Intelligent reagent management method and system based on RFID identification

Abstract

The invention provides an intelligent reagent management method and system based on RFID identification, and the method comprises the steps: synchronously collecting an RFID tag number of a reagent and a first image of a reagent bottle body when the reagent is put in storage, extracting a first image feature vector based on the first image, and generating a first identity signature; when the reagents are checked periodically, extracting a second image feature vector, and generating a second identity signature in combination with the current RFID tag number; calculating a structural difference degree between the second image feature vector and the first image feature vector; on the basis of the structural difference degree, a comparison result of the second identity signature and the first identity signature and a historical recognition record of the reagent, constructing a comprehensive anomaly degree model to calculate a comprehensive anomaly degree; and when the comprehensive abnormality exceeds a preset threshold value, automatically triggering a preset management action. The method is suitable for laboratory reagent management scenes with extremely high requirements on safety and accuracy.

Description

Technical Field

[0001] This invention belongs to the field of intelligent reagent management, and particularly relates to an intelligent reagent management method and system based on RFID identification. Background Technology

[0002] As laboratory management demands increasing precision and automation, the comprehensive management of reagents and consumables throughout the entire process—from warehousing and storage to use and dispensing—is becoming an essential component of laboratory quality control systems. In research, medical testing, biopharmaceuticals, and disease prevention, a large number of reagents are high-value, highly sensitive items. These reagents are diverse, frequently changed in batches, and subject to strict shelf-life management. Relying solely on manual inventory or single-tag identification methods can easily lead to problems such as missing records, delayed ledgers, and inaccurate inventory status. It can also result in serious consequences such as delays in important experiments and interruptions in diagnostic processes. Radio frequency identification (RFID) technology, due to its advantages of being contactless, high-speed, and supporting multi-tag reading, is increasingly being used for laboratory reagent management. However, it still has significant limitations in practical scenarios. For example, when reagent bottles are highly similar, densely packed, or partially obscured, RFID signals are easily interfered with by spatial reflections and metallic environments, leading to misreading or missed readings of tags. Furthermore, RFID tags themselves do not carry physical characteristic information related to the reagent bottle. Once a tag is transferred or counterfeited, the system cannot determine whether the current tag corresponds to the original reagent, thus posing a risk of identity information tampering. Furthermore, reagents in laboratories are used frequently and inventory cycles are short, resulting in fluctuations in the physical state of reagent bottles over time. Traditional identification methods only provide static, one-time readings, failing to continuously monitor changes in reagent identity or abnormal behavior, and unable to respond promptly to suspicious operations. Therefore, achieving highly secure identity binding, robust periodic identification, and automated anomaly management based on identification behavior for reagents without altering existing experimental procedures has become a core challenge in the current intelligentization of reagent management. Summary of the Invention

[0003] The purpose of this invention is to propose an intelligent reagent management method and system based on RFID identification to solve the above-mentioned problems.

[0004] To achieve the above objectives, a first aspect of the present invention provides an intelligent reagent management method based on RFID identification, the method comprising the following steps: When reagents are put into storage, the RFID tag number of the reagent and the first image of the reagent bottle are collected simultaneously. The first image feature vector is extracted based on the first image and combined with the RFID tag number to generate a unique and tamper-proof first identity signature. During periodic reagent inventory, the current RFID tag number of the reagent and a second image of the current reagent bottle are collected. A second image feature vector is extracted based on the second image, and a second identity signature is generated by combining the current RFID tag number. The structural difference between the second image feature vector and the first image feature vector is calculated. Based on the structural difference degree, the comparison result between the second identity signature and the first identity signature, and the historical identification record of the reagent, a comprehensive anomaly degree model is constructed to calculate the comprehensive anomaly degree; when the comprehensive anomaly degree exceeds a preset threshold, a preset management action is automatically triggered.

[0005] Furthermore, the extraction of the first image feature vector based on the first image is encoded using a shallow convolutional network with a fixed structure, and the output vector is processed by global average pooling and normalization to ensure numerical stability and to prevent interference from exposure or slight rotation. The shallow convolutional network consists of five convolutional layers and two max-pooling layers, with a kernel size of [missing value]. With a step size of 1, no training is required, and the parameters are fixed. It is used to capture local shape information and edge structure.

[0006] Furthermore, the content of the first image feature vector is the local contour descriptor of the image.

[0007] Furthermore, the first identity signature is combined by mapping the concatenated data into a fixed-length binary signature string using a standard cryptographic hash function, which is used for unique identity comparison during subsequent reagent identification.

[0008] Furthermore, the feature vector based on the second image is extracted by an image encoder. The image encoder has a fixed structure, consisting of five convolutional layers, and does not include a training process; it directly outputs a 64-dimensional real vector.

[0009] Further, the calculation of the structural difference between the second image feature vector and the first image feature vector specifically involves: A weighted correction term is introduced into the Euclidean distance between the second image feature vector and the first image feature vector to obtain the structural difference degree.

[0010] Furthermore, the weighted correction term is the absolute value of the overall image intensity shift, used to detect structural imbalances caused by occlusion or replacement of the label region.

[0011] Furthermore, based on the structural difference degree, the comparison result between the second identity signature and the first identity signature, and the historical identification records of the reagent, a comprehensive anomaly model is constructed to calculate the comprehensive anomaly degree, specifically as follows: Get the t-th and t-th times The image structure difference degree of the second identification is combined with the time-based decay coefficient and the standard deviation of the difference degree sequence to calculate the degree of image structure fluctuation; wherein, the standard deviation of the difference degree sequence is used to capture whether the overall identification offset is stable; Get the current identity mismatch frequency; The comprehensive anomaly degree is calculated by combining the current identity mismatch frequency, the degree of image structure fluctuation, and the historical maximum structural difference of the image structure difference in the t-th recognition.

[0012] Furthermore, the current identity mismatch frequency is obtained as follows: The second identity signature is precisely compared with the first identity signature; if they do not match, it is marked as a mismatch. The cumulative number of mismatches within each cycle is The mismatch ratio is .

[0013] A second aspect of the invention provides an intelligent reagent management system based on RFID identification, the system comprising: The warehousing binding module is used to simultaneously collect the RFID tag number of the reagent and the first image of the reagent bottle when the reagent is put into storage, extract the first image feature vector based on the first image, and generate a unique and tamper-proof first identity signature by combining the RFID tag number. The periodic identification and comparison module is used to collect the current RFID tag number of the reagent and the second image of the current reagent bottle during periodic inventory of reagents, extract the second image feature vector based on the second image, and generate a second identity signature by combining the current RFID tag number; and calculate the structural difference between the second image feature vector and the first image feature vector. The anomaly detection and response module is used to construct a comprehensive anomaly model based on the structural difference degree, the comparison result between the second identity signature and the first identity signature, and the historical identification record of the reagent, so as to calculate the comprehensive anomaly degree; when the comprehensive anomaly degree exceeds a preset threshold, a preset management action is automatically triggered.

[0014] The beneficial technical effects of the present invention are at least as follows: This invention addresses the core pain points in the aforementioned reagent management scenarios by constructing a complete technical chain from reagent entry binding and periodic identification to abnormal behavior judgment. This achieves an intelligent management method where reagent identity is unforgeable, the identification process is highly robust, and abnormal behavior is automatically triggered. Upon reagent entry, this invention simultaneously collects the RFID number and the reagent bottle's image structure information. It utilizes a shallow convolutional network to extract image features and combine them with the tag number to generate a unique signature, establishing a joint binding relationship between the tag and the reagent bottle from the source, thus solving the problem of traditional RFID tags being copied or transferred. In the periodic identification stage, this invention proposes a method that combines the difference modeling of entry image features and current image features. Through weighted structural deviation terms and overall offset regularization terms, it robustly quantifies changes in bottle appearance, effectively distinguishing between normal placement changes and potential replacement behavior. Furthermore, this invention utilizes time-series windowing analysis to integrate image difference fluctuations, identity signature mismatches, and extreme deviation responses to construct a comprehensive anomaly model. This model enables automatic identification of long-term or short-term abnormal behavior of reagents and immediately triggers management actions such as freezing reagents, sending alarms, and recording anomalies when the anomaly exceeds a threshold, forming an automated closed loop from identification to decision-making. This solution achieves a dual improvement in identity security and identification reliability compared to existing RFID solutions, enabling highly reliable control of reagents even under complex storage conditions. It is suitable for laboratory reagent management scenarios with extremely high requirements for security and accuracy. Attached Figure Description

[0015] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0016] Figure 1 This is a flowchart of the intelligent reagent management method based on RFID identification according to the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] like Figure 1 As shown in the embodiment of the present invention, an intelligent reagent management method based on RFID identification is provided, the method comprising: S1: When reagents are put into storage, the RFID tag number of the reagent and the first image of the reagent bottle are collected simultaneously. The first image feature vector is extracted based on the first image and combined with the RFID tag number to generate a unique and tamper-proof first identity signature. S2: During periodic reagent inventory, the current RFID tag number of the reagent and the second image of the current reagent bottle are collected. The second image feature vector is extracted based on the second image, and a second identity signature is generated by combining the current RFID tag number. The structural difference between the second image feature vector and the first image feature vector is calculated. S3: Based on the structural difference degree, the comparison result between the second identity signature and the first identity signature, and the historical identification record of the reagent, a comprehensive anomaly degree model is constructed to calculate the comprehensive anomaly degree; when the comprehensive anomaly degree exceeds a preset threshold, a preset management action is automatically triggered.

[0019] Specifically, step S1 is used to construct a unique identifier for each reagent bottle during the reagent warehousing process, ensuring that the uniqueness and legality of the reagent can be accurately determined during subsequent identification, thereby effectively preventing problems such as label tampering or bottle replacement. To meet this requirement, the system jointly processes the reagent's RFID code and image appearance features, generating an immutable identity signature through encryption, which serves as the comparison benchmark in all subsequent identification tasks.

[0020] When reagents are placed at the warehousing and filing station, the identification module automatically activates. This module consists of two parts: a radio frequency identification (RFID) reader and an image acquisition unit. The RFID reader is installed at the bottom of the station and equipped with a directional linearly polarized antenna, limiting its reading range to within 30 centimeters to effectively avoid crosstalk between adjacent tags. The reader operates in the 860–960MHz frequency band, the tags conform to the EPCC1G2 protocol, and the internal storage area has a pre-set 128-bit unique number. After a successful read, the number is assigned as a variable. The data type is an unsigned integer.

[0021] Furthermore, the image acquisition unit is installed above the reagent station, employing a fixed-focal-length industrial camera with a CMOS image sensor and an imaging resolution of [resolution missing]. Pixels. To ensure image quality and feature stability, a ring-shaped LED light source with constant brightness is used at the acquisition station, and a diffuse reflector is used to eliminate ambient light interference. Image acquisition is triggered synchronously with RF reading, and the generated image data is denoted as [image data]. Its data type is a three-channel matrix. The image contains structures such as the bottle body, label (including production date), and bottle cap. In particular, the label attachment area and the geometric outline of the bottle mouth are used as the main basis for subsequent image encoding.

[0022] Furthermore, the image The input image feature extraction module uses a shallow convolutional network with a fixed structure for encoding. This network consists of five convolutional layers and two max-pooling layers, with a kernel size of [missing value]. With a stride of 1, no training is required, and the parameters are fixed. This network is used to capture local shape information and edge structure. The image processed by this network outputs a 64-dimensional real vector, denoted as... Its content is a local contour descriptor of the image. This vector undergoes global average pooling and normalization to ensure numerical stability, unaffected by exposure or minor rotation: ; in, Represents image coding operators. For the input image, This is for the output image features. Instead of using a trained network, a static convolutional extractor is used to ensure controllability and determinism. Taking a real reagent as an example, if the label in the image is slightly misaligned, or there are minor scratches on the bottle cap, these minute differences will be detected. This is reflected in subtle changes in edge morphology, thus enabling the image features to be well distinguished between different bottles.

[0023] Subsequently, the system will compare the read RFID number r with the extracted image features. The concatenation is performed to construct a joint feature vector, which is then input into the hash calculation module to generate the identity signature s: ; in, Using a standard cryptographic hash function (such as SHA-256), the concatenated data is mapped to a fixed-length binary signature string s, used for unique identification during subsequent reagent testing. This function is an irreversible mapping, ensuring that even minor changes to the image or serial number will result in a significant change in the resulting signature, providing strong anti-counterfeiting capabilities.

[0024] For example, after a reagent bottle with serial number 82017496 is placed at the filing station, the system captures an image showing a slight scratch on the bottle cap and a slightly tilted label. The extracted first image feature vector is... After concatenation and hashing, an identity signature is generated. The system writes the signature, along with the reagent number and image features, into the database as a read-only entry.

[0025] Specifically, step S2 is used to verify the identity and consistency of each reagent bottle during the periodic inventory process, based on the first image feature vector generated upon entry into the warehouse. Combined with the identity signature s, and the second image features at the time of current identification. and newly generated signature The process involves determining whether a reagent is an original registered object. This step not only implements basic identity verification but also introduces a specific "structural difference modeling method" tailored to the reagent management scenario. This method quantifies the degree of deviation of the reagent image in spatial structure and enhances sensitivity to "replacement anomalies" by adding a custom regularization term. This mechanism balances recognition accuracy with robustness in on-site operations and is a crucial transitional link between static identity binding and dynamic behavior monitoring in this patented system.

[0026] Furthermore, the system first retrieves the original identity signature 's' of the target reagent and the first image feature vector from the database. Both are data items generated and stored in step one, and will not be retrieved again. At this point, the reagent bottle is moved to the inventory area, and the system initiates the identification process, using the top industrial camera to re-capture the current image. And extract the second feature vector through the image encoder. The image encoder has a fixed structure, consisting of five convolutional layers, and does not include a training process; it directly outputs a 64-dimensional real-valued vector. Simultaneously, the system acquires the current RFID tag number via an RFID reader. and with the second image features The common input hash module generates a new second identity signature. This process is exactly the same as step one, and the calculation method is as follows: ; in, The identity signature generated for the current identification process. Using the same standard hash function as in step one, For the current label number, This is the second image feature vector output by the current image encoder. The generation structure of this signature is consistent with that in step one, ensuring consistency in the comparison logic.

[0027] Due to the irreversibility of hash output, the system does not directly compare s with... Instead, it shifts to comparing image features from two stages. and Structural dissimilarity. Considering the potential influences of slight rotation of reagent bottles, uneven lighting, and camera angle deviations in a laboratory environment, using only Euclidean distance is prone to misjudgment. Therefore, this step introduces an image dissimilarity function with a weighted correction term: ; The difference function consists of two parts: the first term is the weighted Euclidean distance, and the second term is the total offset regularization term. Wherein: and The first The values ​​of each image feature component; The weighting coefficients for each dimension are set according to the key regions in the reagent image, such as giving higher weights to the label region and the bottleneck contour region. The second term is the absolute value of the overall image intensity shift, used to detect structural imbalances caused by occlusion or replacement of the label region; The regularization term weight has a range of values. In the experiment, it was set to It can effectively control error balance.

[0028] For example, let's say we have a feature vector for an image of a bottle entering the warehouse. Features extracted during inventory are When calculating the difference, the feature components of the label region are set. Non-critical area settings Regularity coefficient Finally obtained Below the system threshold This indicates that the structure of the reagent image remains consistent.

[0029] Compared to the traditional Euclidean difference function, this difference function takes into account the amplification effect of local changes in key identification areas (such as label pasting surfaces and bottle cap edges) after interference in reagent management scenarios on the overall comparison results, and has higher anomaly sensitivity; at the same time, the total offset term can additionally capture the distribution changes caused by overall morphological variations (such as label obscuring or bottle counterfeiting).

[0030] Specifically, in step S3, after the reagent completes periodic identification, the current identity signature output in step two is used as the basis for step S3. Difference with image features The system determines whether a reagent is in an abnormal state and automatically executes system management actions based on the determination result. Its goal is to accurately identify risky behaviors such as "label replacement," "bottle swapping," "prolonged abnormal placement," and "expiration" without relying on manual intervention, by combining behavioral trajectories during the reagent inventory process. Through a closed-loop response mechanism triggered within the system, it enhances the proactive security and real-time response capabilities of the entire reagent management system.

[0031] Furthermore, in terms of operational procedures, the system first extracts the historical identification records of the reagent from the local database, including previous... Image difference sequence of 1 period and the identity signature sequence generated within the corresponding period. These data are all automatically recorded in the preceding steps, requiring no manual intervention. The image structural difference degree represented by the t-th recognition is obtained through image features. Features of the original image The obtained values ​​were compared; It is composed of image features and radio frequency identification. The concatenated signature, generated by a hash function, has the same format as the original signature s generated in step one. All data is arranged in chronological order to ensure the continuity of behavioral trend analysis.

[0032] The system first calculates a structural fluctuation index. This is used to evaluate the changing trend of image differences over a continuous period. Considering that occasional angular deviations or changes in illumination are normal in real-world environments, the system employs a structure fluctuation function with time decay weights, defined as: ; in, To determine the degree of fluctuation in image structure. It is a time-based decay coefficient; for example, the most recent three values ​​can be set to... , , This ensures that more attention is paid to current trends; The standard deviation of the difference sequence is used to capture whether the overall recognition shift is stable. It is the standard deviation control factor, and the setting range is generally [range missing]. This structural fluctuation index not only reflects whether there are abrupt changes in image features, but also has a certain anti-interference capability, making it suitable for situations where there are minor adjustments to the bottle or unstable recognition in actual use.

[0033] For example, in a typical scenario, the variability of a certain reagent bottle in five cycles was as follows: , , , , If set The system obtains the result through weighted calculation. ,at the same time ,final This value will serve as the first part of the anomaly function, recording the stability of the image structure.

[0034] Secondly, the system tracks signature mismatches. In each cycle, the system analyzes the current signature... With the original signature A precise comparison is performed; if a mismatch is found, it is marked as a single instance of mismatch. The cumulative number of mismatches within each cycle is The mismatch ratio is To enhance the system's sensitivity to malicious packet swapping, an identity mismatch amplification factor is introduced. (Recommended value: 0.4 to 0.6), and construct a comprehensive anomaly function. : ; The formula consists of three parts: Indicates the degree of fluctuation in image structure. Indicates the frequency of identity mismatch. This represents the largest structural difference in history. This is the maximum difference weight; the recommended range is [range to be specified]. This design is specifically optimized for "short-term camouflage" behavior: for example, in some violations, reagent labels may be changed within a cycle and then restored to their original state. Relying solely on the average difference would mask this behavior, while the maximum value term can enhance the ability to detect extreme behaviors.

[0035] System set exception trigger threshold If set to When the calculation yields Greater than or equal to If an abnormal state is detected, management actions will be triggered immediately, including: marking the reagent as "abnormal" and prohibiting it from participating in the dispensing process; sending an abnormal alert to the security administrator via SMS; and writing an abnormal record table to the database, recording the reagent number, abnormal time, and degree of abnormality. Fields such as differential sequence and number of mismatches are displayed; reagents are automatically added to the manual review checklist, prompting staff to conduct on-site inspections.

[0036] To illustrate with a practical example, a reagent bottle experienced three mismatches during continuous identification, with the largest difference in historical images being... Structural volatility has been calculated. ,set up , After substituting, we get Higher than the preset threshold The system immediately locks the reagent and executes all response actions.

[0037] This step outputs two variables: overall anomaly score. and status flag bits ,in Indicates an abnormal state. This indicates a normal state. The entire mechanism's process design heavily relies on the logical continuity of the output variables from the first two steps, especially the degree of difference. and signature This creates a closed loop from reagent documentation and periodic identification to anomaly response, enhancing the system's ability to autonomously assess physical-layer risks. Through this structurally enhanced judgment mechanism, the system maintains high identification accuracy while also possessing excellent anomaly tolerance and automatic control capabilities, making it particularly suitable for laboratory environments with stringent regulatory requirements and high reagent replacement risks.

[0038] This invention also provides an intelligent reagent management system based on RFID identification, the system comprising: The warehousing binding module is used to simultaneously collect the RFID tag number of the reagent and the first image of the reagent bottle when the reagent is put into storage, extract the first image feature vector based on the first image, and generate a unique and tamper-proof first identity signature by combining the RFID tag number. The periodic identification and comparison module is used to collect the current RFID tag number of the reagent and the second image of the current reagent bottle during periodic inventory of reagents, extract the second image feature vector based on the second image, and generate a second identity signature by combining the current RFID tag number; and calculate the structural difference between the second image feature vector and the first image feature vector. The anomaly detection and response module is used to construct a comprehensive anomaly model based on the structural difference degree, the comparison result between the second identity signature and the first identity signature, and the historical identification record of the reagent, so as to calculate the comprehensive anomaly degree; when the comprehensive anomaly degree exceeds a preset threshold, a preset management action is automatically triggered.

[0039] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0040] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or units may be electrical, mechanical, or other forms.

[0041] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0042] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for intelligent reagent management based on RFID identification, characterized in that, The method comprises the following steps: When the reagent is stored, the RFID tag number of the reagent and the first image of the reagent bottle body are synchronously collected, a first image feature vector is extracted based on the first image, and a first identity signature is generated in combination with the RFID tag number; When the reagent is periodically checked, the current RFID tag number of the reagent and the second image of the current reagent bottle body are collected, a second image feature vector is extracted based on the second image, and a second identity signature is generated in combination with the current RFID tag number; and the structural difference degree between the second image feature vector and the first image feature vector is calculated; Based on the structural difference degree, the comparison result of the second identity signature and the first identity signature, and the historical identification record of the reagent, a comprehensive abnormality degree model is constructed to calculate the comprehensive abnormality degree; when the comprehensive abnormality degree exceeds a preset threshold, a preset management action is automatically triggered. 2.The RFID identification based intelligent reagent management method according to claim 1, wherein, The first image feature vector is encoded using a fixed-structure shallow convolutional network, and a vector subjected to global average pooling and normalization processing is outputted to ensure numerical stability and not be disturbed by exposure or slight rotation; The shallow convolutional network is composed of five convolutional layers and two times of maximum pooling layers, kernel size is , step is 1, no training is needed, parameters are fixed, and is used for capturing local shape information and edge structure. 3.The RFID identification based intelligent reagent management method according to claim 1, wherein, The content of the first image feature vector is a local contour descriptor of the image. 4.The RFID identification based intelligent reagent management method according to claim 1, wherein, The first identity signature is mapped into a fixed-length binary signature string through a standard encryption hash function, which is used for subsequent unique identity comparison in reagent identification. 5.The RFID identification based intelligent reagent management method according to claim 1, wherein, The second image feature vector is extracted by an image encoder, and the image encoder is a fixed structure composed of five convolutional layers and does not contain a training process, directly outputting a 64-dimensional real number vector. 6.The RFID identification based intelligent reagent management method according to claim 1, wherein, The structural difference degree between the second image feature vector and the first image feature vector is calculated, specifically as follows: A weighted correction term is introduced into the Euclidean distance between the second image feature vector and the first image feature vector to correct it, and the structural difference degree is obtained. 7.The RFID identification based intelligent reagent management method according to claim 6, wherein, The weighted correction term is the absolute value of the overall intensity shift of the image, which is used to detect the structural imbalance caused by the blocking or replacement of the label area. 8.The RFID identification based intelligent reagent management method according to claim 1, wherein, Based on the structural difference degree, the comparison result of the second identity signature and the first identity signature, and the historical identification record of the reagent, a comprehensive abnormality degree model is constructed to calculate the comprehensive abnormality degree, specifically as follows: Get the t-th and t-th times The image structure difference degree of the second identification is combined with the time-based decay coefficient and the standard deviation of the difference degree sequence to calculate the degree of image structure fluctuation; wherein, the standard deviation of the difference degree sequence is used to capture whether the overall identification offset is stable; The current identity mismatch frequency is obtained; The comprehensive abnormality degree is calculated in combination with the current identity mismatch frequency, the image structure fluctuation degree, and the maximum structural difference in the history of the image structural difference of the tth identification. 9.The RFID identification based intelligent reagent management method according to claim 8, wherein, The current identity mismatch frequency is obtained as follows: comparing the second identity signature to the first identity signature exactly, and marking a mismatch if they are not identical; obtaining the number of mismatches accumulated over a period of time and the mismatch ratio is​​ 10. An intelligent reagent management system based on RFID identification, characterized in that, The system comprises: A storage binding module is configured to, when the reagent is stored, synchronously collect the RFID tag number of the reagent and the first image of the reagent bottle body, extract a first image feature vector based on the first image, and generate a first identity signature in combination with the RFID tag number; The cycle identification and comparison module is configured to, when the reagent is periodically inventoried, collect a current RFID tag number of the reagent and a second image of a current reagent bottle body, extract a second image feature vector based on the second image, and generate a second identity signature in combination with the current RFID tag number; and calculate a structural difference degree between the second image feature vector and the first image feature vector. The abnormality determination and response module is configured to, based on the structural difference degree, a comparison result of the second identity signature and the first identity signature, and a historical identification record of the reagent, construct a comprehensive abnormality degree model to calculate a comprehensive abnormality degree; and when the comprehensive abnormality degree exceeds a preset threshold, automatically trigger a preset management action.

Images