An abnormality recognition algorithm based on stirring magnetic sub and vortex state

Abstract

The application discloses an abnormality recognition algorithm based on stirring magnet and vortex state, comprising: collecting video stream data of a reaction instrument in the horizontal view and the overhead view; extracting the ROI region in the horizontal view and the ROI region in the overhead view; detecting the vortex existence according to the ROI region in the overhead view; detecting the vortex center offset according to the ROI region in the overhead view; detecting the stirring magnet center offset according to the ROI region in the horizontal view; detecting the stirring abnormality through the vortex existence detection, the vortex center offset detection and the stirring magnet center offset detection to obtain the stirring abnormality result; adding the stirring abnormality result into the time queue according to the time state; and judging the abnormal state of the stirring according to the number of continuous frame times in the time queue. The application significantly improves the robustness and detection accuracy of the system by using the overhead camera monitoring and the horizontal camera monitoring, and accurately performs the joint judgment by comprehensively analyzing the vortex center offset and the stirring magnet center offset.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to an anomaly recognition algorithm based on stirring magnets and vortex states. Background Technology

[0002] Stirring is a common operation in chemical experiments. It is crucial for processes such as crystallization, solution clarification, emulsion demulsification, and slurry mixing. However, because stirring time is usually long, real-time manual monitoring is difficult. If the stir bar deviates or falls off, it often goes undetected, potentially leading to uneven stirring, incomplete reactions, or even liquid splashing. Therefore, to improve experimental efficiency and ensure reproducibility, developing a stable and efficient method for identifying anomalies during stirring has become an indispensable key aspect of current chemical experimental research. This method can effectively solve the problems caused by insufficient manual monitoring over long periods, ensuring the smooth progress of experiments. Summary of the Invention

[0003] This invention provides an anomaly identification algorithm based on stirring magnets and vortex states to solve problems in the prior art.

[0004] On one hand, embodiments of the present invention provide an anomaly identification algorithm based on stirring magnets and vortex states, including:

[0005] Video stream data of the reaction instrument was acquired from both eye-level and top-down perspectives;

[0006] Extract the ROI region in the eye-level view and the ROI region in the top-down view;

[0007] The presence of vortices is detected based on the ROI region viewed from above.

[0008] vortex center offset detection is performed based on the ROI region viewed from above.

[0009] The center offset of the stirring magnet is detected based on the ROI region in the eye-view direction.

[0010] The stirring anomaly detection is performed by detecting the presence of vortexes, the offset of vortex centers, and the offset of stirring magnet centers to obtain the stirring anomaly results.

[0011] The results of the stirring anomaly are added to the time queue according to the time status;

[0012] The abnormal state of stirring is determined by the number of consecutive frame times in the time queue.

[0013] In one possible implementation, the extraction of the ROI region in the eye-level view and the ROI region in the top-down view includes:

[0014] Based on the camera's geometrically calibrated position, the image data of the ROI region in the eye-level direction and the image data of the ROI region in the top-down direction of the video stream data are obtained.

[0015] In one possible implementation, vortex presence detection based on the top-view ROI region includes:

[0016] The presence detection of the vortex is performed using the MobileNetv3 deep learning network;

[0017] The MobileNetv3 deep learning network was trained based on offline data augmentation, cosine annealing, and learning rate preheating strategies.

[0018] The image data of the ROI region in the top-view direction is input into the MobileNetv3 network to obtain the vortex presence detection result;

[0019] In one possible implementation, the vortex center offset detection based on the top-view ROI region includes:

[0020] A YOLOv8 network for vortex center detection was trained using image enhancement techniques, a cosine annealing learning rate scheduling strategy, and a preheating training strategy.

[0021] The image data of the ROI region in the top-view direction is input into the YOLOv8 network to obtain the vortex center detection result;

[0022] The vortex center offset detection result is obtained by comparing the vortex center detection result with the preset threshold.

[0023] In one possible implementation, the detection of the stirring magnet center offset based on the ROI region in the horizontal view includes:

[0024] A YOLOv8 network for stirring magnet detection was trained using image enhancement technology, a cosine annealing learning rate scheduling strategy, and a preheating training strategy.

[0025] The image data of the ROI region in the eye-view direction is input into the YOLOv8 network to obtain the detection result of the stirring magnet center;

[0026] The displacement detection result of the stirring magnet center is obtained by comparing the detection result with the center of the stirring magnet according to a preset threshold.

[0027] In one possible implementation, the step of obtaining the stirring anomaly result through the vortex presence detection, vortex center offset detection, and stirring magnet center offset detection includes:

[0028] The state of the reagent is determined by the vortex center offset detection results and the stirring magnet center offset detection results.

[0029] The result of the stirring abnormality is determined based on the described state.

[0030] In one possible implementation, adding the stirring anomaly result to the time queue according to the time state further includes:

[0031] The total length of the time queue is N. When the time queue is full, following the first-in-first-out principle, each time a new frame time of a stirring anomaly result is added, the frame time of the first old stirring anomaly result in the time queue is discarded.

[0032] In one possible implementation, determining the abnormal state of stirring based on the number of consecutive frame times in the time queue includes:

[0033] The set of consecutively existing frame times in the time queue is classified as an abnormal state.

[0034] An anomaly identification algorithm based on stirring magnetic particles and vortex states in this invention has the following advantages:

[0035] (1) The offset of the center of the liquid surface vortex is monitored by using a top-view camera, while the offset of the center of the stirring magnet is monitored by using a head-up camera. This achieves continuous monitoring without contact, and also significantly improves the robustness and detection accuracy of the system, ensuring the stability and reliability of the stirring process.

[0036] (2) By comprehensively analyzing the two key features, namely the vortex center offset and the stirring magnet center offset, a joint judgment is accurately made to achieve stable detection of abnormal states. The multi-feature fusion method effectively avoids the false alarms or false alarms that may be caused by single feature judgment. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart of an anomaly identification algorithm based on stirring magnets and vortex states provided in this application embodiment. Detailed Implementation

[0039] 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.

[0040] Figure 1 This is a flowchart illustrating an anomaly identification algorithm based on stirring magnetic particles and vortex states, provided in an embodiment of the present invention. The embodiment of the present invention provides an anomaly identification algorithm based on stirring magnetic particles and vortex states, comprising:

[0041] Video stream data of the reaction instrument was acquired from both eye-level and top-down perspectives;

[0042] Extract the ROI region in the eye-level view and the ROI region in the top-down view;

[0043] The presence of vortices is detected based on the ROI region viewed from above.

[0044] vortex center offset detection is performed based on the ROI region viewed from above.

[0045] The center offset of the stirring magnet is detected based on the ROI region in the eye-view direction.

[0046] The stirring anomaly detection is performed by detecting the presence of vortexes, the offset of vortex centers, and the offset of stirring magnet centers to obtain the stirring anomaly results.

[0047] The results of the stirring anomaly are added to the time queue according to the time status;

[0048] The abnormal state of stirring is determined by the number of consecutive frame times in the time queue.

[0049] The extraction of the ROI region in the eye-level view and the ROI region in the top-down view includes:

[0050] Based on the camera's geometrically calibrated position, the image data of the ROI region in the eye-level direction and the image data of the ROI region in the top-down direction of the video stream data are obtained.

[0051] vortex presence detection based on the top-view ROI region includes:

[0052] The presence detection of the vortex is performed using the MobileNetv3 deep learning network;

[0053] The MobileNetv3 deep learning network was trained based on offline data augmentation, cosine annealing, and learning rate preheating strategies.

[0054] The image data of the top-view ROI region is input into the MobileNetv3 network to obtain the vortex presence detection result; the vortex center offset detection based on the top-view ROI region includes:

[0055] A YOLOv8 network for vortex center detection was trained using image enhancement techniques, a cosine annealing learning rate scheduling strategy, and a preheating training strategy.

[0056] The image data of the ROI region in the top-view direction is input into the YOLOv8 network to obtain the vortex center detection result;

[0057] The vortex center offset detection result is obtained by comparing the vortex center detection result with the preset threshold.

[0058] The step of detecting the center offset of the stirring magnet based on the ROI region in the horizontal viewing direction includes:

[0059] A YOLOv8 network for stirring magnet detection was trained using image enhancement technology, a cosine annealing learning rate scheduling strategy, and a preheating training strategy.

[0060] The image data of the ROI region in the eye-view direction is input into the YOLOv8 network to obtain the detection result of the stirring magnet center;

[0061] The displacement detection result of the stirring magnet center is obtained by comparing the detection result with the center of the stirring magnet according to a preset threshold.

[0062] The method of detecting stirring anomalies through vortex presence detection, vortex center offset detection, and stirring magnet center offset detection to obtain stirring anomaly results includes:

[0063] The state of the reagent is determined by the vortex center offset detection results and the stirring magnet center offset detection results.

[0064] The result of the stirring anomaly is determined based on the described state.

[0065] Adding the abnormal stirring results to the time queue based on the time status also includes:

[0066] The total length of the time queue is N. When the time queue is full, following the first-in-first-out principle, each time a new frame time of a stirring anomaly result is added, the frame time of the first old stirring anomaly result in the time queue is discarded.

[0067] The step of determining the abnormal state of stirring based on the number of consecutive frame times in the time queue includes:

[0068] The set of consecutively existing frame times in the time queue is classified as an abnormal state.

[0069] For example, a continuous video stream of liquid surface changes is acquired in real time using an industrial camera with a top-down view. Based on the geometric positional relationship between the reagent bottle opening and the camera, the reagent bottle opening region in the current video frame is directly extracted, removing background information and allowing subsequent operations to focus on the changes in the liquid surface. Simultaneously, the center point P of the reagent bottle opening region is calculated. v_center .

[0070] MobileNetv3 is a lightweight convolutional network that captures abstract semantic information of liquid regions as deep semantic features while maintaining computational efficiency, thereby accurately identifying targets. The MobileNetv3 deep learning network for vortex target detection is trained based on offline data augmentation, cosine annealing, and learning rate warm-up strategies.

[0071] The presence of vortices is determined using the MobileNetv3 deep learning network. If no vortex is found, it indicates that stirring has not yet begun or has just started, and the process returns to the video extraction stage. If a vortex is present, it means that stirring has begun and stabilized, and the detection continues.

[0072] The vortex center detection employs a YOLOv8 network model, which achieves multi-scale feature fusion through a CSP structure and a Focus module, and combines a PANet path aggregation network to enhance contextual information extraction capabilities. The YOLOv8 network for vortex center target detection is trained using offline data augmentation, cosine annealing, and learning rate preheating strategies.

[0073] This study uses the YOLOv8 model to detect vortex centers in top-view ROI region image data. The network detects the center of the bounding box P. v_pred That is, the center of the vortex.

[0074] Based on the location of the vortex center, calculate the predicted offset d between the vortex center and the center point of the reagent bottle opening. v .

[0075] d v =|P v_pred -P v_center |

[0076] Based on the calculated vortex offset d v Compare it with the preset threshold d v_th Compare. If d v <d v_th If d indicates a normal state, then the value is determined to be 0. v ≥d v_th If the status is abnormal, it will be judged as 1.

[0077] A continuous video stream of data on the internal changes of a liquid is acquired in real time using an industrial camera with a head-up view.

[0078] Based on the geometric positional relationship between the reagent bottle and the camera, the area around the reagent bottle opening in the current video frame is directly captured, removing interference from irrelevant background information and allowing subsequent operations to focus on changes within the liquid. Simultaneously, the center point P of the stirring magnet under normal conditions is set. s_center That is, the center position at the bottom of the reagent bottle.

[0079] Magnetic particle target detection employs a YOLOv8 network model. The YOLOv8 network achieves multi-scale feature fusion through a CSP structure and a Focus module, and combines this with a PANet path aggregation network to enhance contextual information extraction capabilities. The YOLOv8 network for magnetic particle target detection is trained using offline data augmentation, cosine annealing, and learning rate preheating strategies.

[0080] Based on the YOLOv8 network for magnetic particle targets, stirring magnetic particle detection is performed on the ROI region image data in the eye-level direction. The target box center P detected by the network is... s_pred This is the center of the stirring magnet.

[0081] Calculate the predicted offset d between the center of the stirring magnet and the center of the bottom of the reaction flask. s .

[0082] d s =|P s_pred -P s_center |

[0083] Based on the calculated offset d of the stirring magnet center s Compare it with the preset threshold d s_th Make a comparison. If d s <d s_th If d , it indicates a normal state, and is therefore determined as 0. s ≥d s_th This indicates an abnormal state, and is judged as 1.

[0084] Based on the aforementioned calculations, if at least one of the vortex center or the stirring magnet center exhibits an abnormal state, the time of that current state is added to the time queue. When the time queue is full, following the first-in, first-out (FIFO) principle, the first old frame time in the time queue is discarded each time a new frame time is added.

[0085] Specifically, let the total length of the time queue be N. Determine if there are N consecutive frames in the time queue. If not, that is, the number of consecutive frames is less than N, return to the initial step. If yes, it means that the liquid stirring is in an abnormal state for N consecutive frames, and the stirring is determined to be abnormal.

[0086] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0087] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An anomaly recognition algorithm based on stirring magnetic sub and vortex state, characterized in that, The method comprises the following steps: Collecting video stream data of the reaction instrument in the horizontal and overhead directions; Extracting the horizontal direction ROI region and the overhead direction ROI region; Performing vortex existence detection according to the overhead direction ROI region; Performing vortex center offset detection according to the overhead direction ROI region; Performing stirring magnet center offset detection according to the horizontal direction ROI region; Performing stirring abnormality detection through the vortex existence detection, vortex center offset detection and stirring magnet center offset detection to obtain a stirring abnormality result; Adding the stirring abnormality result to a time queue according to the time state; Determining the abnormality state of stirring according to the number of continuous frame times in the time queue.

2. The anomaly recognition algorithm based on stirring magnetic sub and vortex state according to claim 1, characterized in that, The extraction of the horizontal direction ROI region and the overhead direction ROI region comprises the following steps: Obtaining horizontal direction ROI region picture data and overhead direction ROI region picture data of the video stream data according to the position of camera geometric calibration.

3. The anomaly recognition algorithm based on stirring magnetic sub and vortex state according to claim 1, characterized in that, The vortex existence detection according to the overhead direction ROI region comprises the following steps: The vortex existence detection adopts a MobileNetv3 deep learning network; The MobileNetv3 deep learning network is trained based on an offline data enhancement strategy, a cosine annealing strategy and a learning rate preheating strategy; The overhead direction ROI region picture data is input into the MobileNetv3 network to obtain a vortex existence detection result.

4. The anomaly recognition algorithm based on stirring magnetic sub and vortex state according to claim 1, characterized in that, The vortex center offset detection according to the overhead direction ROI region comprises the following steps: A YOLOv8 network for vortex center detection is trained through an image enhancement technique, a cosine annealing learning rate scheduling strategy and a preheating training strategy; The overhead direction ROI region picture data is input into the YOLOv8 network to obtain a vortex center detection result; A vortex center offset detection result is obtained by comparing the vortex center detection result with a preset threshold.

5. The anomaly recognition algorithm based on stirring magnetic sub and vortex state according to claim 1, characterized in that, The stirring magnet center offset detection according to the horizontal direction ROI region comprises the following steps: A YOLOv8 network for stirring magnet detection is trained through an image enhancement technique, a cosine annealing learning rate scheduling strategy and a preheating training strategy; The horizontal direction ROI region picture data is input into the YOLOv8 network to obtain a stirring magnet center detection result; A stirring magnet center offset detection result is obtained by comparing the stirring magnet center detection result with a preset threshold.

6. The anomaly recognition algorithm based on stirring magnetic sub and vortex state according to claim 1, characterized in that, The stirring abnormality detection through the vortex existence detection, vortex center offset detection and stirring magnet center offset detection to obtain a stirring abnormality result comprises the following steps: The state of the reagent is determined through the vortex center offset detection result and the stirring magnet center offset detection result; The stirring abnormality result is determined according to the state.

7. The anomaly recognition algorithm based on stirring magnetic sub and vortex state according to claim 1, characterized in that, The addition of the stirring abnormality result to the time queue according to the time state further comprises the following steps: The total length of the time queue is N, when the time queue is full, a new stirring abnormality result is added to the time queue, and the first old stirring abnormality result in the time queue is discarded.

8. The anomaly recognition algorithm based on stirring magnetic sub and vortex state according to claim 7, characterized in that, The determination of the abnormality state of stirring according to the number of continuous frame times in the time queue comprises the following steps: The set of continuous frame times in the time queue is listed as an abnormality state.

Images