User biological feature recognition method based on improved deep learning algorithm and safety tear ejector

Abstract

The invention provides a user biological feature recognition method based on an improved deep learning algorithm and a safe tear ejector, and the method comprises the steps: a data collection module carries out fingerprint and skin data collection, feature extraction and conversion, and feature representation and classification of a capsule network; the method comprises the steps of application of an optimization algorithm, system output and user feedback, and system self-learning and updating. The ejector comprises a storage bottle used for storing tear fluid and a holding cover arranged at the top end of the storage bottle, and through the arrangement of the mistaken touch prevention structure, the accidental release probability in the non-use state is greatly reduced, the safety of equipment is improved, and the mistaken touch risk is remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence biometric identification, and in particular to a user biometric identification method based on an improved deep learning algorithm and a safety tear gas sprayer. Background Art

[0002] In the field of police equipment technology, especially in dealing with emergencies, maintaining order or anti-riot operations, tear gas sprayers are widely used as non-lethal weapons to control and disperse crowds. Tear gas sprayers usually spray liquid or gas containing tear gas components to cause temporary visual blurring, respiratory irritation and other discomfort to the target, thereby achieving the effect of control and dispersal. However, the existing tear gas sprayers have the following problems during use:

[0003] Risk of accidental activation: In a tense and chaotic environment, tear gas sprayers may be accidentally triggered, which may not only cause harm to innocent people, but also cause harm to the user himself.

[0004] Safety: Existing tear gas sprayers may lack adequate safety considerations in their design, such as insufficient protective measures, which may result in the accidental release of tear gas when not in use.

[0005] The contradiction between ease of use and accidental touch: In order to ensure that tear gas sprayers can be used quickly and effectively in emergency situations, the equipment is usually designed to be easy to operate. However, this design idea often conflicts with the requirement to prevent accidental touches. How to balance ease of use and safety is a challenge.

[0006] Therefore, there is a problem that needs to be solved in the prior art, namely, how to design a user biometric recognition method and tear gas sprayer that can ensure quick and effective use in emergency situations while effectively avoiding accidental touches and improving safety and target recognition accuracy. Summary of the invention

[0007] In order to overcome the existing problems, the embodiments of the present application provide an identity recognition method based on an improved deep learning mechanism algorithm and a safe tear gas sprayer thereof. By introducing an intelligent algorithm for identity recognition, the probability of accidental release when not in use is greatly reduced, the safety of the device is improved, and the risk of accidental touch is significantly reduced.

[0008] The technical solution adopted by the embodiment of the present application to solve the technical problem is:

[0009] A user biometric recognition method based on an improved deep learning algorithm uses a transformer model and a capsule network algorithm to design a complex and efficient user identity recognition method, which is applied in handheld devices for security operations.

[0010] Furthermore, the specific implementation process of designing a complex and efficient user identity recognition method using the transformer model and capsule network algorithm is as follows:

[0011] Step 1: Data Collection Module

[0012] 1. Fingerprint recognition

[0013] Fingerprint image acquisition: Use a high-resolution fingerprint sensor to collect the user's fingerprint image:

[0014]

[0015] I f is the collected fingerprint image, is the pixel value of the fingerprint image at position (i, j), i is the row index, M is the total number of rows in the fingerprint image matrix, j is the column index, and N is the total number of columns in the fingerprint image matrix;

[0016] 2. Skin resistance measurement

[0017] Resistance measurement: Use a skin resistance sensor to measure the resistance of the user's finger as an auxiliary biometric feature. The measured resistance value will vary depending on the individual's skin characteristics and humidity:

[0018]

[0019] R s is the skin resistance value, V s is the voltage, I s is the current;

[0020] Voltage V s and current I s It is measured by the skin resistance sensor, and the resistance value R s It is calculated by Ohm's law. In order to improve the recognition accuracy, the noise effect can be reduced by taking the average value through multiple measurements:

[0021]

[0022] is the average skin resistance value, To calculate the sum of all measured values ​​divided by the number of measurements n, is the sum from the first measurement to the nth measurement, represents the skin resistance value measured at the kth time;

[0023] Step 2: Feature extraction and conversion

[0024] 1. Input data representation

[0025] Fingerprint images and skin resistance data first need to be converted into an input data form that the model can process:

[0026] Fingerprint image matrix: I f , the size is M×N;

[0027] Skin resistance value: can be considered as a scalar;

[0028] Flatten the fingerprint image into a vector and combine the skin resistance value with the fingerprint feature vector:

[0029]

[0030] x0 is the initial input vector, flatten(I f ) is to flatten the fingerprint image matrix into a one-dimensional vector, is the skin resistance value;

[0031] 2. Self-Attention Mechanism in Transformer Model

[0032] The self-attention mechanism generates weights by calculating the relationship between each element in the input vector and other elements, thereby highlighting important features:

[0033]

[0034] Attention(Q, K, V) is the output of the self-attention mechanism, Q is the query matrix, K is the key matrix, V is the value matrix, and softmax() is the softmax operation on the dot product results of each row. is the matrix obtained after the dot product result is scaled, d k is the dimension of the key vector;

[0035] 3. Converter output

[0036] After being processed by the multi-head self-attention mechanism and the feedforward network layer, the transformer model converts the input data into high-dimensional feature representations. These high-dimensional feature vectors will be passed to the next capsule network for further processing:

[0037] z=Transformer(x0)

[0038] z is the high-dimensional feature vector output by the transformer model, Transformer() is the overall operation process of the transformer model, and x0 is the input vector;

[0039] 4. Transfer of feature vectors

[0040] The high-dimensional feature vector z generated by the transformer model is passed as input to the capsule network for further feature analysis and classification:

[0041] z→CapsuleNetwork

[0042] z is the high-dimensional feature vector output by the transformer model, and CapsuleNetwork is the capsule network;

[0043] Step 3: Feature representation and classification of capsule network

[0044] 1. Main capsule layer

[0045] Feature extraction: The main capsule layer receives the high-dimensional feature vector z from the transformer model and performs preliminary feature extraction through multiple capsules, each of which is a small neural network that processes the input features and generates low-level feature representations:

[0046] u i =W i z+b i

[0047] u i is the output of the ith primary capsule, W i is the weight matrix of the ith primary capsule, z is the high-dimensional feature vector output by the transformer model, and b i is the bias vector of the i-th primary capsule;

[0048] 2. Dynamic Routing Algorithm

[0049] Dynamic routing: low-level feature vector u i It is passed to the sub-capsule layer through the dynamic routing algorithm, which dynamically adjusts the connection weights according to the relevance of the input data to ensure that important features can be better passed to high-level capsules:

[0050]

[0051] v j is the output of the j-th capsule, squash() is a nonlinear activation function, ∑ i is the summation symbol, c ij is the dynamic routing coefficient, u ij The output feature vector u of the main capsule layer i The result after weight transformation;

[0052] 3. Sub-capsule layer

[0053] Feature fusion: The secondary capsule layer not only receives the feature vector v transmitted by the main capsule layer through dynamic routing j , combined with skin resistance data Further processing is performed to generate the final high-level feature representation:

[0054]

[0055] o k is the output of the kth capsule, W k is the weight matrix of the kth capsule, v j is the high-level feature vector generated by the sub-capsule layer through dynamic routing, is the skin resistance data, b k is the bias vector of the kth capsule;

[0056] 4. Classification and identification

[0057] Classification output: final high-level feature representation k Pass it to the classification layer and use the softmax function to classify and determine the identity of the user:

[0058]

[0059] is the classification result, softmax(O) is the softmax function;

[0060] Step 4: Application of optimization algorithm

[0061] 1. Initialization

[0062] Random initialization: Randomly initialize the weights and bias parameters of the capsule network, which are regarded as the initial solution in the black hawk population in BHO:

[0063]

[0064] W(0) is the initial weight matrix, b(0) is the initial bias vector, is a uniform distribution, ∈ is the range of distribution;

[0065] 2. Iterative Optimization

[0066] Black Hawk position update: According to the BHO algorithm, the position of the black hawk is updated each iteration, gradually converging to the optimal parameter set. In BHO, the position of each black hawk is represented as a set of capsule network parameters, including weight W and bias b:

[0067]

[0068] is the new position of the i-th black hawk after the t+1th iteration, is the position of the i-th black hawk at the t-th iteration, α is the learning rate, is the black hole position, β is the random perturbation factor, and r is the random vector;

[0069] 3. Fitness function calculation

[0070] Fitness calculation: After each iteration, the performance of each black hawk position in the capsule network is evaluated. The fitness function is usually the recognition accuracy or error function of the capsule network on the validation set:

[0071] f(X i )=Accuracy(W i , b i )or f(X i )=-Loss(W i , b i )

[0072] f(X i ) is the fitness function value, Accuracy(W i , b i ) is based on the current weight W i and bias b i The recognition accuracy, Loss(W i , b i ) is based on the current weight W i and bias b i The loss function value of

[0073] 4. Termination Conditions

[0074] Termination of optimization: The optimization process terminates when any of the following conditions is met:

[0075] The recognition accuracy reaches the set threshold;

[0076] The maximum number of iterations has been reached;

[0077] Terminate if f(X bh )≥Threshold or t≥t max

[0078] Terminate is the operation to terminate the optimization process, if is the conditional judgment symbol, f(X bh ) is the fitness function value of the current optimal solution, Threshold is the preset fitness threshold, t is the current number of iterations, t max is the maximum number of iterations allowed;

[0079] Step 5 System output and user feedback

[0080] 1. Calculation of identity recognition results

[0081] Classification output: The capsule network finally outputs a vector representing the classification probability of different identities. The softmax function converts these output values ​​into probability distribution to determine the identity of the user:

[0082]

[0083] is the predicted probability of the i-th class, softmax(O i ) is the softmax function, O i is the score value of the i-th class output by the capsule network, is the exponential operator, is to accumulate all categories of j from 1 to C, where C is the total number of categories, o j is the score value of the jth class output by the capsule network;

[0084] 2. Identity Verification and Decision Making

[0085] Identification decision: The system determines the user's identity category based on the maximum probability value. If the category corresponding to the maximum probability value is an authorized identity, the identification is successful and the user is allowed to operate the tear gas sprayer; otherwise, the identification fails and the system locks the sprayer:

[0086]

[0087] Decision is the final decision result of the system, Allow Operation means that the system allows the user to operate the tear gas sprayer (when the system is unlocked, the electromagnetic lock tongue is in the state of being recovered in the electromagnetic lock under the control of the controller), LockSystem means that the system locks the tear gas sprayer (when the system is locked, the lock tongue is engaged in the groove on the protective tube to limit the rotation of the protective tube), if is the conditional judgment symbol, is the maximum probability value in the softmax output, is the predicted probability value of the authorized identity category, δ is the recognition threshold, and otherwise indicates that the condition is not satisfied;

[0088] 3. User Feedback

[0089] Feedback mechanism: The system will feed back the recognition results and decisions to the user. If the recognition is successful, the user can continue to operate the tear gas sprayer. If the recognition fails, the system will lock and prompt the user that the identity authentication failed:

[0090]

[0091] Feedback is the information content fed back to the user by the system, Success Message is the message fed back to the user when the system recognizes successfully, Failure Message is the message fed back to the user when the system recognizes failed, Decision is the final decision result of the system, Allow Operation means the system's decision is to allow the operation (system unlocked state), and LockSystem means the system's decision is to lock the system (system locked state).

[0092] Furthermore, it also includes step 6: system self-learning and updating, the specific process is:

[0093] 1. Adaptive Learning

[0094] Data collection and update: The system regularly collects new user data, including fingerprint images and skin resistance values. Through the continuous learning mechanism, the system uses this new data to update the transformer model and capsule network to ensure that the model can adapt to changes in different environments and users:

[0095]

[0096] For the updated dataset, is the training data set of the system at time t, ∪ is the set union operator,

[0097] is the new data collected at time t+1;

[0098] Model update: use the updated dataset Retrain the transformer model and capsule network and update the model parameters to maintain high performance under the new data distribution:

[0099]

[0100] θ t+1 is the updated model parameter at time t+1, argmin θ To minimize the loss by optimizing the loss function l, θ is the set of model parameters, For the updated dataset, For the updated dataset The number of samples in is the loss function, f θ (x) is the prediction result of the model for the input feature vector x, where x is the input feature vector and y is the target label;

[0101] 2. Enhanced security

[0102] Model parameter re-optimization: In order to prevent the recognition algorithm from being attacked or cracked, the system regularly uses the Black Hawk optimization algorithm to re-optimize the parameters of the transformer model and capsule network to enhance the security of the system:

[0103]

[0104] is the optimized model parameter at time t+1, argmaxθ is the parameter θ that makes the expression in the brackets take the maximum value, BHO is the Black Hawk optimization algorithm, For a given data set Under the condition of , the parameter θ is optimized. is the data set used by the system at time t+1;

[0105] Enhanced defense mechanism: By periodically optimizing model parameters, the system can continuously improve its defense capabilities against potential attacks and ensure the security and reliability of the identification process:

[0106]

[0107] Security t+1 is the security level of the system at time t+1, Function is an abstract function that represents the relationship between system security and model parameters. are the optimized model parameters at time t+1.

[0108] A safe tear gas sprayer applied to a user biometric identification method based on an improved deep learning algorithm of the present invention comprises a storage bottle for storing tear gas, and a holding cap arranged on the top of the storage bottle, wherein the holding cap is connected to the storage bottle and is used to spray the tear gas in the storage bottle;

[0109] Among them, the outer wall of the holding cover is provided with a protective cylinder, the protective cylinder and the holding cover are rotatable structures, a limit plate is provided inside the protective cylinder, the limit plate is located on the upper surface of the holding cover, limits the holding cover, and is used to prevent accidental touch operations. A holding cover is provided, and a protective cylinder is screwed on the outer wall of the holding cover, the protective cylinder protects the holding cover, and an accidental touch plate is provided inside the holding cover, the protective cylinder limits the accidental touch plate, and the anti-accidental touch structure greatly reduces the probability of accidental release in the non-use state, improves the safety of the equipment, and significantly reduces the risk of accidental touch. At the same time, the accidental touch plate inside the holding cover can be flipped over, and a limit plate is provided at the corresponding position of the protective cylinder and the accidental touch plate, the limit plate limits the accidental touch plate, and avoids the problem of flipping of the accidental touch plate during non-operation.

[0110] A miniature electromagnetic lock is built into the inner wall of the protective tube. The electromagnetic lock is connected to a controller with a built-in algorithm as an actuator. The input end of the controller is connected to a fingerprint sensor and a skin resistance sensor, wherein the fingerprint sensor and the skin resistance sensor are in contact with the outer surface of the storage bottle, and there is a fingerprint collection information position prompt on the outer surface of the storage bottle; the controller is built into the storage bottle and the holding cover, and the electromagnetic lock is placed horizontally. The lock tongue of the electromagnetic lock corresponds to the groove on the protective tube (when the lock tongue pops out, it is engaged in the groove to limit the rotation of the protective tube; when the lock tongue rebounds, the groove releases the restriction and the protective tube can rotate). When the system is in a locked state, the lock tongue is used to limit the rotation of the protective tube, and is unlocked after successful identity recognition. When the system is unlocked, the lock tongue is retracted into the electromagnetic lock to unlock the rotation of the protective tube.

[0111] Preferably, the holding cover is provided with a squeeze button at the intersection of the storage bottle, and the holding cover is provided with a card slot at a position corresponding to the squeeze button;

[0112] Among them, a flip-up false touch plate is provided inside the card slot, and the false touch plate is located at the top of the extrusion button. A holding cover is provided, and a protective tube is screwed on the outer wall of the holding cover to protect the holding cover. The false touch plate is provided inside the holding cover to limit the false touch plate. Through the setting of the anti-false touch structure, the probability of accidental release in the non-use state is greatly reduced, the safety of the equipment is improved, and the risk of false touch is significantly reduced.

[0113] Preferably, a spray port is embedded in one end of the holding cover away from the false touch plate, a pipe is provided inside the spray port, and the other end of the pipe is plugged into the storage bottle, the spray port is connected to the raw material inside the storage bottle through the internal pipe, and the pressure change is provided by squeezing a button, so that the raw material inside the storage bottle is sprayed out from the spray port.

[0114] Preferably, the protective tube is a hollow cylindrical structure, the inner cross-sectional diameter of the protective tube is the same as the outer cross-sectional diameter of the holding cover, and a limiting groove is provided at the intersection of the holding cover and the protective tube, so that the protective tube rotates through the limiting groove when the outer wall of the holding cover rotates, thereby avoiding the problem of deviation between the holding cover and the protective tube.

[0115] Preferably, a groove is provided at one end of the protective tube away from the limiting plate, and the size of the groove is larger than the size of the false touch plate, so that after the protective tube is rotated, when the squeezing button is pressed, the limiting effect is avoided by the provided groove.

[0116] Preferably, a protective plate is provided on the inner upper surface of the groove. When the protective tube is connected to the holding cover and not in use, the protective plate inside the protective tube contacts the injection port to protect the injection port. The protective plate is an arc-shaped structure, and the size of the protective plate is larger than the size of the injection port.

[0117] Preferably, the protective tube is provided with a through hole at a position corresponding to the injection port, and the size of the through hole is larger than the size of the injection port. When the protective tube is connected to the holding cover and is not in use, the protective plate inside the protective tube contacts the injection port to protect the injection port. When in use, the protective tube rotates to separate the protective plate from the injection port, and then the through hole rotates to the position corresponding to the injection port, so that the raw material injected from the injection port is ejected from the through hole.

[0118] Preferably, the storage bottle is a hollow cylindrical structure, and an observation window is provided on the surface wall of the storage bottle. Through the setting of the observation window, the capacity inside the storage bottle can be easily observed, which facilitates timely filling.

[0119] Preferably, the holding cover and the storage bottle are detachable structures. If the storage bottle is damaged, the holding cover and the storage bottle can be disassembled for quick repair, thereby improving the service life and convenience of repair.

[0120] The advantages of the embodiments of the present application are:

[0121] 1. A holding cover is provided, and a protective tube is screwed on the outer wall of the holding cover to protect the holding cover. An error-touch plate is provided inside the holding cover to limit the error-touch plate. The anti-error-touch structure greatly reduces the probability of accidental release when not in use, improves the safety of the equipment, and significantly reduces the risk of error-touch.

[0122] 2. A through hole is opened inside the protective tube, and the size of the through hole is larger than the size of the injection port. A protective plate is provided on the side of the protective tube away from the through hole, and the size of the protective plate is larger than the size of the injection port. In this way, the protective plate protects the injection port when the protective tube is prevented from accidental touch. When the protective tube rotates, the through hole inside the protective tube coincides with the injection port, so that the injection port can be sprayed to avoid blockage problems.

[0123] 3. The error touch plate inside the holding cover can be flipped over, and a limit plate is provided at the corresponding position between the protective tube and the error touch plate. The limit plate limits the error touch plate to avoid the error touch plate flipping over during non-operation.

[0124] 4. By introducing intelligent algorithms and adaptive learning mechanisms, the system can optimize the parameter settings of the transformer model and capsule network in real time to ensure efficient identity recognition capabilities under different environments and user conditions. Combined with the re-optimization steps of the Black Hawk optimization algorithm, the security of the system has been further improved, effectively preventing the recognition algorithm from being attacked or cracked, and ensuring the reliability of the tear gas sprayer and the safety of the user. This intelligent design not only improves the protection performance of the equipment, but also enhances the system's adaptability to complex usage scenarios.

[0125] The present invention effectively improves the safety and intelligence level of the tear gas sprayer through a series of innovative designs. First, a holding cover and a protective tube are provided, and the protective tube cooperates with the mis-touch plate to greatly reduce the risk of mis-touch when not in use. Secondly, the through-hole design in the protective tube ensures that the spray port is unobstructed when necessary, avoiding blockage problems that may occur during operation. In addition, the setting of the limit plate prevents the mis-touch plate from accidentally flipping over when not in operation. Finally, by introducing intelligent algorithms, adaptive learning mechanisms and Black Hawk optimization algorithms, the system can not only continuously optimize in different environments, but also effectively prevent external attacks, further improving the reliability and adaptability of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0126] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0127] Figure 1 This is a schematic diagram of the overall structure of the front side of the safety tear gas sprayer of the present invention;

[0128] Figure 2 This is a schematic diagram of the overall structure of the back of the safety tear gas sprayer of the present invention;

[0129] Figure 3 It is a schematic diagram of the overall structure of the front protective tube of the safety tear gas sprayer of the present invention;

[0130] Figure 4 It is a schematic diagram of the rotating structure of the back protective tube of the safety tear gas sprayer of the present invention;

[0131] Figure 5 This is a schematic diagram of the overall structure of the storage bottle in the safe tear gas sprayer of the present invention;

[0132] Figure 6 This is a schematic diagram of the overall structure of the back of the storage bottle in the safe tear gas sprayer of the present invention;

[0133] Figure 7 This is a schematic diagram of the overall structure of the back side of the holding cover of the safety tear gas sprayer of the present invention;

[0134] Figure 8 This is a schematic diagram of the overall structure of the front side of the holding cover of the safety tear gas sprayer of the present invention;

[0135] Fig. 9 This is a schematic diagram of the overall structure of the front side of the protective tube in the safety tear gas sprayer of the present invention;

[0136] Fig.10 It is a schematic diagram of the overall structure of the back side of the protective tube in the safety tear gas sprayer of the present invention;

[0137] Fig.11 The final result diagram for intelligent identification and safety optimization of the present invention;

[0138] Fig.12 This is a flow chart of intelligent identification and security optimization of the present invention.

[0139] Description of main reference numerals:

[0140] 1. Storage bottle; 2. Holding cover; 3. Protective tube; 4. Observation window; 5. Injection nozzle; 6. Mis-touch plate; 7. Squeeze button; 8. Limit plate; 9. Groove; 10. Protective plate; 11. Through hole; 12. Electromagnetic lock. DETAILED DESCRIPTION

[0141] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, for the convenience of description below, the referenced "upper", "lower", "left", and "right" are the same as the upper, lower, left, and right directions of the drawings themselves. The "first", "second", etc. below are distinguished for the description and have no other special meanings.

[0142] The embodiment of the present application solves the problems in the prior art by providing a safe tear gas sprayer. The invention is provided with a holding cover, and a protective tube is screwed on the outer wall of the holding cover to protect the holding cover. An error-touch plate is provided inside the holding cover to limit the error-touch plate. By setting the anti-error-touch structure, the probability of accidental release in a non-use state is greatly reduced, the safety of the device is improved, and the risk of error-touch is significantly reduced. A through hole is provided inside the protective tube, and the size of the through hole is larger than the size of the spray port. A protective plate is provided on the side of the protective tube away from the through hole, and the size of the protective plate is larger than the size of the spray port. In this way, the protective plate protects the spray port when the protective tube is preventing error-touch. When the protective tube rotates, the through hole inside the protective tube coincides with the spray port, so that the spray port can be sprayed to avoid blockage. The error-touch plate inside the holding cover can be flipped, and a limit plate is provided at the corresponding position of the protective tube and the error-touch plate. The limit plate limits the error-touch plate to avoid the error-touch plate flipping during non-operation.

[0143] like Fig. 9 A miniature electromagnetic lock 12 (a miniature ksj-0520 electromagnetic lock is selected, Dongguan Kaisijin Intelligent Technology Co., Ltd.) is built into the inner wall of the protective cylinder 3. The electromagnetic lock is connected to a controller with a built-in algorithm as an actuator. The input end of the controller is connected to a fingerprint sensor and a skin resistance sensor, wherein the fingerprint sensor and the skin resistance sensor are in contact with the outer surface of the storage bottle 1, and the outer surface of the storage bottle 1 has a fingerprint collection information position prompt; the controller is built between the storage bottle 1 and the holding cover 2, the electromagnetic lock is placed horizontally, and the lock tongue of the electromagnetic lock corresponds to the groove on the protective cylinder 3. When the system is in a locked state, the lock tongue is used to limit the rotation of the protective cylinder 3; after the identity recognition is successful, the system is in an unlocked state, and the lock tongue is retracted into the electromagnetic lock to unlock the rotation of the protective cylinder 3.

[0144] The technical solution in the embodiment of the present application is to solve the above problems, and the overall idea is as follows:

[0145] Example 1: Figure 11-12,A user biometric recognition method based on an improved deep learning algorithm, a complex and efficient user identity recognition method is designed using the transformer model and the capsule network algorithm, which is applied in handheld devices for security operations.

[0146] The system needs to accurately identify the user in different environments and user conditions and avoid misoperation. To achieve this goal, the parameters of the transformer model and capsule network must be optimized to maximize the recognition accuracy and reduce the risk of false triggering:

[0147] Converter Model Input and Output

[0148] enter:

[0149] Fingerprint image (I f ) and skin resistance Comprehensive input features of

[0150] The dataset contains 2,000 users, each of whom provides 500 samples, totaling 1 million samples, including samples in a variety of environments and conditions.

[0151] Initial model parameters θ = {W, b}: Contains the weights and biases of the transformer model and capsule network.

[0152] Output:

[0153] The optimized model parameters θ opt : The optimal weights and biases obtained through the Black Hawk optimization algorithm ensure that the system can maintain high recognition accuracy in different environments.

[0154] Algorithm Process

[0155] Initial Model Evaluation:

[0156] Use a 1 million sample dataset to perform preliminary training on the transformer model and capsule network, and calculate the initial recognition accuracy:

[0157] Initial recognition accuracy: 84.2%

[0158] False trigger rate: 1.8%

[0159] Black Hawk Optimization Algorithm Application:

[0160] The iteration number of the Black Hawk optimization algorithm is set to 50 times, and the initial population size is 20 Black Hawk individuals. The goal is to improve the performance of the system by optimizing the objective function.

[0161] Objective function:

[0162] f(θ)=Accuracy(θ)-0.1×Loss(θ)

[0163] f(θ) is the objective function, Accuracy(θ) is the recognition accuracy function, and Loss(θ) is the loss function value.

[0164] Iterative optimization process:

[0165] Initial population: 20 parameter sets θ are randomly generated with initial accuracy ranging from 80% to 85%.

[0166] Iterative update: After each iteration, the current optimal parameter set is selected to update the black hole position, and the positions of other individuals are updated.

[0167] Final optimization result: Fig.11 ,The rising line represents the changing trend of recognition accuracy, and the falling line represents the changing trend of false trigger rate. After 50 iterations, the recognition accuracy of the optimal model reaches 92.5%, and the false trigger rate is reduced to 0.4%.

[0168] This embodiment provides a specific structure of a safe tear gas sprayer, such as Figure 1-10 As shown, it includes a storage bottle 1 for storing tear gas, and a holding cap 2 arranged on the top of the storage bottle 1, the holding cap 2 is connected to the storage bottle 1, and is used to spray the tear gas in the storage bottle 1;

[0169] Among them, a protective cylinder 3 is provided on the outer wall of the holding cover 2, and the protective cylinder 3 and the holding cover 2 are rotatable structures. A limit plate 8 is provided inside the protective cylinder 3, and the limit plate 8 is located on the upper surface of the holding cover 2 to limit the holding cover 2 to prevent accidental touch operations. A holding cover 2 is provided, and a protective cylinder 3 is screwed on the outer wall of the holding cover 2 to protect the holding cover 2. An accidental touch plate 6 is provided inside the holding cover 2 to limit the accidental touch plate 6. Through the setting of the anti-accidental touch structure, the probability of accidental release in the non-use state is greatly reduced, the safety of the equipment is improved, and the risk of accidental touch is significantly reduced. At the same time, the accidental touch plate 6 inside the holding cover 2 can be flipped over, and a limit plate 8 is provided at the corresponding position of the protective cylinder 3 and the accidental touch plate 6. The limit plate 8 limits the accidental touch plate 6 to avoid the problem of flipping of the accidental touch plate 6 during non-operation.

[0170] The holding cover 2 is provided with a squeeze button 7 at the intersection of the storage bottle 1, and the holding cover 2 is provided with a card slot at the position corresponding to the squeeze button 7;

[0171] Among them, a reversible false touch plate 6 is provided inside the card slot, the false touch plate 6 is located at the top of the squeezing button 7, a holding cover 2 is provided, and a protective tube 3 is screwed on the outer wall of the holding cover 2, the protective tube 3 protects the holding cover 2, and an false touch plate 6 is provided inside the holding cover 2, the protective tube 3 limits the false touch plate 6, and through the setting of the anti-false touch structure, the probability of accidental release in the non-use state is greatly reduced, the safety of the equipment is improved, and the risk of false touch is significantly reduced.

[0172] An injection port 5 is embedded in the end of the holding cover 2 away from the false touch plate 6, and a pipe is provided inside the injection port 5, and the other end of the pipe is plugged into the storage bottle 1. The injection port 5 is connected with the raw material inside the storage bottle 1 through the internal pipe, and the pressure change is provided by squeezing the button 7, so that the raw material inside the storage bottle 1 is sprayed out from the injection port 5.

[0173] The protective tube 3 is a hollow cylindrical structure, and the diameter of the inner cross-section of the protective tube 3 is the same as the diameter of the outer cross-section of the holding cover 2. A limiting groove is provided at the intersection of the holding cover 2 and the protective tube 3, so that the protective tube 3 rotates through the limiting groove when the outer wall of the holding cover 2 rotates, thereby avoiding the problem of deviation between the holding cover 2 and the protective tube 3;

[0174] Among them, a groove 9 is provided at one end of the protective cylinder 3 away from the limiting plate 8, and a protective plate 10 is provided on the upper surface of the inner part of the groove 9. When the protective cylinder 3 is connected to the holding cover 2 and is not in use, the protective plate 10 inside the protective cylinder 3 is in contact with the injection port 5 to protect the injection port 5. The protective plate 10 is an arc-shaped structure, and the size of the protective plate 10 is larger than the size of the injection port 5. The size of the groove 9 is larger than the size of the false touch plate 6. In this way, after rotating the protective cylinder 3, when pressing the squeezing button 7, the limiting effect is avoided by the opened groove 9.

[0175] A through hole 11 is provided at the corresponding position of the injection port 5 in the protective tube 3, and the size of the through hole 11 is larger than the size of the injection port 5. When the protective tube 3 is connected to the holding cover 2 and is not in use, the protective plate 10 inside the protective tube 3 contacts the injection port 5 to protect the injection port 5. When in use, the protective tube 3 rotates to separate the protective plate 10 from the injection port 5, and then the through hole 11 rotates to the corresponding position of the injection port 5, so that the raw material sprayed from the injection port 5 is ejected from the through hole 11.

[0176] The storage bottle 1 is a hollow cylindrical structure, and an observation window 4 is provided on the surface wall of the storage bottle 1. Through the setting of the observation window 4, the capacity inside the storage bottle 1 can be easily observed, which is convenient for timely filling.

[0177] The holding cover 2 and the storage bottle 1 are of a detachable structure. If the storage bottle 1 is damaged, the holding cover 2 and the storage bottle 1 can be disassembled for quick repair, thereby increasing the service life and the convenience of repair.

[0178] By adopting the above technical solution:

[0179] A holding cover 2 is provided, and a protective tube 3 is screwed on the outer wall of the holding cover 2, the protective tube 3 protects the holding cover 2, and an error touch plate 6 is provided inside the holding cover 2, the protective tube 3 limits the error touch plate 6. Through the setting of the anti-error touch structure, the probability of accidental release in the non-use state is greatly reduced, the safety of the equipment is improved, and the risk of error touch is significantly reduced. At the same time, the error touch plate 6 inside the holding cover 2 can be flipped, and a limiting plate 8 is provided at the corresponding position of the protective tube 3 and the error touch plate 6, the limiting plate 8 limits the error touch plate 6 to avoid the problem of flipping of the error touch plate 6 during non-operation.

[0180] Embodiment 2: This embodiment provides a specific structure of a safe tear gas sprayer, such as Figure 1-10 As shown, it includes a storage bottle 1 for storing tear gas, an air filling port is provided at the bottom of the storage bottle 1, so that the raw materials inside the storage bottle 1 can be added through the air filling port, thereby improving the practicality, and a holding cap 2 is provided at the top of the storage bottle 1, the holding cap 2 is connected to the storage bottle 1, and is used to spray the tear gas inside the storage bottle 1;

[0181] Among them, a protective cylinder 3 is provided on the outer wall of the holding cover 2, and the protective cylinder 3 and the holding cover 2 are rotatable structures. A limit plate 8 is provided inside the protective cylinder 3, and the limit plate 8 is located on the upper surface of the holding cover 2 to limit the holding cover 2 to prevent accidental touch operations. A holding cover 2 is provided, and a protective cylinder 3 is screwed on the outer wall of the holding cover 2 to protect the holding cover 2. An accidental touch plate 6 is provided inside the holding cover 2 to limit the accidental touch plate 6. Through the setting of the anti-accidental touch structure, the probability of accidental release in the non-use state is greatly reduced, the safety of the equipment is improved, and the risk of accidental touch is significantly reduced. At the same time, the accidental touch plate 6 inside the holding cover 2 can be flipped over, and a limit plate 8 is provided at the corresponding position of the protective cylinder 3 and the accidental touch plate 6. The limit plate 8 limits the accidental touch plate 6 to avoid the problem of flipping of the accidental touch plate 6 during non-operation.

[0182] The holding cover 2 is provided with a squeeze button 7 at the intersection of the storage bottle 1 (the principle of squeezing out the spray liquid between the squeeze button 7 and the storage bottle 1 is the same as the principle of a general tear gas sprayer, and can also be similar to the principle of a mosquito repellent spray, which belongs to common knowledge and does not belong to the invention of the present invention), and the holding cover 2 is provided with a card slot at the position corresponding to the squeeze button 7;

[0183] Among them, a reversible false touch plate 6 is provided inside the card slot, and the false touch plate 6 is located at the top of the squeezing button 7. A holding cover 2 is provided, and a protective tube 3 is screwed on the outer wall of the holding cover 2. The protective tube 3 protects the holding cover 2. The holding cover 2 and the storage bottle 1 are detachable structures. If the storage bottle 1 is damaged, the holding cover 2 and the storage bottle 1 can be disassembled for quick repair, which improves the service life and the convenience of repair. An false touch plate 6 is provided inside the holding cover 2, and the protective tube 3 limits the false touch plate 6. Through the setting of the anti-false touch structure, the probability of accidental release in the non-use state is greatly reduced, the safety of the equipment is improved, and the risk of false touch is significantly reduced.

[0184] An injection port 5 is embedded at one end of the holding cover 2 away from the false touch plate 6, and a pipe is provided inside the injection port 5, and the other end of the pipe is plugged into the storage bottle 1. The injection port 5 is connected with the raw material inside the storage bottle 1 through the internal pipe. The storage bottle 1 is a hollow cylindrical structure, and an observation window 4 is provided on the surface wall of the storage bottle 1. Through the setting of the observation window 4, the capacity inside the storage bottle 1 can be easily observed, which is convenient for timely filling. By squeezing the button 7 to provide a change in pressure, the raw material inside the storage bottle 1 is sprayed out from the injection port 5.

[0185] The protective tube 3 is a hollow cylindrical structure. The inner cross-sectional diameter of the protective tube 3 is the same as the outer cross-sectional diameter of the holding cover 2. A limiting groove is provided at the intersection of the holding cover 2 and the protective tube 3. In this way, the protective tube 3 rotates through the limiting groove when the outer wall of the holding cover 2 rotates, thereby avoiding the problem of deviation between the holding cover 2 and the protective tube 3.

[0186] A groove 9 is provided at one end of the protective tube 3 away from the limiting plate 8, and a protective plate 10 is provided on the upper surface of the groove 9. When the protective tube 3 is connected to the holding cover 2 and is not in use, the protective plate 10 inside the protective tube 3 contacts the injection port 5 to protect the injection port 5. The protective plate 10 is an arc-shaped structure, and the size of the protective plate 10 is larger than the size of the injection port 5. The size of the groove 9 is larger than the size of the false touch plate 6. In this way, after rotating the protective tube 3, when pressing the squeezing button 7, the limiting effect is avoided by the provided groove 9.

[0187] A through hole 11 is provided at the corresponding position of the injection port 5 in the protective tube 3, and the size of the through hole 11 is larger than the size of the injection port 5. When the protective tube 3 is connected to the holding cover 2 and is not in use, the protective plate 10 inside the protective tube 3 contacts the injection port 5 to protect the injection port 5. When in use, the protective tube 3 rotates to separate the protective plate 10 from the injection port 5, and then the through hole 11 rotates to the corresponding position of the injection port 5, so that the raw material sprayed from the injection port 5 is ejected from the through hole 11.

[0188] By adopting the above technical solution: a holding cover 2 is provided, a protective tube 3 is screwed on the outer wall of the holding cover 2, the protective tube 3 protects the holding cover 2, an error touch plate 6 is provided inside the holding cover 2, the protective tube 3 limits the error touch plate 6, and through the setting of the anti-error touch structure, the probability of accidental release in the non-use state is greatly reduced, the safety of the equipment is improved, and the risk of error touch is significantly reduced. At the same time, the error touch plate 6 inside the holding cover 2 can be flipped over, and a limiting plate 8 is provided at the corresponding position of the protective tube 3 and the error touch plate 6. The limiting plate 8 limits the error touch plate 6 to avoid the problem of flipping of the error touch plate 6 during non-operation.

[0189] Finally, it should be noted that: Obviously, the above embodiments are only examples for clearly illustrating the present invention, and are not intended to limit the implementation methods. It is not necessary and impossible to list all implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A user biometric identification method based on an improved deep learning algorithm, characterized in that: A complex and efficient user identification method is designed using the transformer model and capsule network algorithm, which is applied in secure operations in handheld devices.

2. According to claim 1, a user biometric identification method based on an improved deep learning algorithm is characterized in that: The specific implementation process of designing a complex and efficient user identity recognition method using the transformer model and capsule network algorithm is as follows: Step 1: Data Collection Module 1. Fingerprint recognition Fingerprint image acquisition: Use a high-resolution fingerprint sensor to collect the user's fingerprint image: I f is the collected fingerprint image, is the pixel value of the fingerprint image at position (i, j), i is the row index, M is the total number of rows in the fingerprint image matrix, j is the column index, and N is the total number of columns in the fingerprint image matrix; 2. Skin resistance measurement Resistance measurement: Use a skin resistance sensor to measure the resistance of the user's finger as an auxiliary biometric feature. The measured resistance value will vary depending on the individual's skin characteristics and humidity: R s is the skin resistance value, V s is the voltage, I s is the current; Voltage V s and current I s It is measured by the skin resistance sensor, and the resistance value R s It is calculated by Ohm's law. In order to improve the recognition accuracy, the noise effect can be reduced by taking the average value through multiple measurements: is the average skin resistance value, To calculate the sum of all measured values ​​divided by the number of measurements n, is the sum from the first measurement to the nth measurement, represents the skin resistance value measured at the kth time; Step 2: Feature extraction and conversion 1. Input data representation Fingerprint images and skin resistance data first need to be converted into an input data form that the model can process: Fingerprint image matrix: I f , the size is M×N; Skin resistance value: can be considered as a scalar; Flatten the fingerprint image into a vector and combine the skin resistance value with the fingerprint feature vector: x0 is the initial input vector, flatten(I f ) is to flatten the fingerprint image matrix into a one-dimensional vector, is the skin resistance value; 2. Self-Attention Mechanism in the Transformer Model The self-attention mechanism generates weights by calculating the relationship between each element in the input vector and other elements, thereby highlighting important features: Attention(Q,K,V) is the output of the self-attention mechanism, Q is the query matrix, K is the key matrix, V is the value matrix, and softmax() is the softmax operation on the dot product results of each row. is the matrix obtained after the dot product result is scaled, d k is the dimension of the key vector; 3. Converter output After being processed by the multi-head self-attention mechanism and the feedforward network layer, the transformer model converts the input data into high-dimensional feature representations. These high-dimensional feature vectors will be passed to the next capsule network for further processing: z=Transformer(x0) z is the high-dimensional feature vector output by the transformer model, Transformer() is the overall operation process of the transformer model, and x0 is the input vector; 4. Transfer of feature vectors The high-dimensional feature vector z generated by the transformer model is passed as input to the capsule network for further feature analysis and classification: z→CapsuleNetwork z is the high-dimensional feature vector output by the transformer model, and CapsuleNetwork is the capsule network; Step 3: Feature representation and classification of capsule network 1. Main capsule layer Feature extraction: The main capsule layer receives the high-dimensional feature vector z from the transformer model and performs preliminary feature extraction through multiple capsules, each of which is a small neural network that processes the input features and generates low-level feature representations: u i =W i z+b i u i is the output of the ith primary capsule, W i is the weight matrix of the ith primary capsule, z is the high-dimensional feature vector output by the transformer model, and b i is the bias vector of the i-th primary capsule; 2. Dynamic Routing Algorithm Dynamic routing: low-level feature vector u i It is passed to the sub-capsule layer through the dynamic routing algorithm, which dynamically adjusts the connection weights according to the relevance of the input data to ensure that important features can be better passed to high-level capsules: v j is the output of the j-th capsule, squash() is a nonlinear activation function, ∑ i is the summation symbol, c ij is the dynamic routing coefficient, u ij The output feature vector u of the main capsule layer i The result after weight transformation; 3. Sub-capsule layer Feature fusion: The secondary capsule layer not only receives the feature vector v transmitted by the main capsule layer through dynamic routing j , combined with skin resistance data Further processing is performed to generate the final high-level feature representation: o k is the output of the kth capsule, W k is the weight matrix of the kth capsule, v j is the high-level feature vector generated by the sub-capsule layer through dynamic routing, is the skin resistance data, b k is the bias vector of the kth capsule; 4. Classification and identification Classification output: final high-level feature representation k Pass it to the classification layer and use the softmax function to classify and determine the identity of the user: is the classification result, softmax(O) is the softmax function; Step 4: Application of optimization algorithm 1. Initialization Random initialization: Randomly initialize the weights and bias parameters of the capsule network, which are regarded as the initial solution in the black hawk population in BHO: W (0) is the initial weight matrix, b (0) is the initial bias vector, is a uniform distribution, ∈ is the range of distribution; 2. Iterative Optimization Black Hawk position update: According to the BHO algorithm, the position of the black hawk is updated each iteration, gradually converging to the optimal parameter set. In BHO, the position of each black hawk is represented as a set of capsule network parameters, including weight W and bias b: is the new position of the i-th black hawk after the t+1th iteration, is the position of the i-th black hawk at the t-th iteration, α is the learning rate, is the black hole position, β is the random perturbation factor, and r is the random vector; 3. Fitness function calculation Fitness calculation: After each iteration, the performance of each black hawk position in the capsule network is evaluated. The fitness function is usually the recognition accuracy or error function of the capsule network on the validation set: f(X i )=Accuracy(W i ,b i )or f(X i )=-Loss(W i ,b i ) f(X i ) is the fitness function value, Accuracy(W i ,b i ) is based on the current weight W i and bias b i The recognition accuracy, Loss(W i ,b i ) is based on the current weight W i and bias b i The loss function value of 4. Termination Conditions Termination of optimization: The optimization process terminates when any of the following conditions is met: The recognition accuracy reaches the set threshold; The maximum number of iterations has been reached; Terminateiff(X bh )≥Thresholdort≥t max Terminate is the operation to terminate the optimization process, if is the conditional judgment symbol, f(X bh ) is the fitness function value of the current optimal solution, Threshold is the preset fitness threshold, t is the current number of iterations, t max is the maximum number of iterations allowed; Step 5 System output and user feedback 1. Calculation of identity recognition results Classification output: The capsule network finally outputs a vector representing the classification probability of different identities. The softmax function converts these output values ​​into probability distribution to determine the identity of the user: is the predicted probability of the i-th class, softmax(O i ) is the softmax function, O i is the score value of the i-th class output by the capsule network, is the exponential operator, is to accumulate all categories of j from 1 to C, where C is the total number of categories, o j is the score value of the jth class output by the capsule network; 2. Identity Verification and Decision Making Identification decision: The system determines the user's identity category based on the maximum probability value. If the category corresponding to the maximum probability value is an authorized identity, the identification is successful and the user is allowed to operate the tear gas sprayer; otherwise, the identification fails and the system locks the sprayer: Decision is the final decision result of the system, Allow Operation means the system allows the user to operate the tear gas sprayer, Lock System means the system locks the tear gas sprayer, and if is a conditional judgment symbol. is the maximum probability value in the softmax output, is the predicted probability value of the authorized identity category, δ is the recognition threshold, and otherwise indicates that the condition is not satisfied; 3. User Feedback Feedback mechanism: The system will feed back the recognition results and decisions to the user. If the recognition is successful, the user can continue to operate the tear gas sprayer. If the recognition fails, the system will lock and prompt the user that the identity authentication failed: Feedback is the information content fed back to the user by the system, Success Message is the message fed back to the user when the system recognizes successfully, Failure Message is the message fed back to the user when the system recognizes failed, Decision is the final decision result of the system, Allow Operation means the system's decision is to allow the operation, and Lock System means the system's decision is to lock the system.

3. The user biometric identification method based on the improved deep learning algorithm according to claim 2 is characterized in that: It also includes step 6: system self-learning and updating, the specific process is:

1. Adaptive Learning Data collection and update: The system regularly collects new user data, including fingerprint images and skin resistance values. Through the continuous learning mechanism, the system uses this new data to update the transformer model and capsule network to ensure that the model can adapt to changes in different environments and users: For the updated dataset, is the training data set of the system at time t, ∪ is the set union operator, is the new data collected at time t+1; Model update: use the updated dataset Retrain the transformer model and capsule network and update the model parameters to maintain high performance under the new data distribution: θ t+1 is the updated model parameter at time t+1, argmin θ To optimize the loss function Minimize the loss, θ is the set of model parameters, For the updated dataset, For the updated dataset The number of samples in is the loss function, f θ (x) is the prediction result of the model for the input feature vector x, where x is the input feature vector and y is the target label; 2. Enhanced security Model parameter re-optimization: In order to prevent the recognition algorithm from being attacked or cracked, the system regularly uses the Black Hawk optimization algorithm to re-optimize the parameters of the transformer model and capsule network to enhance the security of the system: is the optimized model parameter at time t+1, argmax θ To find the parameter θ that maximizes the expression in brackets, BHO is the Black Hawk optimization algorithm. For a given data set Under the condition of , the parameter θ is optimized. is the data set used by the system at time t+1; Enhanced defense mechanism: By periodically optimizing model parameters, the system can continuously improve its defense capabilities against potential attacks and ensure the security and reliability of the identification process: Security t+1 is the security level of the system at time t+1, Function is an abstract function that represents the relationship between system security and model parameters. are the optimized model parameters at time t+1.

4. The user biometric identification method based on an improved deep learning algorithm according to claim 1, characterized in that: The handheld device is a safe tear gas sprayer.

5. The injector according to claim 4, characterized in that: The sprayer comprises a storage bottle (1) for storing tear gas, and a holding cover (2) arranged at the top of the storage bottle (1), wherein the holding cover (2) is connected to the storage bottle (1) and is used to spray the tear gas in the storage bottle (1); wherein a protective tube (3) is provided on the outer wall of the holding cover (2), wherein the protective tube (3) and the holding cover (2) are rotatable structures, and a limiting plate (8) is provided inside the protective tube (3), wherein the limiting plate (8) is located on the upper surface of the holding cover (2) and limits the holding cover (2) to prevent accidental touch operation; A miniature electromagnetic lock (12) is built into the inner wall of the protective tube (3). The electromagnetic lock is connected to a controller with a built-in algorithm as an execution device. The input end of the controller is connected to a fingerprint sensor and a skin resistance sensor, wherein the fingerprint sensor and the skin resistance sensor are in contact with the outer surface of the storage bottle (1), and the outer surface of the storage bottle (1) has a fingerprint collection information position prompt. The controller is built into between the storage bottle (1) and the holding cover (2). The electromagnetic lock is placed horizontally. The lock tongue of the electromagnetic lock corresponds to the groove on the protective tube (3). When the system is in a locked state, the lock tongue is engaged with the groove on the protective tube (3) to limit the rotation of the protective tube (3). After the identity recognition is successful, the system is in an unlocked state, and the lock tongue is retracted into the electromagnetic lock to unlock the rotation of the protective tube (3).

6. The injector according to claim 5, characterized in that The holding cover (2) is provided with a squeeze button (7) at the intersection of the holding cover and the storage bottle (1), and the holding cover (2) is provided with a card slot at a position corresponding to the squeeze button (7); wherein a reversible mis-touch plate (6) is provided inside the card slot, and the mis-touch plate (6) is located at the top of the squeeze button (7); An injection port (5) is embedded in one end of the holding cover (2) away from the false touch plate (6), a pipe is provided inside the injection port (5), and the other end of the pipe is plugged into the inside of the storage bottle (1); The protective tube (3) is a cylindrical hollow structure, and the diameter of the inner circular cross section of the protective tube (3) is the same as the diameter of the outer circular cross section of the holding cover (2).

7. The injector according to claim 5, characterized in that A groove (9) is provided at one end of the protective tube (3) away from the limiting plate (8), and the size of the groove (9) is larger than the size of the false touch plate (6); A protective plate (10) is provided on the inner upper surface of the groove (9), the protective plate (10) is an arc-shaped structure, and the size of the protective plate (10) is larger than the size of the injection port (5); The protective tube (3) is provided with a through hole (11) at a position corresponding to the injection port (5), and the size of the through hole (11) is larger than the size of the injection port (5).

8. The injector according to claim 5, characterized in that The storage bottle (1) is a hollow cylindrical structure, and an observation window (4) is provided on the surface wall of the storage bottle (1); the holding cover (2) and the storage bottle (1) are detachable structures, and a limiting groove is provided at the intersection of the holding cover (2) and the protective tube (3).

Images