Logistics weighing safety control system based on face recognition

Abstract

The invention discloses a logistics weighing safety control system based on face recognition, and relates to the technical field of logistics transportation management, and the system comprises a collection module which is used for collecting images of a vehicle and a driver when the logistics vehicle is weighed, processing the images and then sending the images to a recognition module, and the recognition module is used for recognizing the images and the driver. The recognition module is used for receiving images sent by the acquisition module, recognizing face images and vehicle images from the images, extracting face feature information and vehicle information and sending the face feature information and the vehicle information to the comparison module, and the storage module is used for storing vehicle and face registration information and receiving the vehicle information and the face feature information transmitted by the comparison module; the logistics weighing safety control system based on face recognition has remarkable technical advantages and practical value in the aspects of improving the identity verification strength, the safety prevention and control capability and the management intelligence level in the logistics weighing link.

Description

Technical Field

[0001] The present invention relates to the technical field of logistics and transportation management, and in particular to a logistics weighing safety control system based on face recognition. Background Art

[0002] In modern logistics and transportation, cargo weighing is a critical step in ensuring logistics management, contract execution, and cargo safety. Truck scales are commonly used to weigh cargo carried by transport vehicles, and the results serve as crucial data for logistics operations. Driver identity and business information verification is often required during the entry and exit of freight transport vehicles, loading and unloading, and weighing processes to ensure the authenticity and validity of weighing data. However, current logistics management processes still suffer from certain loopholes. For example, some transport vehicles rely solely on entry and exit registration information when weighing cargo at the weighing site, lacking effective anti-fraud mechanisms. This makes it easy for individuals to cheat by replacing license plates or using vehicle tare weights, resulting in duplicate transport, under-counting of net cargo weights, or even theft, resulting in illegal profit. This behavior not only seriously interferes with logistics companies' ability to accurately control the flow of goods but also harms the interests of cargo owners and carriers. Furthermore, with the continuous expansion of logistics operations, the management of vehicle and driver information in the transportation process has become increasingly complex. When logistics companies switch vehicles, they often need to re-register, verify, and track the new vehicles to ensure the continuity and accuracy of cargo transportation. This not only increases operational complexity but also the workload of data management, making it easy for oversights in information verification and cargo identification to occur, impacting overall logistics efficiency and increasing the risk of errors. Summary of the Invention

[0003] The purpose of the present invention is to provide a logistics weighing safety control system based on face recognition to solve the problems existing in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a logistics weighing safety control system based on face recognition, the system comprising: The acquisition module is used to collect images of the vehicle and the driver when the logistics vehicle is weighed, and send the processed images to the recognition module; The recognition module receives the image sent by the acquisition module, identifies the human face image and the vehicle image therefrom, extracts the facial feature information and the vehicle information, and sends the facial feature information and the vehicle information to the comparison module, including acquiring three consecutive frames of images, calculating the similarity of the facial features between the consecutive frames, obtaining the standard deviation and taking the inverse of the standard deviation as the stability of the facial feature recognition, similarly processing the vehicle image recognition result to obtain the vehicle image recognition stability, calculating the stability ratio of the human face and vehicle recognition, deciding whether to give priority to the human face result or the vehicle recognition result, and combining the information and sending it to the comparison module; A comparison module queries the vehicle and driver face registration information recorded in the storage module. If there is no vehicle and driver face information that is identical to the currently collected information, the current vehicle information and facial feature information are processed and sent to the storage module for storage. If there is a vehicle and driver face information that is identical to the currently collected information, the received vehicle and driver face information is compared with the vehicle and driver face feature information corresponding to the stored vehicle and driver face information, and the comparison result is sent to the judgment module. The judgment module judges the received comparison results and decides whether to allow the vehicle to pass according to the actual situation. If the judgment is not passed, it will be transferred to manual processing and returned to the acquisition module; The storage module is used to store the registration information of the vehicle and the driver's face and receive the vehicle information and facial feature information transmitted by the comparison module.

[0005] Preferably, when the driver swipes his ID card, the acquisition module collects vehicle and personnel images, calls the image enhancement algorithm to estimate the improvement in recognition success rate after image quality processing, and at the same time measures the computing resource consumption increased to improve image quality, and calculates the marginal utility of image quality. If the marginal utility of image quality is greater than 1, advanced image enhancement is performed and sent to the recognition module; otherwise, the original image is directly sent to save resources.

[0006] Preferably, the specific formula for calculating the marginal utility of image quality is: A=B / C; Among them, A represents the marginal utility of image quality, B represents the improvement in recognition success rate after image quality processing, and C represents the increased computing resource consumption for improving image quality.

[0007] Preferably, the specific formula for calculating the stability ratio of face and vehicle recognition is: R = H / D; Among them, R represents the stability ratio of face and vehicle recognition, H represents the inverse of the standard deviation of similarity changes in consecutive frames, and D represents the stability of vehicle image recognition.

[0008] Preferably, the comparison module monitors the total amount of currently processed comparison requests and the number of comparisons completed at the current moment, sets the maximum comparison capacity, calculates the current comparison efficiency growth trend, and if the current comparison efficiency growth trend is less than 0, switches to a lightweight comparison strategy; otherwise, uses complete features for high-precision comparison, and sends the comparison results to the judgment module.

[0009] Preferably, the specific formula for calculating the current comparison efficiency growth trend is: dV / dt=βV(1-MC); Where dV / dt represents the current comparison efficiency growth trend, β represents the comparison efficiency coefficient, V represents the number of comparisons completed at the current moment, M represents the total number of comparison requests currently being processed, and C represents the set maximum comparison capacity.

[0010] Preferably, the judgment module obtains the similarity score returned by the comparison module, queries the driver's historical pass records in the database, calculates the current driver's historical pass success rate, evaluates the image recognizability score based on the current image acquisition time and lighting conditions, uses experience to set weights, and calculates the total judgment score. If the total judgment score is greater than a preset threshold, the vehicle is automatically released; otherwise, the vehicle is denied passage and an alarm is issued.

[0011] Preferably, the specific formula for calculating the total judgment score is: S=w1×E+w2×F+w3×G; Among them, S represents the total judgment score, w1, w2 and w3 represent weights, satisfying w1 + w2 + w3 = 1, E represents the similarity score, F represents the current driver's historical passage success rate, and G represents the image recognizability score evaluated by the current image acquisition time and lighting conditions.

[0012] Preferably, the storage module regularly scans the access frequency of all driver identities in the database every day, calculates the frequency of appearance of each driver identity, and calculates the identity data distribution entropy value as an identity diversity distribution indicator. If the identity data distribution entropy value decreases, the abnormal data cleaning mechanism is enabled; if the identity data distribution entropy value increases, the index depth and cache space are automatically increased.

[0013] Preferably, the specific method for calculating the identity data distribution entropy value includes calculating the relative frequency of occurrence of each identity for all registered independent identities, then multiplying each frequency by the logarithm of the frequency, summing the multiplication results and taking the opposite, and finally obtaining the value as the identity data distribution entropy value.

[0014] It can be seen from the above technical solution that the present invention has the following beneficial effects: This facial recognition-based logistics weighing safety control system integrates facial recognition with vehicle image information comparison to achieve dual verification of driver identity and vehicle identity, effectively avoiding the risk of fraudulent use caused by verification based solely on registration information, significantly improving the credibility of weighing data. By comparing currently collected facial features with existing registration records, it can identify whether the same driver or vehicle frequently performs abnormal weighing operations, preventing illegal activities such as smuggling and undercounting goods by exploiting tare weight discrepancies. During the logistics vehicle replacement process, by automatically identifying the characteristics of new vehicles and their drivers, the system can quickly complete identity verification and data updates without manual input, reducing manual intervention costs and information entry errors. Through a modular collection, recognition, comparison, and storage mechanism, the system can form a complete identity trajectory and weighing record chain, facilitating subsequent tracing and risk analysis, and enhancing enterprises' control over the flow of goods. Through an automatic information comparison and judgment mechanism, the burden of manual verification is significantly reduced, human judgment omissions are avoided, and the level of logistics operation automation and overall operational efficiency is improved. The system of the present invention has significant technical advantages and practical value in improving the identity verification strength, security control capabilities, and intelligent management level of the logistics weighing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a connection diagram of the system modules of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] like Figure 1 As shown, the present invention provides a technical solution: a logistics weighing safety control system based on face recognition, the system comprising: The acquisition module is used to collect images of the vehicle and the driver when the logistics vehicle is weighed, and send the processed images to the recognition module; The recognition module receives the image sent by the acquisition module, identifies the human face image and the vehicle image therefrom, extracts the facial feature information and the vehicle information, and sends the facial feature information and the vehicle information to the comparison module, including acquiring three consecutive frames of images, calculating the similarity of the facial features between the consecutive frames, obtaining the standard deviation and taking the inverse of the standard deviation as the stability of the facial feature recognition, similarly processing the vehicle image recognition result to obtain the vehicle image recognition stability, calculating the stability ratio of the human face and vehicle recognition, deciding whether to give priority to the human face result or the vehicle recognition result, and combining the information and sending it to the comparison module; A comparison module queries the vehicle and driver face registration information recorded in the storage module. If there is no vehicle and driver face information that is identical to the currently collected information, the current vehicle information and facial feature information are processed and sent to the storage module for storage. If there is a vehicle and driver face information that is identical to the currently collected information, the received vehicle and driver face information is compared with the vehicle and driver face feature information corresponding to the stored vehicle and driver face information, and the comparison result is sent to the judgment module. The judgment module judges the received comparison results and decides whether to allow the vehicle to pass according to the actual situation. If the judgment is not passed, it will be transferred to manual processing and returned to the acquisition module; The storage module is used to store the registration information of the vehicle and the driver's face and receive the vehicle information and facial feature information transmitted by the comparison module.

[0018] This system utilizes a modular design to automatically identify and verify the identities of vehicles and drivers during logistics weighing. When a logistics vehicle is weighed, the acquisition module activates and simultaneously captures images of the vehicle's front and interior, capturing image data containing the vehicle's license plate and the driver's face. The acquisition module then preprocesses the raw images, including image denoising, edge enhancement, and contrast adjustment, to improve subsequent recognition accuracy.

[0019] The recognition module then analyzes the processed image, performing face detection and license plate location based on deep learning algorithms such as convolutional neural networks. Keypoint location algorithms extract the facial features and the character contours of the license plate, calculating and extracting the corresponding feature vectors. To improve recognition stability, the system continuously captures three frames of images and performs feature extraction on each frame. The cosine similarity or Euclidean distance between the feature vectors of adjacent frames is then calculated to produce a set of similarity values. The system further calculates the standard deviation of this set of similarity values ​​and takes its reciprocal as an indicator of recognition stability. Higher recognition stability indicates greater stability of the target within the image sequence.

[0020] The system then compares the stability of facial and vehicle recognition, using a set threshold or weighted algorithm to determine which recognition result should be used as the primary identification basis. The selected identification information (facial or license plate features) is sent along with auxiliary information from the other side to the comparison module. The comparison module retrieves historical registration information from the storage module and constructs a feature index library. It then uses a feature vector comparison algorithm to match the received image information. If no matching record exists, the current image features are automatically bound to the ID card information to generate a new file record. If a matching record exists, a feature comparison is performed to confirm the identity of the driver and vehicle.

[0021] The comparison results are sent to the decision module, which uses fuzzy logic or a rule-based decision tree approach to determine whether the vehicle should be allowed to pass based on factors such as the comparison similarity score, recognition stability, and past recognition history. If the decision is "fail," the system marks the event as an exception and automatically transmits it back to the acquisition module, prompting a manual review to prevent fraudulent use or substitution.

[0022] This invention improves the security and accuracy of driver and vehicle identity verification during logistics weighing. A multi-frame image recognition stability mechanism enhances the system's robustness in complex environments. A dynamic comparison strategy optimizes the priority of face or vehicle recognition, further improving recognition efficiency. The system's automatic comparison and decision-making mechanism significantly reduces the frequency of manual intervention and improves weighing efficiency. Furthermore, the system supports the historical storage and reuse of vehicle information, enhancing its information management capabilities during long-term operation and demonstrating excellent scalability and application prospects.

[0023] When a logistics vehicle is being weighed, the acquisition module collects images of the vehicle and personnel, calls the image enhancement algorithm to estimate the improvement in recognition success rate after image quality processing, and simultaneously measures the computing resource consumption increased to improve image quality and calculates the marginal utility of image quality. If the marginal utility of image quality is greater than 1, advanced image enhancement is performed and sent to the recognition module; otherwise, the original image is directly sent to save resources.

[0024] When a vehicle enters the weighing area, the acquisition module is triggered and begins capturing on-site images. Using its HD camera, the acquisition module simultaneously captures images of the vehicle's exterior and the driver's face. It then uses a pre-defined image enhancement algorithm (such as Retinex, CLAHE, or a deep neural network-based image super-resolution reconstruction model) to assess the quality of the captured images.

[0025] First, the system comprehensively evaluates the captured image based on metrics such as image clarity, brightness balance, and edge detail preservation. Combined with historical image recognition data, it estimates the potential improvement in recognition success rate that would result from this image quality improvement. Next, the system uses a performance analysis module to assess the computing resources required to execute the image enhancement algorithm, including CPU / GPU time usage, memory consumption, and I / O bandwidth usage, quantifying its cost.

[0026] The system uses the improvement in recognition success rate as a benefit indicator and resource consumption as a cost indicator to calculate the marginal utility of the image quality improvement (i.e., the ratio of recognition accuracy improvement per unit resource cost). If the marginal utility value is greater than 1, the system determines that the current resource investment has a positive benefit, performs image optimization processing at a higher level of enhancement, and transmits the enhanced image to the recognition module. If the marginal utility is less than or equal to 1, the image optimization input-output ratio is considered low, and the original image is sent directly to the recognition module, thereby saving computing resources and improving the overall processing efficiency of the system. This mechanism essentially establishes a real-time dynamic balance between image processing benefits and resource consumption, ensuring that recognition accuracy is maintained without wasting resources. It is particularly suitable for edge computing or resource-constrained scenarios.

[0027] This implementation effectively improves the intelligence and resource utilization efficiency of the image acquisition stage. By evaluating the marginal utility of image quality, the system can intelligently decide whether to perform image enhancement operations, thereby avoiding unnecessary image processing when the image quality is high, reducing system load, and improving response speed; while when the image quality is low but recognition is critical, it prioritizes recognition accuracy, improving system reliability and adaptability. This mechanism is highly adaptable and is particularly suitable for actual application scenarios where image quality fluctuates frequently or device computing resources are limited. It helps the system operate stably in the long term and provides a basic guarantee for the stability of subsequent recognition accuracy.

[0028] The specific formula for calculating the marginal utility of image quality is: A = B / C; Among them, A represents the marginal utility of image quality, B represents the improvement in recognition success rate after image quality processing, and C represents the increased computing resource consumption for improving image quality.

[0029] During the image acquisition and preprocessing phase, the system evaluates the image enhancement effect and computational resource consumption of the acquired images, and then uses a marginal utility model to make decisions. The marginal utility of image quality is calculated using the formula A = B / C, where B is the improvement in recognition success rate after the enhancement algorithm is applied compared to the original image. The system can estimate this value by comparing the difference in recognition confidence between the original and enhanced images under existing models or by comparing historical accuracy improvements. C is the computational resource cost required to execute the enhancement algorithm under current hardware conditions, including but not limited to image processing time, CPU / GPU resource percentage, and memory usage. This is expressed as a unified resource consumption metric using a weighted aggregation method. Using this formula, the system calculates the cost-effectiveness of the image enhancement operation, A, representing the improvement in recognition accuracy per unit of computational resource consumption. If A is greater than 1, the enhancement operation has a positive benefit under the current conditions, and the system performs the enhancement. If A is less than or equal to 1, the resource input-output ratio is considered poor, and the system skips the enhancement process and uses the original image directly, improving overall resource efficiency and response speed.

[0030] By introducing a marginal utility calculation model, the system enables quantitative analysis and intelligent decision-making for image processing strategies, avoiding the indiscriminate consumption of resources. While ensuring effective recognition, it effectively reduces the computational burden, improves the flexibility of the image processing stage, and enhances the stability of system operation. The use of a clear formula modeling approach provides a quantitative foundation for subsequent system optimization and resource scheduling algorithm integration, enabling real-time control over the benefits and costs of image processing, and enhancing the transparency and feasibility of algorithmic strategies.

[0031] The specific formula for calculating the stability ratio of face and vehicle recognition is: R = H / D; Among them, R represents the stability ratio of face and vehicle recognition, H represents the inverse of the standard deviation of similarity changes in consecutive frames, and D represents the stability of vehicle image recognition.

[0032] The system incorporates the calculation of a recognition stability ratio, R, in the recognition module to dynamically assess whether face recognition or vehicle image recognition results should be prioritized in recognition tasks. Specifically, after receiving three consecutive frames of image data from the acquisition module, the system extracts features from both the face and vehicle images and calculates the image feature similarity between adjacent frames. The system calculates the standard deviation of the face image similarity results, denoted as S, and then takes the inverse of S, 1 / S, to obtain the face recognition stability H, which measures the consistency of facial features across consecutive frames. The vehicle image processing process follows a similar process, calculating the vehicle recognition stability D. The system then uses the formula R = H / D to calculate the face-to-vehicle recognition stability ratio, R. This value represents the relative stability of face recognition compared to vehicle recognition and guides the selection of the priority decision basis. When the R value is greater than a set threshold (e.g., 1.0), face recognition is more stable than vehicle recognition, and the system will use face recognition as the primary recognition basis, supplemented by vehicle image information. Otherwise, the vehicle recognition result will be the primary basis for subsequent comparisons. This dynamic control mechanism based on R value significantly improves the adaptive ability of the recognition module, ensuring that reasonable recognition judgments can be made based on data stability in different scenarios.

[0033] Introducing the recognition stability ratio R as a decision parameter helps the system make objective priority determinations when processing multi-source recognition information. This method avoids misjudgments caused by differences in image quality or changes in the recognition environment, improving the robustness and accuracy of the face and vehicle recognition processes. By quantifying recognition stability using the inverse standard deviation and further constructing the ratio R, the system develops a decision logic that can adapt in real time without external annotation, demonstrating high engineering practicality and algorithmic scalability. Furthermore, this method facilitates algorithm optimization and hardware resource allocation adjustments, providing a quantitative basis for intelligent upgrades to recognition strategies.

[0034] The comparison module monitors the total number of currently processed comparison requests and the number of comparisons completed at the current moment, sets the maximum comparison capacity, and calculates the current comparison efficiency growth trend. If the current comparison efficiency growth trend is less than 0, it switches to a lightweight comparison strategy. Otherwise, it uses complete features for high-precision comparison and sends the comparison results to the judgment module.

[0035] The system incorporates a dynamic adjustment mechanism in the comparison module, designed to adjust the comparison strategy based on real-time system load and processing performance, thereby optimizing the overall recognition process's responsiveness. During operation, the comparison module monitors the total number of queued comparison requests and the number of completed comparison operations per unit time. Based on the time series, it calculates the comparison processing rate and derives its first-order derivative, representing the comparison efficiency growth trend. The system presets a maximum comparison capacity, representing the optimal parallel processing capability under the current hardware environment. When the comparison efficiency growth trend is positive, it indicates that system processing capacity is still improving or maintaining a stable state. The module continues to use complete features (such as 128-dimensional facial vectors and deep features in vehicle images) for high-precision comparison to ensure recognition accuracy. When the growth trend is less than 0, indicating a decrease in processing efficiency per unit time, the system may be experiencing a performance bottleneck or task backlog. The module immediately switches to a lightweight comparison strategy, such as using low-dimensional feature extraction, downsampling images, and feature template matching to reduce the computational load and alleviate system pressure. After the comparison is completed, regardless of the strategy used, the results are transmitted in real time to the decision module for subsequent passage determination. Through this mechanism, the comparison module can achieve flexible adjustment between ensuring accuracy and efficiency, and improve the overall processing stability and concurrent processing capabilities of the system.

[0036] Introducing the trend of matching efficiency growth as a basis for policy switching enables the system to adaptively adjust computational loads, effectively mitigating the impact of traffic bursts or hardware performance fluctuations on the recognition process. Automatically enabling lightweight policies during high loads significantly reduces response time and prevents system congestion caused by matching task accumulation. High-precision features are used to ensure recognition accuracy when the load is manageable, thus achieving a balance between resource utilization and security control. Furthermore, this policy adjustment mechanism is concise, quantitative, and real-time, making it easy to integrate into various heterogeneous systems and ensuring good engineering feasibility.

[0037] The specific formula for calculating the current comparison efficiency growth trend is: dV / dt=βV(1-MC); Where dV / dt represents the current comparison efficiency growth trend, β represents the comparison efficiency coefficient, V represents the number of comparisons completed at the current moment, M represents the total number of comparison requests currently being processed, and C represents the set maximum comparison capacity.

[0038] To dynamically adjust the comparison strategy, the comparison module uses the formula dV / dt = βV(1-MC) to calculate the current comparison efficiency growth trend. dV / dt represents the rate of change of the comparison efficiency per unit time, reflecting the impact of the system's current load on the comparison capability. V is the number of successfully completed comparisons per unit time at the current moment, a direct indicator of current processing capacity. M is the total number of comparison requests currently awaiting processing, reflecting the system's current load intensity. C is the maximum comparison capacity, representing the system's processing limit under stable operation and typically determined by the system's hardware capabilities and concurrent processing strategy. β is the comparison efficiency coefficient, a positive real number used to adjust system sensitivity and optimized based on historical operating data. This model uses a structure similar to the logistic growth model. When M×C approaches 1, the bracketed term approaches 0, and the rate of change in comparison efficiency slows, indicating that the system is saturated. If M×C is less than 1, the system still has spare capacity, dV / dt is positive, and the comparison efficiency is still increasing. If M×C is greater than 1, dV / dt is negative, indicating that the system load exceeds the upper capacity limit and processing efficiency will decrease. The comparison module dynamically determines whether to switch to a lightweight comparison strategy based on the positive or negative state of dV / dt, thereby ensuring response efficiency while controlling system load.

[0039] By introducing a clear mathematical model, the system can quantify the changing trends in matching efficiency in real time, providing mathematical basis and automated logic for matching strategy switching, rather than relying on manually set empirical values ​​or single threshold judgments. This model provides early warning of system bottlenecks, effectively preventing task accumulation and decreased processing efficiency. Based on the model's results, the system adaptively optimizes the recognition strategy, helping to achieve a balance between processing stability, efficiency, and accuracy. Its concise form and adjustable parameters provide excellent generalization capabilities and engineering deployment value.

[0040] The judgment module obtains the similarity score returned by the comparison module, queries the driver's historical pass records in the database, calculates the current driver's historical pass success rate, evaluates the image recognizability score based on the current image acquisition time and lighting conditions, uses experience to set weights, and calculates the total judgment score. If the total judgment score is greater than the preset threshold, the vehicle is automatically released; otherwise, the vehicle is denied passage and an alarm is issued.

[0041] The system incorporates a multi-factor fusion decision-making mechanism into the decision module to enhance the overall accuracy and adaptability of vehicle and driver identification. Once the comparison module compares the image with the database information, it sends the similarity score to the decision module, which then triggers a multi-dimensional scoring process.

[0042] First, the decision module accesses the historical pass records stored in the storage module to query the current driver's pass history, including whether previous recognition attempts were successful and whether there were any anomalies. Based on the pass records and the total number of attempts, the system calculates the historical pass success rate, which serves as a key parameter for determining the driver's identity credibility. The system then evaluates the current image's recognizability. This evaluation is based on the time period of image acquisition (e.g., daytime, nighttime) and the ambient lighting information collected simultaneously by the system (e.g., brightness and contrast distribution). This score is quantitatively generated to represent the degree to which image input quality influences recognition results.

[0043] The decision module sets weighting parameters for each factor based on experience or training data. For example, it assigns different weights to the similarity score, historical pass success rate, and image recognizability score, and calculates the total decision score through a weighted summation. If the score is above the set safety threshold, the system automatically determines that the identification has passed, triggering a release signal and allowing the control device to allow the vehicle to pass. If the score is below the threshold, the system marks the current event as an anomaly, denies the vehicle passage, and simultaneously sends an alarm signal to the backend system, prompting manual intervention or review. This decision-making mechanism combines static information with dynamic environmental factors to construct a more comprehensive identity determination logic.

[0044] The present invention significantly improves the intelligence of the judgment module and the credibility of the recognition results by introducing historical behavior data and image environment adaptability analysis. The system no longer relies solely on a single similarity value to determine access rights, but instead integrates multi-source data for comprehensive analysis, effectively reducing the false recognition rate and missed recognition rate, especially in complex scenarios such as insufficient lighting and blurred images. It can still maintain stable judgment capabilities. By setting experience weights and total judgment score thresholds, the system can flexibly adapt to different scenarios and user security strategies, improving overall operational efficiency and security. In addition, the alarm mechanism realizes real-time feedback of abnormal events, providing data support for system management and event tracing.

[0045] The specific formula for calculating the total judgment score is: S = w1×E+w2×F+w3×G; Among them, S represents the total judgment score, w1, w2 and w3 represent weights, satisfying w1 + w2 + w3 = 1, E represents the similarity score, F represents the current driver's historical passage success rate, and G represents the image recognizability score evaluated by the current image acquisition time and lighting conditions.

[0046] When the judgment module performs identity determination, the system integrates three types of information: the current recognition similarity E, the driver's historical pass success rate F, and the image recognizability score G. Using a weighted model, it calculates a total judgment score S to determine whether to release the driver. The core calculation model of this module is the formula S = w1 × E + w2 × F + w3 × G, where w1, w2, and w3 are preset weight coefficients, the sum of which is always equal to 1, representing the importance of each factor in the overall judgment. The similarity score E is generated by the comparison module based on the match between the current image and the identity template in the database. It is typically calculated based on the distance between feature vectors and a confidence value ranging from 0 to 1. The historical pass success rate F is calculated by counting the percentage of successful recognitions in the driver's past pass records, reflecting the confidence level of the driver's identity. The image recognizability score G is calculated based on the current image acquisition time (e.g., daytime or nighttime) as well as simultaneously acquired parameters such as light intensity and image contrast, reflecting the degree to which the environment affects image recognition. The decision module sets reasonable weights (e.g., w1=0.5, w2=0.3, and w3=0.2) to combine scores and compare them with a preset threshold. When the S value exceeds the threshold, the system deems the identity trustworthy and automatically triggers a pass signal. If the S value falls below the threshold, an alarm process is triggered and manual processing is initiated. This model implements a clearly structured and responsive access control logic.

[0047] By introducing a weighted scoring model, this decision mechanism avoids over-reliance on a single factor (such as similarity), improving the system's reliability in complex and non-ideal environments. The formula structure offers excellent adjustability, and weights can be adjusted based on historical data and actual scenarios, flexibly adapting to different recognition environments and security strategies. The system enables refined management of decision strategies by controlling weight coefficients and thresholds, effectively reducing false positives and missed positives, and improving the accuracy and rationality of decisions. This makes it particularly suitable for logistics weighing scenarios requiring strong security.

[0048] The storage module regularly scans the access frequency of all driver identities in the database every day, calculates the frequency of each driver identity, and calculates the entropy value of the identity data distribution as an indicator of identity diversity distribution. If the entropy value of the identity data distribution decreases, the abnormal data cleaning mechanism is enabled. If the entropy value of the identity data distribution increases, the index depth and cache space are automatically increased.

[0049] This implementation dynamically monitors and optimizes the distribution of identity data in the storage module by introducing an information entropy model. The storage module sets a daily scheduled task to scan the call frequency of all driver identity records in the database. By counting the number of times each identity is accessed within a certain time window and normalizing the access frequencies of all identities, the probability distribution p(i) for each identity is calculated. The system further uses the information entropy formula H = -Σp(i)log(p(i)) to calculate the overall identity distribution entropy value H, which is used to measure the diversity and distribution balance of identity calls. This entropy value, an indicator of the dynamic structural state of identity data, has the following application logic: When the entropy value continues to decline, it indicates that certain identities are appearing frequently, potentially indicating a concentration of abnormal access, such as malicious facial recognition, identity theft, or operational errors. At this point, the system automatically triggers an abnormal data cleanup mechanism, marking, isolating, or transferring low-frequency or abnormal behavior records to a manual review process to ensure the health and security of the database data structure. When the entropy value rises, it indicates that the distribution of identity access is becoming more uniform, and the system identifies data growth as occurring. To ensure access performance, it automatically increases database index depth (such as enabling multidimensional hash indexes and prefix tree indexes) and cache space allocation, thereby improving data retrieval speed and concurrent processing capabilities. This mechanism, driven by information entropy, enables self-awareness of data status and dynamic adjustment of resource allocation, ensuring that the database maintains high performance and data structure stability even under high concurrent access.

[0050] By introducing the entropy of identity data distribution as a core metric for database management, the system achieves quantitative perception of data structure status and provides a basis for decision-making. Compared to traditional static cleaning or mean-based judgment mechanisms, the information entropy model can reflect more complex distribution characteristics and abnormal concentration phenomena, providing a more sensitive and reliable response. The system proactively cleans potentially abnormal data when identity access is concentrated, avoiding redundant expansion and database pollution. It also automatically adjusts caching and indexing strategies when identity access is balanced and expanded, improving response speed and scalability. This overall mechanism enhances the system's data security, structural controllability, and operational efficiency.

[0051] The specific method for calculating the identity data distribution entropy value includes calculating the relative frequency of each identity for all registered independent identities, then multiplying each frequency by the logarithm of the frequency, summing the multiplication results and taking the opposite, and the final value obtained is used as the identity data distribution entropy value.

[0052] During the data monitoring task performed daily by the storage module, the system scans all registered individual driver IDs in the database and counts the number of accesses for each ID within a specified time window (e.g., 24 hours). The system then divides the number of accesses for each ID by the total number of accesses to calculate its relative frequency, p(i), to form a complete identity access frequency distribution. The system uses the information entropy formula H = -Σp(i)log(p(i)) to calculate the identity data distribution entropy. Here, p(i) is the relative frequency of the i-th identity. The log function typically uses base 2 (i.e., units of bits) or the natural logarithm ln, depending on the system implementation. For each identity, the system calculates the value of p(i) × log(p(i)), sums all the results, and takes the inverse, which is the identity data distribution entropy H for that time period.

[0053] A higher entropy value indicates a more even distribution of identity access. Conversely, a lower entropy value indicates that some identities appear more frequently and are more concentrated. The system uses this value to determine the diversity of the identity data structure and drive cleanup or expansion mechanisms. This entropy value provides a scientific calculation basis for evaluating the health of the database structure and dynamic response.

[0054] This calculation method has clear mathematical principles, simple operation, and good feasibility and scalability. By counting relative frequencies and calculating the sum of their logarithms, the system not only monitors the distribution of identity access behavior but also quantifies its structural diversity trends, providing a scientific basis for subsequent system cleanup, optimization, index adjustment, and other operations. Compared with traditional mean or count threshold judgments, information entropy is more sensitive and accurate in identifying distributed anomalies, making it particularly suitable for handling large-scale, dynamically changing identity data scenarios.

[0055] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A logistics weighing safety control system based on face recognition, characterized in that: The system comprises: The acquisition module is used to collect images of the vehicle and the driver when the logistics vehicle is weighed, and send the processed images to the recognition module; The recognition module receives the image sent by the acquisition module, identifies the human face image and the vehicle image therefrom, extracts the facial feature information and the vehicle information, and sends the facial feature information and the vehicle information to the comparison module, including acquiring three consecutive frames of images, calculating the similarity of the facial features between the consecutive frames, obtaining the standard deviation and taking the inverse of the standard deviation as the stability of the facial feature recognition, similarly processing the vehicle image recognition result to obtain the vehicle image recognition stability, calculating the stability ratio of the human face and vehicle recognition, deciding whether to give priority to the human face result or the vehicle recognition result, and combining the information and sending it to the comparison module; A comparison module queries the vehicle and driver face registration information recorded in the storage module. If there is no vehicle and driver face information that is identical to the currently collected information, the current vehicle information and facial feature information are processed and sent to the storage module for storage. If there is a vehicle and driver face information that is identical to the currently collected information, the received vehicle and driver face information is compared with the vehicle and driver face feature information corresponding to the stored vehicle and driver face information, and the comparison result is sent to the judgment module. The judgment module judges the received comparison results and decides whether to allow the vehicle to pass according to the actual situation. If the judgment is not passed, it will be transferred to manual processing and returned to the acquisition module; The storage module is used to store the registration information of the vehicle and the driver's face and receive the vehicle information and facial feature information transmitted by the comparison module.

2. The logistics weighing safety control system based on face recognition according to claim 1 is characterized in that: When the logistics vehicle is being weighed, the acquisition module collects images of the vehicle and personnel, calls the image enhancement algorithm to estimate the improvement in recognition success rate after image quality processing, and simultaneously measures the computing resource consumption increased to improve image quality, and calculates the marginal utility of image quality. If the marginal utility of image quality is greater than 1, advanced image enhancement is performed and sent to the recognition module; otherwise, the original image is directly sent to save resources.

3. The logistics weighing safety control system based on face recognition according to claim 2 is characterized in that: The specific formula for calculating the marginal utility of image quality is: A=B / C; Among them, A represents the marginal utility of image quality, B represents the improvement in recognition success rate after image quality processing, and C represents the increased computing resource consumption for improving image quality.

4. The logistics weighing safety control system based on face recognition according to claim 1 is characterized in that: The specific formula for calculating the stability ratio of face and vehicle recognition is: R=H / D; Among them, R represents the stability ratio of face and vehicle recognition, H represents the inverse of the standard deviation of similarity changes in consecutive frames, and D represents the stability of vehicle image recognition.

5. The logistics weighing safety control system based on face recognition according to claim 1 is characterized in that: The comparison module monitors the total amount of currently processed comparison requests and the number of comparisons completed at the current moment, sets the maximum comparison capacity, calculates the current comparison efficiency growth trend, and switches to a lightweight comparison strategy if the current comparison efficiency growth trend is less than 0. Otherwise, it uses complete features for high-precision comparison and sends the comparison results to the judgment module.

6. The logistics weighing safety control system based on face recognition according to claim 5, characterized in that: The specific formula for calculating the current comparison efficiency growth trend is: dV / dt=βV(1-MC); Where dV / dt represents the current comparison efficiency growth trend, β represents the comparison efficiency coefficient, V represents the number of comparisons completed at the current moment, M represents the total number of comparison requests currently being processed, and C represents the set maximum comparison capacity.

7. The logistics weighing safety control system based on face recognition according to claim 1 is characterized in that: The judgment module obtains the similarity score returned by the comparison module, queries the driver's historical pass records in the database, calculates the current driver's historical pass success rate, evaluates the image recognizability score based on the current image acquisition time and lighting conditions, uses experience to set weights, and calculates the total judgment score. If the total judgment score is greater than a preset threshold, the driver is automatically allowed to pass; otherwise, the driver is denied passage and an alarm is issued.

8. The logistics weighing safety control system based on face recognition according to claim 7, characterized in that: The specific formula for calculating the total judgment score is: S = w1×E+w2×F+w3×G; Among them, S represents the total judgment score, w1, w2 and w3 represent weights, satisfying w1 + w2 + w3 = 1, E represents the similarity score, F represents the current driver's historical passage success rate, and G represents the image recognizability score evaluated by the current image acquisition time and lighting conditions.

9. The logistics weighing safety control system based on face recognition according to claim 1, characterized in that: The storage module regularly scans the access frequency of all driver identities in the database every day, calculates the frequency of appearance of each driver identity, and calculates the identity data distribution entropy value as an indicator of identity diversity distribution. If the identity data distribution entropy value decreases, the abnormal data cleaning mechanism is enabled. If the identity data distribution entropy value increases, the index depth and cache space are automatically increased.

10. The logistics weighing safety control system based on face recognition according to claim 9, characterized in that: The specific method for calculating the identity data distribution entropy value includes calculating the relative frequency of each identity for all registered independent identities, then multiplying each frequency by the logarithm of the frequency, summing the multiplication results and taking the opposite, and finally obtaining the value as the identity data distribution entropy value.

Images