CN118094461A - Multi-information fusion laboratory monitoring system and method - Google Patents

Abstract

The invention relates to the technical field of laboratory monitoring, in particular to a multi-information fusion laboratory monitoring system and method. The core of the invention is a multi-mode data processing module, which fuses various laboratory environment data and labeling data of a YOLO model. When a potential hazard is detected, the emergency response module initiates an alarm and performs preset emergency measures. The invention realizes the laboratory monitoring of multi-information fusion, improves the identification, prediction and response capability of potential safety threat of the laboratory by fusing various information, applying advanced data processing algorithm and integrating self-adaptive learning mechanism, and provides comprehensive and effective safety guarantee for various laboratory environments.

Description

Multi-information fusion laboratory monitoring system and method

Technical Field

The invention relates to application of a deep learning algorithm and a multi-information fusion technology in the field of laboratory environment monitoring. In particular, the invention relates to a multi-information fusion laboratory monitoring system and method.

Background

An open laboratory means that a student, teacher, etc. can enter the laboratory at any time to learn and study. However, since devices and instruments in a laboratory generally need special rights to be used, students and the like do not have supervision in the experiment, and face a certain potential safety hazard. In addition, environmental factors in the laboratory may also have an effect on the experimental results, such as temperature, humidity, illumination, etc. Therefore, how to effectively monitor and manage an open laboratory becomes a problem to be solved.

At present, although some laboratories have adopted some traditional monitoring systems to monitor safety conditions in the laboratory, such as installing cameras, access control systems, and the like. However, these systems have some problems. Firstly, the monitoring range of the traditional monitoring system is limited, and personnel and equipment in a laboratory cannot be monitored comprehensively; secondly, the monitoring mode of the traditional monitoring system is single, the monitoring can be realized only through video monitoring, the monitoring content is single, only a certain part of the monitoring content can be monitored, and the environmental factors in a laboratory can not be monitored in real time; finally, traditional monitoring systems are complex to operate, require specialized personnel to maintain and manage, and many potential preventable potential safety hazards cannot be found in time.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention aims to provide a multi-information fusion laboratory monitoring system and method, which are used for solving the problems that the monitoring mode and content in the prior art are single, the monitoring range is limited, real-time monitoring cannot be realized, and preventable potential safety hazards are not found in time.

The multi-information fusion laboratory monitoring system disclosed by the invention is based on a deep learning algorithm and a multi-information fusion technology, and can realize real-time environment monitoring and abnormal event detection in a laboratory.

In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:

A multi-information fusion laboratory monitoring system comprising:

the sensor module is used for collecting laboratory environment data;

the video acquisition module acquires laboratory videos to capture the dynamics in a laboratory;

The YOLO model receives the data of the video acquisition module and carries out depth annotation on laboratory video content;

The multi-mode data processing module is used for receiving the environment data and the labeling data of the YOLO model, carrying out abnormality judgment, and obtaining an abnormality signal when the judgment result is abnormal;

the emergency response module is used for receiving the abnormal signal and executing emergency response;

And the network communication module is used for receiving all data of the multi-mode data processing module so as to realize data transmission and communication between the laboratory and the remote monitoring center.

In one embodiment, the sensor module monitors multiple physical quantities of the laboratory environment to accurately capture various environmental changes within the laboratory, the sensor module including a temperature sensor, a humidity sensor, a sound sensor, and a chemical sensor; the temperature sensor is deployed in a region where temperature change is likely to occur in a laboratory; the humidity sensor is deployed in a region sensitive to humidity change in a laboratory; the chemical sensor may be deployed in a laboratory in areas where chemical leakage may occur; the sound sensor is deployed in a region where sound accompanied accidents are likely to occur in a laboratory; the video acquisition module is deployed in the whole laboratory and is independently deployed at a laboratory gate and a laboratory bench to accurately capture various dynamics in the laboratory.

In one embodiment, the YOLO model captures laboratory video as images and performs depth annotation based on a variety of behavioral video or image data under different environmental conditions, including: identification of target behavior, time stamping, and recording of environmental conditions; training by using the marked images, and continuously adjusting model parameters so that the model can be suitable for various changeable laboratory scenes; and after model training is completed, the image of each triggering abnormal signal is incorporated into the newly acquired data, and annotation training is carried out to update the iterative model.

In one embodiment, the multi-mode data processing module comprises a data interface unit, a data preprocessing unit, a data fusion unit, a mode identification unit, a decision logic unit and a communication interface unit;

The data interface unit receives multi-source data streams and performs time stamp synchronization on each data stream; wherein the multi-source data comprises the environmental data and annotation data of the YOLO model;

the data preprocessing unit performs standardization processing on all data, converts the data into a uniform data format, and performs noise filtering and outlier detection;

The data fusion unit adopts a weighted average method, a Kalman filter and a neural network algorithm to deeply integrate the preprocessed data;

The pattern recognition unit analyzes the integrated data by adopting a deep learning model machine algorithm and divides the data into normal and abnormal;

The decision logic unit generates a corresponding response signal according to a preset safety threshold value and a safety rule to trigger the emergency response module, and transmits data to the network communication module through the communication interface unit, so that external alarm transmission is realized.

In one embodiment, the YOLO model, the identifying of the target behavior includes: devices in the video or image use non-canonical behavior or experimental operator's operation non-canonical behavior; transmitting data by taking the position coordinates and the size of the target behavior appearing in the video or the image as transmission data;

The data preprocessing unit converts laboratory environment data into a unified numerical value or text format, and converts labeling data provided by the YOLO model into a unified image format;

The data fusion unit integrates laboratory environment data by adopting a Kalman filter, and reduces the influence of random errors and accidental errors by adopting a weighted average method; a neural network algorithm is used to identify potential abnormal behavior or environmental changes from the data provided by the YOLO model, in combination with laboratory environmental data, to identify whether there is an abnormal behavior or environmental risk.

In one embodiment, the pattern recognition unit analyzes the integrated data using a deep learning model machine algorithm, including:

Laboratory environmental data analysis: laboratory environmental data is used to monitor laboratory environmental conditions, training a deep learning model to identify abnormal patterns in the data that indicate significant changes or potential hazards to the environmental conditions;

and (3) analyzing labeling data of the YOLO model: the deep learning model is trained to analyze the annotation data to identify potential abnormal behaviors or environmental changes, and whether actual security threats exist or not is determined jointly by combining laboratory environment data analysis results.

In one embodiment, the setting of the safety threshold refers to a preset limit value for a specific parameter, and exceeding the limit value indicates that a safety risk exists; the safety rule is a logic guiding principle that the decision logic unit judges whether to trigger an alarm; the security rules are based on a combination of laboratory environment data and labeling data of the YOLO model.

In one embodiment, the emergency response module comprises an alarm mechanism and an automatic control mechanism, and the signal receiving unit of the emergency response module is configured to receive the response signal generated by the decision logic unit, and is responsible for decoding and confirming the received signal and determining the priority and type of the received signal; after confirming the abnormal signal, triggering the alarm mechanism and starting the automatic control mechanism to implement a series of predefined safety measures.

In one embodiment, the network communication module is equipped with NB-IoT, wiFi communication technology, and communication security and protocol management, data transmission and communication between the laboratory and a remote monitoring center, comprising:

OneNET cloud platform connected to the network communication module for realizing data interaction and information sharing between the laboratory and the remote monitoring center, including data storage and history data functions;

the mailbox reminder is connected with the network communication module, and when the multi-mode data processing module transmits an abnormal signal to the emergency response module, a mail is sent to a preset mailbox;

and the mobile phone APP is connected with the network communication module and receives data from the network communication module so as to check the laboratory condition at any time.

In one embodiment, the OneNET cloud platform includes a data storage unit and a historical data analysis unit, and utilizes the stored historical data and real-time data to predict dangerous trend through an RBF neural network model constructed on the platform;

The data storage unit is used for storing all data from the multi-mode data processing module by adopting a distributed database architecture;

the historical data analysis unit adopts an RBF neural network model to conduct data analysis and prediction, and captures complex relations and potential dangerous signals in data; the RBF neural network model firstly carries out data preprocessing, and real-time and historical data are subjected to cleaning, normalization and transformation processing to adapt to the input requirements of the model; then extracting features, namely extracting key features from the preprocessed data through statistical analysis and a machine learning algorithm, wherein the features can represent the state and the change trend of the monitoring environment; training the RBF neural network model, adjusting network weight and bias parameters, and optimizing the prediction accuracy of the model; finally, outputting prediction, analyzing new data input by the RBF model after training, predicting dangerous trend possibly occurring in the future, and feeding back the prediction result to a laboratory safety monitoring system for early warning and risk management;

the dangerous trend prediction comprises trend prediction of dangerous behaviors of common dangerous goods or action states in a laboratory.

The invention also provides a monitoring method of the laboratory monitoring system based on the multi-information fusion, which comprises the following steps:

The video acquisition module acquires data, and performs target detection by acquiring laboratory videos and using a pre-trained YOLO model; meanwhile, collecting laboratory environment data through the sensor module, and monitoring all environmental data in real time;

judging whether dangerous actions are contained in laboratory videos through a YOLO model, if so, firstly alarming, and then transmitting information into a data communication thread in a network communication module; meanwhile, the alarm information is subjected to video recording, the dangerous actions are stored in a material database, the materials in the database are marked, a YOLO model is used for training, the model is continuously updated, each time image information is acquired, if the dangerous actions are recorded, the model is updated, and the data acquired by the next video acquisition module are judged through the new model; if not, directly transmitting the information into a data center; if the laboratory environment data exceeds a set threshold value, alarming is carried out, and then the data is transmitted to a data center; if the set threshold value is not exceeded, directly transmitting the data into a data center;

After all data are transmitted into the data communication threads in the network communication module, the data are directly transmitted to the mobile phone APP and OneNET Internet of things cloud platform, meanwhile, the data are judged, and if abnormal data exist, mail alarm information is sent on the basis of the mobile phone APP and OneNET Internet of things cloud platform.

Compared with the prior art, the invention has the beneficial effects that:

The invention realizes the multi-information fusion technology, not only monitors the environment of laboratory personnel, equipment, temperature, humidity, sound and the like in all aspects, but also adopts the YOLO model to process the collected images of the personnel, the equipment and the like, timely find abnormal conditions and carry out alarm prompt, and simultaneously, laboratory management personnel can also receive alarm information at any time in a mobile phone mailbox, a cloud platform and a mobile phone, thereby being convenient for the management personnel to check the laboratory conditions at any time and any place. The monitoring modes are wide in various ranges, and the monitoring method is not only used for monitoring video acquisition images of personnel and equipment, but also used for monitoring physical sensors of laboratories. Various machine learning algorithms are introduced to enable the system to self-optimize and learn according to the constantly changing laboratory environment. Meanwhile, the invention can store and share experimental data and results in cloud, is convenient for checking and downloading the experimental data and results anytime and anywhere, and can analyze the data archived in the data recording module in the alarm system by adopting an algorithm to discover the problems of potential safety hazards of a laboratory in time, and the like.

Drawings

Fig. 1 is a block diagram of the overall structure of a laboratory monitoring system for multi-information fusion according to an embodiment of the present invention.

FIG. 2 is a flowchart of a YOLO model implementation of a laboratory monitoring system for multi-information fusion according to an embodiment of the present invention.

Fig. 3 is a block diagram of a multi-data fusion module of a laboratory monitoring system for multi-information fusion according to an embodiment of the present invention.

Fig. 4 is a process flow diagram of a monitoring method of a laboratory monitoring system for multi-information fusion according to an embodiment of the present invention.

Detailed Description

In order that the objects and advantages of the invention will become more apparent, the invention will be further described with reference to the following examples; it should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

Preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are merely for explaining the technical principles of the present invention, and are not intended to limit the scope of the present invention.

Aiming at the problems of the traditional safety monitoring means in the laboratory, the invention provides a multi-information fusion laboratory monitoring system. The system adopts the YOLO model, can monitor personnel and equipment in a laboratory in real time, and can identify abnormal behaviors and dangerous situations. Meanwhile, the system is integrated with an environment monitoring module, can monitor environmental factors such as temperature, humidity and illumination in a laboratory in real time, automatically adjust the environmental parameters in the laboratory according to monitoring results, archive data, and analyze the running condition of the laboratory according to archived data to timely find potential safety hazards. In addition, the system also adopts a cloud computing technology, so that remote monitoring and management can be realized, and management personnel can conveniently know the condition in a laboratory at any time.

The following description sets forth a more detailed description of embodiments of the invention.

Referring to fig. 1, the multi-information fusion laboratory monitoring system of the present invention mainly includes:

the sensor module is provided with a plurality of sensors and a plurality of sensors, and is used for comprehensively and accurately monitoring the laboratory environment data and collecting the laboratory environment data.

And the video acquisition module acquires laboratory videos and ensures that the camera can accurately capture various dynamic states in the laboratory.

And the YOLO model receives the data of the video acquisition module and performs depth annotation on laboratory video content.

The multi-mode data processing module is used for receiving the environmental data and the labeling data of the YOLO model collected by the sensor module, carrying out abnormality judgment, and obtaining an abnormality signal when the judgment result is abnormal.

And the emergency response module is used for receiving the abnormal signal and executing emergency response.

And the network communication module is used for receiving all data of the multi-mode data processing module so as to realize data transmission and communication between the laboratory and the remote monitoring center.

In an embodiment of the invention, the sensor module monitors multiple physical quantities of the laboratory environment to accurately capture various environmental changes within the laboratory. The sensor module comprises a plurality of types of sensors such as temperature, humidity, sound, chemistry and the like, is reasonably distributed at each position of a laboratory, comprehensively and accurately monitors the environmental data of the laboratory, and improves the physical environment comprehensiveness of the monitoring range, thereby further realizing the monitoring of the laboratory. Specifically:

The temperature sensor is selected to be a thermocouple digital temperature sensor with high precision, is deployed in a region, such as a chemical storage region, a heating device vicinity and the like, of a laboratory, which is easy to generate temperature change, and is set to be suitable temperature according to the laboratory environment, so that the laboratory temperature can be monitored in real time and with high precision.

Humidity sensor selects high accuracy's capacitive humidity sensor, and response to environmental change is sensitive, can accurate measurement accuse ware humidity, deposits the area deployment sensitive to humidity change such as easily wet material or strict control humidity in the laboratory, sets for humidity alarm threshold according to the specific demand in laboratory, ensures that humidity sensor can provide accurate and continuous humidity.

The chemical sensor is selected for specific harmful gases such as carbon monoxide, hydrogen sulfide and the like, has high sensitivity and strong specificity, can accurately detect trace harmful gases, and can be deployed in areas where chemical leakage possibly occurs such as around a laboratory fume hood and a laboratory table, so that the harmful gases can be accurately and timely detected.

The sound sensor is high in sensitivity, the sound sensor with the capacitance microphone is selected, fine sound wave changes can be captured, a plurality of sound sensors are deployed in a region where sound accompanied accidents are likely to occur in a laboratory, and frequency response adjustment is carried out on the sensors, so that specific types of sounds such as glass breaking sounds, gas leakage sounds and the like can be identified, and the laboratory sounds can be comprehensively and accurately monitored.

In the invention, the environmental changes captured by the sensor module mainly comprise: temperature, humidity, changes in chemical concentration, abnormal sounds captured by a sound sensor, and the like.

In the embodiment of the invention, the video acquisition module adopts a high-definition camera with high definition, large visual angle and high frame rate, is deployed at the whole laboratory and the positions of a gate, a test bed and the like, can accurately capture various dynamic states in the laboratory, and improves the image picture comprehensiveness of a monitoring range, thereby further realizing the monitoring of the laboratory.

In the invention, the video acquisition module mainly captures the dynamics of experimental operation, including: movement of laboratory staff, use of laboratory equipment, operation of the experimental process, etc.

Based on the above dynamics and changes, it is possible to judge safety events and laboratory daily activities, wherein the safety events mainly comprise chemical leakage, equipment failure and other conditions possibly causing safety risks. The daily activities of the laboratory mainly comprise the entrance and exit of laboratory personnel, the normal use of equipment and the like.

In the embodiment of the invention, the training and workflow of the YOLO model is shown in fig. 2, and the model adopts YOLOv algorithm with the advantages of high precision, high speed, high robustness and the like. Firstly, capturing laboratory videos into images, and performing depth annotation on image contents based on various behavior videos or image data under different environmental conditions by adopting an advanced image annotation technology, wherein the method comprises the following steps: the identification of target behavior, time stamping, and recording of environmental conditions provides a rich set of features for the YOLOv algorithm. Training is carried out by using the marked images, and model parameters are continuously adjusted, so that the model is ensured to be suitable for various changeable laboratory scenes. The cross verification technology is adopted to ensure that the model has wide generalization capability, high recognition accuracy can be maintained on a non-training data set, the training is stopped after the model performance is stable for preventing the model from excessive fitting training, and an independent test set is adopted to evaluate the model, so that the model is ensured to have performance indexes such as high accuracy, high recall rate, high F1 score and the like. After model training is completed, the image of each triggering abnormal signal is brought into the newly acquired data, annotation training is carried out, the iteration model is updated, the application of the model to new behavior modes and environmental changes is ensured, and reliable decision support is provided for the multi-mode data fusion unit. The monitoring of the states of laboratory personnel and equipment of different types is accurately, efficiently and quickly realized, so that the monitoring of the laboratory is further realized.

Illustratively, the identifying the target behavior by the YOLO model according to the present embodiment includes: devices in the video or image use non-canonical behavior or experimental operator's operation non-canonical behavior; and transmitting data with the coordinates and size of the location where the target behavior appears in the video or image. The YOLO model can also identify a specific object in the image, such as a certain equipment of a laboratory, an experiment operator, etc. For example, a rectangular frame of XY coordinates of a worker in a laboratory image (e.g., upper left corner coordinates (50, 100), lower right corner coordinates (150, 200), and the size of the rectangular frame (100 pixels wide, 100 pixels high).

In the embodiment of the invention, the multi-mode data processing module can judge the data and transmit the data into the network communication module, and if the data exceeds a set safety threshold value, an abnormal signal is sent out at the same time, so that the emergency response module is started. The multi-mode data processing module comprises a data interface unit, a data preprocessing unit, a data fusion unit, a mode identification unit, a decision logic unit and a communication interface unit, and realizes multi-information fusion of a laboratory through the multi-mode data processing module, so that monitoring of the laboratory is further realized. As shown in fig. 3, the processing flow and functions of the multi-mode data processing module are described as follows:

first, the data interface unit receives multi-source data streams and performs time stamp synchronization on each data stream to ensure the consistency of the multi-source data in time. The multi-source data here includes environmental data from the sensor module and annotation data from the YOLO model.

And then, the data preprocessing unit performs normalization processing on all data, converts the data into a unified data format, and performs preliminary noise filtering and outlier detection. In particular, it converts laboratory environment data provided by different sensors of temperature, humidity, sound, chemical composition, etc. into a unified numerical or text format for further analysis and processing, e.g. temperature data may be converted into data within a specific numerical range, humidity data may also be processed according to the same criteria. For data provided by the YOLO model, the YOLO model is mainly used for object detection of image or video data. These data include processed images or video frames, or image features and detection results. In the normalization process, the data may be converted into a unified image format, or the detection result may be converted into a standard data format, such as JSON or XML, which contains information about the type, position, size, etc. of the object. In this way, data of different sources and types are converted into one or more uniform formats, which facilitates subsequent data fusion and analysis

And then, the data fusion unit adopts a weighted average method, a Kalman filter and a neural network algorithm to deeply integrate the preprocessed data, and the algorithm can complement the defects of the sensor when the multisource data is processed, so that the accuracy and the robustness of the data are enhanced. For example, processing of laboratory environmental data collected by the sensor module may employ Kalman filters to integrate and efficiently estimate future conditions, and weighted averaging may be employed to reduce the effects of random and occasional errors. If multiple sensors provide similar data, a Kalman filter may integrate these data sources to provide a more accurate and stable estimate. For example, if two temperature sensors provide slightly different readings, a Kalman filter may combine the readings to provide a weighted average of the temperature values based on the historical accuracy and current measurement uncertainty of each sensor, thereby reducing the effects of random and occasional errors. Labeling data for the YOLO model may identify potential abnormal behavior or environmental changes from complex data patterns using neural network algorithms to combine with laboratory environmental data to identify whether abnormal behavior or environmental risk exists. These image data can be analyzed using neural network algorithms, combined with data from other sensors such as temperature, humidity, chemical detection, etc., to identify if there is abnormal behavior or environmental risk. For example, if the YOLO model identifies a particular chemical operation in the image, and the associated chemical sensor also detects a potentially dangerous chemical leak, the neural network may fuse this information to confirm whether there is an actual safety risk in the laboratory. The algorithm can complement the defects of the sensor when processing the multi-source data, and the accuracy and the robustness of the data are enhanced. In this process, different types of data are integrated, providing more comprehensive and accurate laboratory environmental monitoring and analysis, and a good fusion of the data.

The integrated data is then analyzed using a pattern recognition unit equipped with a deep learning model machine algorithm and the data is classified as normal and abnormal so that when a new pattern, which is significantly different from the historical data, appears, it can be quickly detected and marked as a potential security threat.

And finally, the decision logic unit generates a corresponding response signal according to a preset safety threshold value and a preset safety rule to trigger the emergency response module, and transmits data to the network communication module through the communication interface unit, so that external alarm transmission is realized. Obviously, when a potential security threat occurs, an exception signal is triggered.

In an embodiment of the present invention, the pattern recognition unit analyzes the integrated data using a deep learning model machine algorithm, and specifically includes:

Laboratory environmental data analysis: laboratory environmental data of temperature, humidity, chemical composition, etc. are used to monitor laboratory environmental conditions, training a deep learning model to identify abnormal patterns in the data that indicate significant changes or potential hazards to the environmental conditions; for example, if the temperature sensor suddenly records an abnormally high temperature, the humidity sensor also shows an abnormal reading, which may indicate the occurrence of a fire or other dangerous condition. The deep learning model will compare these readings to historical data and if a significant deviation is detected, it will flag this as abnormal.

And (3) analyzing labeling data of the YOLO model: the YOLO model is used for image data and can identify and annotate various objects and behaviors within the laboratory. The deep learning model is trained to analyze the annotation data to identify potential abnormal behaviors or environmental changes, and whether actual security threats exist or not is determined jointly by combining laboratory environment data analysis results. For example, if the YOLO model identifies a dangerous chemical experimental operation in an image, where the chemical sensor also detects the release of dangerous chemicals, the deep learning model may analyze a combination of these information to determine if an actual security threat exists.

In both cases, the deep learning model will use data from different sources to perform a comprehensive analysis, comparing the data to historical patterns to identify new patterns that differ significantly from the historical data. If such a pattern is detected, the system will quickly mark it as a potential security threat. By adopting the method, the system can utilize the advantages of multi-source data, and the detection capability and response speed of potential risks are improved.

In embodiments of the present invention, safety thresholds and safety rules are based on the understanding of the system for normal and abnormal conditions, which are used to distinguish between normal laboratory operations and potentially dangerous situations. Specifically, the setting of the safety threshold refers to preset limit values of specific parameters, such as temperature, humidity, chemical substance concentration and the like, and if the preset limit values are exceeded, safety risks are indicated; for example, if the temperature detected by the temperature sensor exceeds a set safety threshold, such as 35 ℃, this indicates that there is an overheating condition in the laboratory, which may be due to equipment failure or fire, etc. The safety rule is a logic guiding principle that the decision logic unit judges whether to trigger an alarm; the security rules may be based on a combination of laboratory environment data and labeling data of the YOLO model. For example, if the YOLO model identifies a hazardous chemical operation in the image data while the chemical sensor also detects the release of hazardous chemicals, the decision logic unit will trigger an alarm based on this information.

The data fusion unit integrates the environment data and the data provided by the YOLO model. The data fusion unit not only integrates the two data respectively, but also performs double judgment during use. For example, when sensor environmental data shows a sudden rise in chemical concentration, and the YOLO model simultaneously recognizes that an area in the laboratory has irregular operation, the data fusion unit combines the information, and comprehensively determines that chemical leakage or other safety risks occur in the laboratory.

In the embodiment of the invention, the emergency response module is started when the abnormal signal is transmitted by the multi-mode data processing module, and comprises an alarm mechanism and an automatic control mechanism, and the emergency response module is provided with a signal receiving unit which is used for receiving the response signal generated by the decision logic unit, is responsible for decoding and confirming the received signal and determining the priority and the type of the received signal. After confirming the abnormal signal, a built-in alarm mechanism is triggered, including a flashing light, an audible alarm, an emergency notification on a system interface and other visual and audible warning systems. At the same time, an automatic control mechanism is started, and the mechanism implements a series of predefined safety measures, such as opening air conditioning equipment for cooling, closing a key system of a laboratory, starting a fire-fighting system and the like, so that the rapidity and the safety coefficient of monitoring of the laboratory are improved, and the monitoring of the laboratory is further realized.

In an embodiment of the invention, the emergency response module is activated once the system detects a potential hazard according to preset thresholds and rules. The system combines data from different sensors and image recognition modules, and utilizes deep learning and pattern recognition techniques to ensure rapid and accurate response to potential security threats

In the embodiment of the invention, the network communication module is provided with NB-IoT, wiFi communication technology and communication security and protocol management, and is used for realizing communication between the laboratory security monitoring system and OneNET cloud platform, and between the mobile phone APP and the mailbox, so that real-time, comprehensive and reliable laboratory data transmission is ensured, and further monitoring of a laboratory is realized. In this embodiment, data transmission and communication between the laboratory and the remote monitoring center includes:

the OneNET cloud platform is connected with the network communication module so as to realize data interaction and information sharing between the laboratory and the remote monitoring center, and the functions of data storage and historical data are included, so that an administrator can check the condition of the laboratory anytime and anywhere and timely find the potential safety hazard of the laboratory, and further realize monitoring of the laboratory. In the embodiment, a wireless transmission NB-IoT module with high penetration and wide coverage is adopted, a low-power-consumption wide area network connection is established between a system and a OneNET cloud platform, a stable communication link is established, data packets are sent and received through the NB-IoT network, and low power consumption and high efficiency of data transmission are ensured.

The mailbox reminding module is connected with the network communication module, and when the multi-mode data processing module transmits an abnormal signal to the emergency response module, the mail is sent to the preset mailbox, so that an administrator is ensured to know the situation in time, and the monitoring of a laboratory is further realized.

The mobile phone APP is connected with the network communication module, and data from the network communication module is received, so that the laboratory condition is checked at any time, the working pressure of a manager is reduced, and the laboratory is further monitored.

The WiFi communication module of the embodiment aims at the mobile phone APP and the mailbox to remind, and is used for establishing high-speed wireless local area network connection in a laboratory, so that the system can communicate with high data transmission rate in a local range, and the system is suitable for rapid uploading of real-time video streams and large-capacity data.

The communication security and protocol management of the present embodiment is used to ensure the security of data transmitted through NB-IoT and WiFi modules, encrypt, authenticate and verify the integrity of the transmitted data, and ensure that all outgoing data is protected from unauthorized access and tampering.

The OneNET cloud platform of the embodiment has the functions of data storage and historical data analysis, correspondingly comprises a data storage unit and a historical data analysis unit, and utilizes stored historical data and real-time data to predict dangerous trend through an RBF neural network model constructed on the platform.

The data storage unit adopts a distributed database architecture and is used for storing all multi-mode data from a laboratory safety monitoring system, has a data index and query mechanism, supports quick retrieval and processing of a large amount of data, and also comprises data backup and recovery functions, so that the durability and reliability of the data are ensured.

The historical data analysis unit adopts an RBF neural network model to conduct data analysis and prediction, the RBF neural network model on the OneNET cloud platform uses a radial basis function as an activation function of neurons of an hidden layer, and network parameters are trained by utilizing historical data, so that complex relations and potential dangerous signals in the data are captured. Firstly, data preprocessing is carried out, and real-time and historical data are subjected to cleaning, normalization and transformation processing to adapt to the input requirement of an RBF neural network model; then extracting features, namely extracting key features from the preprocessed data through statistical analysis and a machine learning algorithm, wherein the features can represent the state and the change trend of the monitoring environment; training the RBF neural network model by utilizing the characteristics and the historical data set, adjusting network weight and bias parameters, and optimizing the prediction accuracy of the model; and finally, outputting a prediction, analyzing new data input by the trained RBF neural network model, predicting dangerous trend possibly occurring in the future, and feeding back the prediction result to a laboratory safety monitoring system for early warning and risk management.

The dangerous trend prediction comprises trend predictions such as dangerous goods or dangerous actions of action states of some laboratories. For example:

and (3) predicting abnormal temperature, namely identifying fire, equipment faults and the like according to environmental data of a temperature laboratory, and predicting abnormal temperature change.

Humidity abnormality pre-warning, the humidity change of a laboratory is observed by analyzing the data of a humidity sensor, and the abnormal change of the humidity is predicted.

Chemical leakage or harmful gas detection, and analyzing the concentration trend of chemical substances by using an RBF neural network model to predict potential chemical leakage or harmful gas accumulation.

Personnel safety risk: using video monitoring data and other laboratory environment data, such as proximity sensors, the RBF neural network model can predict personnel security risks in a laboratory, including unauthorized access of personnel in hazardous areas or potential personnel status.

The mailbox reminder sends a mailbox to an administrator when the multi-mode data processing module transmits an abnormal signal to the emergency response module. Firstly, setting a mailbox server, and determining the address, port number, mailbox account number and other information of the mailbox server; opening an SMTP port of the mailbox server and setting an authorization code; setting mailbox reminding rules in a multi-mode data processing module, wherein the mailbox reminding rules comprise information such as receivers, triggering conditions, mail contents and the like; when the triggering condition is met, the multi-mode data processing module automatically sends mails to appointed receivers through the network communication module.

The mobile phone APP receives the data from the network communication module, so that the laboratory condition can be conveniently checked at any time. Developing an interface based on a mobile phone APP, and determining a keyword search interface address of the APP; setting a mobile phone APP reminding rule in a multi-mode data processing module, wherein the mobile phone APP reminding rule comprises information such as keywords, triggering conditions, reply content and the like; when the triggering condition is met, the multi-mode data processing module automatically sends reply content to the appointed user through the network communication module.

Referring to fig. 4, the monitoring method of the laboratory monitoring system for multi-information fusion further realizes comprehensive fusion of laboratory information, and the monitoring method is various, comprehensive and real-time, and timely discovers potential safety hazards which can be prevented by a laboratory, thereby further realizing monitoring of the laboratory. The method is divided into 6 stages as a whole, wherein a camera acquires data and a sensor acquires data as data acquisition stages, a data processing stage is judged by a YOLO model and data analysis is carried out, whether dangerous actions are judged and exceeding a threshold value is judged as a dangerous judgment stage, an alarm is an alarm and control stage, a data center transmits data and judges whether abnormal data is a data transmission stage, and a cloud platform analysis stage is obtained after the data is uploaded to a OneNET cloud platform. And the other action of the alarm is that the system carries out video recording on the alarm information, stores the dangerous action into a material database, marks the materials in the database, uses the YOLO model for training, continuously updates the model, and if the dangerous action is the dangerous action, the model is recorded and updated, and the next time the acquired data of the camera is judged through the new model. The following examples were analyzed in accordance with the five stages.

And a data acquisition stage: and acquiring data acquired by the sensor and the camera by using a sensor module and a video acquisition module of the whole system block diagram. The sensor module deploys various types of sensors at key positions of a laboratory and is used for collecting environmental data, including temperature, humidity, sound, chemical substances and the like; the video acquisition module adopts a camera with high definition, large visual angle and high frame rate, and is deployed at the positions of the whole laboratory, a doorway, a test bed and the like so as to ensure that various dynamic states in the laboratory can be comprehensively and accurately captured.

And a data processing stage: the multi-mode data fusion module using the whole system block diagram comprises a data interface unit, a data preprocessing unit, a data fusion unit, a mode identification unit, a decision logic unit and a communication interface unit.

Firstly, a data interface unit receives multi-source data streams marked by a sensor module and a YOLO model, and performs time stamp synchronization on each data stream to ensure the consistency of the multi-source data in time;

Then, the data preprocessing unit performs standardization processing on all data, converts the data into a unified data format, and performs primary noise filtering and outlier detection;

And then the data fusion unit adopts a weighted average method, a Kalman filter and a neural network algorithm to deeply integrate the data, and the algorithm can complement the defects of the sensor when the multi-source data is processed, so that the accuracy and the robustness of the data are enhanced. For example, processing of temperature and humidity sensors may employ a Kalman filter to effectively estimate future states, and image data annotated to the YOLO model may use a neural network algorithm to identify potential abnormal behavior or environmental changes from complex data patterns.

The pattern recognition unit equipped with the deep learning model machine algorithm is employed to analyze the integrated data, separating the data into normal and abnormal so that when a new pattern, which is significantly different from the historical data, occurs, it can be quickly detected and marked as a potential security threat.

When potential security threat occurs, the decision logic unit generates a corresponding response signal according to a preset security threshold and rules to trigger the emergency response module, and transmits data to the network communication module through the communication interface unit, so that external alarm transmission is realized.

And (3) a danger judging stage: and using a multi-mode data processing module of the whole system block diagram, detecting abnormal data in the multi-mode data processing module, and immediately triggering an emergency response module.

And a data transmission stage: the system comprises a network communication module, a multi-mode data processing module, a mobile phone APP and a mailbox, wherein the network communication module is used for receiving all data from the multi-mode data processing module and is used for realizing data transmission and communication between a laboratory and a remote monitoring center, and the module is provided with NB-IoT, wiFi communication technology and communication security and protocol management and is used for realizing communication between the laboratory security monitoring system and a OneNET cloud platform, the mobile phone APP and a mailbox.

The wireless transmission NB-IoT module with high penetration and wide coverage is specially used for establishing low-power consumption wide area network connection between the system and the OneNET cloud platform, establishing a stable communication link, sending and receiving data packets through the NB-IoT network, and ensuring low power consumption and high efficiency of data transmission.

The WiFi communication module is specially used for reminding a mobile phone APP and a mailbox, and is used for establishing high-speed wireless local area network connection in a laboratory, so that the system can carry out high-data-transmission-rate communication in a local range, and the system is suitable for rapid uploading of real-time video streams and large-capacity data.

Communication security and protocol management to ensure security of data transmitted through NB-IoT and WiFi modules, encrypt, authenticate and integrity check the transmitted data, ensure that all outgoing data is protected from unauthorized access and tampering.

Alarm and control phase: the emergency response module of the whole system block diagram is used, and is started when the abnormal signal is transmitted by the multi-mode data processing module, and comprises an alarm mechanism and an automatic control mechanism, wherein the alarm mechanism comprises a signal receiving unit which is used for receiving a response signal generated by a decision logic unit, and is responsible for decoding and confirming the received signal and determining the priority and the type of the received signal. After confirming the abnormal signal, a built-in alarm mechanism is triggered, including visual and audible warning systems such as flashing lights, audible alarms and emergency notifications on a system interface. At the same time, the system will activate an automatic control mechanism that implements a series of predefined safety measures, such as turning on the air conditioning equipment cool down, turning off critical systems in the laboratory, activating fire protection systems, etc.

Cloud platform analysis: and using OneNET cloud platform modules, having data storage and historical data analysis functions, and carrying out dangerous trend prediction by using stored historical data and real-time data through an RBF neural network model constructed on the platform.

The data storage adopts a distributed database architecture for storing multi-mode data from a laboratory safety monitoring system, can perform a data index and query mechanism, supports quick retrieval and processing of a large amount of data, also comprises data backup and recovery functions, and ensures the durability and reliability of the data.

And (3) historical data analysis, namely carrying out data analysis and prediction by adopting an RBF neural network model, wherein the RBF model on the OneNET cloud platform uses a radial basis function as an activation function of neurons of an hidden layer, and training network parameters by utilizing the historical data, so that complex relations and potential dangerous signals in the data are captured. Firstly, data preprocessing is carried out, and real-time and historical data are subjected to cleaning, normalization and transformation processing to adapt to the input requirement of an RBF neural network model; then extracting features, namely extracting key features from the preprocessed data through statistical analysis and a machine learning algorithm, wherein the features can represent the state and the change trend of the monitoring environment; training the RBF neural network model by utilizing the characteristics, training the RBF neural network model by utilizing a historical data set, adjusting network weight and bias parameters, and optimizing the prediction accuracy of the model; and finally, outputting a prediction, analyzing the new data input by the trained RBF model, predicting dangerous trend possibly occurring in the future, and feeding back the prediction result to a laboratory safety monitoring system for early warning and risk management.

The dangerous trend prediction comprises trend predictions such as dangerous goods or dangerous actions of action states of some laboratories. For example, temperature anomaly prediction, based on temperature laboratory environment data, the platform can identify fire, equipment failure, etc., predicting an abnormal temperature change;

Humidity abnormality early warning, namely, observing humidity changes of a laboratory by analyzing data of a humidity sensor, and predicting abnormal changes of humidity;

Chemical leakage or harmful gas detection, namely, analyzing the concentration trend of chemical substances by utilizing an RBF neural network model built on a cloud platform, and predicting potential chemical leakage or harmful gas accumulation;

Personnel safety risk: using video monitoring data and other laboratory environment data, such as proximity sensors, the RBF neural network model can predict personnel security risks in a laboratory, including unauthorized access of personnel in hazardous areas or potential personnel status.

Specific implementation of the invention application scenario example 1: laboratory chemical leakage monitoring

Sensor deployment: in the chemical storage area of the laboratory, high precision temperature, humidity and chemical sensors are deployed. Temperature and humidity sensors are used to monitor the environmental conditions of the storage area, while chemical sensors are used exclusively to detect the concentration of certain harmful gases, such as ammonia, hydrogen sulfide, etc.;

And a data acquisition stage: the sensor collects temperature, humidity and chemical gas concentration data of the storage area in real time. The data are transmitted to a multi-mode data processing module in real time for data processing for further analysis;

And a data processing stage: the multi-mode data processing module is used for receiving the data from the sensor and comprehensively processing the data. The system uses an algorithm (such as Kalman filtering) to perform data fusion by integrating data from a chemical sensor, such as gas leakage detection, and data from an environmental sensor, such as a temperature and humidity sensor, so as to accurately identify a chemical leakage event;

And (3) a danger judging stage: when the multi-mode data processing module detects abnormal data, the abnormal data is in a dangerous state, an emergency response module is immediately triggered, and an alarm and control stage is entered;

Alarm and control phase: when the system detects harmful gas leakage through the sensor, the multi-mode data processing module can immediately trigger an alarm mechanism, give out an alarm, flash dangerous light and the like, the automatic control mechanism carries out corresponding processing, activates emergency response measures, such as opening a fire-fighting channel, automatically starting a ventilation system to reduce the concentration of chemical substances, and closing key equipment of a laboratory to prevent further accidents;

And a data transmission stage: the system sends alarm information to a cloud platform, a mobile phone APP of a laboratory manager and a mailbox through a network communication module. The mobile APP will display the real-time alert and the mailbox will send detailed alert information and suggested countermeasures. If no dangerous situation occurs, an administrator can monitor the environmental state of the chemical storage area in real time through the mobile phone APP, and check historical data and alarm records;

Cloud platform analysis: and through connecting to the OneNET cloud platform, data are stored and analyzed through historical data, and the fused data are analyzed through models such as RBF neural networks, so that the possibility of chemical leakage and the potential influence of the chemical leakage are predicted.

Specific implementation application scenario example 2: laboratory temperature anomaly monitoring

Temperature sensor deployment: a high-precision temperature sensor is installed near a heating device of a laboratory and the laboratory is deployed as a whole. The sensor of the heating device accessory is designed to monitor the temperature of the heating zone in real time, the laboratory is integral to monitor the laboratory temperature in real time and accurately detect any abnormal temperature changes.

And a data acquisition stage: each temperature sensor continuously collects temperature data of a heating area and environmental temperature data of a laboratory, and transmits the data to a multi-mode data processing module in real time for data processing;

And a data processing stage: receiving data from each sensor by using a multi-mode data processing module, and comprehensively processing the data;

And (3) a danger judging stage: when the multi-mode data processing module detects abnormal data, the abnormal data is in a dangerous state, an emergency response module is immediately triggered, and an alarm and control stage is entered;

Alarm and control phase: when the environmental data of the laboratory is abnormal, the multi-mode processing module monitors the abnormal data and triggers an alarm mechanism to give out an alarm, flash dangerous light and the like, and meanwhile, the system automatically starts the emergency response module and automatically opens the air conditioning equipment to reduce the temperature in the laboratory and prevent dangers caused by overhigh temperature;

And a data transmission stage: the system sends alarm information to a cloud platform, a mobile phone APP of a laboratory manager and a mailbox through a network communication module. The mobile APP will display the real-time alert and the mailbox will send detailed alert information and suggested countermeasures. If no dangerous condition occurs, an administrator can monitor the temperature change of the laboratory in real time through the mobile phone APP, and check historical data and alarm records;

Cloud platform analysis: and utilizing the data storage analysis and prediction functions of the cloud platform, the cloud platform uses the built RBF neural network model architecture to identify potential reasons of abnormal temperature and predict dangerous trends possibly occurring in the future. The system can identify the potential cause of the abnormal temperature and predict the dangerous trend that may occur in the future.

Specific implementation application scenario example 3: laboratory personnel monitoring

And (3) deploying a video acquisition module: the camera is deployed at the positions of a laboratory gate, a laboratory bench, a laboratory peripheral angle and the like, so that the camera can accurately capture various dynamics in the laboratory.

And a data acquisition stage: the method comprises the steps of acquiring data such as figures and equipment through a video acquisition module, and transmitting the data into a YOLO model;

And a data processing stage: analyzing the acquired images in real time by using a YOLO model, performing deep learning, pattern recognition and the like on personal behaviors, marking abnormal behaviors, such as unsafe illegal operations, physical discomfort of personnel in a laboratory or other behaviors which do not meet the laboratory specifications, and transmitting the abnormal behaviors to a multi-mode data fusion module to judge whether the abnormal behaviors are abnormal behaviors;

And (3) a danger judging stage: when abnormal data in the multi-mode data fusion module in the data processing stage is found to be in a dangerous state, immediately triggering an emergency response module, and entering an alarm and control stage;

Alarm and control phase: once the emergency response module is triggered, an alarm and automatic control mechanism is braked, firstly, an alarm such as sound, light flashing and the like is carried out, and the automatic control module executes safety measures such as closing related experimental equipment, opening a fire-fighting channel or starting other emergency response programs;

And a data transmission stage: the system sends alarm information to a cloud platform, a mobile phone APP of a laboratory manager and a mailbox through a network communication module. The mobile APP will display real-time alerts, while the mailbox will send detailed alert information such as time, place and possibly offending descriptions and suggested countermeasures. If no dangerous situation occurs, an administrator can monitor the condition of a laboratory in real time through a mobile phone APP and check historical data and alarm records;

Cloud platform analysis: through being connected to OneNET cloud platform, store and historical data analysis to the data, use the data storage function, can make things convenient for the analysis afterwards, utilize models analysis such as RBF neural network to fuse the data, predict the possibility and the potential influence of chemistry leakage.

Thus far, the technical solution of the present invention has been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of protection of the present invention is not limited to these specific embodiments. Equivalent modifications and substitutions for related technical features may be made by those skilled in the art without departing from the principles of the present invention, and such modifications and substitutions will be within the scope of the present invention.

The foregoing description is only of the preferred embodiments of the invention and is not intended to limit the invention; various modifications and variations of the present invention will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. The laboratory monitoring system of multiple information fusion, characterized by comprising:

the sensor module is used for collecting laboratory environment data;

the video acquisition module acquires laboratory videos to capture the dynamics in a laboratory;

The YOLO model receives the data of the video acquisition module and carries out depth annotation on laboratory video content;

The multi-mode data processing module is used for receiving the environment data and the labeling data of the YOLO model, carrying out abnormality judgment, and obtaining an abnormality signal when the judgment result is abnormal;

the emergency response module is used for receiving the abnormal signal and executing emergency response;

And the network communication module is used for receiving all data of the multi-mode data processing module so as to realize data transmission and communication between the laboratory and the remote monitoring center.

2. The multi-information fusion laboratory monitoring system of claim 1, wherein the sensor module monitors a laboratory environment for multiple physical quantities to accurately capture various environmental changes within the laboratory, the sensor module comprising a temperature sensor, a humidity sensor, a sound sensor, and a chemical sensor; the temperature sensor is deployed in a region where temperature change is likely to occur in a laboratory; the humidity sensor is deployed in a region sensitive to humidity change in a laboratory; the chemical sensor may be deployed in a laboratory in areas where chemical leakage may occur; the sound sensor is deployed in a region where sound accompanied accidents are likely to occur in a laboratory; the video acquisition module is deployed in the whole laboratory and is independently deployed at a laboratory gate and a laboratory bench to accurately capture various dynamics in the laboratory.

3. The multi-information fusion laboratory monitoring system of claim 1, wherein the YOLO model captures laboratory video as images and performs depth annotation based on multiple behavioral videos or image data under different environmental conditions, comprising: identification of target behavior, time stamping, and recording of environmental conditions; training by using the marked images, and continuously adjusting model parameters so that the model can be suitable for various changeable laboratory scenes; after model training is completed, the image of each triggering abnormal signal is brought into the newly acquired data, annotation training is carried out, and an iteration model is updated; the identifying of the target behavior includes: devices in the video or image use non-canonical behavior or experimental operator's operation non-canonical behavior; and transmitting data with the coordinates and size of the location where the target behavior appears in the video or image.

4. The multi-information fusion laboratory monitoring system of claim 1, wherein the multi-modal data processing module comprises a data interface unit, a data preprocessing unit, a data fusion unit, a pattern recognition unit, a decision logic unit, and a communication interface unit;

The data interface unit receives multi-source data streams and performs time stamp synchronization on each data stream; wherein the multi-source data comprises the environmental data and annotation data of the YOLO model;

the data preprocessing unit performs standardization processing on all data, converts the data into a uniform data format, and performs noise filtering and outlier detection;

The data fusion unit adopts a weighted average method, a Kalman filter and a neural network algorithm to deeply integrate the preprocessed data;

The pattern recognition unit analyzes the integrated data by adopting a deep learning model machine algorithm and divides the data into normal and abnormal;

The decision logic unit generates a corresponding response signal according to a preset safety threshold value and a safety rule to trigger the emergency response module, and transmits data to the network communication module through the communication interface unit, so that external alarm transmission is realized.

5. The multiple information fusion laboratory monitoring system of claim 4, wherein;

The data preprocessing unit converts laboratory environment data into a unified numerical value or text format, and converts labeling data provided by the YOLO model into a unified image format;

The data fusion unit integrates laboratory environment data by adopting a Kalman filter, and reduces the influence of random errors and accidental errors by adopting a weighted average method; identifying potential abnormal behaviors or environmental changes from data provided by the YOLO model by adopting a neural network algorithm, and combining the potential abnormal behaviors or environmental changes with laboratory environment data to identify whether abnormal behaviors or environmental risks exist;

the pattern recognition unit analyzes the integrated data by adopting a deep learning model machine algorithm, and comprises the following steps:

Laboratory environmental data analysis: laboratory environmental data is used to monitor laboratory environmental conditions, training a deep learning model to identify abnormal patterns in the data that indicate significant changes or potential hazards to the environmental conditions;

and (3) analyzing labeling data of the YOLO model: the deep learning model is trained to analyze the annotation data to identify potential abnormal behaviors or environmental changes, and whether actual security threats exist or not is determined jointly by combining laboratory environment data analysis results.

6. The multiple information fusion laboratory monitoring system of claim 4, wherein the setting of the safety threshold refers to a preset limit on a particular parameter, exceeding the limit indicating a safety risk; the safety rule is a logic guiding principle that the decision logic unit judges whether to trigger an alarm; the security rules are based on a combination of laboratory environment data and labeling data of the YOLO model.

7. The multiple information fusion laboratory monitoring system of claim 4, wherein the emergency response module comprises an alarm mechanism and an automatic control mechanism, and wherein the signal receiving unit of the emergency response module is configured to receive the response signal generated by the decision logic unit, and is responsible for decoding and confirming the received signal and determining the priority and type thereof; after confirming the abnormal signal, triggering the alarm mechanism and starting the automatic control mechanism to implement a series of predefined safety measures.

8. The multi-information converged laboratory monitoring system of claim 1, wherein the network communication module is equipped with NB-IoT, wiFi communication technology and communication security and protocol management, and wherein data transmission and communication between the laboratory and remote monitoring center comprises:

OneNET cloud platform connected to the network communication module for realizing data interaction and information sharing between the laboratory and the remote monitoring center, including data storage and history data functions;

the mailbox reminder is connected with the network communication module, and when the multi-mode data processing module transmits an abnormal signal to the emergency response module, a mail is sent to a preset mailbox;

and the mobile phone APP is connected with the network communication module and receives data from the network communication module so as to check the laboratory condition at any time.

9. The multi-information fusion laboratory monitoring system according to claim 1, wherein the OneNET cloud platform comprises a data storage unit and a historical data analysis unit, and the stored historical data and real-time data are utilized to predict dangerous trend through an RBF neural network model built on the platform;

The data storage unit is used for storing all data from the multi-mode data processing module by adopting a distributed database architecture;

the historical data analysis unit adopts an RBF neural network model to conduct data analysis and prediction, and captures complex relations and potential dangerous signals in data; the RBF neural network model firstly carries out data preprocessing, and real-time and historical data are subjected to cleaning, normalization and transformation processing to adapt to the input requirements of the model; then extracting features, namely extracting key features from the preprocessed data through statistical analysis and a machine learning algorithm, wherein the features can represent the state and the change trend of the monitoring environment; training the RBF neural network model, adjusting network weight and bias parameters, and optimizing the prediction accuracy of the model; finally, outputting prediction, analyzing new data input by the RBF model after training, predicting dangerous trend possibly occurring in the future, and feeding back the prediction result to a laboratory safety monitoring system for early warning and risk management;

the dangerous trend prediction comprises trend prediction of dangerous behaviors of common dangerous goods or action states in a laboratory.

10. A monitoring method of a laboratory monitoring system based on the multi-information fusion of claim 1, comprising:

The video acquisition module acquires data, and performs target detection by acquiring laboratory videos and using a pre-trained YOLO model; meanwhile, collecting laboratory environment data through the sensor module, and monitoring all environmental data in real time;

judging whether dangerous actions are contained in laboratory videos through a YOLO model, if so, firstly alarming, and then transmitting information into a data communication thread in a network communication module; meanwhile, the alarm information is subjected to video recording, the dangerous actions are stored in a material database, the materials in the database are marked, a YOLO model is used for training, the model is continuously updated, each time image information is acquired, if the dangerous actions are recorded, the model is updated, and the data acquired by the next video acquisition module are judged through the new model; if not, directly transmitting the information into a data center; if the laboratory environment data exceeds a set threshold value, alarming is carried out, and then the data is transmitted to a data center; if the set threshold value is not exceeded, directly transmitting the data into a data center;

After all data are transmitted into the data communication threads in the network communication module, the data are directly transmitted to the mobile phone APP and OneNET Internet of things cloud platform, meanwhile, the data are judged, and if abnormal data exist, mail alarm information is sent on the basis of the mobile phone APP and OneNET Internet of things cloud platform.