LIMS laboratory process maintenance method based on adaptive learning model

Abstract

The present application relates to the technical field of process maintenance, and specifically discloses a LIMS laboratory process maintenance method based on an adaptive learning model, comprising the following steps: step S1: recording a standard operation video of an experimental personnel, dividing into two stages of preparation and operation, respectively extracting a standard limb feature sequence and an operation limb feature sequence, and determining a starting limb feature from the latter; step S2: when a user is preparing, acquiring a limb feature sequence of the user in real time, comparing the sequence with the standard limb feature sequence, and calculating a personalized adjustment coefficient; step S3: using the adjustment coefficient to correct the standard operation sequence, generating a simulation limb feature sequence and a monitoring starting point that are exclusive to the user; in actual operation of the user, once a real-time sequence matches the monitoring starting point, the sequence is tracked in real time, and through comparison with the simulation sequence, it is judged whether the operation is abnormal; the method realizes whole-process assistance from individualized standard formulation, real-time preparation posture guidance to operation process abnormality monitoring.

Description

Technical Field

[0001] This invention relates to the field of process maintenance technology, and more specifically to a LIMS laboratory process maintenance method based on an adaptive learning model. Background Technology

[0002] Laboratory Information Management Systems (LIMS) play a vital role in modern laboratories, primarily managing laboratory samples, data, processes, and resources, significantly improving laboratory operational efficiency and data reliability.

[0003] However, traditional LIMS systems mainly focus on the management of information flow, such as sample tracking, result entry, and report generation. They lack effective real-time monitoring and maintenance methods to ensure that experimental procedures are executed correctly and in accordance with regulations. For example, in terms of ensuring compliance with experimental procedures, most laboratories still rely on manual supervision or simple video recording review. Manual supervision is not only costly and inefficient, but also difficult to maintain uninterrupted focus throughout the process. It is prone to oversights due to subjective fatigue and other factors, and cannot intervene and correct errors in time, which may lead to experimental failure, sample damage, or even safety accidents. Summary of the Invention

[0004] The purpose of this invention is to provide a LIMS laboratory process maintenance method based on an adaptive learning model, and to solve the following technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: The LIMS laboratory process maintenance method based on an adaptive learning model includes the following steps: Step S1: The experimenters demonstrate the experimental procedure and record a standard video of the experimental procedure demonstration. The experimental procedure is divided into an experimental preparation stage and an experimental operation stage in the standard video. The standard limb feature sequence of the experimenters in the experimental preparation stage is obtained, and the limb feature sequence of the operation stage is obtained. The starting limb feature is extracted from the limb feature sequence of the operation stage. Step S2: When the user enters the experiment preparation stage, the experiment process image of the user is acquired in real time, and the user's limb feature sequence is obtained in real time based on the experiment process image. The limb feature sequence is compared with the standard limb feature sequence to obtain the adjustment coefficient. Step S3: Based on the adjustment coefficient, the limb feature sequence of the experimental operation stage is corrected to obtain the simulated limb feature sequence of the user in the experimental process, and the monitoring starting point is extracted. The system acquires images of the user's actions during each stage of the experiment in real time, obtaining the user's real-time limb features. When the similarity between the real-time limb features and the monitoring starting point exceeds a threshold, a limb feature sequence is generated in real time. The real-time limb feature sequence is then compared with a simulated limb feature sequence to determine whether the user exhibits any abnormalities during the experimental operation.

[0006] As a further aspect of the present invention: the experimental preparation stage is the time period occupied by the experimenter placing the experimental equipment on the experimental operating table, and the experimental operation stage is the time period occupied by the experimenter performing the experimental operation.

[0007] As a further aspect of the present invention, the process of dividing the experimental procedure into an experimental preparation stage and an experimental operation stage includes: Acquire image information of each experimental device, including front, side and top views of the experimental device; divide the standard process video into several image frames, capture the image of the experimental operating table in real time, and record it as the operating table image; when all experimental devices are identified in the operating table image based on image recognition technology, the image frame at this time is designated as a stage segment frame. The video segment consisting of all image frames between the first frame and the stage segment frame in the standard process video is designated as the experimental preparation stage, and the video segment consisting of all image frames between the stage segment frame and the last frame is designated as the experimental operation stage.

[0008] As a further aspect of the present invention: the process of obtaining the standard limb feature sequence of the experimental personnel in the experimental preparation stage includes: All image frames in the experimental preparation phase are denoted as preparation phase frames. Key point detection is performed on the experimental personnel in the preparation phase frames to identify several key points, including skeletal joint points. A coordinate system is established on the preparation phase frames, and the preparation phase frames are divided into several pixels. The coordinates corresponding to each pixel in the coordinate system are obtained and denoted as pixel coordinates. The pixel coordinates corresponding to the key points are obtained and denoted as the positions of the key points. Based on the changes in the positions of the key points in several consecutive preparation phase frames, the movement speed and direction of the key points are obtained. The movement speed and direction of the key points are denoted as key point information. The key point information of each key point is fused to obtain a standard limb feature sequence.

[0009] As a further aspect of the present invention: the process of fusing the key point information of each key point includes: The key point information of each key point in the same preparation stage frame is converted into running speed vector and motion direction vector, and the running speed vector and motion direction vector of each key point are concatenated to obtain a high-dimensional comprehensive feature vector; the comprehensive feature vector of each preparation stage frame is numbered according to the time sequence to obtain a standard limb feature sequence.

[0010] As a further aspect of the present invention: the process of obtaining the adjustment coefficient includes: For any limb feature and comprehensive feature vector with the same number in the limb feature sequence and the standard limb feature sequence, obtain the vector difference between the limb feature and comprehensive feature vector under this number, and finally obtain the vector difference under each number. Obtain the average value of all vector differences, and denot it as the average vector difference. Denote any limb feature, comprehensive feature vector and average vector difference with the same number as a sample combination. Obtain the sample data from the sample combinations of all numbers. An initial regression model is established, which takes the comprehensive feature vector and vector difference as input and limb features as output. The initial regression model is iteratively trained using the sample data to obtain the final regression model KC+V. S =V y Where C is the average vector difference, K is the adjustment coefficient obtained from iterative training, and V S For the comprehensive feature vector, V y It refers to physical characteristics.

[0011] As a further aspect of the present invention: the process of extracting the monitoring starting point includes, for any simulated limb feature in the simulated limb feature sequence, obtaining the simulated limb feature with the highest similarity to the starting point limb feature, and recording it as the monitoring starting point.

[0012] As a further aspect of the present invention: the process of determining whether the user exhibits any abnormalities during the experimental operation phase includes: The latest simulated limb feature with the latest number in the real-time limb feature sequence is acquired in real time and recorded as the latest simulated limb feature. Simulated limb features with the same numerical value as the latest number are selected from the simulated limb feature sequence and recorded as the latest simulated limb features. If the similarity between the latest simulated limb features is lower than a preset similarity threshold, then the user has an anomaly during the experimental operation phase.

[0013] The beneficial effects of this invention are: This invention introduces an adaptive learning model to analyze user limb feature sequences in real time during the experimental preparation phase and generate personalized adjustment coefficients. Based on these adjustment coefficients, the standard operating sequence is dynamically corrected, thereby transforming the monitoring benchmark from a rigid, uniform standard into a personalized expectation that aligns with the user's individual characteristics. This fundamentally solves the industry problem of misjudgments caused by individual differences. Furthermore, this method meticulously divides the experimental process into a preparation phase and an operation phase, designating the preparation phase as an active calibration step. This enables the pre-construction of monitoring benchmarks and problem prevention, forming a paradigm shift from passive supervision to proactive maintenance. Combined with computer vision technology, it achieves full automation from personalized calibration to real-time sequence comparison and immediate anomaly alerts, significantly reducing the cost of manual supervision and enabling effective intervention in the early stages of errors. This greatly improves the laboratory's operational standardization, result reliability, and safety management level. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram of the steps of the LIMS laboratory process maintenance method based on the adaptive learning model of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 As shown, this invention is a LIMS laboratory process maintenance method based on an adaptive learning model, comprising the following steps: Step S1: The experimenters demonstrate the experimental procedure and record a standard video of the experimental procedure demonstration. The experimental procedure is divided into an experimental preparation stage and an experimental operation stage in the standard video. The standard limb feature sequence of the experimenters in the experimental preparation stage is obtained, and the limb feature sequence of the operation stage is obtained. The starting limb feature is extracted from the limb feature sequence of the operation stage. Specifically, experienced experimenters demonstrate the correct experimental procedure in a standard experimental environment and use a fixed-position camera to film the experimental procedure, generating a standard procedure video. The moment when the experimenter places the last key piece of experimental equipment on the operating table and prepares to start the operation is taken as the stage node. The video segments before this point are divided into the experimental preparation stage, and the video segments after this point are divided into the experimental operation stage. The OpenPose pose recognition algorithm was used to detect human key points frame by frame in the preparation stage video, extracting the coordinates of 15 core joints, including the shoulder, elbow, and wrist. The motion velocity and motion direction features were obtained by calculating the joint displacement between consecutive frames. The features of each key point were then concatenated into a high-dimensional feature vector in time sequence, ultimately forming a standard limb feature sequence that represents the dynamic features of the preparation stage. The same feature extraction process was then performed on the experimental operation stage video to generate an operational limb feature sequence. The first feature vector in this sequence, i.e., the limb pose feature at the start of the experimental operation, was extracted as the starting limb feature. In a preferred embodiment of the present invention, the experimental preparation stage is the time period occupied by the experimenter placing the experimental equipment on the experimental operating table, and the experimental operation stage is the time period occupied by the experimenter performing the experimental operation. In a preferred embodiment of the present invention, the process of dividing the experimental procedure into an experimental preparation stage and an experimental operation stage includes: Acquire image information of each experimental device, including front, side and top views of the experimental device; divide the standard process video into several image frames, capture the image of the experimental operating table in real time, and record it as the operating table image; when all experimental devices are identified in the operating table image based on image recognition technology, the image frame at this time is designated as a stage segment frame. The video segment consisting of all image frames between the first frame and the stage segment frame in the standard process video is recorded as the experimental preparation stage, and the video segment consisting of all image frames between the stage segment frame and the last frame is recorded as the experimental operation stage. In a preferred embodiment of the present invention, the process of obtaining the standard limb feature sequence of the experimenter during the experimental preparation stage includes: All image frames in the experimental preparation phase are denoted as preparation phase frames. Key point detection is performed on the experimental personnel in the preparation phase frames to identify several key points, including skeletal joint points. A coordinate system is established on the preparation phase frames, and the preparation phase frames are divided into several pixels. The coordinates corresponding to each pixel on the coordinate system are obtained and denoted as pixel coordinates. The pixel coordinates corresponding to the key points are obtained and denoted as the positions of the key points. Based on the changes in the positions of the key points in several consecutive preparation phase frames, the movement speed and direction of the key points are obtained. The movement speed and direction of the key points are denoted as key point information. The key point information of each key point is fused to obtain a standard limb feature sequence. The process of integrating the key point information from each key point includes: The key point information of each key point in the same preparation stage frame is converted into running speed vector and motion direction vector, and the running speed vector and motion direction vector of each key point are concatenated to obtain a high-dimensional comprehensive feature vector; the comprehensive feature vector of each preparation stage frame is numbered according to the time sequence to obtain a standard limb feature sequence. In a preferred embodiment of the present invention, the process of obtaining the limb feature sequence during the experimental operation phase includes: All image frames in the experimental operation phase are obtained and denoted as operation phase frames. The comprehensive feature vectors of the experimental personnel in each operation phase frame are obtained. Each comprehensive feature vector is numbered sequentially to obtain the operation limb feature sequence. In a preferred embodiment of the present invention, the starting limb feature is the first comprehensive feature vector in the sequence of operating limb features; Step S2: When the user performs the experimental procedure and enters the experimental preparation stage, the experimental procedure image of the user is acquired in real time. Based on the experimental procedure image, the user's limb feature sequence is obtained in real time. The limb feature sequence is compared with the standard limb feature sequence to obtain the adjustment coefficient. Specifically, when the user starts the experiment and enters the preparation stage, the user's operation images are collected in real time by a camera deployed above the experimental platform. After human body detection is performed using the YOLOv7 model, the user's real-time limb feature sequence is constructed. The real-time limb feature sequence is then time-aligned and similarity is calculated with the standard limb feature sequence. The adjustment coefficient K of the best fit is obtained by iterative optimization using the gradient descent method through the established initial regression model. In a preferred embodiment of the present invention, the process of obtaining the user's limb feature sequence in real time is as follows: the comprehensive feature vector of the user in the experimental process image is obtained in real time and recorded as the user's limb feature; the limb features of each experimental process image are numbered to obtain the limb feature sequence. In a preferred embodiment of the present invention, the process of obtaining the adjustment coefficient includes: For any limb feature and comprehensive feature vector with the same number in the limb feature sequence and the standard limb feature sequence, obtain the vector difference between the limb feature and comprehensive feature vector under this number, and finally obtain the vector difference under each number. Obtain the average value of all vector differences, and denot it as the average vector difference. Denote any limb feature, comprehensive feature vector and average vector difference with the same number as a sample combination. Obtain the sample data from the sample combinations of all numbers. An initial regression model is established, which takes the comprehensive feature vector and vector difference as input and limb features as output. The initial regression model is iteratively trained using the sample data to obtain the final regression model KC+V. S =V yWhere C is the average vector difference, K is the adjustment coefficient obtained from iterative training, and V S For the comprehensive feature vector, V y Physical characteristics; Specifically, the user's real-time collected limb feature sequence is temporally aligned with the standard limb feature sequence. For each feature vector with the same number, the Euclidean distance is calculated as the vector difference. Then, the vector differences under all numbers are counted and the arithmetic mean is calculated to obtain the average vector difference C. Sample data containing sample combinations under all numbers is constructed. Each sample combination consists of the user's limb features, comprehensive feature vector, and average vector difference. Based on the sample data, the least squares method is used for parameter fitting. The gradient descent algorithm is used to iteratively optimize the model parameters with the mean squared error as the loss function, and finally the optimal adjustment coefficient is obtained. Step S3: Based on the adjustment coefficient, the operation limb feature sequence of the experimental operation stage is corrected to obtain the simulated limb feature sequence of the user in the experimental process, and the monitoring starting point is extracted from the simulated limb feature sequence. The system acquires images of the user's actions during each stage of the experiment in real time, obtaining the user's real-time limb features. When the similarity between the real-time limb features and the monitoring starting point exceeds a threshold, a limb feature sequence is generated in real time. The real-time limb feature sequence is then compared with a simulated limb feature sequence to determine whether the user exhibits any abnormalities during the experimental operation. Specifically, based on the adjustment coefficient K, a linear transformation is performed on each feature vector in the standard limb feature sequence to generate a simulated limb feature sequence that conforms to the user's personal characteristics. In the simulated limb feature sequence, the cosine similarity calculation method is used to select the feature point with the highest similarity to the standard starting point limb feature as the personalized monitoring starting point. In the experimental operation phase, skeletal joint features are extracted from the user operation images collected in real time. When the similarity between the real-time feature and the monitoring starting point reaches the 85% threshold, the dynamic time warping algorithm is immediately started to synchronize the sequence. The real-time generated limb feature sequence is compared with the simulated sequence frame by frame. If the similarity of 5 consecutive frames is lower than 70%, it is judged as an operation abnormality. In a preferred embodiment of the present invention, the process of correcting the limb feature sequence during the experimental operation phase includes: for any comprehensive feature vector in the limb feature sequence, based on the regression model and according to the adjustment coefficient, correcting the comprehensive feature vector to the user's simulated limb features; and after correcting all comprehensive feature vectors in the limb feature sequence, obtaining the simulated limb feature sequence. In a preferred embodiment of the present invention, the process of extracting the monitoring starting point includes: for any simulated limb feature in the simulated limb feature sequence, obtaining the simulated limb feature with the highest similarity to the starting point limb feature, and recording it as the monitoring starting point; In a preferred embodiment of the present invention, the real-time limb features are a comprehensive feature vector of the user obtained based on the operation image; In a preferred embodiment of the present invention, the limb feature sequence is composed of real-time limb features acquired in real time after the user enters the experimental operation phase. In a preferred embodiment of the present invention, the process of determining whether the user exhibits any abnormality during the experimental operation phase includes: The latest simulated limb feature with the latest number in the real-time limb feature sequence is acquired in real time and recorded as the latest simulated limb feature. Simulated limb features with the same numerical value as the latest number are selected from the simulated limb feature sequence and recorded as the latest simulated limb features. If the similarity between the latest simulated limb features is lower than a preset similarity threshold, then the user has an anomaly during the experimental operation phase.

[0018] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A LIMS laboratory process maintenance method based on an adaptive learning model, characterized in that, Includes the following steps: Step S1: The experimenters demonstrate the experimental procedure and record a standard video of the demonstration. In the standard process video, the experimental process is divided into an experimental preparation stage and an experimental operation stage. The standard limb feature sequence of the experimenters in the experimental preparation stage is obtained, and the operational limb feature sequence in the experimental operation stage is obtained. The starting limb feature is extracted from the operational limb feature sequence. Step S2: When the user enters the experiment preparation stage, the experiment process image of the user is acquired in real time, and the user's limb feature sequence is obtained in real time based on the experiment process image. The limb feature sequence is compared with the standard limb feature sequence to obtain the adjustment coefficient. Step S3: Based on the adjustment coefficient, the limb feature sequence of the experimental operation stage is corrected to obtain the simulated limb feature sequence of the user in the experimental process, and the monitoring starting point is extracted. The system acquires images of the user's actions during each stage of the experiment in real time, obtaining the user's real-time limb features. When the similarity between the real-time limb features and the monitoring starting point exceeds a threshold, a limb feature sequence is generated in real time. The real-time limb feature sequence is then compared with the simulated limb feature sequence to determine whether the user has any abnormalities during the experimental operation stage. In step S1, the experimental preparation stage is the time period occupied by the experimenter placing the experimental equipment on the experimental workbench, and the experimental operation stage is the time period occupied by the experimenter performing the experimental operation. In step S1, the process of dividing the experimental procedure into the experimental preparation stage and the experimental operation stage includes: Acquire image information of each experimental instrument, including front, side and top views of the experimental instrument; The standard process video is divided into several image frames. The image of the experimental operating table in the image frame is captured in real time and recorded as the operating table image. When all the experimental equipment is identified in the operating table image based on image recognition technology, the image frame at this time is called the stage segment frame. The video segment consisting of all image frames between the first frame and the stage segment frame in the standard process video is designated as the experimental preparation stage, and the video segment consisting of all image frames between the stage segment frame and the last frame is designated as the experimental operation stage.

2. The LIMS laboratory process maintenance method based on an adaptive learning model according to claim 1, characterized in that, In step S1, the process of obtaining the standard limb feature sequences of the experimental personnel during the experimental preparation phase includes: All image frames in the experimental preparation phase are denoted as preparation phase frames. Key point detection is performed on the experimental personnel in the preparation phase frames to identify several key points, including skeletal joint points. A coordinate system is established on the preparation phase frames, and the preparation phase frames are divided into several pixels. The coordinates corresponding to each pixel in the coordinate system are obtained and denoted as pixel coordinates. The pixel coordinates corresponding to the key points are obtained and denoted as the positions of the key points. Based on the changes in the positions of the key points in several consecutive preparation phase frames, the movement speed and direction of the key points are obtained. The movement speed and direction of the key points are denoted as key point information. The key point information of each key point is fused to obtain a standard limb feature sequence.

3. The LIMS laboratory process maintenance method based on an adaptive learning model according to claim 2, characterized in that, In step S1, the process of fusing the key point information of each key point includes: The key point information of each key point in the same preparation stage frame is converted into running speed vector and motion direction vector, and the running speed vector and motion direction vector of each key point are concatenated to obtain a high-dimensional comprehensive feature vector; the comprehensive feature vector of each preparation stage frame is numbered according to the time sequence to obtain a standard limb feature sequence.

4. The LIMS laboratory process maintenance method based on an adaptive learning model according to claim 3, characterized in that, In step S2, the process of obtaining the adjustment coefficient includes: For any limb feature and comprehensive feature vector with the same number in the limb feature sequence and the standard limb feature sequence, obtain the vector difference between the limb feature and comprehensive feature vector under this number, and finally obtain the vector difference under each number. Obtain the average value of all vector differences, and denot it as the average vector difference. Denote any limb feature, comprehensive feature vector and average vector difference with the same number as a sample combination. Obtain the sample data from the sample combinations of all numbers. An initial regression model is established, which takes the comprehensive feature vector and vector difference as input and limb features as output. The initial regression model is iteratively trained using the sample data to obtain the final regression model KC+V. S =V y Where C is the average vector difference, K is the adjustment coefficient obtained from iterative training, and V S For the comprehensive feature vector, V y It refers to physical characteristics.

5. The LIMS laboratory process maintenance method based on an adaptive learning model according to claim 1, characterized in that, In step S3, the process of extracting the monitoring starting point includes: for any simulated limb feature in the simulated limb feature sequence, obtaining the simulated limb feature with the highest similarity to the starting point limb feature, and recording it as the monitoring starting point.

6. The LIMS laboratory process maintenance method based on an adaptive learning model according to claim 1, characterized in that, In step S3, the process of determining whether there is any abnormality in the user's experimental operation includes: The latest simulated limb feature with the latest number in the real-time limb feature sequence is acquired in real time and recorded as the latest simulated limb feature. Simulated limb features with the same numerical value as the latest number are selected from the simulated limb feature sequence and recorded as the latest simulated limb features. If the similarity between the latest simulated limb features is lower than a preset similarity threshold, then the user has an anomaly during the experimental operation phase.

Images