and the registration signature based on the L1 norm.

[0078] The two first long short-term memory network layers mentioned above serve as the starting layers of the Siamese network model. They receive the denoised feature sequences of the input signature and the denoised feature sequences of the registration signature, respectively. Through their internal memory units and gating mechanisms, they capture early time-series features and local patterns in the signature sequences, and perform preliminary feature extraction and processing on the input data.

[0079] The two second long short-term memory network layers mentioned above can integrate the feature information output by the corresponding first long short-term memory network layers. Through their own recurrent structure, they continuously update and refine the features, enabling the Siamese network model to learn more discriminative features in the denoised feature sequence, providing more valuable information for subsequent processing.

[0080] The first fully connected layer further integrates and transforms the features extracted by the second long short-term memory network layer, mapping the features to a feature space that is more suitable for subsequent similarity measurement and classification tasks. It also plays a role in dimensionality reduction or dimensionality adjustment of the features, enabling the Siamese network model to better learn and distinguish different features.

[0081] The similarity calculation layer determines the similarity between the entry signature and the registration signature based on the L1 norm. The L1 norm calculates the sum of the absolute values ​​of the differences between corresponding elements of two vectors. This layer quantifies the degree of difference between the feature vectors of the entry signature and the registration signature, obtaining a numerical value to represent their similarity. This provides an important basis for subsequent determination of whether the entry signature and the registration signature belong to the same person.

[0082] The similarity calculation layer calculates similarity based on the L1 norm. Specifically:

[0083] (15)

[0084] In the formula, The first person to sign autographs upon entry One noise reduction feature; For the registration signature One noise reduction feature.

[0085] The third long short-term memory network layer further processes and extracts features from the sequence information related to the similarity measure (if there is other relevant sequence information in the input data), uncovering deeper patterns and relationships hidden in the similarity measure values, and enhancing the accuracy of the Siamese network model in judging signature similarity.

[0086] The flattening layer flattens the multidimensional feature tensor output by the third long short-term memory network layer into a one-dimensional vector, so that it can be processed by the subsequent second fully connected layer. Inputting the flattened one-dimensional vector into the second fully connected layer makes it easier for the second fully connected layer to integrate all features.

[0087] The second fully connected layer, as the last layer of the Siamese network model, uses the sigmoid activation function to make the final classification decision based on the feature information obtained from the previous layers, and outputs the result of whether the signature matches (e.g., outputting 0 or 1 to indicate whether it does not match or matches), thus completing the online signature network authentication task.

[0088] The process for determining whether a user is a legitimate user using a twin network model in this embodiment of the invention is as follows:

[0089] 1) Obtain the denoised feature sequence of the entry signature :

[0090] 2) Determine the entire signature feature :

[0091] (16)

[0092] 3) Use the above method to determine the entire signature feature Z corresponding to the denoised feature sequence of the registered signature, and determine the similarity S between H and Z;

[0093] 4) Judgment If the condition is met, the user is deemed a legitimate user and allowed entry; otherwise, the user is deemed an illegitimate user and an alarm is triggered.

[0094] A second aspect of the present invention provides a smart security system for hydropower stations, which uses the signature authentication method of the aforementioned smart security system for hydropower stations for authentication.

[0095] This invention employs an online signature authentication method, authenticating identity based on user signature behavior characteristics. Because these behavioral characteristics are difficult to imitate, when an unauthorized intruder attempts to forge a signature, they can only mimic the outline of the genuine user's signature; features such as pen speed during signing are difficult to imitate, thus providing strong resistance to attacks. Furthermore, compared to authentication methods based on physiological characteristics, this method can compensate for authentication failures caused by changes in physiological characteristics (such as sweaty fingers, peeling skin, or wearing face coverings). It is highly versatile, has a high authentication accuracy rate, and can be applied to intelligent security systems in hydropower stations, demonstrating broad application prospects.

[0096] This invention addresses noise interference from two aspects: first, it employs a trajectory reconstruction method, and then it applies the FISTA algorithm to the reconstructed signature. When environmental interference is significant at the power plant entrance, this method can effectively filter out interference caused by external noise. By reconstructing the user's trajectory, deep-seated intrinsic features of the user's signature can be extracted; simultaneously, the FISTA algorithm, through multiple iterations, allows the noisy signal to continuously approximate the original signal, thereby achieving a high authentication accuracy.

[0097] The twin network model used in this invention performs authentication based on a set of similarity metrics of signatures, rather than learning features from training samples. Therefore, it is not affected by the amount of signature data or the number of negative signature samples during training. The model can be trained based on a small number of samples, which solves the problem of not being able to collect a large amount of data on-site and the imbalance of positive and negative samples. It can be extended to the classification of signatures with few negative samples.

[0098] The above descriptions are merely a few embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any modifications or alterations made by those skilled in the art without departing from the scope of the technical solution of the present invention using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A signature authentication method for a smart security system in a hydropower station, characterized in that, include: The user's entry signature is reconstructed to obtain the reconstruction speed and reconstruction trajectory. Based on the reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature, a continuous feature sequence is constructed, and the continuous feature sequence is denoised to obtain the denoised feature sequence of the entry signature. The denoised feature sequence of the entry signature and the denoised feature sequence of the registration signature are both input into the Siamese network model to obtain the similarity between the entry signature and the registration signature, and the user is determined to be a legitimate user based on the similarity. Based on the reconstruction speed, reconstruction trajectory, and signature parameters of the entry signature, a continuous feature sequence is constructed, specifically including: Obtain the changes in reconstruction speed, reconstruction trajectory, and entry signature parameters respectively; The reconstruction speed, reconstruction trajectory, signature parameters of the entry signature, and the changes are constructed into a feature sequence; Determine the transitional feature values ​​between adjacent features in the feature sequence to obtain a continuous feature sequence.

2. The signature authentication method for the intelligent security system of a hydropower station according to claim 1, characterized in that, The twin network model includes two identical subnetworks and a common network connected to the outputs of the two subnetworks.

3. The signature authentication method for the intelligent security system of a hydropower station according to claim 2, characterized in that, Each of the sub-networks includes a first long short-term memory network layer, a second long short-term memory network layer, and a first fully connected layer connected in sequence; The output of the first fully connected layer is connected to the public network.

4. The signature authentication method for the intelligent security system of a hydropower station according to claim 3, characterized in that, The public network includes a similarity calculation layer, a third long short-term memory network layer, a flattening layer, and a second fully connected layer connected in sequence. The similarity calculation layer determines the similarity between the entry signature and the registration signature based on the L1 norm.

5. The signature authentication method for the intelligent security system of a hydropower station according to claim 1, characterized in that, The entry signature trajectory is reconstructed to obtain the reconstruction speed and reconstruction trajectory, as follows: The features of each stroke of the entry signature are obtained, and the speed and angular position of each stroke are determined based on the features of each stroke. Determine based on the speed and angular position of each stroke. , Reconstruction speed in the direction; according to , The reconstruction speed in a given direction determines the reconstruction trajectory in that direction.

6. The signature authentication method for the intelligent security system of a hydropower station according to claim 5, characterized in that, Each stroke is characterized by amplitude, start time, logarithmic delay, logarithmic response time, start angle, and end angle.

7. The signature authentication method for the intelligent security system of a hydropower station according to claim 1, characterized in that, The signature parameters include timestamp, status, azimuth angle, and... At least one of shaft angle and pressure.

8. The signature authentication method for the intelligent security system of a hydropower station according to claim 1, characterized in that, The continuous feature sequence is denoised as follows: The FISTA algorithm is used to denoise the continuous feature sequence.

9. A smart security system for a hydropower station, characterized in that, Authentication is performed using the signature authentication method of the intelligent security system for hydropower stations as described in any one of claims 1 to 8.

Images