Laboratory safety behavior detection system based on deep learning

Abstract

The invention provides a laboratory safety behavior detection system based on deep learning. The system comprises an image detection module, a behavior recognition module, a behavior safety judgment module and a system automatic recording and optimizing module. Through image acquisition, preprocessing and storage, behaviors of laboratory personnel are classified and identified by using a deep learning model, the behaviors are compared with preset safety behavior standards, the safety of the behaviors is judged, alarm information is generated and fed back to a data center, and related personnel are reminded to pay attention to and correct unsafe behaviors in time. The system automatically records laboratory personnel behavior data, regularly performs data cleaning and labeling, continuously optimizes model performance through incremental learning and hyper-parameter tuning, and ensures high efficiency, accuracy and robustness of the system. The system can monitor personnel behaviors in a laboratory in real time, effectively improves the safety management level of the laboratory, and guarantees the smooth implementation of scientific research work.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring and safety management, and in particular to a laboratory safety behavior detection system based on deep learning. Background Art

[0002] The laboratory is an important place for scientific experiments and research. The safety behavior of personnel directly affects the safety of the laboratory and the safety of life and property of personnel. Traditional safety monitoring methods mainly rely on manual monitoring, which has problems such as untimely detection, limited coverage, and high false alarm rate. Therefore, it is of great practical significance to develop an automated laboratory safety behavior detection method and system based on deep learning technology. With the rapid development of science and technology, the importance of laboratories in scientific research, teaching, and industrial production has become increasingly prominent. As the main place for the research and development of high-precision technologies and new materials, laboratories often involve hazardous chemicals, complex mechanical equipment, and high-energy physics experiments. Therefore, laboratory safety management has become a vital issue. In recent years, laboratory safety accidents have occurred frequently, seriously affecting the life safety of scientific researchers and the smooth progress of scientific research. Traditional laboratory safety management mainly relies on manual inspections and regular training, but this method has the disadvantages of low efficiency, insufficient coverage, and high human resource consumption. It is impossible to monitor the behavior of personnel and environmental changes in the laboratory in real time and accurately, and it is difficult to discover and deal with safety hazards in a timely manner.

[0003] In recent years, with the development of artificial intelligence technology, especially the breakthrough of deep learning algorithms, intelligent monitoring systems have been widely used in the field of security. By installing surveillance cameras in the laboratory and combining deep learning technology, real-time monitoring and automatic recognition of laboratory personnel behavior can be achieved, thereby improving the safety management level of the laboratory. Deep learning algorithms can automatically learn features from a large amount of image data and classify and identify complex behavior patterns, so they have significant advantages in behavior detection and recognition. Existing behavior recognition systems based on deep learning are mostly used for security monitoring in public places, and there are few studies on laboratory environments. The laboratory environment has its unique characteristics, such as complex layout in the laboratory, diverse personnel activities, changeable behavior patterns, and a large number of professional equipment and materials. These characteristics put forward higher requirements for the behavior recognition system, which not only requires high-precision behavior recognition capabilities, but also requires judgment and feedback on safety behavior standards in different areas. Summary of the invention

[0004] The present invention aims to at least solve the technical problems existing in the prior art, and particularly innovatively proposes a laboratory safety behavior detection system based on deep learning. Through the collection and preprocessing of image data, the training and application of deep learning algorithms, the safety judgment and feedback of behavior, and the automatic recording and optimization of the system, real-time monitoring and safety management of laboratory personnel behavior can be achieved, the safety level of the laboratory can be improved, and the smooth progress of scientific research can be guaranteed.

[0005] In order to achieve the above-mentioned object of the present invention, the present invention provides a laboratory safety behavior detection system based on deep learning, comprising: an image detection module, a behavior recognition module and a behavior safety judgment module, wherein the data output end of the image detection module is connected to the data input end of the behavior recognition module, and the data output end of the behavior recognition module is connected to the data input end of the behavior safety judgment module;

[0006] Image detection module: collects the behavior images of laboratory personnel in real time through multiple high-definition cameras in the laboratory, and pre-processes the behavior images to generate pre-processed images;

[0007] Behavior recognition module: extract features from the pre-processed images, use a trained deep learning model to recognize the behavior of laboratory personnel, the deep learning model is a combination of a convolutional neural network (CNN) and a long short-term memory network (LSTM); and classify into several behavior types; the behavior types include operating equipment, wearing protective equipment, and carrying items;

[0008] Behavior safety judgment module: compares and analyzes the identified behavior with the preset safety behavior standards to determine whether the behavior meets the safety standards and generates corresponding safety judgment results.

[0009] In a preferred embodiment of the present invention, the image detection module includes: an image acquisition module, an image preprocessing module, and a data storage module;

[0010] The image acquisition module uses multiple high-definition cameras to collect behavioral images in the laboratory in real time;

[0011] An image preprocessing module performs cropping, denoising, enhancement, scaling and format conversion on the behavior image to obtain preprocessed image data;

[0012] Data storage module: stores preprocessed image data, transfers the image data to the local storage device (NAS) or cloud storage platform installed in the laboratory for storage, and stores the real-time collected image data for subsequent training and recognition, while establishing an image data set of laboratory personnel behavior.

[0013] In a preferred embodiment of the present invention, the image detection module also includes a communication module: the pre-processed image data is connected to the camera, data processing center and storage device through a wired or wireless network to realize real-time data transmission, and encryption technology is used to encrypt the behavioral image data and transmit it using the SSL / TLS protocol to ensure the security of the image data during transmission and prevent the data from being stolen or tampered with, so as to carry out subsequent behavior recognition and analysis.

[0014] In a preferred embodiment of the present invention, the behavior recognition module includes: a data preparation module and a model training module;

[0015] a data preparation module configured to annotate the preprocessed image data and improve the diversity of the data through data augmentation techniques;

[0016] Model training module, which trains deep learning models to recognize and classify the behaviors of laboratory personnel.

[0017] In a preferred embodiment of the present invention, the behavior recognition module also includes a model evaluation and optimization module: evaluating the model performance on the test set, calculating the accuracy (Accuracy), precision (Precision), recall (Recall) and F1 score (F1 Score) evaluation indicators.

[0018] In a preferred embodiment of the present invention, the behavior safety judgment module includes: a safety behavior standard definition module, a behavior identification and standard comparison module and a safety judgment and safety feedback module;

[0019] Safety behavior standard definition module, which formulates safety behavior standards for different areas according to the specific needs and potential risks of the laboratory;

[0020] A behavior recognition and standard comparison module matches and analyzes the recognized behavior with the safety behavior standard;

[0021] The safety judgment and feedback module generates safety judgment results and transmits alarm information to the remote data center via a wireless network to alert laboratory managers.

[0022] In a preferred embodiment of the present invention, it also includes a system automatic recording and optimization module: recording the behavior data and its related metadata, storing them in a high-performance database, and backing them up regularly. At the same time, through data cleaning, labeling, data enhancement and model optimization, the system's behavior recognition accuracy and safety judgment capabilities are continuously improved.

[0023] In a preferred embodiment of the present invention, the system automatic recording and optimization module includes: a data recording and storage module, a data processing and analysis module and a model optimization module;

[0024] The data recording and storage module automatically records the behavior data and related metadata of laboratory personnel, stores them in a high-performance database, and regularly performs incremental and full backups to ensure data security and recoverability;

[0025] The data processing and analysis module uses data cleaning tools to clean the recorded data and improves the diversity of data and the robustness of the model through data enhancement technology;

[0026] The data cleaning tool (OpenRefine) is configured to clean the recorded data and remove noise data, duplicate data, and abnormal data; the data annotation tool (VGG Image Annotator, VIA) is configured to manually annotate the unlabeled data to ensure the accuracy and completeness of the data.

[0027] The model optimization module monitors the performance of the model in the production environment in real time, uses Keras Tuner to automatically tune hyperparameters, and continuously trains and updates the model through incremental learning technology. The optimized model is exported to a deployable format to replace the old model, ensuring that the system uses the latest model for behavior recognition in real time, continuously optimizing model performance, and improving the system's behavior recognition accuracy and safety judgment capabilities.

[0028] The hyperparameter search space of the hyperparameter tuning is configured to define the hyperparameter search space of the deep learning model, and the Bayesian optimization algorithm is selected for automatic tuning to find the optimal hyperparameter combination;

[0029] The incremental learning technology is configured to select new representative samples from the recorded data as the data set for incremental learning, preprocess and annotate the newly added data, generate a new training data set and retrain, update model parameters, generate a new model version, and evaluate its performance on the validation set.

[0030] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0031] This method and system automatically identify the behavior of laboratory personnel by utilizing image detection technology and deep learning algorithms, and make safety judgments and feedback based on the safety behavior standards of different areas of the laboratory, thereby realizing intelligent and automated safety management of the laboratory and improving the laboratory's safety level and management efficiency.

[0032] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0034] Figure 1 It is a system module framework diagram of the present invention.

[0035] Figure 2 It is a flow chart of a specific implementation mode of the present invention. DETAILED DESCRIPTION

[0036] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0037] The present invention proposes a laboratory safety behavior detection system based on deep learning, such as Figures 1-2 As shown, it includes: an image detection module, a behavior recognition module, a behavior safety judgment module and a system automatic recording and optimization module.

[0038] The image detection module is responsible for the collection and preprocessing of image data, and includes the following four main modules: image acquisition module, image preprocessing module, data storage module, and communication module.

[0039] Step 1: The image acquisition module covers all areas that need to be monitored through multiple high-definition cameras in the laboratory. These cameras have night vision capabilities and can capture clear images at a high frame rate. The camera equipment uses a camera control system, including a camera control interface (adjusting camera parameters such as angle, focal length and aperture) to ensure the best image quality; the video stream processor is responsible for converting the video stream collected by the camera into image frames for the next step of processing.

[0040] Step 2: Image preprocessing module, crop the image frame obtained in step 1, retain the key area, remove the redundant part, use Gaussian filtering to reduce the noise in the image, enhance the clarity of the image, and improve the brightness and contrast of the image through histogram equalization to make the details in the image more obvious, thereby improving the image quality and recognition accuracy. According to the input size requirements of the deep learning model, scale the image to ensure that the image size is consistent with the model input, and convert the image into the format required by the deep learning model (grayscale image).

[0041] Step 3: The data storage module is responsible for storing the image data preprocessed in step 2, transferring the data to the local storage device (NAS) or cloud storage platform installed in the laboratory for storage, and storing the real-time collected image data for subsequent training and recognition, and establishing an image dataset of laboratory personnel behavior.

[0042] Step 4: The communication module is responsible for the transmission and communication of image data. It connects the camera, data processing center and storage device through a wired or wireless network to realize real-time data transmission. It also uses encryption technology to encrypt the behavioral image data and transmit it using the SSL / TLS protocol to ensure the security of the image data during transmission and prevent the data from being stolen or tampered with, so as to facilitate subsequent behavior recognition and analysis.

[0043] The method of encrypting behavior image data using encryption technology is as follows:

[0044] The pixel values ​​corresponding to each pixel point in the behavioral images taken in the laboratory are converted into binary pixel values; the number of bits of each binary pixel value should be consistent. If not, add 0 in front of the binary pixel value to make the number of bits K of each binary pixel value consistent.

[0045] Obtain an encrypted string of the cloud storage platform, which is composed of 62-base digital characters 0 to 9, lowercase characters a to z, and uppercase characters A to Z; after obtaining the encrypted string of the cloud storage platform, convert the encrypted string into a binary encrypted string; take the 1st to Kth characters in the binary encrypted string as the intercepted string;

[0046] The pixel value corresponding to each pixel in the behavior image is XORed with the intercepted string to obtain the encrypted behavior image. The encrypted behavior image is transmitted to the storage platform using the SSL / TLS protocol. The storage platform uses the encrypted string to obtain the behavior image actually shot in the laboratory:

[0047] The platform converts the pixel value corresponding to each pixel point in the encrypted behavior image into a binary pixel value; the number of bits of each binary pixel value should be consistent. If not, add 0 in front of the binary pixel value to make the number of bits K of each binary pixel value consistent.

[0048] Convert the encrypted string stored in the platform into a binary encrypted string; take the 1st to Kth characters in the binary encrypted string as the intercepted string;

[0049] The decrypted behavior image can be obtained by performing an XOR operation on the pixel value corresponding to each pixel point in the encrypted image and the intercepted string. The decrypted behavior image is the behavior image taken in the laboratory and is stored on the cloud storage platform.

[0050] The behavior recognition module involves the following five main modules: data preparation module, model training module, model evaluation and optimization module, and model deployment module.

[0051] Step 5: Data preparation module, through the preprocessed image data transmitted in step 4, the data preparation module extracts representative samples from the preprocessed image data set, covering various behaviors of laboratory personnel. Use the VGG Image Annotator (VIA) labeling tool to annotate the image data, annotate the behavior category and the corresponding area. Then, the image data is enhanced by randomly cropping, randomly rotating, color changing, randomly flipping, and standardizing the image to enhance the robustness and generalization ability of the model.

[0052] VGG Image Annotator (VIA) is used to annotate objects and their attributes in images, and the generated annotation data is used to train computer vision models. The main functions of VIA are: ① Multiple annotation types: Supports multiple annotation types such as rectangles, circles, polygons, points and lines, suitable for different types of annotation needs. ② Attribute annotation: You can add attributes to the annotated objects, such as category labels, description text, etc. ③ Batch operations: Support batch import and export of images and annotation data. VIA can efficiently annotate image data and provide high-quality training data for deep learning models.

[0053] Step 6: Use the train_test_split function of the scikit-learn library to split the dataset obtained in step 5 into training set, validation set, and test set with a ratio of 70:20:10 (70% training set, 20% validation set, and 10% test set) to better evaluate the model performance and prevent overfitting.

[0054] Step 7: The behavior recognition and model training module in the system architecture first uses a hybrid model (convolutional neural network (CNN) + multi-task learning (MTL) model) to process the pre-processed data.

[0055] The low-level and high-level features of the image are extracted through the convolution layer and the pooling layer. The convolution layer captures the spatial features of the image, and the pooling layer reduces the dimension and prevents overfitting. The image data after feature extraction is flattened through the feature map, and the convolution layer output is flattened into a one-dimensional vector as the input of the LSTM. The LSTM layer based on multi-task learning processes the temporal features and captures the temporal dependency of the behavior. Finally, it is classified through the fully connected layer and the Softmax layer to identify the specific behavior category.

[0056] Step 8: The model training module is responsible for training the deep learning model. The specific implementation steps are as follows: First, set the hyperparameters, including selecting the Adam optimizer, the cross entropy loss function, and the accuracy evaluation index, and set the learning rate (0.001), batch size (32), and the number of training rounds (50). Then, define the training and validation functions, and perform forward propagation, loss calculation, backpropagation, and parameter update in each epoch. During the training process, use the EarlyStopping callback function to monitor the validation loss to prevent overfitting. Use PyTorch to load the data, and use data augmentation techniques (random cropping, rotation, translation, scaling, and color transformation) to improve the robustness and generalization ability of the model. The data loading and preprocessing steps ensure that the data can be efficiently batched into the model and the input data is standardized. After starting training, forward propagation, loss calculation, backpropagation, and parameter update are performed in each epoch, and the training loss and accuracy are recorded in real time to monitor the learning progress of the model. The model performance is evaluated on the validation set, and the model structure and hyperparameters are adjusted according to the evaluation results. Finally, the model with the best performance is selected as the final model, and the accurate detection of laboratory safety behaviors is finally achieved. The trained model will be saved in a deployable format for subsequent deployment and use.

[0057] Step 9: The model evaluation and optimization module is responsible for evaluating the model performance and optimizing the model. The specific implementation steps are as follows: Evaluate the model performance on the test set and calculate the accuracy, precision, recall and F1 score evaluation indicators. Accuracy is the ratio of the number of correctly classified samples to the total number of samples; precision is the proportion of samples classified as positive that are actually positive; recall is the proportion of samples that are actually positive that are correctly classified; F1 score is the harmonic mean of precision and recall.

[0058] Use the confusion matrix to analyze the performance of the model on various behaviors and identify the classification errors and deviations of the model in each category. Through these steps, the performance of the model is comprehensively evaluated, and the classification performance of the model in each category is analyzed in detail, so as to optimize the model in a targeted manner and improve the accuracy and robustness of the laboratory safety behavior detection system. Optimization methods include improving model performance through hyperparameter tuning, increasing training data, and improving model architecture. In the specific optimization steps, Keras Tuner is used for automatic hyperparameter search, and algorithms such as Bayesian optimization or random search are used to automatically search for the optimal hyperparameter configuration to improve the performance of the model on specific tasks. Finally, through comprehensive evaluation and optimization steps, the potential of the model can be explored and utilized more effectively, so that it can adapt to different data and problem scenarios, and improve the overall performance and reliability of the laboratory safety behavior detection system.

[0059] Step 10: The model deployment module is responsible for deploying the trained model to the production environment to achieve real-time laboratory safety behavior identification.

[0060] The specific technical implementation is as follows: Export the trained model to ONNX (Open Neural Network Exchange) format, because ONNX has cross-platform compatibility and is suitable for a variety of hardware and software environments. Then, deploy the exported ONNX model to the cloud server and use ONNX Runtime for inference services. Develop an API interface on the server to receive real-time image data transmitted by the laboratory camera, call the model for inference, and return the behavior recognition results.

[0061] The specific implementation includes using FastAPI to build an API service, receiving image data, calling ONNX Runtime for inference after preprocessing, and returning behavior recognition results. This can effectively achieve real-time monitoring and identification of laboratory safety behaviors to ensure laboratory safety. Through these detailed steps and technical implementations, the model deployment module can efficiently apply deep learning models to actual scenarios, achieve real-time monitoring of laboratory behaviors, and improve the safety management level of laboratories.

[0062] The behavior safety judgment module is responsible for feeding back the judgment results to the data center to remind relevant personnel to pay attention to unsafe behaviors. It includes the following main parts: safety behavior standard definition module, behavior identification and standard comparison module, and safety judgment and safety feedback module.

[0063] Step 11: Safety behavior standard definition module. In the process of laboratory management, in order to ensure the safety of personnel, safety behavior standards for different areas need to be formulated. The specific implementation steps are as follows: First, formulate safety behavior standards according to the specific needs and potential risks of the laboratory. Safety behavior standards usually include the following: ① Equipment operation standards, which describe the correct methods of operating experimental equipment to ensure standardized and safe operation; ② Personal protection standards, which stipulate the protective equipment that must be worn in different areas, such as goggles, laboratory coats, gloves, etc., to ensure the personal safety protection of personnel; ③ Activity specification standards, which stipulate the activities in the laboratory, such as walking routes, correct postures for carrying items, etc., to ensure personnel behavior norms and laboratory order; ④ Emergency handling standards, which describe the evacuation and emergency handling methods in emergency situations, to ensure that personnel can evacuate and handle them quickly and safely in emergency situations. Then, these standards can be recorded in the form of documents and displayed in the laboratory by area so that all laboratory personnel can clearly understand and comply with them. At the same time, the standards can be reviewed and updated regularly to ensure that they are consistent with the latest safety requirements and the actual situation of the laboratory, so as to continuously improve the safety management level of the laboratory.

[0064] Step 12: Behavior recognition and standard comparison module, after identifying the specific behaviors of laboratory personnel through the deep learning algorithm model, these behaviors are compared and analyzed with the preset safety behavior standards. The specific implementation steps are as follows: First, use the deep learning model (CNN+LSTM) trained in steps 5 to 10 to classify the personnel behaviors in the image data and identify specific behavior types, such as "operating equipment", "wearing protective equipment", "carrying items", etc. Next, according to the functions and layout of different areas in the laboratory, the laboratory is divided into areas, such as "experimental area", "storage area", "channel area", etc., and different safety behavior standards are set for each area. Then, the identified behavior is matched with the safety behavior standards of the area to determine whether the behavior meets the safety specifications. For example, in the "experimental area", the "operating equipment" behavior is identified and checked to see if it meets the safety standard of "wearing protective equipment". If it is found that the behavior does not meet the safety standard, the system will generate a warning message and remind the relevant personnel to make corrections through the data center. Through this module, it is ensured that the behavior of laboratory personnel always meets the safety standards and improves the overall safety of the laboratory.

[0065] Step 13: The safety judgment and safety feedback module makes a safety judgment on the behavior according to the matching results of step 12. The specific implementation steps are as follows: First, determine whether the identified behavior meets the preset safety behavior standards, and identify possible abnormal behaviors or unsafe behaviors. Perform risk assessment on the identified unsafe behaviors to determine the possible safety risks and severity. Then, the safety judgment results are fed back to the data center, and relevant personnel are reminded to pay attention to unsafe behaviors. According to the safety judgment results, corresponding alarm information is generated. If unsafe behavior is detected, the system will generate an alarm and indicate the specific behavior and area. Then, the system transmits the alarm information to the remote data center via a wireless network, and the data center sends the alarm information to the terminal devices of relevant personnel (such as mobile phones, computers or laboratory management systems). In this way, the system can provide feedback to laboratory managers and relevant personnel in real time, reminding them to immediately pay attention to and correct unsafe behaviors, thereby ensuring the safety of the laboratory and the safety of personnel.

[0066] The system automatic recording and optimization module is responsible for automatically recording and storing behavioral data, and continuously optimizing model performance through deep learning algorithms. The system automatic recording and optimization module includes: data recording and storage module, data processing and analysis module, and model optimization module.

[0067] Step 14: Data recording and storage module, automatically records the behavior data of laboratory personnel in all the above steps, including time, location, behavior category, safety judgment results and other information, and records metadata related to the behavior data, such as camera number, image acquisition time, preprocessing steps, etc. In order to ensure efficient storage and management of data, a high-performance database MySQL is selected to store these behavior data and metadata. The database design is divided into three parts: behavior data table, metadata table and safety judgment result table, which record different types of data respectively. In order to ensure the security and recoverability of data, the database is backed up regularly, using a combination of incremental backup and full backup. Use high-performance storage devices such as SSD to ensure data read and write speed and storage capacity. At the same time, a cold and hot data separation storage strategy is adopted to store recently frequently accessed data on fast storage media, while historical data is stored on large-capacity storage media to optimize the utilization efficiency of storage resources.

[0068] Step 15: Data processing and analysis module, repeat the operation of step 5, use the data cleaning tool OpenRefine to clean the recorded data, remove noise data, duplicate data and abnormal data. Use VGG ImageAnnotator (VIA) to manually annotate the unlabeled data to ensure the accuracy and completeness of the data. Ensure the quality of the labeled data through cross-validation and manual review. Use the data enhancement library of the deep learning framework TensorFlow to achieve data enhancement, and use data enhancement techniques (random cropping, random rotation, color transformation, etc.) to increase the diversity of the data, thereby improving the robustness and generalization ability of the model.

[0069] Step 16: Model optimization module, repeat step 9, monitor the performance of the model in the production environment in real time, record the model prediction results and actual results, use Keras Tuner to automatically tune the hyperparameters, and find the optimal hyperparameter combination. Define the hyperparameter search space, select the Bayesian optimization algorithm, perform automatic tuning, and find the optimal hyperparameter combination; use incremental learning technology (online learning, semi-supervised learning) to continuously train and update the model. Select new representative data samples from the recorded data as the data set for incremental learning. Preprocess and annotate the new data, generate a new training data set, retrain the model with the new training data set, update the model parameters, generate a new model version, and evaluate the performance of the retrained model on the validation set to ensure the effectiveness and stability of the model. Export the optimized model to a deployable format, replace the old model with the new model, and ensure that the system can use the latest model for behavior recognition in real time. By looping the above steps, the model performance is continuously optimized to improve the system's behavior recognition accuracy and safety judgment capabilities.

[0070] Through the above detailed steps and technical implementations, the present invention can realize the system to automatically record and store behavioral data, and continuously optimize the model performance through deep learning algorithms, thereby improving the accuracy and robustness of the laboratory safety behavior detection system.

[0071] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A laboratory safety behavior detection system based on deep learning, characterized in that: include: An image detection module, a behavior recognition module and a behavior safety judgment module, wherein the data output end of the image detection module is connected to the data input end of the behavior recognition module, and the data output end of the behavior recognition module is connected to the data input end of the behavior safety judgment module; Image detection module: collects the behavior images of laboratory personnel in real time through multiple high-definition cameras in the laboratory, and pre-processes the behavior images to generate pre-processed images; Behavior recognition module: extract features from the pre-processed images, use the trained deep learning model to recognize the behavior of laboratory personnel, the deep learning model is a combination of convolutional neural network and long short-term memory network; and classify into several behavior types; the behavior types include operating equipment, wearing protective equipment, and carrying items; Behavior safety judgment module: compares and analyzes the identified behavior with the preset safety behavior standards to determine whether the behavior meets the safety standards and generates corresponding safety judgment results.

2. A laboratory safety behavior detection system based on deep learning according to claim 1, characterized in that: The image detection module includes: an image acquisition module, an image preprocessing module, and a data storage module; The image acquisition module uses multiple high-definition cameras to collect behavioral images in the laboratory in real time; An image preprocessing module performs cropping, denoising, enhancement, scaling and format conversion on the behavior image to obtain preprocessed image data; Data storage module: stores preprocessed image data, transfers image data to local storage devices or cloud storage platforms installed in the laboratory for storage, and stores real-time collected image data for subsequent training and recognition, while establishing an image data set of laboratory personnel behavior.

3. A laboratory safety behavior detection system based on deep learning according to claim 2, characterized in that: The image detection module also includes a communication module: the pre-processed image data is connected to the camera, data processing center and storage device through a wired or wireless network to achieve real-time data transmission, and encryption technology is used to encrypt the behavioral image data and transmit it using the SSL / TLS protocol to ensure the security of the image data during transmission and prevent the data from being stolen or tampered with, so as to carry out subsequent behavior recognition and analysis.

4. A laboratory safety behavior detection system based on deep learning according to claim 1, characterized in that: The behavior recognition module includes: a data preparation module and a model training module; a data preparation module configured to annotate the preprocessed image data and improve the diversity of the data through data augmentation techniques; Model training module, which trains deep learning models to recognize and classify the behaviors of laboratory personnel.

5. A laboratory safety behavior detection system based on deep learning according to claim 4, characterized in that: The behavior recognition module also includes a model evaluation and optimization module: evaluating the model performance on the test set, calculating the accuracy, precision, recall and F1 score evaluation indicators.

6. A laboratory safety behavior detection system based on deep learning according to claim 1, characterized in that: The behavior safety judgment module includes: a safety behavior standard definition module, a behavior identification and standard comparison module, and a safety judgment and safety feedback module; Safety behavior standard definition module, which formulates safety behavior standards for different areas according to the specific needs and potential risks of the laboratory; A behavior recognition and standard comparison module matches and analyzes the recognized behavior with the safety behavior standard; The safety judgment and feedback module generates safety judgment results and transmits alarm information to the remote data center via a wireless network to alert laboratory managers.

7. A laboratory safety behavior detection system based on deep learning according to claim 1, characterized in that: It also includes a system automatic recording and optimization module: recording the behavioral data and its related metadata, storing them in a high-performance database, and backing them up regularly. At the same time, through data cleaning, labeling, data enhancement and model optimization, the system's behavior recognition accuracy and security judgment capabilities are continuously improved.

8. A laboratory safety behavior detection system based on deep learning according to claim 7, characterized in that: The automatic recording and optimization module of the system includes: a data recording and storage module, a data processing and analysis module and a model optimization module; The data recording and storage module automatically records the behavior data and related metadata of laboratory personnel, stores them in a high-performance database, and regularly performs incremental and full backups to ensure data security and recoverability; The data processing and analysis module uses data cleaning tools to clean the recorded data and improves the diversity of data and the robustness of the model through data enhancement technology; A data cleaning tool is configured to clean the recorded data to remove noise data, duplicate data, and abnormal data; a data labeling tool is configured to manually label the unlabeled data to ensure the accuracy and completeness of the data; Model optimization module: monitors the performance of the model in the production environment in real time, uses Keras Tuner to automatically tune hyperparameters, and continuously trains and updates the model through incremental learning technology. It also exports the optimized model to a deployable format and replaces the old model to ensure that the system uses the latest model for behavior recognition in real time, continuously optimizes model performance, and improves the system's behavior recognition accuracy and safety judgment capabilities. The hyperparameter search space of the hyperparameter tuning is configured to define the hyperparameter search space of the deep learning model, and the Bayesian optimization algorithm is selected for automatic tuning to find the optimal hyperparameter combination; The incremental learning technology is configured to select new representative samples from the recorded data as the data set for incremental learning, preprocess and annotate the newly added data, generate a new training data set and retrain, update model parameters, generate a new model version, and evaluate its performance on the validation set.

Images