Safety cabinet control system based on fingerprint identification

Abstract

The invention relates to the technical field of fingerprint identification, and particularly discloses a safety cabinet control system based on fingerprint identification, which is characterized in that after a pressing event is detected, the dynamic change in a fingerprint pressing process is monitored by continuously collecting a fingerprint grayscale image sequence and a time sequence capacitance matrix data sequence in a preset time window; and further introducing a deep learning algorithm, and carrying out time sequence fluctuation modeling and cross-modal interaction response analysis on the two to mine time sequence collaborative association between the fingerprint state and the capacitance characteristic in the pressing process so as to judge whether the pressing event is a real fingerprint or not. And when the fingerprint is judged to be a real fingerprint, a high-quality fingerprint image is called to be compared with the authorized user fingerprint template database so as to verify the identity of the user, and unlocking control of the safety cabinet is carried out. According to the method, by dynamically analyzing the fingerprint characteristics in the pressing process of the user, the accuracy and anti-cheating ability of living fingerprint identification can be effectively improved, and the overall safety of the safety cabinet is improved.

Description

Technical Field

[0001] The present application relates to the field of fingerprint recognition technology, and more specifically to a safety cabinet control system based on fingerprint recognition. Background Art

[0002] With the development of society and the improvement of security awareness, people have an increasing demand for the storage and safekeeping of important documents, valuables, and restricted items. Safes, as an effective physical security device, are widely used in homes, offices, financial institutions, and other places. Traditional safes usually use mechanical locks or password locks as a means of access control, but these methods have many inconveniences. For example, mechanical keys are at risk of loss and duplication, while password systems are vulnerable to peeping attacks and memory burdens. In order to provide more convenient and secure access methods, safe control solutions based on biometric recognition technology have emerged. Among them, fingerprint recognition technology has become the mainstream choice due to its uniqueness, stability, and convenience. Through fingerprint recognition, users no longer need to carry keys or remember passwords. A simple touch can achieve identity authentication, greatly improving the user experience and security of the safe.

[0003] However, existing fingerprint recognition-based safe control solutions mostly use static fingerprint images for recognition, typically relying on capturing a one-time fingerprint image and comparing it with a pre-stored authorized fingerprint template. While static fingerprint recognition offers some convenience, it faces a serious security challenge: susceptibility to fingerprint spoofing attacks. Attackers can create artificial fake fingerprints from various materials (such as silicone, gelatin, and resin) that mimic the texture characteristics of real fingerprints, thereby deceiving static fingerprint recognition systems. This vulnerability allows safes to be opened illegally, seriously threatening the safety of stored items.

[0004] Therefore, we look forward to an optimized fingerprint recognition-based safe cabinet control system that can effectively distinguish real live fingerprints from artificial forged fingerprints to achieve more secure and reliable identity authentication. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a safe control system based on fingerprint recognition, which monitors the dynamic changes in the fingerprint pressing process by continuously collecting the fingerprint grayscale image sequence and the time-series capacitance matrix data sequence within a preset time window after detecting the pressing event, and further introduces a deep learning algorithm, and performs time-series fluctuation modeling and cross-modal interactive response analysis on the two to explore the time-series collaborative correlation between the fingerprint state and the capacitance characteristics during the pressing process, so as to determine whether the pressing event is a real fingerprint. Furthermore, when it is judged to be a real fingerprint, a high-quality fingerprint image is retrieved and compared with the authorized user fingerprint template database to verify the user's identity and perform safe cabinet unlocking control. This method can effectively improve the accuracy and anti-deception ability of live fingerprint recognition by dynamically analyzing the fingerprint characteristics of the user during the pressing process, thereby improving the overall security of the safe.

[0006] Accordingly, according to one aspect of the present application, a fingerprint recognition-based safety cabinet control system is provided, which includes:

[0007] A press trigger module, configured to trigger an acquisition start signal in response to a sensor detecting an initial press event;

[0008] A data acquisition module, configured to acquire a fingerprint grayscale image sequence and a time-series capacitance matrix data sequence within a preset time window after receiving the acquisition start signal;

[0009] A liveness detection module, configured to perform a cross-modal time series joint analysis on the fingerprint grayscale image sequence and the time series capacitance matrix data sequence to obtain a liveness detection result;

[0010] a high-quality fingerprint image extraction module, configured to extract a high-quality fingerprint image from the fingerprint grayscale image sequence in response to the liveness detection result being true;

[0011] The fingerprint recognition module is used to perform fingerprint matching on the high-quality fingerprint image to determine whether to generate a safe unlocking instruction.

[0012] Compared with the prior art, the fingerprint recognition-based safe control system provided by the present application monitors the dynamic changes in the fingerprint pressing process by continuously collecting the fingerprint grayscale image sequence and the time-series capacitance matrix data sequence within a preset time window after detecting the pressing event, and further introduces a deep learning algorithm to conduct time-series fluctuation modeling and cross-modal interactive response analysis on the two to explore the time-series synergistic correlation between the fingerprint state and the capacitance characteristics during the pressing process, thereby determining whether the pressing event is a real fingerprint. Furthermore, when it is determined to be a real fingerprint, a high-quality fingerprint image is retrieved and compared with the authorized user fingerprint template database to verify the user's identity and perform safe cabinet unlocking control. This method can effectively improve the accuracy and anti-deception ability of live fingerprint recognition by dynamically analyzing the fingerprint characteristics of the user during the pressing process, thereby improving the overall security of the safe. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0014] Figure 1 4 is a block diagram of a fingerprint recognition-based safe control system according to an embodiment of the present application.

[0015] Figure 2 Schematic diagram of data flow of a fingerprint recognition-based safe control system according to an embodiment of the present application.

[0016] Figure 3 4 is a block diagram of a liveness detection module in a safe control system based on fingerprint recognition according to an embodiment of the present application.

[0017] Figure 4 4 is a block diagram of a cross-modal feature fusion unit in a fingerprint recognition-based safe control system according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0019] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0020] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0022] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0023] In response to the technical problems described in the above background technology, this application proposes an optimized safe control system based on fingerprint recognition. After detecting a pressing event, it monitors the dynamic changes in the fingerprint pressing process by continuously collecting a sequence of fingerprint grayscale images and a sequence of time-series capacitance matrix data within a preset time window, and further introduces a deep learning algorithm to conduct time-series fluctuation modeling and cross-modal interactive response analysis on the two to explore the time-series synergistic correlation between the fingerprint state and the capacitance characteristics during the pressing process, thereby determining whether the pressing event is a real fingerprint. Furthermore, when it is determined to be a real fingerprint, a high-quality fingerprint image is retrieved and compared with the authorized user fingerprint template database to verify the user's identity and perform safe unlocking control. This method can effectively improve the accuracy and anti-deception ability of live fingerprint recognition by dynamically analyzing the fingerprint characteristics of the user during the pressing process, thereby improving the overall security of the safe.

[0024] Figure 1 4 is a block diagram of a fingerprint recognition-based safe control system according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of a safe control system based on fingerprint recognition according to an embodiment of the present application. Figure 1 and Figure 2As shown, the fingerprint recognition-based safe control system 100 includes: a press trigger module 110, which is used to trigger an acquisition start signal in response to the sensor detecting an initial press event; a data acquisition module 120, which is used to acquire a fingerprint grayscale image sequence and a time-series capacitance matrix data sequence within a preset time window after receiving the acquisition start signal; a liveness detection module 130, which is used to perform a cross-modal time-series joint analysis on the fingerprint grayscale image sequence and the time-series capacitance matrix data sequence to obtain a liveness detection result; a high-quality fingerprint image extraction module 140, which is used to extract a high-quality fingerprint image from the fingerprint grayscale image sequence in response to the liveness detection result being true; and a fingerprint recognition module 150, which is used to perform fingerprint matching on the high-quality fingerprint image to determine whether to generate a safe unlocking instruction.

[0025] In the above-mentioned safe control system based on fingerprint recognition, the press trigger module 110 is used to trigger the acquisition start signal in response to the sensor detecting an initial press event. It should be understood that the user needs to start the fingerprint recognition process by making physical contact with the fingerprint sensor. Therefore, based on the principle of physical contact perception, the present application uses the fingerprint sensor to detect surface pressure and capacitance changes, generates an initial press event signal, and triggers the subsequent fingerprint acquisition process. In this way, the system only starts high-precision data acquisition when a valid contact event occurs, significantly reducing unnecessary energy consumption while avoiding erroneous operations caused by environmental noise.

[0026] In practice, as the finger gradually applies pressure, the pressure value sensed on the sensor surface changes. This change not only reflects the presence of contact but also contains information about the force of the contact. Advanced sensors can even analyze this information to make a preliminary assessment of the quality of the contact. For example, if the pressure value falls within a certain range, it may indicate a valid contact. Conversely, if the pressure value is too low or too high, it may indicate poor contact or an anomaly, such as an attempt to spoof a non-living material.

[0027] At the same time, capacitance changes have also become a key detection indicator. Due to the natural difference in charge distribution between the finger's skin and the sensor's dielectric layer, once a finger touches the sensor surface, the capacitance on each electrode in the electrode array will change accordingly. This change is not static, but dynamically adjusts based on factors such as the finger's pressure, contact area, and skin moisture. Therefore, by monitoring these capacitance changes in real time, not only can the exact moment of finger contact be quickly captured, but the subtle changes that occur during the contact process can also be further analyzed. This provides an important basis for subsequent judgment of whether it is a real finger.

[0028] The sensor integrates a series of precision components specifically designed to measure changes in the aforementioned physical quantities. For example, pressure sensing utilizes highly sensitive pressure sensing elements capable of responding to minute pressure fluctuations in a very short time. These elements are typically arranged across the entire sensor surface, forming a dense grid structure to comprehensively cover all possible contact points. Every time an object touches the sensor surface, regardless of its severity, it causes a localized increase in pressure, which is then captured by the corresponding sensing element. After a series of signal processing steps, such as amplification and filtering, the raw pressure signal is converted into digital form for further analysis by the system.

[0029] The same is true for detecting changes in capacitance. Capacitive sensors utilize a carefully arranged array of electrodes to reflect finger contact by measuring the instantaneous change in capacitance across each electrode. When a finger first contacts the sensor surface, the electric field distribution between the electrodes is immediately affected, causing a significant change in capacitance. Subsequently, as the finger gradually applies more pressure or adjusts its position, these capacitance values ​​continue to fluctuate. The system's built-in circuit design samples these capacitance values ​​at an extremely high frequency, ensuring that even the most subtle changes are not missed. By comparing and analyzing continuously acquired capacitance data streams, the exact moment of first contact can be pinpointed, serving as the key basis for triggering the acquisition start signal.

[0030] To improve the reliability and anti-interference capabilities of the overall system, a combination of various technical approaches is often necessary. In addition to directly monitoring pressure and capacitance, auxiliary functions such as temperature sensing can also be introduced. Because human skin typically maintains a relatively stable temperature range, detecting temperature changes at the contact site can also help confirm whether it is a genuine finger.

[0031] In the above-mentioned fingerprint recognition-based safe control system, the data acquisition module 120 is used to collect fingerprint grayscale image sequences and time-series capacitance matrix data sequences within a preset time window after receiving the acquisition start signal. It should be understood that this application takes into account that the existing static fingerprint recognition scheme is vulnerable to fingerprint spoofing attacks and cannot distinguish between real live fingers and artificial forged fingerprints. During the pressing process, live fingers are affected by various physiological factors such as pulse, pressing force, skin elasticity, etc., and their fingerprint imaging texture and the capacitance response characteristics between the finger epidermis and the sensor dielectric layer will change accordingly. Therefore, in order to capture the unique time-varying characteristics of living organisms, this application is based on the principle of multimodal data time-series synchronous acquisition. Through the collaborative work of optical sensors and capacitive sensors, fingerprint grayscale image sequences and time-series capacitance matrix data sequences (16×16 electrode arrays) are continuously collected at fixed intervals (such as 10ms) within a preset time window (such as 200ms) to fully record the dynamic changes of fingerprints during the pressing process. During this process, the optical sensor obtains the grayscale changes of the fingerprint ridge texture through near-infrared light source reflection imaging, and the capacitive sensor measures the dynamic response of the charge distribution between the epidermis and the dielectric layer through the electrode array. By synchronously recording the spatiotemporal evolution of the fingerprint texture and the changes in the dielectric properties of the subcutaneous tissue, it can provide multimodal complementary time series data support for subsequent liveness detection, thereby achieving more accurate live fingerprint recognition.

[0032] In practice, when a user's finger first touches the sensor surface, the trigger module immediately responds and sends a collection start signal to the data acquisition module. At this point, the data acquisition module quickly activates the optical and capacitive sensors, preparing to begin synchronous collection. To ensure that the collected data is rich and representative, the entire collection process is limited to a short time window, such as 200 milliseconds. During this time, the sensor samples at a frequency of once every 10 milliseconds, ensuring that every subtle change during the press is captured.

[0033] Specifically, the optical sensor relies on near-infrared light source reflection imaging technology. When the user's finger touches the sensor surface, near-infrared light is emitted and reflected from the finger surface. Because moisture and other components in the finger's skin have different absorption and reflectivity rates for near-infrared light, information about the fingerprint ridge texture can be obtained by detecting the reflected light. In this process, each sample generates a grayscale image reflecting the current state of the fingerprint ridges. Over time, as the finger gradually applies pressure and eventually presses fully on the sensor, a series of these grayscale images form a complete sequence of fingerprint grayscale images. These images not only reveal the shape and distribution of the fingerprint ridges, but also include subtle variations caused by factors such as the force and angle of the finger's pressure, providing rich material for subsequent analysis.

[0034] At the same time, the capacitive sensor operates synchronously. This sensor is based on an electrode array design, typically employing a 16×16 electrode layout. When a finger approaches or touches the sensor surface, the charge distribution between the finger's skin and the sensor's dielectric layer changes, causing the capacitance on each electrode to change accordingly. By precisely measuring these changes, a set of data describing the internal structural characteristics of the finger's contact area can be obtained. During the acquisition process, the capacitive sensor also samples at fixed intervals. Each sampling result forms a capacitance matrix, representing the charge distribution between the finger and the sensor at that moment. As the finger presses, these capacitance matrices combine to form a time-series capacitance matrix data sequence. Notably, this capacitance-based data acquisition method not only reflects the morphological characteristics of the finger surface but also detects deeper tissue characteristics, such as blood flow beneath the skin. This is crucial for distinguishing live fingerprints from forged ones.

[0035] Throughout the acquisition phase, optical and capacitive sensors work closely together. The optical sensor focuses on capturing visual information from the fingerprint surface, while the capacitive sensor reveals the electrical properties of the finger's interior. These two sensors complement each other, simultaneously recording both the external morphological changes and internal physiological responses during the fingerprint press process, creating a comprehensive and three-dimensional sequence of fingerprint grayscale images and time-series capacitance matrix data.

[0036] In the above-mentioned safe control system based on fingerprint recognition, the liveness detection module 130 is used to perform cross-modal time series joint analysis on the fingerprint grayscale image sequence and the time series capacitance matrix data sequence to obtain a liveness detection result. Figure 3 FIG is a block diagram of a liveness detection module in a safe control system based on fingerprint recognition according to an embodiment of the present application. Figure 3As shown, the liveness detection module 130 includes: a timing analysis unit 131, which is used to perform sequence analysis on the fingerprint grayscale image sequence and the timing capacitance matrix data sequence to obtain a fingerprint image texture timing fluctuation feature coding vector and a capacitance timing fluctuation feature coding vector; a cross-modal feature fusion unit 132, which is used to perform cross-modal feature fusion on the fingerprint image texture timing fluctuation feature coding vector and the capacitance timing fluctuation feature coding vector to obtain a liveness state multi-modal timing joint coding feature vector; and a liveness detection result generation unit 133, which is used to input the liveness state multi-modal timing joint coding feature vector into the trained liveness detection model to obtain the liveness detection result, which is used to indicate whether it is a real fingerprint.

[0037] Specifically, in a specific example of the present application, the timing analysis unit 131 is used to: input the fingerprint grayscale image sequence and the timing capacitance matrix data sequence into a sequence encoder based on a CNN-RNN hybrid model to perform timing fluctuation feature extraction to obtain the fingerprint image texture timing fluctuation feature encoding vector and the capacitance timing fluctuation feature encoding vector. That is, in order to effectively model the dynamic texture fluctuations (such as grayscale changes caused by ridge micro-displacement and sweat gland secretion) and the timing changes of capacitance response characteristics (such as capacitance fluctuations caused by skin elasticity and blood flow) during the fingerprint pressing process, so as to achieve effective distinction between real live fingerprints and artificial forged fingerprints, the present application adopts a CNN (convolutional neural network) and RNN (recurrent neural network) hybrid architecture to construct a sequence encoder, so as to utilize the powerful spatial feature extraction capability of the CNN model and the sensitivity and memory capability of the RNN model for time series data, perform timing modeling analysis on the fingerprint grayscale image sequence and the timing capacitance matrix data sequence, and capture the subtle dynamic changes during the fingerprint pressing process. Specifically, first, the CNN layer of the sequence encoder performs spatial convolution encoding on the single-frame fingerprint grayscale image and the time-series capacitance matrix data to capture the detailed texture features in the fingerprint grayscale image (such as the shape, direction and relative position relationship of the ridges and valleys) and the local charge distribution features in the time-series capacitance matrix, and generate the time series of the fingerprint image texture features and the time series of the dielectric distribution features. Then, the RNN layer is used to track the temporal fluctuation laws of the fingerprint image texture features and the capacitance distribution features between consecutive frames, and the temporal dependence and dynamic fluctuation pattern of the fingerprint texture features and the dielectric distribution features during the user's pressing process are extracted, thereby generating a fingerprint image texture temporal fluctuation feature encoding vector and a capacitance temporal fluctuation feature encoding vector with high discriminability. In this way, the unique biodynamic characteristics of live fingerprints can be accurately characterized, providing a strong feature basis for subsequent liveness detection and identity authentication.

[0038] Specifically, the cross-modal feature fusion unit 132 is used to perform cross-modal feature fusion on the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector to obtain a live state multi-modal temporal joint coding feature vector. Specifically, due to the nonlinear coupling relationship between the dynamic deformation of the fingerprint texture and the capacitance dielectric response (such as the elastic deformation of the epidermis causing the reconstruction of the local capacitance gradient), simple feature splicing or weighted fusion is difficult to capture the fine-grained collaborative association between the two. Therefore, in order to make full use of the complementarity between multimodal data and deeply explore the nonlinear coupling relationship between the two, the present application proposes a cross-modal feature fusion method, which performs local feature interaction response analysis on the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector to capture the fine-grained collaborative change pattern between the two, and dynamically updates and aggregates the cross-modal global joint features between the fingerprint texture and the capacitance dielectric response through a dynamic reasoning mechanism, thereby generating a live state multi-modal temporal joint coding feature vector with a higher degree of discrimination for forged fingerprints. Among them, Figure 4 FIG is a block diagram of a cross-modal feature fusion unit in a fingerprint recognition-based safe control system according to an embodiment of the present application. Figure 4 As shown, the cross-modal feature fusion unit 132 includes: a local segmentation subunit 1321, which is used to perform local segmentation on the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector to obtain a sequence of fingerprint image texture local temporal fluctuation feature ordered coding vectors and a sequence of capacitance local temporal fluctuation feature ordered coding vectors; a temporal fluctuation interactive response inference subunit 1322, which is used to input each group of corresponding fingerprint image texture local temporal fluctuation feature ordered coding vectors and capacitance local temporal fluctuation feature ordered coding vectors in the sequence of fingerprint image texture local temporal fluctuation feature ordered coding vectors and the sequence of capacitance local temporal fluctuation feature ordered coding vectors into the temporal fluctuation interactive response inference unit to obtain a sequence of fingerprint image texture-capacitance local temporal fluctuation interactive response coding matrices; a local transfer coding subunit 1323, which is used to perform local transfer coding on the sequence of fingerprint image texture-capacitance local temporal fluctuation interactive response coding matrices to obtain the living state multimodal temporal joint coding feature vector.

[0039] More specifically, the local segmentation subunit 1321 is configured to perform an ordered arrangement based on the eigenvalues ​​of the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector to obtain an ordered arrangement coding vector of the fingerprint image texture temporal fluctuation feature and an ordered arrangement coding vector of the capacitance temporal fluctuation feature, which can be expressed as follows:

[0040] V1'=sort(V1)

[0041] V2'=sort(V2)

[0042] Among them, V1 represents the encoding vector of the temporal fluctuation feature of the fingerprint image texture, V2 represents the encoding vector of the temporal fluctuation feature of the capacitance, sort(·) represents the sorting operation on the vector elements, V1' represents the ordered arrangement encoding vector of the temporal fluctuation feature of the fingerprint image texture, and V2' represents the ordered arrangement encoding vector of the temporal fluctuation feature of the capacitance.

[0043] That is, by sorting the internal elements of the fingerprint image texture temporal fluctuation feature encoding vector and the capacitance temporal fluctuation feature encoding vector according to the numerical size of the feature elements, the subsequent interactive analysis process is given a certain invariance or equivariance to the input arrangement, so that the interactive analysis can focus more on the distribution and relative strength of the eigenvalues ​​rather than their initial position in the input, thereby improving the standardization of the feature representation and the robustness of the model to the features.

[0044] More specifically, the local segmentation subunit 1321 is further configured to perform equal-granularity feature segmentation on the ordered arrangement coding vector of the fingerprint image texture temporal fluctuation feature and the ordered arrangement coding vector of the capacitance temporal fluctuation feature to obtain a sequence of ordered coding vectors of the fingerprint image texture local temporal fluctuation feature and a sequence of ordered coding vectors of the capacitance local temporal fluctuation feature, which can be expressed as follows:

[0045] Split(V1')={x1,x2,...,x i ,...,x n}

[0046] Split(V2')={y1,y2,...,y i ,...,y n}

[0047] Among them, x1, x2, x i and x n They represent the first, second, i-th and n-th ordered coding vectors of local temporal fluctuation characteristics of fingerprint image texture in the sequence of ordered coding vectors of local temporal fluctuation characteristics of fingerprint image texture, respectively, n is the number of ordered coding vectors of local temporal fluctuation characteristics of fingerprint image texture, y1, y2, y i and y n represent the first, second, i-th, and n-th ordered coding vectors of the local timing fluctuation characteristics of capacitance, respectively, and Split(·) represents the feature splitting function.

[0048] That is, by synchronously cutting the sorted fingerprint image texture temporal fluctuation feature ordered arrangement coding vectors and the capacitance temporal fluctuation feature ordered arrangement coding vectors along the feature dimension, both are converted into continuous sub-vector segments with the same multiple dimensions to form a sequence of fingerprint image texture local temporal fluctuation feature ordered coding vectors and a sequence of capacitance local temporal fluctuation feature ordered coding vectors. This helps to analyze the temporal collaborative correlation between features at a finer granularity, improve the model's ability to capture subtle dynamic changes during fingerprint pressing, and provide more targeted local feature units for subsequent cross-modal interactive response analysis. At the same time, it can also balance the analysis fineness and computational efficiency by controlling the segmentation granularity, enhance the model's ability to distinguish real fingerprints from forged fingerprints in the dynamic changes of local features, and ensure the comparability and fusion of cross-modal features through a unified segmentation dimension.

[0049] More specifically, the time series fluctuation interactive response reasoning subunit 1322 is expressed as follows:

[0050]

[0051] Among them, W i represents the linear transformation matrix, ReLU(·) represents the ReLU activation function, Represents matrix multiplication, M i Represents x i with y i The fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix.

[0052] That is, by quantizing and encoding the interactive response patterns between the ordered encoding vectors of the local temporal fluctuation characteristics of the fingerprint image texture and the ordered encoding vectors of the local temporal fluctuation characteristics of the capacitance, rich and structured interaction details are captured and retained rather than simple scalar scores, providing more comprehensive information for subsequent analysis. Through deep modeling and quantization encoding, a sequence of fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrices is obtained that can encapsulate detailed interaction information within the corresponding local area, thereby enhancing the model's ability to understand and capture complex collaborative associations and improving the ability to distinguish real fingerprints from forged fingerprints based on their interaction characteristics.

[0053] More specifically, the local transfer coding subunit 1323 is used to input the sequence of the fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix into the LSTM transfer coding module with attention mechanism to obtain the living state multimodal temporal joint coding feature vector, which is expressed as follows:

[0054] vec(M i )=v i

[0055]

[0056] Among them, vec(·) represents the matrix flattening operation, v i Indicates M i The expanded fingerprint image texture-capacitance local temporal fluctuation interaction response encoding vector is obtained, where exp(·) represents the exponential function operation with e as the base. is the feature modulation function based on the attention mechanism, M1 represents the fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix between x1 and y1, and M n Represents x n with y n The fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix, LSTM(·) represents the LSTM model, v f Represents the multimodal temporal joint encoding feature vector of the living state.

[0057] That is, by utilizing the LSTM model's ability to capture the temporal dependencies of sequential data, combined with the attention mechanism to focus on key temporal fluctuation interaction information, the sequence of the fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix is ​​effectively associated with modeling and integration, thereby generating a multimodal temporal joint encoding feature vector of the live state that can comprehensively and accurately characterize the global temporal domain live state during the fingerprint pressing process, providing a deep, highly discriminative feature basis for liveness detection. Through the processing of the LSTM transfer coding module, the unique temporal fluctuation characteristics of real live fingerprints that change over time during the pressing process and the dependencies between cross-modal features can be captured, forming a more robust and discriminative feature expression of the live state, thereby effectively distinguishing real live fingerprints from artificial forged fingerprints, and improving the accuracy and reliability of liveness detection results.

[0058] In particular, considering that the segmentation granularity of the above-mentioned equal-granularity feature segmentation affects the regional granularity size of the fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix, it will also directly affect the mutual representation system between the sequence of the corresponding fingerprint image texture temporal fluctuation feature ordered arrangement coding vector and the capacitor temporal fluctuation feature ordered arrangement coding vector. Therefore, in a preferred example of the present application, the local transfer coding subunit 1323 is used to: based on the temporal fluctuation interaction response strength between each group of corresponding fingerprint image texture local temporal fluctuation feature ordered coding vectors and capacitor local temporal fluctuation feature ordered coding vectors, each fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix in the sequence of the fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix is ​​subjected to feature sparsification constraints to obtain a sequence of optimized fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrices; and then the sequence of the optimized fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix is ​​input into the LSTM transfer coding module with attention mechanism to obtain the multimodal temporal joint coding feature vector of the living state.

[0059] Specifically, since the fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix serves as a cross-domain mutual construction response between the two domains, its row vector is essentially used as a benchmark to characterize the mutual construction space measure, and the row vector modulus is also a representation of the neighborhood constraint factor. If the neighborhood constraint factor, that is, the row vector modulus, is introduced as the neighborhood constraint strength, then the low-rank space measure representation of the fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix, that is, the F norm ‖M i ‖ F It should follow the Poisson distribution relationship:

[0060]

[0061] where (·)! represents factorial, ‖·‖ F represents the Frobenius norm, L represents M i The row vector modulus, e (·) represents an exponential function with a natural constant as the base, and λ represents the Poisson distribution parameter, which characterizes the expected frequency of the interaction between fingerprint image texture and local capacitance temporal fluctuations. From this, the Poisson distribution parameter λ can be solved.

[0062] Thus, when the Poisson distribution process with the local interaction constraint strength L and the mean expectation λ is introduced, ‖x i ‖2 and ‖y i ‖2 describes the boundary correlation representation of the interaction space metric, and further determines the temporal fluctuation interaction response intensity between the two local fingerprint image texture temporal features and the capacitance temporal features as follows:

[0063]

[0064] Where ρ represents x i and y i The cross-domain mutual construction interaction response connection probability between them, |·| represents the absolute value operation, and ‖·‖2 represents the two-norm.

[0065] Then, the local interaction constraint strength L is iteratively modified using the cross-domain mutual construction interaction response connection probability:

[0066] L′=ρL

[0067] Wherein, L′ represents the corrected L.

[0068] Finally, the optimized L′ is used to re-constrain the fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix M i Sparsity:

[0069]

[0070] Among them, M′ i Indicates M i The corresponding optimized fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix.

[0071] In this way, while strictly maintaining the expected connectivity L′, regularization constraints are implemented through the mean mutual construction probability fluctuations of each row to determine the regularization process of the global mutual construction space measure, so that the structured mutual information within the inter-domain mutual construction system can avoid the risk of neighborhood overfitting and enhance the overall representation efficiency of the sequence of the fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix.

[0072] Specifically, in a specific example of the present application, the liveness detection result generating unit 133 is configured to: perform fully-connected encoding on the liveness state multimodal temporal joint encoding feature vector using the fully-connected layer of the liveness detection model to obtain a liveness state multimodal temporal joint fully-connected encoding feature vector; input the liveness state multimodal temporal joint fully-connected encoding feature vector into the softmax classification function of the liveness detection model to obtain probability values ​​of the liveness state multimodal temporal joint encoding feature vector belonging to each classification label, wherein the classification labels include true and false; and determine the classification label corresponding to the largest of the probability values ​​as the liveness detection result. That is, in order to determine whether the input fingerprint is from a real live finger or an artificially forged fingerprint based on the rich biodynamic information contained in the liveness state multimodal temporal joint encoding feature vector, the present application utilizes the classification principle of machine learning to input the liveness state multimodal temporal joint encoding feature vector into the trained liveness detection model for classification judgment, thereby achieving high-precision live fingerprint recognition. Specifically, the liveness detection model is a classification model based on a neural network architecture. It is pre-trained on a data set containing live and various forged samples, and has learned the differences between real live fingerprints and forged fingerprints in biodynamic characteristics such as dynamic deformation of fingerprint texture and capacitance dielectric response. Therefore, it can accurately identify the authenticity of the input fingerprint. During the liveness detection process, the trained liveness detection model learns the feature pattern of the multimodal temporal joint encoding feature vector of the live state, and combines the decision boundary learned in the training to accurately judge whether the input fingerprint is from a real live finger or an artificial forged fake fingerprint, and outputs the corresponding liveness detection result, such as "true" (live) or "false" (forged). In this way, an intelligent liveness discrimination mechanism based on multimodal dynamic features is effectively established, which can effectively filter out fingerprint deception attempts and greatly improve the security and reliability of the fingerprint recognition system.

[0073] In the above-mentioned safe control system based on fingerprint recognition, the high-quality fingerprint image extraction module 140 is used to extract a high-quality fingerprint image from the fingerprint grayscale image sequence in response to the liveness detection result being true. Specifically, since high-precision identity authentication is still required after the liveness detection is passed, there may be image quality fluctuations caused by motion blur or uneven lighting during the dynamic acquisition process. Therefore, in order to screen out the optimal fingerprint image for identity matching, this application is based on the principle of multi-frame preferential enhancement. By performing a quality assessment on each frame image in the fingerprint grayscale image sequence, the image clarity (based on Laplacian gradient value), contrast (based on grayscale histogram entropy) and integrity (based on ridge continuity detection) are comprehensively considered. The comprehensive quality score of each frame of the fingerprint grayscale image is determined by calculating the weighted sum of various quality indicators, thereby selecting the image with the highest score as the high-quality fingerprint image for subsequent fingerprint feature extraction and comparison. Through this screening process, the high-quality fingerprint image not only has a clear ridge and valley structure, but also accurately reflects the detailed features of the fingerprint, providing a reliable basis for subsequent identity authentication.

[0074] In the above-mentioned safe control system based on fingerprint recognition, the fingerprint recognition module 150 is used to perform fingerprint matching on the high-quality fingerprint image to determine whether to generate a safe unlocking instruction. In a specific example of the present application, the fingerprint recognition module 150 is used to compare the high-quality fingerprint image with the authorized user fingerprint template database to obtain a fingerprint recognition result. That is, in order to achieve fast and accurate identity authentication based on liveness detection, the present application adopts a matching algorithm based on deep learning (Deep Learning-based Matching), and by constructing a deep neural network, performs deep feature extraction and representation learning on the fingerprint features in the high-quality fingerprint image to generate a fingerprint feature vector with high robustness and discriminability. At the same time, in the authorized user registration stage, the fingerprint sample of each authorized user is pre-collected, quality screened and pre-processed, and its fingerprint features are extracted using the same deep neural network architecture, and an authorized user fingerprint template database containing the fingerprint feature vectors of all authorized users is constructed. During the comparison phase, the fingerprint feature vector of the high-quality fingerprint image to be authenticated is compared one by one with each fingerprint feature vector in the fingerprint template database, and the degree of match between the fingerprint to be authenticated and the authorized user's fingerprint is measured by calculating the similarity between the fingerprint feature vectors (such as cosine similarity). When the similarity between the fingerprint to be authenticated and the fingerprint of an authorized user exceeds the preset similarity threshold, the match is considered successful; if the similarity between the fingerprint to be authenticated and the fingerprints of all authorized users does not reach the preset similarity threshold, the match is considered to have failed, and the identity authentication request is rejected. Based on this, this application realizes dual verification of the input fingerprint by combining liveness detection with high-precision fingerprint comparison, thereby ensuring the security and accuracy of identity authentication.

[0075] In a specific example of the present application, the fingerprint recognition module 150 is further used to: in response to the fingerprint recognition result being a successful match, generate the safe unlocking instruction. That is, after confirming that the visitor is a real authorized user, the unlocking operation is triggered, and the safe unlocking instruction is generated to the control unit connected to the safe lock, driving the lock's actuator (for example, an electromagnet, a motor or other electronic latch) to release the locked state, so that the safe door can be opened. Through this process, the electronic authentication process is converted into actual physical access permission, completing the entire process of safe control based on fingerprint recognition, ensuring that only users who have passed strict verification can open the safe.

[0076] In summary, the fingerprint recognition-based safe control system based on the embodiment of the present application is explained. After detecting a pressing event, it monitors the dynamic changes in the fingerprint pressing process by continuously collecting the fingerprint grayscale image sequence and the time-series capacitance matrix data sequence within a preset time window, and further introduces a deep learning algorithm. By performing time-series fluctuation modeling and cross-modal interactive response analysis on the two, the time-series synergistic correlation between the fingerprint state and the capacitance characteristics during the pressing process is mined, thereby determining whether the pressing event is a real fingerprint. Furthermore, when it is determined to be a real fingerprint, a high-quality fingerprint image is retrieved and compared with the authorized user fingerprint template database to verify the user's identity and perform safe cabinet unlocking control. This method can effectively improve the accuracy and anti-deception ability of live fingerprint recognition by dynamically analyzing the fingerprint characteristics of the user during the pressing process, thereby improving the overall security of the safe.

[0077] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in the present invention are merely illustrative and non-limiting, and should not be construed as necessarily possessed by each embodiment of the present invention. Furthermore, the specific details of the above embodiments are provided for illustrative purposes and to facilitate understanding, and are not intended to be limiting. These details do not necessarily limit the present invention to being implemented using these specific details.

[0078] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0079] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0080] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0081] Finally, it should be noted that the above description has been provided for purposes of illustration and description. Furthermore, the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. The safety cabinet control system based on fingerprint recognition is characterized by: include: A press trigger module, configured to trigger an acquisition start signal in response to a sensor detecting an initial press event; A data acquisition module, configured to acquire a fingerprint grayscale image sequence and a time-series capacitance matrix data sequence within a preset time window after receiving the acquisition start signal; A liveness detection module, configured to perform a cross-modal time series joint analysis on the fingerprint grayscale image sequence and the time series capacitance matrix data sequence to obtain a liveness detection result; a high-quality fingerprint image extraction module, configured to extract a high-quality fingerprint image from the fingerprint grayscale image sequence in response to the liveness detection result being true; The fingerprint recognition module is used to perform fingerprint matching on the high-quality fingerprint image to determine whether to generate a safe unlocking instruction.

2. The fingerprint recognition-based safe control system according to claim 1, characterized in that: The living body detection module includes: A time series analysis unit, configured to perform sequence analysis on the fingerprint grayscale image sequence and the time series capacitance matrix data sequence to obtain a fingerprint image texture time series fluctuation feature coding vector and a capacitance time series fluctuation feature coding vector; A cross-modal feature fusion unit, configured to perform cross-modal feature fusion on the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector to obtain a living state multi-modal temporal joint coding feature vector; The liveness detection result generating unit is used to input the liveness state multimodal time series joint encoding feature vector into the trained liveness detection model to obtain the liveness detection result, and the liveness detection result is used to indicate whether it is a real fingerprint.

3. The fingerprint recognition-based safe control system according to claim 2, characterized in that: The timing analysis unit is used to: The fingerprint grayscale image sequence and the temporal capacitance matrix data sequence are input into a sequence encoder based on a CNN-RNN hybrid model to extract temporal fluctuation features to obtain the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector.

4. The fingerprint recognition-based safe control system according to claim 3 is characterized in that: The cross-modal feature fusion unit includes: A local segmentation subunit, configured to perform local segmentation on the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector to obtain a sequence of fingerprint image texture local temporal fluctuation feature ordered coding vectors and a sequence of capacitance local temporal fluctuation feature ordered coding vectors; a timing fluctuation interaction response inference subunit, configured to input each corresponding group of fingerprint image texture local timing fluctuation feature ordered coding vectors and capacitor local timing fluctuation feature ordered coding vectors in the sequence of fingerprint image texture local timing fluctuation feature ordered coding vectors and the sequence of capacitor local timing fluctuation feature ordered coding vectors into the timing fluctuation interaction response inference unit to obtain a sequence of fingerprint image texture-capacitance local timing fluctuation interaction response coding matrices; The local transfer coding subunit is used to perform local transfer coding on the sequence of the fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix to obtain the living state multimodal temporal joint coding feature vector.

5. The fingerprint recognition-based safe control system according to claim 4 is characterized in that: The local area segmentation subunit is configured to: Performing an ordered arrangement of the fingerprint image texture temporal fluctuation feature coding vector and the capacitance temporal fluctuation feature coding vector based on the size of the eigenvalues ​​to obtain a fingerprint image texture temporal fluctuation feature ordered arrangement coding vector and a capacitance temporal fluctuation feature ordered arrangement coding vector; The ordered arrangement coding vector of the fingerprint image texture temporal fluctuation feature and the ordered arrangement coding vector of the capacitance temporal fluctuation feature are subjected to equal-granularity feature segmentation to obtain a sequence of the ordered coding vector of the fingerprint image texture local temporal fluctuation feature and a sequence of the ordered coding vector of the capacitance local temporal fluctuation feature.

6. The fingerprint recognition-based safe control system according to claim 5, characterized in that: The local area transmission encoding subunit is used to: The sequence of the fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix is ​​input into the LSTM transfer encoding module with an attention mechanism to obtain the living state multimodal temporal joint encoding feature vector.

7. The fingerprint recognition-based safe control system according to claim 5, characterized in that: The local area transmission encoding subunit is used to: Based on the temporal fluctuation interaction response strength between each group of corresponding fingerprint image texture local temporal fluctuation feature ordered coding vectors and capacitance local temporal fluctuation feature ordered coding vectors, feature sparsification constraints are performed on each fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrix in the sequence of fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrices to obtain a sequence of optimized fingerprint image texture-capacitance local temporal fluctuation interaction response coding matrices; The sequence of the optimized fingerprint image texture-capacitance local temporal fluctuation interaction response encoding matrix is ​​input into an LSTM transfer encoding module with an attention mechanism to obtain the living state multimodal temporal joint encoding feature vector.

8. The fingerprint recognition-based safe control system according to claim 2, characterized in that: The living body detection result generating unit is used to: Using the fully connected layer of the liveness detection model to perform fully connected encoding on the liveness state multimodal temporal joint encoding feature vector to obtain the liveness state multimodal temporal joint fully connected encoding feature vector; Inputting the liveness state multimodal time series joint fully connected encoding feature vector into the Softmax classification function of the liveness detection model to obtain probability values ​​of the liveness state multimodal time series joint encoding feature vector belonging to each classification label, wherein the classification labels include true and false; The classification label corresponding to the largest probability value among the probability values ​​is determined as the living body detection result.

9. The fingerprint recognition-based safe control system according to claim 1, characterized in that: The fingerprint recognition module is used to: Comparing the high-quality fingerprint image with a database of authorized user fingerprint templates to obtain a fingerprint recognition result; In response to the fingerprint recognition result being a successful match, the safe unlocking instruction is generated.

Images