Laboratory hazard source identification system and method based on machine vision

Abstract

The invention provides a laboratory hazard source identification system and method based on machine vision. The laboratory hazard source identification system based on machine vision comprises a BIM modeling module, an algorithm training module, a rule engine module and an early warning module. The laboratory hazard source identification method based on machine vision comprises the following steps: S1, establishing a laboratory hazard source model; s2, obtaining a training set and a test set; s3, training and evaluating the target detection algorithm model; s4, using the trained target detection algorithm model to identify a hazard source; according to the method provided by the invention, the problem that in the prior art, a new dangerous source in the laboratory cannot be input and identified in time is solved, corresponding control measures can be taken in time, and the safety management of the laboratory and the risk assessment accuracy of the dangerous source of the laboratory are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a laboratory hazard source identification system and method based on machine vision. Background Art

[0002] At present, the management of laboratory hazards mainly relies on laboratory administrators to regularly enter the system and manually enter the hazard information into the system. However, this static management method cannot adapt to changes in the laboratory environment in a timely manner, especially when new hazards are brought about by experiments conducted by different research groups. It is impossible to enter and identify new hazards in the laboratory in a timely manner, resulting in serious lags in laboratory safety management, inaccurate risk assessment of laboratory hazards, lack of dynamic management and comprehensive risk assessment, and difficulty in taking corresponding control measures in a timely manner. This may cause laboratory personnel to face potential safety risks and pose a threat to the safety of the laboratory. Summary of the invention

[0003] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a laboratory hazard source identification system and method based on machine vision.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A laboratory hazard identification system based on machine vision, including a BIM modeling module, an algorithm training module, a rule engine module, and an early warning module; The BIM modeling module is used to create a three-dimensional model of the laboratory; The algorithm training module trains the target detection algorithm model based on the test set and training set obtained by the simulation software; The rule engine module includes defined usage rules for dangerous sources, compares the usage rules for dangerous sources with the usage methods of known dangerous sources in the acquired surveillance video, and determines whether the usage methods of known dangerous sources in the acquired surveillance video comply with the usage rules. If not, the early warning module is triggered; The early warning module warns of dangerous sources and operations that do not comply with dangerous source usage rules.

[0005] The present invention also proposes a laboratory hazard source identification method based on machine vision, comprising the following steps: S1: Establish laboratory hazard source model; The following sub-steps are included: S11: Collect hazard source information; Engineers collect information on various types of hazards in the laboratory, including the name, type, storage requirements, images, etc. of the hazard source; the types of hazard sources include high-temperature equipment, high-pressure equipment, low-temperature equipment, flammable and explosive substances, toxic and radioactive substances, corrosive substances, etc.; The high-temperature equipment includes muffle furnaces, autoclaves, etc.; the high-pressure equipment includes high-pressure reactors, hydrogen cylinders, nitrogen cylinders, etc.; the low-temperature equipment includes liquid nitrogen tanks, etc.; the flammable and explosive substances include flammable gases, flammable chemicals, etc.; the toxic and radioactive substances include sulfuric acid, hydrochloric acid, sodium hydroxide solution, radioactive substances, etc.; the corrosive substances include strong acids, strong alkalis and other chemicals; The collected hazard source information is saved in a hazard source information database, wherein the hazard source information database is bound to a laboratory hazard source identification system based on machine vision; S12: Create a three-dimensional model of the laboratory; Use a laser scanner to scan the laboratory and obtain the 3D point cloud data of the laboratory; the BIM modeling module creates a 3D model of the laboratory based on the obtained 3D point cloud data; S13: Generate laboratory hazard source model; Based on the hazard source information collected in step S1, the hazard sources included in the laboratory three-dimensional model are marked to obtain a laboratory hazard source model.

[0006] S2: Get training set and test set; The engineer imports the laboratory hazard source model obtained in step S13 into the simulation software, and the simulation software simulates the hazard source of the laboratory and generates a simulation video; The simulation software includes ANSYS, Unity 3D, etc.; the scenarios include normal operation scenarios, illegal operation scenarios, accident scenarios, etc.; the normal operation scenarios refer to the normal operation of equipment and hazardous sources according to standard operating procedures; the illegal operation scenarios refer to the simulation of operators not operating according to the specifications, such as overheating equipment, wrong operation of chemicals, etc.; the accident scenarios include chemical leakage, etc.; The generated simulation video is disassembled into n video clips in chronological order, and each video clip is labeled, where n is a positive integer, and the hazard sources and corresponding scenes contained therein are labeled; the labeled video clips are saved in the data set and randomly divided into training set and test set according to the actual preset ratio of the engineer; S3: Train and evaluate the target detection algorithm model; Use the training set to train the target detection algorithm model. After the training is completed, the target detection algorithm model is evaluated through the test set. Specifically, the training script of the target detection algorithm model YOLOv8 is configured through the YOLOv8 library. After the configuration is completed, the algorithm training module uses the video clips in the training set to train the target detection algorithm model YOLOv8; after the training is completed, the test set is used to verify whether the average precision map and recall rate Recall of the target detection algorithm model YOLOv8 on the test set meet the standards. If the average precision map is greater than 95% and the recall rate Recall is greater than 99%, the target detection algorithm model YOLOv8 meets the standards; otherwise, it does not meet the standards, and training is performed again by adjusting the parameters in the target detection algorithm model YOLOv8, increasing the number of video clips in the training set, etc., until the target detection algorithm model YOLOv8 meets the standards.

[0007] S4: Use the trained target detection algorithm model to identify the source of danger; There are cameras installed in the lab; By training the target detection algorithm model that meets the standards, the surveillance video acquired by the camera in the laboratory in real time is tested to detect the hazardous sources contained in the laboratory; The specific detection methods are as follows: If the object detection algorithm model detects that the items in the laboratory are exactly the same as the hazardous sources included in the hazardous source information database, they are directly identified as known hazardous sources; the rule engine module is triggered, and the rule engine module compares the usage methods of the known hazardous sources in the acquired surveillance video according to the defined usage rules of the hazardous sources. If they do not meet the preset usage rules, the early warning module is triggered to remind; If the target detection algorithm model detects that the items in the laboratory are not completely the same as the hazards included in the hazard source information database, the image of the item in the surveillance video is extracted, and the similarity between the image of the item and the image of the hazard source included in the hazard source information database is calculated in sequence through the deep learning architecture to obtain the similarity score between each item image and each hazard source image in the hazard source information database; based on the similarity score, the hazard source and its category that are most similar between the item and the hazard source information database are obtained; Engineers set similarity thresholds in advance based on actual needs; If the similarity score is higher than the similarity threshold, the item is judged to be a potential new hazard source, and the hazard source is added to the database and the early warning module issues an early warning reminder, thereby realizing dynamic management of the hazard source information database.

[0008] Compared with the prior art, the present invention has the following beneficial effects: The method provided by the present invention models a laboratory through BIM, marks the hazard sources in the laboratory model, obtains the laboratory hazard source model, obtains a simulation video of the laboratory hazard source through simulation software, disassembles and marks the simulation video, and divides it into a training set and a test set, and trains and evaluates the target detection algorithm model through the training set and the test set; finally, the trained target detection algorithm model is used to detect the hazard sources contained in the laboratory; when the identified object is different from the hazard source contained in the database, the similarity between the object and the hazard source in the database is calculated in turn, and according to the similarity score, the most similar hazard source and its category in the object and the hazard source information database are obtained; and according to the similarity score, it is judged whether it is a potential new hazard source, and if so, the hazard source is added to the database to realize dynamic management of the hazard source information database.

[0009] The present invention solves the problem in the prior art that new hazardous sources in the laboratory cannot be timely entered and identified, facilitates timely adoption of corresponding control measures, and improves the accuracy of laboratory safety management and risk assessment of laboratory hazardous sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 The present invention is a flowchart of the steps of a laboratory hazard source identification method based on machine vision. DETAILED DESCRIPTION

[0011] In order to provide a further understanding of the purpose, structure, features, and functions of the present invention, the following detailed description is given in conjunction with the embodiments.

[0012] A laboratory hazard identification system based on machine vision, including a BIM modeling module, an algorithm training module, a rule engine module, and an early warning module; The BIM modeling module is used to create a three-dimensional model of the laboratory; The algorithm training module trains the target detection algorithm model based on the test set and training set obtained by the simulation software; The rule engine module includes defined usage rules for hazardous sources, including the number of gas cylinders stacked, the number of fire extinguishers, the door status of hazardous chemicals cabinets, etc., and determines the hazardous sources based on the usage rules; The early warning module warns of dangerous sources and operations that do not comply with dangerous source usage rules.

[0013] like Figure 1 As shown, the present invention also proposes a laboratory hazard source identification method based on machine vision, comprising the following steps: S1: Establish laboratory hazard source model; The following sub-steps are included: S11: Collect hazard source information; Engineers collect information on various types of hazards in the laboratory, including the name, type, storage requirements, images, etc. of the hazard source; the types of hazard sources include high-temperature equipment, high-pressure equipment, low-temperature equipment, flammable and explosive substances, toxic and radioactive substances, corrosive substances, etc.; The high-temperature equipment includes muffle furnaces, autoclaves, etc.; the high-pressure equipment includes high-pressure reactors, hydrogen cylinders, nitrogen cylinders, etc.; the low-temperature equipment includes liquid nitrogen tanks, etc.; the flammable and explosive substances include flammable gases, flammable chemicals, etc.; the toxic and radioactive substances include sulfuric acid, hydrochloric acid, sodium hydroxide solution, radioactive substances, etc.; the corrosive substances include strong acids, strong alkalis and other chemicals; The collected hazard source information is saved in a hazard source information database, wherein the hazard source information database is bound to a laboratory hazard source identification system based on machine vision; S12: Create a three-dimensional model of the laboratory; Use a laser scanner to scan the laboratory and obtain the 3D point cloud data of the laboratory; the BIM modeling module creates a 3D model of the laboratory based on the obtained 3D point cloud data; S13: Generate laboratory hazard source model; Based on the hazard source information collected in step S1, the hazard sources included in the laboratory three-dimensional model are marked to obtain a laboratory hazard source model.

[0014] Further, collect the hazard source image contained in the laboratory hazard source model, obtain the width, height, pixel value of the hazard source image and the width, height, pixel value of the hazard source image collected by the camera; The similarity between the hazard source image included in the laboratory hazard source model and the actual hazard source image taken by the camera is calculated by a similarity comparison formula; The similarity comparison formula is as follows: ; Among them, MES is the similarity, , are the hazard source images included in the laboratory hazard source model and the actual hazard source images taken by the camera, m and n are the width and height of the image, respectively. , For images , The pixel value at position (i, j); Only when the mean square error of the two images is less than 0.05, the accuracy of the model meets the requirements. Otherwise, the BIM modeling module adjusts the parameters to recreate the laboratory model until it meets the requirements.

[0015] Improve the accuracy of the model through similarity calculation.

[0016] S2: Get training set and test set; The engineer imports the laboratory hazard source model obtained in step S13 into the simulation software, and the simulation software simulates the hazard source of the laboratory and generates a simulation video; The simulation software includes ANSYS, Unity 3D, etc.; the scenarios include normal operation scenarios, illegal operation scenarios, accident scenarios, etc.; the normal operation scenarios refer to the normal operation of equipment and hazardous sources according to standard operating procedures; the illegal operation scenarios refer to the simulation of operators not operating according to the specifications, such as overheating equipment, wrong operation of chemicals, etc.; the accident scenarios include chemical leakage, etc.; The generated simulation video is disassembled into n video clips in chronological order, and each video clip is labeled, where n is a positive integer, and the hazard sources and scenes contained in it are labeled; the labeled video clips are saved in the data set and randomly divided into training set and test set according to the actual preset ratio of the engineer; S3: Train and evaluate the target detection algorithm model; Use the training set to train the target detection algorithm model. After the training is completed, the target detection algorithm model is evaluated through the test set. Specifically, the training script of the target detection algorithm model YOLOv8 is configured through the YOLOv8 library. After the configuration is completed, the algorithm training module uses the video clips in the training set to train the target detection algorithm model YOLOv8; after the training is completed, the test set is used to verify whether the average precision map and recall rate Recall of the target detection algorithm model YOLOv8 on the test set meet the standards. If the average precision map is greater than 95% and the recall rate Recall is greater than 99%, the target detection algorithm model YOLOv8 meets the standards; otherwise, it does not meet the standards, and training is performed again by adjusting the parameters in the target detection algorithm model YOLOv8, increasing the number of video clips in the training set, etc., until the target detection algorithm model YOLOv8 meets the standards.

[0017] S4: Use the trained target detection algorithm model to identify the source of danger; There are cameras installed in the lab; By training the target detection algorithm model that meets the standards, the surveillance video acquired by the camera in the laboratory in real time is tested to detect the hazardous sources contained in the laboratory; The specific detection methods are as follows: If the object detection algorithm model detects that the items in the laboratory are exactly the same as the hazardous sources included in the hazardous source information database, they are directly identified as known hazardous sources; the rule engine module is triggered, and the rule engine module compares the usage methods of the known hazardous sources in the acquired surveillance video according to the defined usage rules of the hazardous sources. If they do not meet the preset usage rules, the early warning module is triggered to remind; If the target detection algorithm model detects that the items in the laboratory are not completely the same as the hazards included in the hazard source information database, the image of the item in the surveillance video is extracted, and the similarity between the image of the item and the image of the hazard source included in the hazard source information database is calculated in sequence through the deep learning architecture to obtain the similarity score between each item image and each hazard source image in the hazard source information database; based on the similarity score, the hazard source and its category that are most similar between the item and the hazard source information database are obtained; Engineers set similarity thresholds in advance based on actual needs; If the similarity score is higher than the similarity threshold, the item is judged to be a potential new hazard source, and the hazard source is added to the database and the early warning module issues an early warning reminder, thereby realizing dynamic management of the hazard source information database.

[0018] After receiving the early warning reminder, the staff will take corresponding control measures in a timely manner.

[0019] This method reduces the potential safety risks faced by laboratory personnel and improves the safety of the laboratory.

[0020] The present invention has been described by the above-mentioned relevant embodiments, however, the above-mentioned embodiments are only examples for implementing the present invention. It must be pointed out that the disclosed embodiments do not limit the scope of the present invention. On the contrary, changes and modifications made without departing from the spirit and scope of the present invention are all within the scope of patent protection of the present invention.

Claims

1. A method for identifying laboratory hazard sources based on machine vision, characterized in that: The following steps are involved: S1: Establish laboratory hazard source model; The following sub-steps are included: S11: Collect hazard source information; S12: Create a three-dimensional model of the laboratory; S13: Generate laboratory hazard source model; S2: Get training set and test set; The laboratory hazard source model obtained in step S13 is imported into the simulation software, and the simulation software simulates the hazard source of the laboratory and generates a simulation video; Disassemble and annotate the simulation video, and divide it into training set and test set according to the ratio preset by the engineer; S3: Train and evaluate the target detection algorithm model; Use the training set to train the target detection algorithm model. After the training is completed, the target detection algorithm model is evaluated through the test set. S4: Use the trained target detection algorithm model to identify the source of danger; Cameras are installed in the laboratory. By training a qualified target detection algorithm model, the surveillance video acquired by the camera in the laboratory in real time is detected to detect the hazardous sources contained in the laboratory. If the object detection algorithm model detects that the items in the laboratory are exactly the same as the hazardous sources included in the hazardous source information database, they are directly identified as known hazardous sources; the rule engine module is triggered, and the rule engine module compares the usage methods of the known hazardous sources in the acquired surveillance video according to the defined usage rules of the hazardous sources. If they do not meet the preset usage rules, the early warning module is triggered to remind; If the target detection algorithm model detects that the items in the laboratory are not completely the same as the hazards included in the hazard source information database, the image of the item in the surveillance video is extracted, and the similarity between the image of the item and the image of the hazard source included in the hazard source information database is calculated in sequence through the deep learning architecture to obtain the similarity score between each item image and each hazard source image in the hazard source information database; based on the similarity score, the hazard source and its category that are most similar between the item and the hazard source information database are obtained; Engineers set similarity thresholds in advance based on actual needs; If the similarity score is higher than the similarity threshold, the object is determined to be a potential new hazard source.

2. The method for identifying laboratory hazards based on machine vision according to claim 1, characterized in that: The specific content of step S1 is as follows: S11: Collect hazard source information; Engineers collect information on various types of hazardous sources in the laboratory, including the name, type, storage requirements, and image of the hazardous source; the types of hazardous sources include high-temperature equipment, high-pressure equipment, low-temperature equipment, flammable and explosive substances, toxic and radioactive substances, and corrosive substances; The collected hazard source information is saved in a hazard source information database, wherein the hazard source information database is bound to a laboratory hazard source identification system based on machine vision; S12: Create a three-dimensional model of the laboratory; Use a laser scanner to scan the laboratory and obtain the 3D point cloud data of the laboratory; the BIM modeling module creates a 3D model of the laboratory based on the obtained 3D point cloud data; S13: Generate laboratory hazard source model; Based on the hazard source information collected in step S1, the engineer marks the hazard sources included in the laboratory three-dimensional model to obtain a laboratory hazard source model.

3. The method for identifying laboratory hazards based on machine vision according to claim 1, characterized in that: In step S2, the scenarios include normal operation scenarios, illegal operation scenarios, and accident scenarios; the normal operation scenarios refer to the normal operation of equipment and hazardous sources according to standard operating procedures; the illegal operation scenarios refer to the simulated operator's non-standard operation; the accident scenarios refer to the equipment, hazardous sources, or operating environment experiencing abnormalities, failures, or loss of control, causing the equipment to deviate from the normal state, resulting in safety accidents or potential dangers; The generated simulation video is decomposed into n video clips in chronological order, and each video clip is labeled, where n is a positive integer, and the hazard sources and corresponding scenes contained therein are labeled; the labeled video clips are saved in the data set and randomly divided into training set and test set according to the actual preset ratio of the engineer.

4. The method for identifying laboratory hazards based on machine vision according to claim 1, characterized in that: In step S3, the training script of the target detection algorithm model YOLOv8 is configured through the YOLOv8 library. After the configuration is completed, the algorithm training module uses the video clips in the training set to train the target detection algorithm model YOLOv8; After the training is completed, the test set is used to verify whether the average precision map and recall rate Recall of the target detection algorithm model YOLOv8 on the test set meet the standards. If the average precision map is greater than 95% and the recall rate Recall is greater than 99%, the target detection algorithm model YOLOv8 meets the standards; otherwise, it does not meet the standards. The target detection algorithm model YOLOv8 is trained again by adjusting the parameters in the target detection algorithm model YOLOv8 and increasing the number of video clips in the training set until the target detection algorithm model YOLOv8 meets the standards.

5. The method for identifying laboratory hazards based on machine vision according to claim 1, characterized in that: In step S4, while the hazard source is added to the database, the early warning module issues an early warning reminder, thereby realizing dynamic management of the hazard source information database.

6. A laboratory hazard source identification system based on machine vision, used to implement the laboratory hazard source identification method based on machine vision according to any one of claims 1 to 5, characterized in that: Including BIM modeling module, algorithm training module, rule engine module, and early warning module; The BIM modeling module is used to create a three-dimensional model of the laboratory; The algorithm training module trains the target detection algorithm model based on the test set and training set obtained by the simulation software; The rule engine module includes defined usage rules for dangerous sources, compares the usage rules for dangerous sources with the usage methods of known dangerous sources in the acquired surveillance video, and determines whether the usage methods of known dangerous sources in the acquired surveillance video comply with the usage rules. If not, the early warning module is triggered; The early warning module warns of dangerous sources and operations that do not comply with dangerous source usage rules.

Images