Electronic scale dynamic anti-cheating calibration method based on AI

Abstract

The invention relates to the technical field of weighing calibration, in particular to an AI-based electronic scale dynamic anti-cheating calibration method. The method comprises the following steps: firstly, acquiring an attack window to determine an attack pulse signal; matching the electronic scale with cheating devices of different models based on the noise of different offsets of the electronic scale during the real-time weighing period, and determining the maximum signal-to-noise ratio of the matched cheating device as a crystal oscillator noise coefficient; based on the waveform jitter characteristic of the attack pulse signal, matching the attack pulse signal with the simulated attack pulse signal in the contrast library, and determining a pulse gene variation coefficient according to the waveform abnormity of the matched simulated attack pulse signal; determining a behavior layer abnormal value according to the number of times of position transfer of the attack window in different state stages; and according to the crystal oscillator noise coefficient, the pulse gene variation coefficient and the behavior layer abnormal value, respectively correcting the baseline, the cut-off frequency and the sampling interval of the electronic scale. The anti-interference precision of the electronic scale is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of weighing calibration, and particularly to an AI-based dynamic anti-cheating calibration method for electronic scales. Background Art

[0002] Traditional electronic scales are easily illegally tampered with, which damages market fairness. The dynamic anti-cheating calibration algorithm for electronic scales can continuously monitor the original weighing signal (AD value) of the electronic scale, convert it into a second weight value, and dynamically compare it with the first weight value displayed. When the deviation exceeds the preset threshold, an abnormal alarm is triggered. Combined with remote instructions, it controls the built-in calibration weights or compensation algorithm to automatically repair the cheating parameters, and uses an anti-disassembly and lockout circuit to block physical cheating. Although this dynamic comparison algorithm can detect abnormal display values, it lacks real-time protection for the intermediate links of the signal chain.

[0003] Anti-cheating technology is a dynamic game of offense and defense. The algorithm should not only adapt to static weighing scenarios but also guard against dynamic attacks. For example, wireless remote control interference can cover the sensor analog signal by injecting co-frequency noise. The existing AD value sampling frequency is usually 100 - 200Hz, which is difficult to capture nanosecond-level transient interference pulses. When cheaters use Pulse Width Modulation (PWM) technology to superimpose false voltages during the weighing moment, the displayed value and the second weight value compared by the algorithm will be distorted synchronously, forming a "attack-distortion-synchronization" camouflage effect, resulting in the failure of the dynamic comparison mechanism. Summary of the Invention

[0004] In order to solve the technical problem that it is difficult to detect cheating when cheaters use Pulse Width Modulation technology to superimpose false voltages during the weighing moment, the purpose of the present invention is to provide an AI-based dynamic anti-cheating calibration method for electronic scales. The specific technical solutions adopted are as follows: In a first aspect, an embodiment of the present invention provides an AI-based dynamic anti-cheating calibration method for electronic scales, which includes: Obtain the AD value of the electronic scale and the attack window; the AD values within the attack window are those with a sudden increase in signal energy at consecutive moments; Determine the attack pulse signal according to the AD values in different frequency bands within the attack window, the AD values of the electronic scale in the no-load state and the reference weighing state; Based on the noise with different offsets during the real-time weighing of the electronic scale, match the electronic scale with different models of cheaters, and determine the maximum signal-to-noise ratio of the matched cheater as the crystal oscillator noise coefficient; Based on the waveform jitter characteristics of the attack pulse signal, match the attack pulse signal with the simulated attack pulse signals in the comparison library, and determine the pulse gene mutation coefficient according to the waveform abnormality of the matched simulated attack pulse signal; Divide the AD values of the electronic scale into different state stages, and determine the outlier at the behavior layer according to the number of position transfers of the attack window in different state stages. Correct the baseline of the electronic scale according to the crystal oscillator noise coefficient, correct the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient, and correct the sampling interval of the electronic scale according to the outlier at the behavior layer.

[0005] Furthermore, the method for obtaining the attack window is as follows: Slide a window on the AD value sequence arranged in time sequence, take the derivative of the AD values within the sliding window. When the derivative of consecutive sampling points is greater than the preset sliding threshold, the window range where the consecutive sampling points are located is used as the candidate window; the preset sliding threshold is a fixed multiple of the standard deviation of the data within the sliding window; the number of consecutive sampling points is greater than the preset consecutive threshold. Calculate the ratio of the maximum AD value in the candidate window to the AD value at the previous moment as the mutation rate. When the mutation rate is greater than the preset mutation threshold and the sampling duration corresponding to the candidate window belongs to the preset duration range, determine the candidate window as the attack window.

[0006] Furthermore, determining the attack pulse signal according to the AD values in different frequency bands within the attack window and the AD values of the electronic scale in the no-load state and the reference weighing state includes: Compare the AD value sequences of the electronic scale in the no-load state and the reference weighing state to obtain the abnormal determination value of the attack window; determine the asymmetry factor of the attack window according to the total duration of the pulse rising edge and the total duration of the pulse falling edge of the AD values in the attack window; combine the abnormal determination value and the asymmetry factor to determine the initial abnormal determination factor. Determine the attack pulse characteristics of the attack window according to the AD values in different frequency bands within the attack window. Generate a time-frequency spectrum within the attack window, obtain the time-sequence change sequence of the energy at the 20 kHz frequency point, and determine the verification factor according to the correlation coefficient between the time-sequence change sequence of the energy at the 20 kHz frequency point and the time-sequence change sequence of the AD values. Determine the decision coefficient according to the initial abnormal determination factor, the attack pulse characteristics, and the verification factor. When the decision coefficient is greater than or equal to the preset decision threshold, determine that there is a PWM attack in the attack window; take the AD values within the longest consecutive attack window with PWM attack during weighing as the attack pulse signal.

[0007] Furthermore, determining the attack pulse characteristics of the attack window according to the AD values in different frequency bands within the attack window includes: For any attack window, perform 8-layer wavelet packet decomposition in the frequency band of 100 kHz - 10 MHz, and calculate the energy entropy at different frequency positions. Perform short-time Fourier transform on the frequency bands of 20 kHz and within a preset range to determine the proportion of the narrowband duration; Combine the energy entropy and the proportion of the narrowband duration to determine the attack pulse characteristics of the attack window.

[0008] Further, based on the noise of different magnitudes of offset during real-time weighing of the electronic scale, match the electronic scale with different models of cheating devices, and determine the maximum signal-to-noise ratio of the matched cheating device as the crystal oscillator noise coefficient, including: Based on the noise of different magnitudes of frequency offset during real-time weighing of the electronic scale, determine the signal-to-noise ratio corresponding to different frequency offsets, and construct a signal-to-noise ratio vector; Match the signal-to-noise ratio vectors of the electronic scale and different models of cheating devices, screen out the cheating devices, and determine the maximum signal-to-noise ratio corresponding to the cheating device as the crystal oscillator noise coefficient.

[0009] Further, based on the waveform jitter characteristics of the attack pulse signal, match the attack pulse signal with the simulated attack pulse signals in the reference library, and determine the pulse gene mutation coefficient according to the waveform abnormality of the matched simulated attack pulse signal, including: Extract the pulse waveform chain of the attack pulse signal, and the pulse waveform chain is used to represent the waveform jitter characteristics of the attack pulse signal; Based on the pulse waveform chain, match the attack pulse signal with the simulated attack pulse signals in the reference library, screen out the reference attack pulse signal type from the pulse waveform types to which the matched simulated attack pulse signals belong, and determine the pulse gene mutation coefficient according to the waveform abnormality of the simulated attack pulse signals within the reference attack pulse signal type.

[0010] Further, divide the AD value of the electronic scale into different state stages, and determine the outlier at the behavior layer according to the number of position transfers of the attack window in different state stages, including: The state stages include: stable period, light unloading period, and heavy unloading period; Determine the current state stage where the attack window is located and the next adjacent state stage of the attack window; Take the ratio of the number of attack windows in the next adjacent state stage to the current state stage as the transition probability from the current state stage to the next adjacent state stage; Obtain the information entropy of all transition probabilities; calculate the sum of the transition probabilities from the stable period to the light unloading period and from the stable period to the heavy unloading period; Take the ratio of the sum of the transition probabilities to the information entropy as the outlier at the behavior layer.

[0011] Further, the method for correcting the baseline of the electronic scale according to the crystal oscillator noise coefficient includes: Obtain the envelope of the AD value of the electronic scale, and use the average amplitude of the envelope as the baseline value; Weight the baseline value with the normalized crystal oscillator noise coefficient to obtain the corrected baseline compensation value.

[0012] Further, the method for correcting the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient includes: Weight the default cut-off frequency with the normalized pulse gene mutation coefficient to obtain the corrected compensation coefficient.

[0013] Further, the method for correcting the sampling interval of the electronic scale according to the outlier in the behavior layer includes: Perform a negative correlation mapping on the normalized outlier in the behavior layer, and use the mapped result value as the weight to weight the default sampling interval to obtain the corrected sampling interval.

[0014] In a second aspect, an AI-based dynamic anti-cheating calibration system for an electronic scale is provided. The system includes the following modules: A signal acquisition module, configured to acquire the AD value of the electronic scale and the attack window; the AD values with sudden increase in signal energy at continuous moments are within the attack window; An attack determination module, configured to determine the attack pulse signal according to the AD values in different frequency bands within the attack window, the AD values of the electronic scale in the no-load state and the reference weighing state; A crystal oscillator analysis module, configured to match the electronic scale with different models of cheating devices based on the noise with different offsets during real-time weighing of the electronic scale, and determine the maximum signal-to-noise ratio of the matched cheating device as the crystal oscillator noise coefficient; A pulse analysis module, configured to match the attack pulse signal with the simulated attack pulse signals in the comparison library based on the waveform jitter characteristics of the attack pulse signal, and determine the pulse gene mutation coefficient according to the waveform abnormality of the matched simulated attack pulse signal; A behavior analysis module, configured to divide the AD value of the electronic scale into different state stages, and determine the outlier in the behavior layer according to the number of position transfers of the attack window in different state stages; A correction module, configured to correct the baseline of the electronic scale according to the crystal oscillator noise coefficient, correct the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient, and correct the sampling interval of the electronic scale according to the outlier in the behavior layer.

[0015] In a third aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor. An executable code is stored in the memory. When the processor executes the executable code, the embodiments of all possible implementations in the first aspect are implemented.

[0016] Fourthly, an embodiment of the present invention provides a computer program product, which includes computer program code. When the computer program code runs on a computer, the computer is enabled to execute the method in the first aspect or any possible implementation manner of the first aspect as described above.

[0017] Fifthly, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed in a computer, the computer is enabled to execute each possible implementation example in the first aspect.

[0018] The embodiments of the present invention have at least the following beneficial effects: In this application, more than 90% of conventional PWM attacks are blocked through high-frequency sampling and signal chain reinforcement. First, in combination with a multi-modal dynamic detection mechanism, the attack window is accurately located based on the sudden increase in signal energy; the crystal oscillator phase noise and pulse jitter characteristics are introduced, and through behavior analysis, the AD values of the electronic scale are divided into different state stages, and the crystal oscillator noise coefficient, pulse gene mutation coefficient, and behavior layer outliers are respectively determined. Through these three characteristics, the baseline compensation, cut-off frequency, and sampling interval are dynamically adjusted in combination, improving the anti-interference accuracy of the weighing signal, and at the same time, the model number of the cheating device can be traced; effectively solving the problem of concealed professional cheating faced by electronic scales in the fields of logistics, trade, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 It is a method flowchart of an AI-based dynamic anti-cheating calibration method for an electronic scale provided by an embodiment of the present invention; Figure 2 It is a system block diagram of an AI-based dynamic anti-cheating calibration system for an electronic scale provided by an embodiment of the present invention; Figure 3 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in combination with the drawings and preferred embodiments, describe in detail the specific implementation manner, structure, features, and effects of the AI-based dynamic anti-cheating calibration method for an electronic scale proposed according to the present invention.

[0022] In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0023] Wherein, in the description of the embodiments of the present invention, unless otherwise specified, " / " means "or". For example, A / B may mean A or B. The "and / or" in the text is merely a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B may mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of the present invention, "a plurality of" means two or more than two.

[0024] Hereinafter, the terms "first" and "second" are only used for descriptive purposes and should not be construed as implying or suggesting relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this invention belongs.

[0026] The embodiments of the present invention will be described below with reference to the accompanying drawings. As those of ordinary skill in the art can know, with the development of technology and the emergence of new scenarios, the technical solutions provided by the embodiments of the present invention are equally applicable to similar technical problems.

[0027] The embodiments of the present invention provide a specific implementation method for an AI-based dynamic anti-cheating calibration method for electronic scales, which is applicable to the dynamic calibration scenario of electronic scales. To solve the problem that the analog signal (AD value) of the weighing sensor is easily interfered by a wireless remote control device during transmission, and cheaters inject PWM pulses during the weighing instant (within 0.1 - 0.5 s) to synchronously amplify the AD value and the display value. The present invention improves the anti-interference accuracy of the weighing signal by dynamically jointly adjusting the baseline compensation, cut-off frequency, and sampling interval, and at the same time, the model of the cheating device can be traced; effectively solving the problem of concealed professional cheating faced by electronic scales in the fields of logistics, trade, etc.

[0028] The specific solution of the AI-based dynamic anti-cheating calibration method for electronic scales provided by the present invention will be specifically described below with reference to the accompanying drawings.

[0029] Please refer to Figure 1 , which shows a flowchart of the steps of an AI-based dynamic anti-cheating calibration method for electronic scales provided by an embodiment of the present invention. The method includes the following steps: Step S100, obtaining the AD value of the electronic scale and the attack window; the AD values with sudden increase in signal energy at consecutive moments are within the attack window.

[0030] Adopt the dual-ADC interleaved sampling technique. For example, two 100 MHz ADCs are used to increase the effective sampling rate to 200 MHz. In cooperation with the analog-to-digital converter, 24-bit high-resolution signal conversion is achieved, and nanosecond-level pulse interference can be captured. Then, a programmable gain amplifier (PGA) is integrated in the sensor analog front end, and the signal chain is adaptively optimized through dynamic gain adjustment (such as AD8253): the gain is automatically increased to 100 times at the beginning of weighing to amplify weak deformation signals, and the gain is reduced to 1 time after weighing is stable to avoid saturation.

[0031] Deploy a multi-stage LC filter circuit and a magnetic bead isolator to block the high-frequency coupling path of PWM interference. Among them, the cut-off frequency of the LC filter circuit is 500 MHz. And differential signal transmission (such as RS485) is used to replace single-ended transmission. In cooperation with the shielded twisted pair and the ground plane design, the common-mode rejection ratio is increased to more than 80 dB.

[0032] High-frequency sampling can capture PWM attack pulses more accurately, and the filter circuit and isolator can block most of the attack pulses, and differential signal transmission can suppress interference signals.

[0033] However, its protection ability is still limited. If cheaters use methods such as variable-frequency pulse impact and multi-band combination, they may break through the above protection. Therefore, the recognition algorithm for attack behaviors still needs to be strengthened and optimized.

[0034] When the electronic scale is in the no-load state, collect the AD signal for 1 minute, that is, the AD value, and calculate the standard deviation σ and kurtosis coefficient K of the noise amplitude.

[0035] Slide a window on the AD value sequence arranged in time sequence, and take the first derivative of the AD values within the sliding window. When the derivative of consecutive sampling points is greater than the preset sliding threshold, the window range where the consecutive sampling points are located is used as the candidate window; the preset sliding threshold is a fixed multiple of the standard deviation of the data within the sliding window; the number of consecutive sampling points is greater than the preset continuous threshold.

[0036] In the embodiment of the present invention, the preset sliding threshold is three times the standard deviation of the data within the sliding window; the value of the preset continuous threshold is 10.

[0037] Calculate the ratio of the maximum AD value in the candidate window to the AD value at the previous moment as the mutation rate; when the mutation rate is greater than the preset mutation threshold and the sampling duration corresponding to the candidate window belongs to the preset duration range, determine that the candidate window is an attack window. In the embodiment of the present invention, the value of the preset mutation threshold is 5; the preset duration range is 0.1 s to 0.5 s.

[0038] Step S200: Determine the attack pulse signal based on the AD values at different frequency bands within the attack window and the AD values of the electronic scale in the no-load state and the reference weighing state.

[0039] Obtain the waveforms of the AD values of the last 5 legal weighings in the baseline library, take the average value and record it as the normal loading waveform, that is, the AD value sequence in the reference weighing state; compare the AD value sequences of the electronic scale in the no-load state and the reference weighing state to obtain the abnormal determination value of the attack window, and this abnormal determination value is used to characterize the waveform distortion degree.

[0040] In some embodiments, the method for obtaining the abnormal determination value is as follows: obtain the AD value sequence of the electronic scale in the no-load state and the AD value sequence in the reference weighing state; use the DTW algorithm to align the two AD value sequences, and then calculate the average value of the DTW distances at all times of the two AD value sequences as the abnormal determination value; the DTW distance of normal difference is less than 5, while the DTW distance of cheating attack may be greater than 20, indicating that there are sudden rises and slow descents of pulses resulting in sudden changes in waveform morphology.

[0041] Then, extract the total duration of the pulse rising edge and the total duration of the pulse falling edge of the AD value in each attack window. Due to the hysteresis of the sensor response, the rising edge time length of the cheating pulse is usually less than the falling edge time length. Determine the asymmetry factor of the attack window according to the duration of the pulse rising edge and the duration of the pulse falling edge of the AD value in the attack window. Specifically: calculate the absolute value of the difference between the total duration of the pulse rising edge and the total duration of the pulse falling edge as the asymmetry factor of the corresponding attack window.

[0042] Obtain the abnormal determination value and the asymmetry factor that characterize the waveform distortion degree, both of which express the abnormality of the time-domain waveform of the AD signal. Combine the abnormal determination value and the asymmetry factor to determine the initial abnormal determination factor. Normalize the abnormal determination value and the asymmetry factor respectively, and take the product of the two after normalization as the initial abnormal determination factor.

[0043] To avoid false alarms in time-domain determination, supplementary determination is required. If the waveform distortion degree of the abnormal determination value > 15, start the high-frequency band analysis mode (100 kHz - 10 MHz), otherwise skip this step to reduce false alarms.

[0044] Furthermore, determine the attack pulse characteristics of the attack window according to the AD values at different frequency bands within the attack window. Specifically: For any attack window, perform 8-layer wavelet packet decomposition in the frequency band of 100 kHz - 10 MHz, and calculate the energy entropy at different frequency positions; Because the frequency band energy distribution is uniform during normal loading, while the PWM carrier energy of cheating attack is concentrated, and the energy entropy may suddenly drop at the 20 kHz node.

[0045] Therefore, short-time Fourier transform is performed on the frequency band within 20 kHz and the preset range to determine the proportion of the narrowband duration.

[0046] Since the attack pulse needs to be maintained by a stable carrier, that is, it has a long duration, and the lower the energy entropy, the more concentrated the energy appears abnormally. Therefore, the ratio of the narrowband duration proportion to the energy entropy is used as the attack pulse feature extracted in the attack window.

[0047] To ensure the credibility of the cross-domain determination between the time domain and the frequency domain, the result needs to be verified. A time-frequency spectrum is generated within the attack window, and the time series change sequence of the energy at the 20 kHz frequency point is obtained. According to the correlation coefficient between the time series change sequence of the energy at the 20 kHz frequency point and the time series change sequence of the AD value, the verification factor is determined.

[0048] In the embodiment of the present invention, the correlation coefficient is the Pearson correlation coefficient.

[0049] Furthermore, according to the initial anomaly determination factor, the attack pulse feature, and the verification factor, a decision coefficient is determined.

[0050] In some embodiments, the decision coefficient The calculation formula is: ; where is the initial anomaly determination factor; is the attack pulse feature; is the verification factor; e is the natural constant; is the anomaly determination value.

[0051] Where represents the use of a frequency-domain logarithmic amplifier to perform non-linear compression on the attack pulse feature to prevent the energy concentration in the frequency band from solely dominating the decision when the noise is dense; represents the joint verification of the correlation coefficient and the DTW distance. The higher the anomaly determination value, the closer the current waveform is to being abnormal. At this time, a slightly lower verification factor is allowed, that is, when the anomaly determination value is larger, the denominator is smaller, and it can amplify , allowing a slightly lower correlation coefficient between the signal amplitude and the energy at the core frequency point (20 kHz).

[0052] When the decision coefficient is greater than or equal to the preset decision threshold, it is determined that there is a PWM attack in the attack window; the AD value within the attack window with the longest continuous PWM attack during the weighing period is used as the attack pulse signal. In the embodiment of the present invention, the value of the preset decision threshold is 1.8, and in other embodiments, the implementer can adjust this value according to the actual situation.

[0053] In step S200, through progressive analysis, the feature evidence chains of multiple dimensions are concatenated, and finally the high-confidence attack determination is realized by the fusion model, which not only avoids the limitations of single features but also ensures the rigor of the logical link.

[0054] Step S300: Based on the noise of different offsets during the real-time weighing of the electronic scale, match the electronic scale with different models of cheating devices, and determine the maximum signal-to-noise ratio of the matched cheating device as the crystal oscillator noise coefficient.

[0055] After determining the PWM attack, the detailed anti-cheating calibration strategy cannot be determined yet. This is because different models of remote controls, different types of pulse methods, such as intermittent pulses, and different attack methods will all result in different pulse attack methods and different ways of weighing signal offset. Therefore, it is still necessary to track the cheating behavior and formulate the best calibration strategy.

[0056] The phase noise characteristics of the cheating remote control oscillator are determined by the crystal oscillator defects. Due to the production process differences, different crystal oscillators will emit electromagnetic waves with unique noises, and the crystal oscillator defects of different brand remote controls result in a unique depression-peak mode in the phase noise spectrum. Therefore, the type of attack device can be locked by matching the noise power characteristics.

[0057] A high-impedance probe is implanted in the sensor power supply circuit to extract the phase noise of the PWM carrier. Among them, the input impedance of the high-impedance probe > 1MΩ.

[0058] Obtain the center frequency of the AD value during weighing and preset the frequency offset , in the embodiment of the present invention, three gears of 1kHz, 10kHz, and 100kHz are taken, that is, the noise magnitudes at three key point positions 1kHz, 10kHz, and 100kHz away from the center frequency are measured respectively. For example, if the center frequency is 20kHz, the frequency offsets are at three positions of 20 + 1, 20 + 10, and 20 + 100.

[0059] Calculate the signal-to-noise ratio of the single-sideband phase : ; where, lg is the logarithmic function with base 10; is the center frequency of the attack pulse signal; is the frequency offset; is the noise power at the frequency offset; is the carrier power at the center frequency of the attack pulse signal.

[0060] It realizes the logarithmic conversion of the ratio of the noise power of the three frequency offsets to the carrier power of the main power respectively, and obtains the signal-to-noise ratios of the three frequency offset magnitudes, that is, different frequency offset magnitudes can obtain their respective corresponding different signal-to-noise ratios.

[0061] Construct a signal-to-noise ratio vector from the signal-to-noise ratios corresponding to different frequency offset amounts; where the elements in the signal-to-noise ratio vector are signal-to-noise ratios.

[0062] For all types of cheating devices, extract their respective signal-to-noise ratio vectors and establish a reference library.

[0063] Match the signal-to-noise ratio vector of the electronic scale with those of different types of cheating devices, screen out the cheating devices, and determine the maximum signal-to-noise ratio corresponding to the cheating device as the crystal oscillator noise coefficient. Specifically: Use the calculated signal-to-noise ratio vector of the electronic scale to match with the signal-to-noise ratio vectors of all known types of cheating devices respectively, that is, calculate the cosine similarity of the two signal-to-noise ratio vectors respectively. The higher the cosine similarity, the more the attacked electromagnetic fingerprint of the AD signal matches the type of this remote control. In the embodiment of the present invention, a similarity threshold of 0.65 is preset. When the cosine similarity of the two signal-to-noise ratio vectors is greater than the preset similarity threshold, the possible type of cheating device can be determined, and the maximum signal-to-noise ratio corresponding to the cheating device is determined as the crystal oscillator noise coefficient. It should be noted that if the cosine similarity of no cheating device is greater than the preset similarity threshold, the maximum signal-to-noise ratio of the cheating device corresponding to the current maximum cosine similarity is used as the crystal oscillator noise coefficient.

[0064] Process defects in the crystal oscillator will cause abnormal noise power at specific frequency offsets. For example, the noise power of pirated remote controls at a 10 kHz frequency offset is 8 - 12 dB higher than that of genuine products.

[0065] Step S400: Based on the waveform jitter characteristics of the attack pulse signal, match the attack pulse signal with the simulated attack pulse signals in the reference library, and determine the pulse gene mutation coefficient according to the waveform abnormality of the matched simulated attack pulse signal.

[0066] The parasitic inductance and capacitance of the attack circuit form unique resonance characteristics, that is, when the cheating pulse passes through the circuit, it will leave unique waveform characteristics. Different circuit designs of different manufacturers will form different unique waveform characteristics.

[0067] Therefore, first extract the pulse waveform chain of the attack pulse signal, and the pulse waveform chain is used to represent the waveform jitter characteristics of the attack pulse signal. Specifically: First, perform fifth-order differential processing on the attack pulse signal to obtain a fifth-order differential signal, which realizes 5 times of amplification of the pulse, making the original tiny jitter become obvious. For example, making the circuit resonance characteristics become obvious.

[0068] Then, extract the extreme points from the fifth-order differential signal, perform polynomial fitting on these extreme points, and then solve the coefficients using the least squares method. The finally obtained coefficient vector is called the pulse waveform chain.

[0069] The attack circuits of different manufacturers result in different jitter patterns after differentiation and different fitting results. By simulating the hardware circuits of known cheater models, various types of attack circuits (including parasitic inductance / capacitance) are built in the laboratory. An intermittent / continuous pulse is output through a programmable signal generator to simulate the test and output several pulse waveforms of simulated attacks. These pulse waveform chains are extracted and a reference library is also established.

[0070] Based on the pulse waveform chain, the attack pulse signal is matched with the simulated attack pulse signals in the reference library. The reference attack pulse signal type is selected from the pulse waveform types to which the matched simulated attack pulse signals belong. According to the waveform anomalies of the simulated attack pulse signals within the reference attack pulse signal type, the pulse gene mutation coefficient is determined.

[0071] The screening process of the reference attack pulse signal type is as follows: The pulse waveform chain of the attack pulse signal and the pulse waveform chains of the simulated attack pulse waveforms in the reference library are respectively calculated for cosine similarity. When the cosine similarity is greater than 0.7, the pulse waveform type to which the corresponding simulated attack pulse waveform belongs is used as the reference attack pulse signal type. It should be noted that if the cosine similarity of the simulated attack pulse waveforms is not greater than 0.7, the pulse waveform type corresponding to the current maximum cosine similarity is used as the reference attack pulse signal type.

[0072] Furthermore, according to the waveform anomalies of the reference attack pulse signal type, the pulse gene mutation coefficient is determined.

[0073] For the reference attack pulse signal type, the Prony algorithm is used to decompose the oscillating waveform at its pulse falling edge; Prony is a mathematical model that uses a linear combination of complex exponential decays to fit equally spaced sampled data, and can directly estimate the frequency, decay factor, amplitude, and phase of a given signal, and can express the characteristics of transient signals more comprehensively than traditional Fourier analysis methods.

[0074] First, the Prony fitting oscillation equation is: ; where A represents the initial amplitude, represents the damping coefficient, [[ID=2,0]] represents the oscillation frequency, represents the initial phase angle. The damping coefficient , oscillation frequency , and initial phase angle that best match the actual waveform are found by solving the equation.

[0075] Furthermore, the pulse gene mutation coefficient of the attack pulse signal is calculated. The calculation formula of this pulse gene mutation coefficient is: ; among them, the numerator represents the vibration energy accumulation ability of the system per second, and the damping coefficient in the denominator represents the energy loss rate. Therefore, the larger this value is, the higher the energy stability of the attack pulse.

[0076] Even if the attacker adjusts the pulse amplitude or duration, as long as the parameters of its hardware circuit remain unchanged, the Q value is determined. Take the value corresponding to the reference attack pulse signal type as the pulse gene mutation coefficient.

[0077] Step S500: Divide the AD value of the electronic scale into different state stages, and determine the outlier in the behavior layer according to the number of position transfers of the attack window in different state stages.

[0078] Professional cheaters show specific parameter adjustment inertia, and the professional characteristics are as follows: the first acceleration adjustment, at this time the first derivative is large; then the second fine-tuning deceleration, at this time the second derivative is negative.

[0079] By establishing an adversarial experiment model, using professional and amateur groups, respectively test a number of (more than 200 times) random attack events; take different remote control types and operation behaviors as random variables to generate a large amount of experimental data, and each piece of experimental data is a weighing process + cheating operation.

[0080] According to the collected experimental data, by means of manual division or designing thresholds, etc., divide the reading display of each weighing process into multiple state stages. The state stages at least include: stable period, light unloading period, and heavy unloading period, and at most are divided into: ① no-load period, ② loading period, ③ stable period, ④ light unloading period, ⑤ heavy unloading period, ⑥ abnormal vibration period.

[0081] For example, the reading is stable near 0 in the no-load state; the weight continuously rises in the loading state; the weight fluctuation is less than 0.1% and lasts for more than 2 seconds in the stable state; the weight slowly drops during light unloading; the weight drops suddenly during heavy unloading; the reading jumps frequently under abnormal vibration. Note that not every piece of experimental data must be divided into 6 states, nor is it divided into states in a fixed time sequence. Each weighing operation may be different. For example, there may be no ④ light unloading period or no ② loading period; it may also be that the ⑤ heavy unloading period occurs first and then the ④ light unloading period.

[0082] Use the method in step S100 to measure the position of the attack window where the attack event is located; then observe the state stage where the attack window is located and the state stage where the weighing signal is located after the attack window, and then it can be obtained which state the weighing signal transfers to after being attacked. That is, determine the current state stage where the attack window is located and the next adjacent state stage of the attack window.

[0083] The state transition situations of all sub-tests can be obtained, and the occurrence probabilities of cheating in each stage can be statistically analyzed. The calculation method is as follows: The ratio of the number of attack windows in the next adjacent state stage and the current state stage is used as the transition probability from the current state stage to the next adjacent state stage. That is, the number of times the attack window transfers from the current state stage to the next adjacent state stage is used as the numerator, and the number of times the attack window transfers from the previous adjacent state stage to the current state stage is used as the denominator. The ratio composed of the numerator and the denominator is used as the transition probability. The number of transfer times here is the number of attack windows.

[0084] The transition probabilities between all state combinations can be obtained.

[0085] Compared with the weight gain stage, the weight loss stage is more concealed and often not noticed. When the display reading on the electronic scale has been locked in the stable state, the actual weighing data during the background transmission process can be tampered with through a cheating device during the unloading period. Attack behaviors in this situation are more difficult to guard against.

[0086] Therefore, professional cheating behaviors may occur concentratedly between ③ the stable period → ④⑤ the unloading period, showing abnormal transition probabilities, while amateur cheating behaviors show a uniform distribution in terms of state transition probabilities, that is, they may be randomly distributed in each stage; By calculating the information entropy of the transition probabilities of all state combinations in each experimental data, the larger the entropy value, the more random the distribution of the transition probabilities, and the sum of the transition probabilities of ③ the stable period → ④ the light unloading period and ③ the stable period → ⑤ the heavy unloading period. The ratio of the sum of the transition probabilities to the information entropy is used as the outlier at the behavior layer.

[0087] Step S600, correct the baseline of the electronic scale according to the crystal oscillator noise coefficient, correct the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient, and correct the sampling interval of the electronic scale according to the outlier at the behavior layer.

[0088] Based on these three feature vectors of the crystal oscillator noise coefficient, the pulse gene mutation coefficient, and the outlier at the behavior layer, a three-dimensional decision space is constructed: X-axis: crystal oscillator noise coefficient; Y-axis: pulse gene mutation coefficient; Z-axis: outlier at the behavior layer.

[0089] From the large number of experimental sample data obtained through the experimental model in step S500, the above three feature vectors of each experimental sample data are extracted, and then all the experimental sample data are density-clustered in the three-dimensional decision space to divide and obtain several attacker ethnic groups: Output the feature profiles of each attacker ethnic group, including the median values of the three feature vectors in each ethnic group, as the attack reference vectors; the anti-cheating system of the electronic scale can classify the target weighing process according to the similarity with the attack reference vectors of different ethnic groups; After normalizing the three attack reference vectors of each attack group, they are converted into feature weights, i.e., X + Y + Z = 1; the three types of attack features correspond to three calibration strategies. Determine that the three feature vectors corresponding to the current electronic scale participate in density clustering to obtain the attacker group to which the current electronic scale belongs, and use the normalized result values of the attack reference vectors of the attacker group as the normalized value X of the crystal oscillator noise coefficient, the normalized value Y of the pulse gene mutation coefficient, and the normalized value Z of the outlier value at the behavior layer corresponding to the current electronic scale.

[0090] (1) Calibrate the signal baseline offset according to the crystal oscillator noise coefficient; Obtain the envelope of the AD value of the electronic scale in the no-load state. According to the average amplitude of the envelope, use it as the baseline value V. Using the normalized crystal oscillator noise coefficient as the weight, weight the baseline value to obtain the corrected baseline compensation value. Subtract the baseline compensation value from the AD value of the electronic scale in the current no-load state to obtain the corrected AD value of the electronic scale. This corrected AD value is the AD value of the electronic scale after baseline correction, realizing the correction of the baseline of the electronic scale. Among them, the corrected baseline compensation value is: .

[0091] This is because the greater the crystal oscillator noise power ratio, the more significant the baseline voltage offset, and the zero-point voltage needs to be directly compensated.

[0092] (2) Correct the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient; Obtain the default cut-off frequency of the filter in the electronic scale , and then set the bandwidth attenuation factor to . Using the normalized pulse gene mutation coefficient as the weight, weight the default cut-off frequency to obtain the corrected compensation coefficient , specifically: .

[0093] This is because if the pulse waveform stability is lower, the pulse gene mutation coefficient is smaller, and the filter bandwidth needs to be reduced to suppress high-frequency resonance noise.

[0094] (3) Correct the sampling interval of the electronic scale according to the outlier value at the behavior layer; Obtain the default sampling interval . Perform a negative correlation mapping on the normalized outlier value at the behavior layer, and use the mapped result value as the weight to weight the sampling interval to obtain the corrected sampling interval, obtaining the compensated sampling interval , specifically: ; Because the more professional the cheating behavior, the shorter the sampling interval needs to be to improve the capture probability of instantaneous attacks.

[0095] By tracking three weights of the attack behavior, simultaneously regulating the baseline compensation value of the weighing signal (AD value), the filter cut-off frequency, and the sampling interval; performing multi-modal calibration on the weighing signal of the electronic scale.

[0096] During calibration, the calibration weights (accuracy ±0.01% FS) are automatically released for hardware-level verification. If the deviation is still greater than 0.3%, the locking circuit is triggered and the forensic data is uploaded to the supervision platform.

[0097] According to the above attacker feature profile, search for cheating evidence directionally, which can assist in locking the cheating object.

[0098] Please refer to Figure 2 , which shows the system block diagram of the AI-based dynamic anti-cheating calibration system for electronic scales provided by an embodiment of the present invention. The system includes the following modules: A signal acquisition module for acquiring the AD value of the electronic scale and the attack window; the AD values with sudden increase in signal energy at consecutive moments are within the attack window; An attack determination module for determining the attack pulse signal according to the AD values in different frequency bands within the attack window, the AD values of the electronic scale in the no-load state and the reference weighing state; A crystal oscillator analysis module for matching the electronic scale with different models of cheating devices based on the noise of different magnitudes of offsets during real-time weighing of the electronic scale, and determining the maximum signal-to-noise ratio of the matching cheating device as the crystal oscillator noise coefficient; A pulse analysis module for matching the attack pulse signal with the simulated attack pulse signals in the comparison library based on the waveform jitter characteristics of the attack pulse signal, and determining the pulse gene mutation coefficient according to the waveform abnormality of the matching simulated attack pulse signal; A behavior analysis module for dividing the AD value of the electronic scale into different state stages, and determining the outlier at the behavior layer according to the number of position transfers of the attack window in different state stages; A correction module for correcting the baseline of the electronic scale according to the crystal oscillator noise coefficient, correcting the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient, and correcting the sampling interval of the electronic scale according to the outlier at the behavior layer.

[0099] Optionally, the transmission medium can be a wired link, such as but not limited to, coaxial cable, optical fiber, and digital subscriber line, etc., or a wireless link, such as but not limited to, Wireless Fidelity (WIFI), Bluetooth, and mobile device network, etc.

[0100] It should be noted that: For the device provided in the above embodiments, only the division of the above functional modules is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above.

[0101] Figure 3 It is a schematic structural diagram of a computer device provided by an embodiment of the present invention. Exemplarily, as Figure 3 shown, the computer device 700 includes: a memory 710, a processor 720, and a computer program 730 stored in the memory 710 and running on the processor 720. When the processor 720 executes the computer program 730, the computer device can execute any of the above-described AI-based dynamic anti-cheating calibration methods for electronic scales.

[0102] In addition, an embodiment of the present invention also protects a device, which may include a memory and a processor. Among them, an executable program code is stored in the memory, and the processor is used to call and execute the executable program code to execute the AI-based dynamic anti-cheating calibration method provided by an embodiment of the present invention.

[0103] The embodiment of the present invention can divide the functions of the device according to the above method examples. For example, it can correspond to each functional module, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is illustrative, only a logical function division, and there may be other division methods in actual implementation.

[0104] In the case of dividing each module according to each corresponding function, the device may further include a signal uploading module, a determination module, an adjustment module, etc. It should be noted that all relevant contents of each step involved in the above method embodiment can be cited in the function description of the corresponding functional module, and will not be repeated here.

[0105] It should be understood that the device provided by the embodiment of the present invention is used to execute the above AI-based dynamic anti-cheating calibration method for electronic scales, so the same effect as the above implementation method can be achieved.

[0106] In the case of adopting integrated units, the device may include a processing module and a storage module. Among them, when the device is applied to a device, the processing module can be used to control and manage the actions of the device. The storage module can be used to support the device in executing mutual program codes, etc. Among them, the processing module can be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits described in connection with the disclosure of the present invention. The processor can also be a combination that realizes computing functions, such as including a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module can be a memory.

[0107] In addition, the device provided by the embodiments of the present invention can specifically be a chip, a component, or a module. The chip may include a connected processor and a memory; among them, the memory is used to store instructions. When the processor calls and executes the instructions, the chip can execute the method for dynamically anti-cheating calibration of an electronic scale based on AI provided in the above embodiments.

[0108] The embodiments of the present invention also provide a computer-readable storage medium. Computer program code is stored in the computer-readable storage medium. When the computer program code runs on a computer, the computer is enabled to execute the above-related method steps to implement the method for dynamically anti-cheating calibration of an electronic scale based on AI provided in the above embodiments.

[0109] The embodiments of the present invention also provide a computer program product. When the computer program product runs on a computer, the computer is enabled to execute the above-related steps to implement the method for dynamically anti-cheating calibration of an electronic scale based on AI provided in the above embodiments.

[0110] Among them, the device, computer-readable storage medium, computer program product, or chip provided by the embodiments of the present invention are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be elaborated here. Through the description of the above embodiments, those skilled in the art can understand that for the convenience and conciseness of description, only the above-mentioned division of each functional module is used as an example for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In the embodiments provided by the present invention, it should be understood that the disclosed device and method can be implemented in other ways.

[0111] The device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical or other forms.

[0112] It should also be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or terminal device. Without more limitations, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or terminal device including the said element.

[0113] It should be noted that the above sequence of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0114] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key point of each embodiment is to illustrate the differences from other embodiments.

[0115] The above content is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. An AI-based dynamic anti-cheating calibration method for electronic scales, characterized in that, The method includes the following steps: Obtain the AD value of the electronic scale and the attack window; the AD values with sudden increase in signal energy at consecutive moments are within the attack window; Determine the attack pulse signal according to the AD values in different frequency bands within the attack window, the AD values of the electronic scale in the no-load state and the reference weighing state; Based on the noise with different offsets during the real-time weighing of the electronic scale, match the electronic scale with different models of cheaters, and determine the maximum signal-to-noise ratio of the matched cheater as the crystal oscillator noise coefficient; Based on the waveform jitter characteristics of the attack pulse signal, match the attack pulse signal with the simulated attack pulse signals in the comparison library, and determine the pulse gene mutation coefficient according to the waveform abnormality of the matched simulated attack pulse signal; Divide the AD values of the electronic scale into different state stages, and determine the outlier of the behavior layer according to the number of position transfers of the attack window in different state stages; Correct the baseline of the electronic scale according to the crystal oscillator noise coefficient, correct the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient, and correct the sampling interval of the electronic scale according to the outlier of the behavior layer.

2. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, wherein, The method for obtaining the attack window is as follows: Slide a window on the AD value sequence arranged in time sequence, take the derivative of the AD values within the sliding window, and when the derivative of consecutive sampling points is greater than the preset sliding threshold, take the window range where the consecutive sampling points are located as the candidate window; the preset sliding threshold is a fixed multiple of the standard deviation of the data within the sliding window; the number of consecutive sampling points is greater than the preset consecutive threshold; Calculate the ratio of the maximum AD value in the candidate window to the AD value at the previous moment as the mutation rate; When the mutation rate is greater than the preset mutation threshold and the sampling duration corresponding to the candidate window belongs to the preset duration range, determine the candidate window as the attack window.

3. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, characterized in that, The determination of the attack pulse signal according to the AD values in different frequency bands within the attack window, the AD values of the electronic scale in the no-load state and the reference weighing state includes: Compare the AD value sequences of the electronic scale in the no-load state and the reference weighing state to obtain the abnormal determination value of the attack window; determine the asymmetry factor of the attack window according to the total duration of the pulse rising edge and the total duration of the pulse falling edge of the AD values in the attack window; combine the abnormal determination value and the asymmetry factor to determine the initial abnormal determination factor; Determine the attack pulse characteristics of the attack window according to the AD values in different frequency bands within the attack window; Generate a time-frequency spectrum within the attack window, obtain the time-sequence change sequence of the energy at the 20 kHz frequency point, and determine the verification factor according to the correlation coefficient between the time-sequence change sequence of the energy at the 20 kHz frequency point and the time-sequence change sequence of the AD values; Determine the decision coefficient according to the initial abnormal determination factor, the attack pulse characteristics, and the verification factor; When the decision coefficient is greater than or equal to the preset decision threshold, determine that there is a PWM attack in the attack window; take the AD values within the longest consecutive attack window with PWM attack during weighing as the attack pulse signal.

4. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 3, wherein, The determination of the attack pulse characteristics of the attack window according to the AD values in different frequency bands within the attack window includes: For any attack window, perform 8-layer wavelet packet decomposition in the frequency band of 100 kHz - 10 MHz, and calculate the energy entropy at different frequency positions; Perform short-time Fourier transform on the frequency band of 20 kHz and within a preset range, and determine the proportion of the narrowband duration; Combine the energy entropy and the proportion of the narrowband duration to determine the attack pulse characteristics of the attack window.

5. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, characterized in that Based on the noise of different offsets during real-time weighing of the electronic scale, match the electronic scale with different models of cheating devices, and determine the maximum signal-to-noise ratio of the matched cheating device as the crystal oscillator noise coefficient, including: Based on the noise of different frequency offsets during real-time weighing of the electronic scale, determine the signal-to-noise ratio corresponding to different frequency offsets, and construct a signal-to-noise ratio vector; Match the signal-to-noise ratio vectors of the electronic scale and different models of cheating devices, screen out the cheating devices, and determine the maximum signal-to-noise ratio corresponding to the cheating device as the crystal oscillator noise coefficient.

6. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, wherein Based on the waveform jitter characteristics of the attack pulse signal, match the attack pulse signal with the simulated attack pulse signals in the reference library, and determine the pulse gene mutation coefficient according to the waveform abnormality of the matched simulated attack pulse signal, including: Extract the pulse waveform chain from the attack pulse signal, and the pulse waveform chain is used to represent the waveform jitter characteristics of the attack pulse signal; Based on the pulse waveform chain, match the attack pulse signal with the simulated attack pulse signals in the reference library, screen out the reference attack pulse signal type from the pulse waveform types to which the matched simulated attack pulse signals belong, and determine the pulse gene mutation coefficient according to the waveform abnormality of the simulated attack pulse signals within the reference attack pulse signal type.

7. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, wherein Divide the AD value of the electronic scale into different state stages, and determine the behavior layer outlier according to the number of position transfers of the attack window in different state stages, including: The state stages include: stable period, light unloading period, and heavy unloading period; Determine the current state stage where the attack window is located and the next adjacent state stage of the attack window; Use the ratio of the number of attack windows in the next adjacent state stage to the current state stage as the transition probability from the current state stage to the next adjacent state stage; Obtain the information entropy of all transition probabilities; calculate the sum of the transition probabilities from the stable period to the light unloading period and the transition probability from the stable period to the heavy unloading period; Use the ratio of the sum of the transition probabilities to the information entropy as the behavior layer outlier.

8. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, characterized in that, The method for correcting the baseline of the electronic scale according to the crystal oscillator noise coefficient includes: Obtain the envelope of the AD value of the electronic scale, and use the average amplitude of the envelope as the baseline value; Weight the baseline value with the normalized crystal oscillator noise coefficient as the weight to obtain the corrected baseline compensation value.

9. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, characterized in that The method for correcting the cut-off frequency of the electronic scale according to the pulse gene mutation coefficient includes: Weight the default cut-off frequency with the normalized pulse gene mutation coefficient as the weight to obtain the corrected compensation coefficient.

10. The AI-based dynamic anti-cheating calibration method for electronic scales according to claim 1, wherein The method for correcting the sampling interval of the electronic scale according to the behavior layer outlier includes: Perform negative correlation mapping on the normalized behavior layer outlier, and weight the default sampling interval with the result value of the mapping as the weight to obtain the corrected sampling interval.

Images