Anti-cheating intelligent weighing system and method based on multi-source signals

Abstract

The present invention provides an anti-cheating intelligent weighing system and method based on multi-source signals, which relates to the field of weighing technology. The system comprises the following steps: based on real-time synchronously collected pressure distribution time series data, object three-dimensional contour and center of gravity position trajectory, and deformation displacement field data, and in combination with a preset conduction reference library, dynamically calculating and generating multi-source features; based on the force conduction deviation value, contour confidence value, and pressure distribution entropy value in the multi-source features, dynamically calculating and generating a first weight, a second weight, and a third weight; based on the first weight, the second weight, and the third weight, and in combination with a geometric out-of-bounds ratio value and a force conduction deviation value, dynamically calculating and generating a dynamic trust threshold; based on the conduction reference library, calculating the global consistency index within the current weighing cycle; based on the global consistency index and the dynamic trust threshold, generating an anti-cheating result according to preset judgment conditions. The present invention realizes high-precision and high-reliability weighing anti-cheating through multi-source signal fusion and dynamic analysis.

Description

Technical Field

[0001] The present invention relates to the field of weighing technology, and in particular to an anti-cheating intelligent weighing system and method based on multi-source signals. Background Art

[0002] In the field of weighing technology, traditional weighing equipment has problems such as insufficient detection capabilities and poor accuracy when faced with complex cheating methods. With the development of science and technology, cheating methods have become more covert and diverse, such as changing the placement of objects and using external interference to affect sensor data. Most existing anti-cheating methods for weighing are based on a single signal source (such as relying solely on pressure sensor data), which makes it difficult to fully capture various abnormal information during the weighing process. Multi-source signals contain rich weighing-related features, but lack effective fusion and analysis methods, and cannot fully utilize the advantages of multi-source data to improve anti-cheating capabilities. At the same time, existing technologies lack flexible response strategies for the dynamic changes of various parameters during the weighing process, making it difficult to accurately determine whether there is cheating behavior based on real-time working conditions. This results in limited anti-cheating effects and an inability to meet the needs of high-precision and high-reliability weighing anti-cheating in practical applications.

[0003] Therefore, it is necessary to provide an anti-cheating intelligent weighing system and method based on multi-source signals to solve the above technical problems. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides an anti-cheating intelligent weighing system and method based on multi-source signals. Through multi-source signal fusion and dynamic analysis, high-precision and high-reliability weighing anti-cheating is achieved, ensuring the fairness and accuracy of weighing transactions.

[0005] The present invention provides an anti-cheating intelligent weighing method based on multi-source signals, the method comprising the following steps:

[0006] Based on the real-time synchronous collection of pressure distribution time series data, object 3D contour and center of gravity position trajectory, and deformation displacement field data during the current weighing cycle, and combined with the preset conduction reference library, the system dynamically calculates and generates geometric out-of-bounds ratio value, force conduction deviation value, contour confidence value and pressure distribution entropy value;

[0007] Based on the force conduction deviation value, the profile confidence value, and the pressure distribution entropy value, dynamically calculating and generating a first weight representing volume variation, a second weight representing force conduction, and a third weight representing delay sensitivity;

[0008] Dynamically calculating and generating a dynamic credible threshold based on the first weight, the second weight, and the third weight, and in combination with the geometric out-of-bounds ratio value and the force conduction deviation value;

[0009] Calculating a global consistency index within a current weighing cycle based on the conductive reference library;

[0010] Based on the global consistency index and the dynamic trust threshold, an anti-cheating result is generated according to preset judgment conditions.

[0011] Preferably, the dynamic calculation of the geometric out-of-bounds ratio value, the force conduction deviation value, the profile confidence value and the pressure distribution entropy value includes:

[0012] Based on the three-dimensional contour of the object, the ratio of the object boundary exceeding the projection area of ​​the scale platform is calculated, and the geometric out-of-bounds ratio value is output;

[0013] Based on the pressure distribution time series data and the deformation displacement field data, a real-time mapping relationship between the pressure distribution gradient matrix and the deformation displacement characteristic vector is generated, and a preset conduction reference library is matched to output a force conduction deviation value;

[0014] Calculating a volume variation coefficient of a point cloud of the three-dimensional contour of the object in three-dimensional space and an edge fuzziness in an image domain, and generating a contour confidence value by combining the volume variation coefficient and the edge fuzziness;

[0015] Based on the pressure distribution time series data, the Shannon entropy of the pressure distribution is calculated according to the time window division strategy preset in the conduction reference library as the pressure distribution entropy value.

[0016] Preferably, the generation of the first weight, the second weight and the third weight specifically includes:

[0017] Based on the force conduction deviation value and the contour confidence value, dynamically generating a first weight representing volume variation through the Sigmoid function transformation rule and conduction attenuation constraint strategy configured in the conduction benchmark library;

[0018] Combining the pressure distribution entropy value with the deviation attenuation parameter output by the conduction reference library to construct a joint gain factor, driving the generation of a second weight representing force conduction;

[0019] Through the weight compensation rule defined in the conduction reference library, the first weight is used as a feedback item and dynamically coupled with the contour confidence value and the force conduction deviation value to generate a third weight that characterizes delay sensitivity.

[0020] Preferably, the dynamic calculation of the dynamic trust threshold specifically includes:

[0021] Generate a boundary penalty term based on the first weight and the geometric out-of-bounds ratio value;

[0022] Generate a conduction compensation term by coupling calculation of the second weight and the force conduction deviation value, combined with the deviation sensitivity coefficient preset in the conduction reference library;

[0023] Adaptively weighting the real-time dynamically generated modified delay value based on the third weight to generate a delay impact item;

[0024] A dynamic credible threshold is generated based on the boundary penalty term, the conduction compensation term, and the delay impact term.

[0025] Preferably, the calculation formula of the dynamic trust threshold is:

[0026]

[0027] in, represents the dynamic trust threshold, represents the boundary penalty term, represents the conduction compensation term, Indicates the delay impact item.

[0028] Preferably, the calculating of the global consistency index in the current weighing cycle based on the conductive reference library includes:

[0029] Obtain historical operating condition data within the current weighing cycle from the conductive reference library, and calculate the average vibration intensity of the weighing platform, the rate of change of ambient temperature and humidity, and the sensor voltage fluctuation coefficient;

[0030] Based on the vibration weight, temperature and humidity weight, and voltage weight configured in the conductive reference library, a weighted calculation is performed, and the result of the weighted calculation is transformed and outputted through a hyperbolic tangent function to obtain a global consistency index.

[0031] Preferably, the generating of the anti-cheating result based on the global consistency index and the dynamic trust threshold according to the preset judgment conditions includes:

[0032] Inputting the global consistency index into a pre-trained hidden Markov model in the conduction benchmark library, and outputting a state transition probability vector;

[0033] Normalizing the entropy value of the state transition probability vector based on the dynamic trustworthy threshold to generate a modified state transition sequence;

[0034] Calculating the distance between the modified state transition sequence and the typical cheating pattern features in the historical cheating behavior pattern feature library stored in the conductive benchmark library;

[0035] When the distance value is less than the pattern matching threshold set in the conductive reference library, a cheating alarm result is output; otherwise, an approved weighing result is output.

[0036] The present invention also provides an anti-cheating intelligent weighing system based on multi-source signals, which is used to execute an anti-cheating intelligent weighing method based on multi-source signals. The system includes:

[0037] The feature calculation module is used to dynamically calculate and generate the geometric out-of-bounds ratio value, force conduction deviation value, contour confidence value and pressure distribution entropy value based on the pressure distribution time series data, object three-dimensional contour and center of gravity position trajectory, and deformation displacement field data collected in real time during the current weighing cycle, combined with the preset conduction reference library;

[0038] a weight generation module, configured to dynamically calculate and generate a first weight representing volume variation, a second weight representing force conduction, and a third weight representing delay sensitivity based on the force conduction deviation value, the profile confidence value, and the pressure distribution entropy value;

[0039] a threshold calculation module, configured to dynamically calculate and generate a dynamic credible threshold based on the first weight, the second weight, and the third weight, in combination with the geometric out-of-bounds ratio value and the force conduction deviation value;

[0040] An index calculation module, configured to calculate a global consistency index within a current weighing cycle based on the conductive reference library;

[0041] The result generation module is used to generate an anti-cheating result according to preset judgment conditions based on the global consistency index and the dynamic trust threshold.

[0042] Compared with related technologies, the anti-cheating intelligent weighing system and method based on multi-source signals provided by the present invention has the following beneficial effects:

[0043] The present invention collects multi-source signals such as pressure distribution time series data, object three-dimensional contour and center of gravity position trajectory in real time and synchronously, and dynamically calculates multiple eigenvalues ​​to comprehensively capture key information in the weighing process. By using multi-source data fusion to calculate different weights, and then generating a dynamic trust threshold, combined with the global consistency index, it can flexibly adjust the judgment criteria according to the real-time working conditions and effectively identify various complex cheating methods. When calculating eigenvalues ​​such as geometric out-of-bounds ratio values ​​and force conduction deviation values, targeted algorithms are used to ensure the accuracy and reliability of the data. The state transition sequence is processed by methods such as hidden Markov model and entropy normalization, and matched with the historical cheating behavior pattern feature library, which greatly improves the accuracy and timeliness of cheating behavior identification. This method effectively solves the shortcomings of traditional weighing anti-cheating technology, provides a more accurate and reliable anti-cheating solution, and ensures the fairness and accuracy of weighing transactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of an anti-cheating intelligent weighing method based on multi-source signals provided by the present invention;

[0045] Figure 2 This is a structural diagram of an anti-cheating intelligent weighing system based on multi-source signals provided by the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all of the structures. Furthermore, the embodiments of the present invention and the features of the embodiments may be combined with one another unless there is a conflict.

[0047] It should also be noted that, for ease of description, only portions relevant to the present invention are shown in the accompanying drawings, rather than all of the contents. Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the various operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. In addition, the order of the various operations can be rearranged. The process can be terminated when its operations are completed, but may also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0048] Example 1

[0049] The present invention provides an anti-cheating intelligent weighing method based on multi-source signals. Figure 1 As shown, the method includes the following steps:

[0050] S1: Based on the real-time synchronous collection of pressure distribution time series data, object three-dimensional contour and center of gravity position trajectory, and deformation displacement field data during the current weighing cycle, and combined with the preset conduction reference library, the geometric out-of-bounds ratio value, force conduction deviation value, contour confidence value and pressure distribution entropy value are dynamically calculated and generated.

[0051] Specifically, in step S1, the dynamic calculation of the geometric out-of-bounds ratio value, the force transmission deviation value, the profile confidence value, and the pressure distribution entropy value includes:

[0052] Based on the three-dimensional contour of the object, a ratio value of the object boundary exceeding the projection area of ​​the scale platform is calculated, and a geometric out-of-bounds ratio value is output.

[0053] In this embodiment, the generation of the geometric out-of-bounds ratio value first obtains the point cloud data of the object on the scale platform through the 3D vision sensor. ,(in , indicating the coordinates of the point cloud).

[0054] Then calculate the center coordinates of the scale platform As a reference benchmark, calculate the distance function for each point cloud: (where 、 、 The scale boundary equation coefficients provided by the conductive reference library) and the center point distance are calculated The conditional counting formula is used to count the number of out-of-bounds points:

[0055] ;

[0056] in, It is a sign function. When the point cloud is outside the scale boundary, the count is 1, and the final output is the geometric out-of-bounds ratio value. , which represents the proportion of the object that exceeds the scale boundary.

[0057] Based on the pressure distribution time series data and the deformation displacement field data, a real-time mapping relationship between the pressure distribution gradient matrix and the deformation displacement characteristic vector is generated, and a preset conduction reference library is matched to output a force conduction deviation value.

[0058] In this embodiment, the calculation of the force conduction deviation value begins with the calculation of the pressure distribution gradient component: (Implemented by Sobel operator, Indicates the pressure change, represents the spatial direction), Indicates the pressure distribution in The spatial gradient of the direction (the change in pressure per unit distance), Indicates the pressure distribution in The spatial gradient of the direction, together with the pressure gradient matrix G, reflects the spatial change rate of the pressure on the scale. Then the displacement feature vector is constructed. , the specific expression is:

[0059] ;

[0060] in, represents the average value of the displacement field, represents the standard deviation of displacement, Indicates the maximum displacement, represents the displacement change rate, represents the displacement change between adjacent time steps, Represents the time step interval.

[0061] Matching standard models from the Conductive Benchmark Library: ( represents the Frobenius norm of the matrix, i.e. the sum of the squares of the differences between the matrix elements). Finally, the force conduction deviation value is calculated. , the formula is: ;

[0062] in is the material attenuation coefficient, is the L2 norm, which reflects the degree of deviation between the actual force conduction and the benchmark model. Represents the standard displacement feature vector stored in the conductive reference library, which contains the displacement statistical characteristics under ideal working conditions. is the standard pressure gradient matrix in the benchmark library, which represents the pressure distribution pattern during normal weighing.

[0063] The volume variation coefficient of the point cloud of the three-dimensional contour of the object in the three-dimensional space and the edge fuzziness in the image domain are calculated, and the contour confidence value is generated by combining the volume variation coefficient and the edge fuzziness.

[0064] In this embodiment, the contour confidence value calculation first divides the space into 5mm cubic pixel grids. Calculate the volume variation coefficient :

[0065] ;

[0066] in, Indicates the proportion of valid voxels, that is, the proportion of voxels containing point clouds; Represents the standard deviation of voxel density, reflecting the uniformity of point cloud distribution.

[0067] Extract the contour edge through the Canny operator and calculate the average gradient amplitude (indicates edge clarity), from which edge blur is calculated , the formula is as follows:

[0068] ;

[0069] in, Larger values ​​indicate sharper edges and less blur.

[0070] Get sensor weights from a conductive reference library and environmental weights (Automatically adapted according to sensor type and ambient lighting). The final output contour confidence value:

[0071] ;

[0072] This value comprehensively reflects the reliability of the 3D contour. The closer it is to 1, the higher the confidence. The volume variation coefficient reflects the distribution uniformity of the point cloud in the 3D space (such as the standard deviation of the voxel density). Its error mainly comes from the measurement accuracy of the sensor itself (such as the point cloud sampling noise and depth measurement error of the 3D vision sensor). The sensor weight Amplify the impact of sensor hardware defects on confidence. For example, low-precision sensors can lead to sparse or unevenly distributed point clouds ( Increase), at this time This will increase the penalty weight of this parameter, highlighting the limitation of sensor performance on the credibility of the contour.

[0073] Edge blur depends on the image edge gradient, and its error is mainly caused by the degradation of optical imaging quality due to environmental interference (such as insufficient lighting, reflection, and fog), and has nothing to do with the sensor hardware.

[0074] The environmental weight It is possible to quantify the interference of environmental factors on the definition of contours. For example, strong light reflections can cause edge blurring ( Increase), at this time It will enhance the suppression weight of environmental noise, reflecting the negative impact of environmental conditions on contour extraction.

[0075] Based on the pressure distribution time series data, the Shannon entropy of the pressure distribution is calculated according to the time window division strategy preset in the conduction reference library as the pressure distribution entropy value.

[0076] In this embodiment, the pressure distribution entropy value is generated by loading the time window configuration from the conductive reference library. (in , typical values ​​are [0.2s, 0.5s, 1.0s], representing different analysis time domains).

[0077] Shannon entropy is calculated for each time window. The specific formula is:

[0078] ;

[0079] in, Indicates the Normalized value of the pressure zone, is the total number of partitions, and the logarithm is used to quantify the uncertainty of the pressure distribution).

[0080] Then multi-scale entropy fusion is performed, the specific formula is:

[0081] ;

[0082] ;

[0083] in, is the time window weight coefficient, and the long time window has a greater weight.

[0084] Finally, get the entropy limit parameters from the benchmark library and , output the final entropy value , and its calculation formula is:

[0085] ; This value represents the complexity of the pressure distribution. Cheating behavior usually leads to an abnormal increase in entropy.

[0086] S2: Based on the force conduction deviation value, the contour confidence value and the pressure distribution entropy value, dynamically calculate and generate a first weight representing volume variation, a second weight representing force conduction and a third weight representing delay sensitivity.

[0087] Specifically, in step S2, the dynamic calculation to generate the first weight, the second weight, and the third weight specifically includes:

[0088] Based on the force conduction deviation value and the contour confidence value, a first weight characterizing volume variation is dynamically generated through the Sigmoid function transformation rule and conduction attenuation constraint strategy configured in the conduction reference library.

[0089] In this embodiment, the first weight characterizing the volume variation Dynamically generated by the following steps:

[0090] First, obtain the force conduction deviation value (larger values ​​indicate more severe force conduction abnormalities), and contour confidence values (A larger value indicates a more reliable 3D contour).

[0091] Then, perform Sigmoid function transformation, specifically obtain Sigmoid function parameters from the conduction benchmark library and calculate the confidence offset , the formula is:

[0092]

[0093] Applying Sigmoid transformation, the expression is:

[0094]

[0095] in, Represents the Sigmoid curve steepness coefficient (the default value of the benchmark library is 2.5). Indicates the confidence deviation threshold (the default value of the benchmark library is 0.6).

[0096] Next, conduction attenuation constraints are performed, including obtaining the attenuation coefficient from the conduction reference library and offset , calculate the attenuation factor , the calculation formula is:

[0097]

[0098] Apply the exponential decay constraint and normalize to get the first weight , where the exponential decay constraint is applied as:

[0099] .

[0100] The pressure distribution entropy value is combined with the deviation attenuation parameter output by the conduction reference library to construct a joint gain factor to drive the generation of a second weight that characterizes force conduction.

[0101] In this embodiment, the second weight representing the force conduction Dynamically generated by the following steps:

[0102] Parameter preparation: Get the pressure distribution entropy value Sentropy and get the deviation attenuation parameter from the conductivity reference library (Environmental stability coefficient).

[0103] Joint gain factor construction, specifically including: calculating entropy offset : ( is the standard entropy value in the benchmark library).

[0104] Construction gain factor : .

[0105] Force conduction weight drive, specifically including: obtaining force conduction reference weight from the benchmark library , calculate the basic weight , applying logarithmic constraints, and finally generating .

[0106] Through the weight compensation rule defined in the conduction reference library, the first weight is used as a feedback item and dynamically coupled with the contour confidence value and the force conduction deviation value to generate a third weight that characterizes delay sensitivity.

[0107] In this embodiment, the third weight characterizing delay sensitivity Dynamically generated by the following steps:

[0108] First, obtain the coupling coefficients from the benchmark library , calculate the confidence-deviation coupling term:

[0109] ;

[0110] Apply the feedback compensation rule to process the confidence-bias coupling term and obtain the third weighted intermediate value , specifically;

[0111] ;

[0112] Secondly, delay sensitivity enhancement is performed, specifically including obtaining the current delay value , calculate the delay correction ,in is the delay correction function (the benchmark library is defined as ), apply delay weighting and perform normalization to obtain the third weight , where the expression for applying delay weighting is:

[0113] .

[0114] S3: Based on the first weight, the second weight and the third weight, and in combination with the geometric out-of-bounds ratio value and the force conduction deviation value, dynamically calculate and generate a dynamic credible threshold.

[0115] Specifically, in step S3, the dynamic calculation of the dynamic trust threshold includes the following steps:

[0116] A boundary penalty term is generated based on the first weight and the geometric out-of-bounds ratio value.

[0117] In this embodiment, the generation of the boundary penalty term includes the following steps:

[0118] Get the geometric out-of-bounds ratio value and the first weight , query the boundary sensitivity coefficient from the conductive benchmark library (Default value is 1.5).

[0119] Calculate the penalty base: the geometric out-of-bounds ratio value and boundary sensitivity coefficient Multiply to get the penalty base .

[0120] Application weight enhancement: The specific enhancement formula is ,in The penalty base, represents the exponential function, is the boundary penalty reinforcement value.

[0121] Get the upper bound of the penalty from the benchmark library (Default 0.3).

[0122] Output the final boundary penalty term: take the smaller value of the weighted reinforcement result and the boundary penalty upper limit.

[0123] The conduction compensation term is generated by coupling calculation of the second weight and the force conduction deviation value, combined with the deviation sensitivity coefficient preset in the conduction reference library.

[0124] In this embodiment, the force conduction deviation value is obtained and the second weight , calculate the deviation sensitivity coefficient :According to the formula Dynamic Generation

[0125] Calculate the base compensation: deviate from the sensitivity coefficient Multiply by (1 force conduction deviation value).

[0126] Apply a logarithmic transformation: compute the natural logarithm of (1 + basis offset).

[0127] Get the lower limit of the compensation range from the reference library and upper limit .

[0128] Output final conduction compensation term : Ensure that the result is within the compensation range. This compensation term suppresses the influence of abnormal force conduction. The second weight Dynamically regulate sensitivity to conduction deviations.

[0129] The modified delay value dynamically generated in real time is adaptively weighted based on the third weight to generate a delay impact item.

[0130] In this embodiment, generating the delay impact item includes the following steps:

[0131] Get the third weight and system real-time delay value , calling the delay correction function from the conduction benchmark library .

[0132] Calculate the delay correction: Input the real-time delay value into the delay correction function.

[0133] Calculate the intermediate variable: multiply the third weight by the delay correction.

[0134] Apply the hyperbolic tangent constraint: multiply the intermediate variable by 2.5 and take the hyperbolic tangent value

[0135] Output the final delay impact term, which captures the delay characteristics of cheating behavior, the third weight Determines the system's sensitivity to delay.

[0136] A dynamic credible threshold is generated based on the boundary penalty term, the conduction compensation term, and the delay impact term.

[0137] Specifically, the calculation formula of the dynamic trust threshold is:

[0138]

[0139] in, represents the dynamic trust threshold, represents the boundary penalty term, represents the conduction compensation term, Indicates the delay impact item.

[0140] In this embodiment, the synthesis of dynamic trustworthy threshold is a multi-stage calculation process, the core of which is the comprehensive boundary penalty term , conduction compensation term and delay impact Generate quantitative indicators of impact.

[0141] First calculate the total penalty compensation ,This sum represents the combined impact of boundary crossing, force conduction anomaly, and time delay anomaly on the data credibility.

[0142] Then calculate the basic credibility value ,The calculation follows the principle of “1 is completely credible” and obtains a preliminary ,credibility assessment by subtracting the sum of negative impact factors from the ,ideal state.

[0143] Then implement the lower limit protection. When it is less than 0.05, the correction threshold is taken as 0.05, otherwise the correction threshold is taken as ,This mechanism ensures that the trust threshold is not lower than 0.05, maintaining the minimum ,fault tolerance.

[0144] Finally, an upper bound constraint is applied to obtain the final threshold T = min (0.95, the modified threshold). This step locks the upper bound of the credible threshold to 0.95 to avoid over-optimization and ignoring risk. Physically, T close to 0.95 indicates high credibility, while close to 0.05 indicates severe unreliability. Intermediate values ​​correspond to different risk levels. This process enables those skilled in the art to convert multi-source anomaly signals into a single quantitative indicator, requiring only basic arithmetic and comparison operations.

[0145] S4: Calculating a global consistency index within the current weighing cycle based on the conductive reference library.

[0146] Specifically, in step S4, the calculation of the global consistency index includes the following steps:

[0147] The historical working condition data in the current weighing cycle is obtained from the conductive reference library, and the average vibration intensity of the weighing platform, the rate of change of the ambient temperature and humidity, and the sensor voltage fluctuation coefficient are calculated.

[0148] In this embodiment, the average vibration intensity of the weighing platform is first calculated: the three-axis acceleration sampling values ​​at all time points in the cycle are extracted from the reference library (sampling frequency is 100 Hz, and there are 300 sampling points in a typical weighing cycle of 3 seconds); the three-axis composite vector modulus (i.e., vibration intensity value) is calculated for each sampling point, and the formula is the square root of the sum of the squares of the X / Y / Z three-axis accelerations of each sampling point; finally, the arithmetic mean of the vibration intensity values ​​of all sampling points is calculated and the output is the vibration intensity average.

[0149] Then calculate the rate of change of ambient temperature and humidity: obtain the temperature and humidity values ​​at the start and end of the cycle from the reference library; calculate the temperature change rate and humidity change rate respectively (the temperature change rate is the average change per second of the absolute value of the temperature difference between the end and start times, and the humidity change rate is the average change per second of the absolute value of the humidity difference); output the comprehensive change rate (take the square root of the sum of the squares of the temperature change rate and the humidity change rate).

[0150] Finally, the sensor voltage fluctuation coefficient is calculated: the voltage sequence of all pressure sensors within the cycle is extracted from the reference library; the voltage fluctuation coefficient of each sensor is calculated (that is, the standard deviation of the sensor voltage sequence divided by the mean); and the global fluctuation coefficient is output (the maximum value of all sensor voltage fluctuation coefficients is taken).

[0151] Based on the vibration weight, temperature and humidity weight, and voltage weight configured in the conductive reference library, a weighted calculation is performed, and the result of the weighted calculation is transformed and outputted through a hyperbolic tangent function to obtain a global consistency index.

[0152] In this embodiment, after obtaining the three operating condition indicators, weight configuration and weighted calculation are performed:

[0153] Obtain vibration weight, temperature and humidity weight, and voltage weight from the conductive benchmark library. These weights satisfy the normalization condition that the sum is 1.

[0154] Calculate the weighted composite value (the average vibration intensity multiplied by the vibration weight plus the combined temperature and humidity change rate multiplied by the temperature and humidity weight plus the global voltage fluctuation coefficient multiplied by the voltage weight).

[0155] Then, a hyperbolic tangent transformation is performed: a normalized transformation function is applied (the weighted composite value is multiplied by a scaling factor, the hyperbolic tangent value is taken, and then the result is subtracted from 1 and divided by 2). The default scaling factor, 2.0, is obtained from the benchmark library. This function has the characteristic that when the weighted composite value is 0, the output is 0.5 (ideal stability), and as the weighted composite value increases, the output approaches 0 (increasing environmental disturbance). Finally, an exponential normalization is performed: the result of the hyperbolic forward transformation is multiplied by 2 to obtain the global consistency index, which ranges from 0 to 1 (approaching 1 indicates a highly stable environment, approaching 0 indicates a highly fluctuating environment).

[0156] S5: Based on the global consistency index and the dynamic trust threshold, an anti-cheating result is generated according to preset judgment conditions.

[0157] Specifically, step S5 includes the following steps:

[0158] The global consistency index is input into a pre-trained hidden Markov model in the conductive benchmark library, and a state transition probability vector is output.

[0159] In this embodiment, a pre-trained Markov state model (i.e., hidden Markov model) is first used for state analysis. The model architecture contains three key components: the state space defines three hidden states: normal, suspicious, and abnormal; the observation space is divided into three observation symbols: stable, fluctuating, and violent based on the global consistency index; and the parameter matrix includes the state transition matrix, the observation probability matrix, and the initial state distribution. The model pre-training process is completed on a historical dataset: using the global consistency index time series data of the past 6 months (sampling interval 0.5 seconds), the parameters are iteratively optimized using the EM algorithm, specifically:

[0160] The E step calculates the posterior probability, and the M step re-estimates the parameters until the log-likelihood change is less than ±10% Gaussian noise was added during the training process to enhance robustness. The final model was stored in the conduction benchmark library after ensuring that the recall rate of abnormal states was greater than 95% through cross-validation.

[0161] In the state transition probability vector generation phase, the global consistency index generated in real time is input into the pre-trained state analysis model and processed as follows:

[0162] First, the observation symbol conversion is performed, specifically including accurately mapping the current global consistency index into discrete observation symbols according to preset rules: when the index value exceeds 0.8, it is judged as a stable symbol; when it is in the range of 0.4 to 0.8, it is marked as a fluctuating symbol; when it is lower than 0.4, it is marked as a violent symbol.

[0163] Secondly, the time series is constructed, which specifically includes creating an observation sequence consisting of three consecutive weighing cycles: integrating the observation symbols of the current cycle and the two cycles before it to form an analysis unit with temporal continuity (covering the standard 1.5-second time window).

[0164] Finally, the probability dynamic calculation is performed, which specifically includes a forward recursive algorithm based on the hidden Markov model. The algorithm process includes the following stages:

[0165] Initialization phase: Calculate the basic probability by combining the initial state distribution and the first observation value;

[0166] Iterative evolution stage: Dynamic update of probability distribution is achieved through the state transition probability matrix;

[0167] Convergence output stage: Generates a probability distribution vector containing three core indicators: the probability of being in a normal state (the probability of being in a compliant weighing state), the probability of being in a suspicious state (the probability of detecting a minor anomaly but not reaching the cheating threshold), and the probability of being in an abnormal state (the probability of being highly suspected of cheating).

[0168] The state transition probability vector is entropy normalized based on the dynamic credibility threshold to generate a modified state transition sequence.

[0169] In this embodiment, the initial probability distribution is intelligently optimized at this stage.

[0170] First, the information entropy of the state probability is calculated, and the Shannon entropy formula is used to precisely quantify the distribution uncertainty. The entropy peaks (>1.5) when the probability distribution is uniform, and approaches 0 when the distribution is skewed. A dynamic trust threshold is then integrated to generate an entropy scaling factor: the lower the trust threshold, the stronger the scaling becomes exponentially. After introducing an engineering protection mechanism to eliminate division-by-zero risk, nonlinear distribution reconstruction is performed. A power transformation algorithm is used to enhance the weight of abnormal signals. For example, if the original abnormal probability is 0.3, it can be exponentially amplified to above 0.6. Conservation normalization is also used to ensure that the sum of the probabilities remains constant at 1. This enhanced three-dimensional probability vector is output, significantly improving the sensitivity to detecting covert cheating patterns.

[0171] The distance value between the modified state transition sequence and the typical cheating pattern features in the historical cheating behavior pattern feature library stored in the conductive reference library is calculated.

[0172] When the distance value is less than the pattern matching threshold set in the conductive reference library, a cheating alarm result is output; otherwise, an approved weighing result is output.

[0173] This phase completes multi-dimensional feature fusion and hierarchical decision-making. Feature extraction captures the essence of the behavior: The persistent anomaly indicator calculates the proportion of abnormal states in the time series; the state transition frequency counts the number of state transitions per second; and the distribution skewness analyzes the left / right skewness of the probability distribution.

[0174] These three dimensions form complementary feature vectors, among which the frequency feature is sensitive to the "intermittent stepping on the scale" cheating pattern, and the skewness feature can identify "gradual pressure" type cheating.

[0175] The pattern matching phase utilizes 21 pre-stored cheating templates in the conductive benchmark library and employs the Mahalanobis distance algorithm to eliminate inter-feature correlation interference. This algorithm first calculates the inverse of the feature covariance matrix (generated through offline training on a historical cheating dataset) and then performs a normalized spatial mapping, resulting in a 63% reduction in error compared to the Euclidean distance algorithm. The hierarchical decision engine operates in real time: a cheating alarm is triggered immediately when the minimum matching distance falls below the 0.3 threshold; manual review is initiated if the dynamic trust threshold falls below the 0.15 safety threshold; and the weight is output as approved under normal circumstances. For high-entropy scenarios (entropy > 1.5), a cosine similarity backup algorithm is automatically activated: secondary verification is performed through vector space angle calculation, and a fuse mechanism is combined to trigger sensor calibration within 10 milliseconds, forming a closed-loop risk control system.

[0176] Example 2

[0177] The present invention also provides an anti-cheating intelligent weighing system based on multi-source signals, which is used to execute an anti-cheating intelligent weighing method based on multi-source signals. Figure 2 As shown, the system includes:

[0178] The feature calculation module 100 is used to dynamically calculate and generate a geometric out-of-bounds ratio value, a force conduction deviation value, a contour confidence value, and a pressure distribution entropy value based on the pressure distribution time series data, the object's three-dimensional contour and center of gravity position trajectory, and the deformation displacement field data collected in real time and synchronously during the current weighing cycle, in combination with a preset conduction reference library.

[0179] The weight generation module 200 is used to dynamically calculate and generate a first weight representing volume variation, a second weight representing force conduction, and a third weight representing delay sensitivity based on the force conduction deviation value, the contour confidence value, and the pressure distribution entropy value.

[0180] The threshold calculation module 300 is used to dynamically calculate and generate a dynamic credible threshold based on the first weight, the second weight and the third weight, and in combination with the geometric out-of-bounds ratio value and the force conduction deviation value.

[0181] The index calculation module 400 is used to calculate the global consistency index in the current weighing cycle based on the conductive reference library.

[0182] The result generation module 500 is configured to generate an anti-cheating result based on the global consistency index and the dynamic trust threshold according to preset judgment conditions.

[0183] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0184] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, magnetic disk storage, or magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0185] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

Claims

1. An anti-cheating intelligent weighing method based on multi-source signals, characterized in that: The method comprises the following steps: Based on the real-time synchronous collection of pressure distribution time series data, object 3D contour and center of gravity position trajectory, and deformation displacement field data during the current weighing cycle, and combined with the preset conduction reference library, the system dynamically calculates and generates geometric out-of-bounds ratio value, force conduction deviation value, contour confidence value and pressure distribution entropy value; This step specifically includes: Based on the three-dimensional contour of the object, the ratio of the object boundary exceeding the projection area of ​​the scale platform is calculated, and the geometric out-of-bounds ratio value is output; Based on the pressure distribution time series data and the deformation displacement field data, a real-time mapping relationship between the pressure distribution gradient matrix and the deformation displacement characteristic vector is generated, and a preset conduction reference library is matched to output a force conduction deviation value; Calculating a volume variation coefficient of a point cloud of the three-dimensional contour of the object in three-dimensional space and an edge fuzziness in an image domain, and generating a contour confidence value by combining the volume variation coefficient and the edge fuzziness; Based on the pressure distribution time series data, the Shannon entropy of the pressure distribution is calculated according to the time window division strategy preset in the conductive reference library as the pressure distribution entropy value; Based on the force conduction deviation value, the profile confidence value, and the pressure distribution entropy value, dynamically calculating and generating a first weight representing volume variation, a second weight representing force conduction, and a third weight representing delay sensitivity; This step specifically includes: Based on the force conduction deviation value and the contour confidence value, dynamically generating a first weight representing volume variation through the Sigmoid function transformation rule and conduction attenuation constraint strategy configured in the conduction benchmark library; Combining the pressure distribution entropy value with the deviation attenuation parameter output by the conduction reference library to construct a joint gain factor, driving the generation of a second weight representing force conduction; Using the weight compensation rule defined in the conduction reference library, the first weight is used as a feedback item, dynamically coupled with the contour confidence value and the force conduction deviation value to generate a third weight representing delay sensitivity; Dynamically calculating and generating a dynamic credible threshold based on the first weight, the second weight, and the third weight, and in combination with the geometric out-of-bounds ratio value and the force conduction deviation value; This step specifically includes: Generate a boundary penalty term based on the first weight and the geometric out-of-bounds ratio value; Generate a conduction compensation term by coupling calculation of the second weight and the force conduction deviation value, combined with the deviation sensitivity coefficient preset in the conduction reference library; Adaptively weighting the real-time dynamically generated modified delay value based on the third weight to generate a delay impact item; Based on the boundary penalty term, the conduction compensation term, and the delay impact term, a dynamic trustworthy threshold is generated, wherein the calculation formula of the dynamic trustworthy threshold is: T = min (0.95, 1-(B + C + D)) Where T represents the dynamic trust threshold, B represents the boundary penalty term, C represents the conduction compensation term, and D represents the delay impact term. Calculating a global consistency index within a current weighing cycle based on the conductive reference library; This step specifically includes: Obtain historical operating condition data within the current weighing cycle from the conductive reference library, and calculate the average vibration intensity of the weighing platform, the rate of change of ambient temperature and humidity, and the sensor voltage fluctuation coefficient; Perform weighted calculation based on the vibration weight, temperature and humidity weight, and voltage weight configured in the conductive reference library, and transform and output the weighted calculation result through a hyperbolic tangent function to obtain a global consistency index; Based on the global consistency index and the dynamic trust threshold, generating an anti-cheating result according to preset judgment conditions; This step specifically includes: Inputting the global consistency index into a pre-trained hidden Markov model in the conduction benchmark library, and outputting a state transition probability vector; Normalizing the entropy value of the state transition probability vector based on the dynamic trustworthy threshold to generate a modified state transition sequence; Calculating the distance between the modified state transition sequence and the typical cheating pattern features in the historical cheating behavior pattern feature library stored in the conductive benchmark library; When the distance value is less than the pattern matching threshold set in the conductive reference library, a cheating alarm result is output; otherwise, an approved weighing result is output.

2. An anti-cheating intelligent weighing system based on multi-source signals, used to execute the anti-cheating intelligent weighing method based on multi-source signals according to claim 1, characterized in that: The system comprises: The feature calculation module is used to dynamically calculate and generate the geometric out-of-bounds ratio value, force conduction deviation value, contour confidence value and pressure distribution entropy value based on the pressure distribution time series data, object three-dimensional contour and center of gravity position trajectory, and deformation displacement field data collected in real time during the current weighing cycle, combined with the preset conduction reference library; a weight generation module, configured to dynamically calculate and generate a first weight representing volume variation, a second weight representing force conduction, and a third weight representing delay sensitivity based on the force conduction deviation value, the profile confidence value, and the pressure distribution entropy value; a threshold calculation module, configured to dynamically calculate and generate a dynamic credible threshold based on the first weight, the second weight, and the third weight, in combination with the geometric out-of-bounds ratio value and the force conduction deviation value; An index calculation module, configured to calculate a global consistency index within a current weighing cycle based on the conductive reference library; The result generation module is used to generate an anti-cheating result according to preset judgment conditions based on the global consistency index and the dynamic trust threshold.

Images