Abnormality recognition algorithm based on stirring magneton and vortex state

Abstract

The invention discloses an anomaly recognition algorithm based on a stirring magneton and vortex state. The anomaly recognition algorithm comprises the steps of collecting video stream data of a reaction instrument in the head-up direction and the top-down direction; extracting an ROI (Region of Interest) in the head-up direction and an ROI in Vortex existence detection is carried out according to the ROI in the overlook direction; vortex center offset detection is carried out according to the ROI in the overlook direction; performing stirring magneton center offset detection according to the ROI in the head-up direction; performing stirring anomaly detection through vortex existence detection, vortex center offset detection and stirring magneton center offset detection to obtain a stirring anomaly result; adding the stirring abnormal result into a time queue according to the time state; and judging the abnormal state of stirring according to the number of continuous frame time in the time queue. According to the method, the overlook camera and the head-up camera are used for monitoring, so that the robustness and the detection precision of the system are remarkably improved, and combined judgment is accurately performed by comprehensively analyzing the two key characteristics of the vortex center offset and the stirring magneton center offset.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to an abnormality recognition algorithm based on stirring magnets and vortex states. Background Art

[0002] In the process of chemical experiments, stirring is a very common operation. This operation is crucial for processes such as crystallization, solution clarification, emulsion demulsification, and slurry mixing. However, since the stirring time is usually long, manual real-time monitoring is difficult. Once the stirring magnet deviates or falls off, it is often not discovered in time, which may lead to problems such as uneven stirring, insufficient reaction, and even liquid splashing. Therefore, in order to improve experimental efficiency and ensure the reproducibility of results, developing a stable and efficient method to identify abnormal conditions during stirring has become an indispensable key link in current chemical experimental research. Through this method, the problems caused by insufficient manual monitoring for a long time can be effectively solved to ensure the smooth progress of the experiment. Summary of the Invention

[0003] The embodiment of the present invention provides an abnormality recognition algorithm based on stirring magnets and vortex states to solve the problems in the prior art.

[0004] On the one hand, an embodiment of the present invention provides an abnormality recognition algorithm based on stirring magnets and vortex states, including:

[0005] Collect video stream data of the reaction instrument in the horizontal and downward directions;

[0006] Extract the ROI area in the horizontal direction and the ROI area in the downward direction;

[0007] Performing vortex presence detection according to the ROI area in the top-view direction;

[0008] Perform vortex center offset detection according to the ROI area in the top-view direction;

[0009] Performing a center offset detection of the stirring magnet according to the horizontal direction ROI area;

[0010] Performing stirring abnormality detection by detecting the vortex existence, the vortex center deviation, and the stirring magnet center deviation to obtain a stirring abnormality result;

[0011] Add the abnormal stirring result to the time queue according to the time status;

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

[0013] In a possible implementation, extracting the ROI region in the horizontal direction and the ROI region in the top-down direction includes:

[0014] The ROI area image data in the horizontal direction and the ROI area image data in the top-down direction of the video stream data are obtained according to the position of the camera geometric calibration.

[0015] In a possible implementation, detecting the presence of a vortex according to the top-view ROI area includes:

[0016] The vortex presence detection adopts the MobileNetv3 deep learning network;

[0017] The MobileNetv3 deep learning network is trained based on offline data enhancement strategy, cosine annealing strategy and learning rate warm-up strategy.

[0018] Input the top-view ROI area image data into the MobileNetv3 network to obtain the vortex presence detection result;

[0019] In a possible implementation, the detecting the vortex center offset according to the top-view ROI area includes:

[0020] The YOLOv8 network for vortex center detection is trained through image enhancement technology, cosine annealing learning rate scheduling strategy, and warm-up training strategy;

[0021] Input the top-view ROI area image data 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 according to a preset threshold.

[0023] In a possible implementation, the detecting the center offset of the stirring magnet according to the horizontal direction ROI area includes:

[0024] The YOLOv8 network for stirred magneton detection was trained using image enhancement technology, cosine annealing learning rate scheduling strategy, and warm-up training strategy.

[0025] Input the horizontal direction ROI area image data into the YOLOv8 network to obtain the stirring magnetic subcenter detection result;

[0026] The stirring magnet center detection result is compared with the preset threshold value to obtain the stirring magnet center offset detection result.

[0027] In a possible implementation, the stirring abnormality detection performed by the vortex presence detection, the vortex center offset detection, and the stirring magnet center offset detection to obtain the stirring abnormality result includes:

[0028] The status of the reagent is determined by the vortex center deviation detection result and the stirring magnetic subcenter deviation detection result;

[0029] The stirring abnormality result is determined according to the state.

[0030] In a possible implementation, adding the abnormal stirring result to a 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, the first-in-first-out principle is followed. Each time a new frame time of an abnormal stirring result is added, the first old frame time of an abnormal stirring result in the time queue is discarded.

[0032] In a possible implementation, determining the abnormal state of stirring according to the number of consecutive frame times in the time queue includes:

[0033] The set of frame times that continuously exist in the time queue is listed as an abnormal state.

[0034] The present invention provides an abnormality recognition algorithm based on stirring magnets and vortex states, which has the following advantages:

[0035] (1) A top-down camera is used to monitor the center offset of the vortex on the liquid surface, and a head-up camera is used to monitor the center offset of the stirring magnet. This achieves continuous monitoring without contact, significantly improves the robustness and detection accuracy of the system, and ensures the stability and reliability of the stirring process.

[0036] (2) By comprehensively analyzing the two key features, the vortex center offset and the stirring magnet center offset, an accurate joint judgment is made to achieve stable detection of abnormal conditions. The multi-feature fusion method effectively avoids the false positives or negative negatives that may be caused by single feature judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 A flowchart of an abnormality identification algorithm based on stirring magnets and vortex states provided in an embodiment of the present application. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] Figure 1 A schematic flow chart of an abnormality identification algorithm based on stirring magnets and vortex states provided in an embodiment of the present invention; an abnormality identification algorithm based on stirring magnets and vortex states provided in an embodiment of the present invention includes:

[0041] Collect video stream data of the reaction instrument in the horizontal and downward directions;

[0042] Extract the ROI area in the horizontal direction and the ROI area in the downward direction;

[0043] Performing vortex presence detection according to the ROI area in the top-view direction;

[0044] Perform vortex center offset detection according to the ROI area in the top-view direction;

[0045] Performing a center offset detection of the stirring magnet according to the horizontal direction ROI area;

[0046] Performing stirring abnormality detection by detecting the vortex existence, the vortex center deviation, and the stirring magnet center deviation to obtain a stirring abnormality result;

[0047] Add the abnormal stirring result to the time queue according to the time status;

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

[0049] The extracting of the horizontal viewing direction ROI area and the top viewing direction ROI area includes:

[0050] The ROI area image data in the horizontal direction and the ROI area image data in the top-down direction of the video stream data are obtained according to the position of the camera geometric calibration.

[0051] Detecting the presence of a vortex according to the ROI area in the top-view direction includes:

[0052] The vortex presence detection adopts the MobileNetv3 deep learning network;

[0053] The MobileNetv3 deep learning network is trained based on offline data enhancement strategy, cosine annealing strategy and learning rate warm-up strategy.

[0054] Inputting the image data of the top-view ROI area into the MobileNetv3 network to obtain a vortex presence detection result; performing vortex center offset detection based on the top-view ROI area includes:

[0055] The YOLOv8 network for vortex center detection is trained through image enhancement technology, cosine annealing learning rate scheduling strategy, and warm-up training strategy;

[0056] Input the top-view ROI area image data 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 according to a preset threshold.

[0058] The detecting of the center offset of the stirring magnet according to the horizontal direction ROI area includes:

[0059] The YOLOv8 network for stirred magneton detection was trained using image enhancement technology, cosine annealing learning rate scheduling strategy, and warm-up training strategy.

[0060] Input the horizontal direction ROI area image data into the YOLOv8 network to obtain the stirring magnetic subcenter detection result;

[0061] The stirring magnet center detection result is compared with the preset threshold value to obtain the stirring magnet center offset detection result.

[0062] The stirring abnormality detection by the vortex presence detection, vortex center offset detection and stirring magnet center offset detection to obtain the stirring abnormality result includes:

[0063] The status of the reagent is determined by the vortex center deviation detection result and the stirring magnetic subcenter deviation detection result;

[0064] The stirring abnormality result is determined according to the state.

[0065] Adding the abnormal stirring result to the time queue according to the time state also includes:

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

[0067] The determining of the abnormal state of stirring according to the number of consecutive frame times in the time queue includes:

[0068] The set of frame times that continuously exist in the time queue is listed as an abnormal state.

[0069] For example, a continuous video stream of the liquid surface change process is collected in real time by an industrial camera with a top-down perspective. Based on the geometric position relationship between the reagent bottle mouth and the camera, the reagent bottle mouth area in the current video frame is directly intercepted to remove the interference of irrelevant background information, so that subsequent operations can focus on the changes in the liquid surface. At the same time, the center point P of the reagent bottle mouth area is calculated. v_center .

[0070] MobileNetv3 is a lightweight convolutional network that captures abstract semantic information about liquid regions while maintaining computational efficiency, serving as deep semantic features for accurate target identification. The MobileNetv3 deep learning network for vortex target detection is trained using an offline data augmentation strategy, cosine annealing, and a learning rate warmup strategy.

[0071] The MobileNetv3 deep learning network detects whether vortices exist. If no vortices exist, stirring is not in progress or has just begun, and the process returns to the video extraction stage. If vortices exist, stirring has begun and stabilized, and detection continues.

[0072] Vortex center detection uses the YOLOv8 network model, which uses a CSP architecture and a Focus module to achieve multi-scale feature fusion, combined with a PANet path aggregation network to enhance contextual information extraction. The YOLOv8 network for vortex center target detection is trained using an offline data augmentation strategy, a cosine annealing strategy, and a learning rate warmup strategy.

[0073] The vortex center is detected based on the YOLOv8 model for the ROI area image data in the top view direction. The target box center P detected by the network v_pred , which is the center of the vortex.

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

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

[0076] Based on the calculated vortex offset d v , and compare it with the preset threshold d v_th Compare. If d v <d v_th , indicating that the status is normal, then the judgment is 0. If d v ≥d v_th , indicating that the state is abnormal, then it is judged as 1

[0077] Continuous video stream data of the internal changes of the liquid is collected in real time through an industrial camera with a head-up perspective.

[0078] Based on the geometric position relationship between the reagent bottle and the camera, the reagent bottle mouth area in the current video frame is directly intercepted to remove the interference of irrelevant background information, so that subsequent operations can focus on the changes inside the liquid. At the same time, the center point P of the stirring magnet under normal conditions is set s_center , which is the center of the bottom of the reagent bottle.

[0079] Magnetic sub-target detection uses the YOLOv8 network model. The YOLOv8 network uses a CSP architecture and a Focus module to achieve multi-scale feature fusion, combined with a PANet path aggregation network to enhance contextual information extraction. The YOLOv8 network for magnetic sub-target detection is trained using an offline data augmentation strategy, a cosine annealing strategy, and a learning rate warm-up strategy.

[0080] Based on the YOLOv8 network for magnetic targets, the image data of the ROI area in the horizontal direction is subjected to stirring magnetic detection. The target box center P detected by the network s_pred , which 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 bottle s .

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

[0083] Based on the calculated center offset d of the stirring magnet s , and compare it with the preset threshold d s_th For comparison. If d s <d s_th , indicating that the status is normal, the judgment is 0. If d s ≥d s_th , indicating that the status is abnormal, and the judgment is 1.

[0084] Based on the above calculation results, if at least one abnormal state result occurs at the vortex center or the stirring magnetic subcenter, the time of the current state result is added to the time queue. When the time queue is full, the first-in-first-out principle is followed, and each time a new frame time is added, the first old frame time in the time queue is discarded.

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

[0086] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including all changes and modifications that fall within the scope of the present invention and the preferred embodiments.

[0087] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. An abnormality recognition algorithm based on stirring magnets and vortex states, characterized in that: include: Collect video stream data of the reaction instrument in the horizontal and downward directions; Extract the ROI area in the horizontal direction and the ROI area in the downward direction; Performing vortex presence detection according to the ROI area in the top-view direction; Perform vortex center offset detection according to the ROI area in the top-view direction; Performing a center offset detection of the stirring magnet according to the horizontal direction ROI area; Performing stirring abnormality detection by detecting the vortex existence, the vortex center deviation, and the stirring magnet center deviation to obtain a stirring abnormality result; Add the abnormal stirring result to the time queue according to the time status; The abnormal state of stirring is determined based on the number of consecutive frame times in the time queue.

2. The abnormality identification algorithm based on stirring magnets and vortex states according to claim 1 is characterized in that: The extracting of the horizontal viewing direction ROI area and the top viewing direction ROI area includes: The ROI area image data in the horizontal direction and the ROI area image data in the top-down direction of the video stream data are obtained according to the position of the camera geometric calibration.

3. The abnormality identification algorithm based on stirring magnets and vortex states according to claim 1 is characterized in that: Detecting the presence of a vortex according to the ROI area in the top-view direction includes: The vortex presence detection adopts the MobileNetv3 deep learning network; The MobileNetv3 deep learning network is trained based on offline data enhancement strategy, cosine annealing strategy and learning rate warm-up strategy. The ROI area image data in the top-view direction is input into the MobileNetv3 network to obtain the vortex presence detection result.

4. The abnormality identification algorithm based on stirring magnets and vortex states according to claim 1 is characterized in that: The detecting of vortex center offset according to the top-view ROI area includes: The YOLOv8 network for vortex center detection is trained through image enhancement technology, cosine annealing learning rate scheduling strategy, and warm-up training strategy; Input the top-view ROI area image data into the YOLOv8 network to obtain the vortex center detection result; The vortex center offset detection result is obtained by comparing the vortex center detection result according to a preset threshold.

5. The abnormality identification algorithm based on stirring magnets and vortex states according to claim 1 is characterized in that: The detecting of the center offset of the stirring magnet according to the horizontal direction ROI area includes: The YOLOv8 network for stirred magneton detection was trained using image enhancement technology, cosine annealing learning rate scheduling strategy, and warm-up training strategy. Input the horizontal direction ROI area image data into the YOLOv8 network to obtain the stirring magnetic subcenter detection result; The stirring magnet center detection result is compared with the preset threshold value to obtain the stirring magnet center offset detection result.

6. The abnormality identification algorithm based on stirring magnets and vortex states according to claim 1 is characterized in that: The stirring abnormality detection by the vortex presence detection, vortex center offset detection and stirring magnet center offset detection to obtain the stirring abnormality result includes: The status of the reagent is determined by the vortex center deviation detection result and the stirring magnetic subcenter deviation detection result; The stirring abnormality result is determined according to the state.

7. The abnormality identification algorithm based on stirring magnets and vortex states according to claim 1 is characterized in that: Adding the abnormal stirring result to the time queue according to the time state also includes: The total length of the time queue is N. When the time queue is full, the first-in-first-out principle is followed. Each time a new frame time of an abnormal stirring result is added, the first old frame time of an abnormal stirring result in the time queue is discarded.

8. The abnormality identification algorithm based on stirring magnets and vortex states according to claim 7 is characterized in that: The determining of the abnormal state of stirring according to the number of consecutive frame times in the time queue includes: The set of frame times that continuously exist in the time queue is listed as an abnormal state.

Images