Safety monitoring method and system for industrial park

Abstract

The invention relates to the technical field of safety monitoring, in particular to a safety monitoring method and system for an industrial park, and the method comprises the steps: carrying out the preprocessing of monitoring data through a dynamic dust suppression method, a light reflection compensation method and motion region enhancement; pre-processing data subjected to dust suppression processing, light reflection compensation and motion area enhancement are obtained; identifying and analyzing the preprocessed data by adopting a preset multi-task joint detection method to obtain a security detection result of multi-task integration; prediction analysis is performed based on the monitoring data, the safety detection result and the task list information to obtain safety early warning information, so that a supervisor performs targeted processing according to the safety early warning information, and meanwhile, the problem that existing AI model training data mostly aims at a general scene and is inconvenient to carry out is solved. The problems that special interference (such as high-concentration dust and light reflection) of a powder plant area in a chemical plant area is not optimized, and when the method is applied to the powder plant area, the false alarm rate of a general safety monitoring method is high are solved.

Description

Technical Field

[0001] The present invention relates to the field of security monitoring technology, and in particular to a security monitoring method and system for an industrial park. Background Art

[0002] With the gradual development of IoT technology and monitoring devices, the safety monitoring capabilities of industrial parks have greatly improved, effectively reducing the likelihood of safety accidents. However, in industrial park safety monitoring, existing AI model training data is mostly targeted at general scenarios and is not optimized for interference specific to powder processing plants (such as high dust concentrations and light reflections). When applied to powder processing plants, common safety monitoring methods have a high false alarm rate. Frequent false alarms can lead to a decrease in management trust and the possibility of real hidden dangers being overlooked. While methods that rely on fixed thresholds to adjust data acquisition accuracy when monitoring devices are obscured by dust have been proposed to improve safety monitoring efficiency, these methods are unable to detect changes in obscuration thickness in real time, resulting in blind spots and still low safety monitoring accuracy. Furthermore, abnormal events rely on manual judgment and experience, lacking pre-defined, tiered response plans (such as automatically shutting down associated equipment and activating dust suppression systems). Furthermore, accident impact assessment relies on static maps, without integrating real-time dust dispersion simulations and personnel location data to enable timely incident response and avoid escalation caused by delayed emergency response.

[0003] Therefore, the present invention provides a security monitoring method and system for an industrial park to solve the above problems. Summary of the Invention

[0004] In response to the above situation, in order to overcome the defects of the existing technology, the present invention provides a safety monitoring method and system for industrial parks to solve the problem that the training data of the above-mentioned existing AI models are mostly targeted at general scenarios and are not optimized for the interference specific to powder plants in chemical plants (such as high concentrations of dust and light reflections). When applied to powder plants, the general safety monitoring method has a high false alarm rate.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A security monitoring method for an industrial park, comprising:

[0007] Acquiring monitoring data, wherein the monitoring data includes monitoring data, environmental data, and audio data;

[0008] Preprocessing the monitoring data to obtain preprocessed data;

[0009] Use the preset multi-task joint detection method to identify and analyze the pre-processed data to obtain security detection results;

[0010] Based on monitoring data, safety inspection results and task order information, predictive analysis is carried out to obtain safety warning information.

[0011] Preferably, the preprocessing of the monitoring data to obtain preprocessed data includes: processing the monitoring data using a preset dynamic dust suppression method to suppress the optical interference of dust; processing the monitoring data using a transmitted light model to compensate for the powder shielding effect; and processing the audio data using wavelet frequency domain filtering to suppress high-frequency noise. The formula of the transmitted light compensation model is:

[0012] Where C is the dust concentration, T is the transmittance, and a and b are on-site calibration coefficients.

[0013] Preferably, the preprocessing of the monitoring data to obtain preprocessed data includes: processing the monitoring data in the monitoring data using a preset light reflection compensation method to achieve light reflection compensation; specifically including: constructing a Stokes parameter matrix using a linear polarizer group to solve the surface reflection component; based on a physical BRDF model, inferring the real surface texture to obtain corrected monitoring data; wherein the polarization filtering formula is:

[0014]

[0015] Among them, S is the image after surface reflection suppression, is the image intensity at different polarization angles;

[0016] The inverse rendering modeling formula is:

[0017] v=κ·cosθ·(1+α·N·H n ),

[0018] Among them, κ is the material refractive index, θ is the incident angle, N and H are the surface normal and the unit vector of the illumination direction, α and n are empirical parameters obtained through sample training, and v is the reflection intensity.

[0019] Preferably, the preprocessing of the monitoring data to obtain preprocessed data includes: using a sparse optical flow method to process the monitoring data in the monitoring data to distinguish between dynamic areas and static areas; using a saliency measurement method to identify dynamic areas to mark significant motion areas; performing weight processing on significant motion areas to suppress static background; and performing morphological correction and time domain filtering on significant motion areas to obtain an enhanced image.

[0020] Preferably, the preset multi-task joint detection method is used to identify and analyze the pre-processed data to obtain a security detection result, including: fusing the monitoring data, audio data and environmental data in the pre-processed data to obtain fused data; using the improved YOLOv5 backbone network algorithm to extract features from the fused data to obtain a feature vector; using the multi-task output head algorithm to detect and analyze the feature vector to obtain a detection result; using a multi-task loss function to adjust the detection result and optimize the multi-task joint detection method to output the multi-task detection result.

[0021] Preferably, the improved YOLOv5 backbone network algorithm is used to extract features from the fused data to obtain a feature vector, including: using a convolutional layer and a spatial pyramid pooling method to process the fused data, capturing the spatial features and semantic information of the image, and generating a shared feature map containing the feature vector.

[0022] Preferably, the multi-task output head algorithm is used to detect and analyze the feature vector to obtain the detection result, including: using the Sigmoid function to analyze and process the corresponding feature vector to obtain the flame confidence and the smoke confidence; using the Softmax activation function to analyze and process the corresponding feature vector to obtain the temperature anomaly distribution information; using the preset shared positioning formula to analyze and process the corresponding feature vector to obtain the bounding box coordinate information; and integrating the flame confidence, smoke confidence, temperature anomaly distribution information and bounding box coordinate information to obtain the detection result.

[0023] Preferably, the use of a multi-task loss function to adjust the detection results and optimize the multi-task joint detection method to output the multi-task detection results includes: using each preset loss function to analyze the detection results and the true label to determine the loss information; adjusting the corresponding detection results according to the loss information; and determining the model optimization parameters according to the parameter information corresponding to the minimized total loss value of back propagation.

[0024] Preferably, the predictive analysis based on monitoring data, safety detection results and task order information to obtain safety warning information includes: using a digital twin warning model constructed based on digital twins to process the monitoring data, safety detection results and task order information, simulating dust diffusion information and calculating dust concentration and duration; determining the processing strategy based on the matching relationship between dust concentration, duration and a preset strategy library; and generating safety warning information based on the dust concentration and processing strategy.

[0025] The present invention also provides a security monitoring system for an industrial park, which includes a data acquisition module, a preprocessing module, a security detection module and a security warning module; the security monitoring system implements the security monitoring method described in any one of the above technical solutions through mutual cooperation between the modules.

[0026] The beneficial effects of the present invention are:

[0027] 1. The present invention pre-processes the monitoring data through a dynamic dust suppression method, a light reflection compensation method and a motion area enhancement method to obtain pre-processed data that has undergone dust suppression processing, light reflection compensation and motion area enhancement, which is convenient for subsequent analysis and processing and improves the accuracy of data processing; then, a preset multi-task joint detection method is used to identify and analyze the pre-processed data to obtain a multi-task integrated safety detection result; finally, a predictive analysis is performed based on the monitoring data, safety detection results and task order information to obtain safety warning information, so that supervisors can carry out targeted processing according to the safety warning information and avoid the occurrence of safety accidents in advance. At the same time, it helps to solve the problem that the existing AI model training data is mostly targeted at general scenarios, and is not optimized for the unique interference of powder plant areas in chemical plant areas (such as high concentration of dust, light reflection). When applied to powder plant areas, the general safety monitoring method has a high false alarm rate.

[0028] 2. The present invention uses a multi-task loss function to adjust the detection results and optimize the multi-task joint detection method, thereby improving the accuracy of multi-task classification while ensuring positioning accuracy.

[0029] 3. Based on monitoring data, safety detection results and task order information, the present invention can simulate the movement state of dust in the digital twin early warning model and calculate the dust concentration and duration. When the early warning conditions are triggered, the corresponding safety protection measures are activated and early warning information is sent to the supervisor. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a schematic flow chart of a security monitoring method for an industrial park according to the present invention;

[0031] Figure 2 This is a schematic module diagram of a security monitoring system for an industrial park according to the present invention. DETAILED DESCRIPTION

[0032] The following will refer to the attached Figure 1 To the attached Figure 2 The embodiments of the present invention are described in detail. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0033] A security monitoring method for an industrial park, as shown in the attached Figure 1 As shown, the following steps are included:

[0034] Step S11: Acquire monitoring data, which includes monitoring data, environmental data, and audio data.

[0035] To obtain monitoring data, environmental data, and audio data, an industrial park consisting of a powder plant needs to deploy many sensors within the plant to obtain the corresponding data. For example, the heterogeneous sensor network deployment of a powder plant is as follows: explosion-proof 4K ultra-high-definition cameras (frame rate ≥ 30fps) and infrared thermal imagers (temperature measurement accuracy ±2°C) are used to obtain monitoring data; beta-ray dust concentration meters (range 0-5000mg / m 3 ) and a laser particle counter (particle size resolution 0.1 μm) to collect environmental data; an array microphone (frequency response 20 Hz-20 kHz) and an ultrasonic detector (sensitivity -40 dB) to collect audio data.

[0036] Then build an NTP clock synchronization network to achieve time alignment of monitoring data, environmental data and audio data; it is convenient for subsequent analysis and processing of monitoring data at the same time

[0037] Step S12: pre-processing the monitoring data to obtain pre-processed data.

[0038] The environmental data and audio data are cleaned and denoised to complete the preprocessing process.

[0039] Step S13: using a preset multi-task joint detection method to identify and analyze the pre-processed data to obtain a safety detection result.

[0040] Step S14: Perform prediction analysis based on monitoring data, safety detection results and task order information to obtain safety warning information.

[0041] The present invention pre-processes the monitoring data through a dynamic dust suppression method, a light reflection compensation method and a motion area enhancement method to obtain pre-processed data that has undergone dust suppression processing, light reflection compensation and motion area enhancement, which is convenient for subsequent analysis and processing and improves the accuracy of data processing; then a preset multi-task joint detection method is used to identify and analyze the pre-processed data to obtain a multi-task integrated safety detection result; finally, a predictive analysis is performed based on the monitoring data, safety detection results and task order information to obtain safety warning information, so that supervisors can carry out targeted processing according to the safety warning information and avoid the occurrence of safety accidents in advance. At the same time, it helps to solve the problem that the existing AI model training data is mostly targeted at general scenarios, and is not optimized for the unique interference of powder plant areas in chemical plant areas (such as high concentration of dust, light reflection). When applied to powder plant areas, the general safety monitoring method has a high false alarm rate.

[0042] Furthermore, in one embodiment of the present invention, the preprocessing of the monitoring data to obtain preprocessed data includes: processing the monitoring data using a preset dynamic dust suppression method to suppress optical interference from dust; processing the monitoring data using a transmitted light model to compensate for the powder shielding effect; and processing the audio data using wavelet frequency domain filtering to suppress high-frequency noise. The formula of the transmitted light compensation model is:

[0043] Where C is the dust concentration, T is the transmittance, and a and b are on-site calibration coefficients.

[0044] Specifically, when the preset dynamic dust suppression method is used to process the monitoring data to suppress the optical interference of dust, a transmitted light compensation algorithm is used to establish a dust concentration-transmittance model, and the on-site calibration coefficients a and b are corrected online; the video frames of the video stream are decomposed by wavelet packets to achieve frequency domain filtering, retain the effective frequency band, and eliminate high-frequency dust flicker noise; then the improved histogram equalization algorithm is used to dynamically adjust the window width to achieve the effect of adaptive contrast enhancement; the monitoring data is processed by combining the above steps to complete dynamic dust suppression.

[0045] Furthermore, in one embodiment of the present invention, the preprocessing of the monitoring data to obtain the preprocessed data includes: processing the monitoring data in the monitoring data using a preset light reflection compensation method to achieve light reflection compensation; specifically including: constructing a Stokes parameter matrix using a linear polarizer group to solve the surface reflection component; based on a physical BRDF model, inferring the real surface texture to obtain corrected monitoring data; wherein the polarization filtering formula is:

[0046]

[0047] Among them, S is the image after surface reflection suppression, is the image intensity at different polarization angles;

[0048] The inverse rendering modeling formula is:

[0049] v=κ·cosθ·(1+α·N·H n ),

[0050] Among them, κ is the material refractive index, θ is the incident angle, N and H are the surface normal and the unit vector of the illumination direction, α and n are empirical parameters obtained through sample training, and v is the reflection intensity.

[0051] Specifically, when using polarization filtering to separate surface reflection components, the polarization characteristics of light are utilized to obtain images of different polarization states through multi-angle polarization imaging (such as 0°, 45°, and 90° polarization directions). Based on the principle that surface reflected light (such as specular reflection from a smooth surface) generally has partial polarization characteristics, while diffuse reflected light (from the object itself) has a lower degree of polarization, a polarization filtering algorithm is used to separate the surface reflection components, and a reflection intensity map is generated through inverse rendering modeling. A simplified bidirectional reflectance distribution function (BRDF) model is then used to describe the surface reflection characteristics. BRDF parameters are fitted using a calibration plate or reference area, and the theoretical reflection distribution is calculated in combination with real-time lighting conditions (such as light source direction and intensity). The image brightness is reversely corrected based on the BRDF model to eliminate brightness unevenness caused by surface tilt or changes in lighting angle. This helps quantify the modulation effect of ambient light on surface reflection and correct image brightness distortion. Then, reflection component separation is used to decompose the image into diffuse reflection layer and specular reflection layer, and a dynamic weighted fusion formula is used to adjust the weights of the diffuse reflection layer and specular reflection layer to balance the diffuse and specular reflection contributions. The weights are updated according to the changes in dust, which helps to restore the real texture of the monitoring data.

[0052] The formula for dynamically updating weights is: I c =w d I d +w s I s , where I c is the image intensity balance value, which is a constant and can be adjusted according to actual conditions. d is the image intensity of the diffuse reflection layer, I s is the image intensity of the specular reflection layer, w d is the dynamic weight corresponding to the diffuse reflection layer, w s is the dynamic weight corresponding to the specular reflection layer; finally, the compensated real texture image is output for subsequent processing and analysis.

[0053] Through the configuration of this embodiment, the present invention can solve the image distortion problem caused by surface reflection interference in a high-dust environment by combining physical models with data-driven methods, and provide a reliable texture basis for subsequent analysis.

[0054] Furthermore, in one embodiment of the present invention, the preprocessing of the monitoring data to obtain preprocessed data includes: using a sparse optical flow method to process the monitoring data in the monitoring data to distinguish between dynamic areas and static areas; using a saliency measurement method to identify dynamic areas to mark significant motion areas; performing weight processing on significant motion areas to suppress static background; and performing morphological correction and time domain filtering on significant motion areas to obtain an enhanced image.

[0055] Specifically, the sparse optical flow method is used to process monitoring data, calculate local motion variables through feature point matching, and combine polarization filtering to suppress false optical flow caused by dust scattering; and dynamically adjust the optical flow algorithm window size, such as the number of pyramid layers, to adapt to different motion speeds. The calculation formula of the sparse optical flow method is:

[0056]

[0057] Among them, u is the horizontal optical flow component of the pixel point (x, y), v is the vertical optical flow component of the pixel point (x, y), λ is the smoothing term weight, To obtain the minimum function, I t , I t+1 is the image time scale at different time points, is the change in the horizontal optical flow component from time t to time t+1, and V(x, y) is the optical flow velocity from time t to time t+1. Dynamic and static areas in the surveillance data are distinguished by calculating the motion vector of each pixel. Binarization is then performed based on the relationship between the optical flow velocity modulus and the corresponding preset threshold to mark salient areas.

[0058] Weight processing is performed on the significant motion area to suppress the static background, including: calculating the basic weight, time accumulation weight and spatial attenuation weight based on the monitoring data adjusted for the dynamic area; averaging the basic weight, time accumulation weight and spatial attenuation weight to generate the final dynamic weight map, and then applying the dynamic weight map to the monitoring data to achieve the effect of enhancing the motion area and suppressing background noise. Among them, the basic weight W b The calculation formula is: Among them, M m is the result value of binary differentiation, η is the depth attenuation coefficient, ranging from (0.5-2), σ z is the standard deviation of the depth map.

[0059] Time cumulative weight W tzThe calculation formula is:

[0060] Among them, M m (x, y, t z ) is the tth z The result value of the frame binarization distinction, min is the minimum value, T z is the total number of historical frames.

[0061] Spatial attenuation weight W sp The calculation formula is:

[0062] Among them, d is the distance from the current pixel to the nearest moving pixel obtained based on morphological distance transformation, σ d is the spatial attenuation coefficient.

[0063] The morphological correction and time-domain filtering are performed on the significant motion area to obtain the enhanced image in the following manner: the monitoring data of the enhanced motion area and the suppressed background noise are processed using the morphological denoising method and the time-domain moving average filter to eliminate noise holes and generate a stable motion mask; finally, the image is enhanced using the weight fusion method, the motion area is processed using histogram equalization, and the binary motion mask is Gaussian blurred to generate a probability map; after image enhancement, histogram equalization, and binary motion mask processing, the enhanced monitoring data is output, i.e., the preprocessed data.

[0064] The monitoring data is processed through the motion area enhancement algorithm to highlight the dynamic area and suppress static background interference, thereby improving the robustness of subsequent analysis and processing.

[0065] In one embodiment of the present invention, the use of a preset multi-task joint detection method to identify and analyze preprocessed data to obtain security detection results includes: fusing the monitoring data, audio data, and environmental data in the preprocessed data to obtain fused data; using an improved YOLOv5 backbone network algorithm to extract features from the fused data to obtain a feature vector; using a multi-task output head algorithm to detect and analyze the feature vector to obtain a detection result; using a multi-task loss function to adjust the detection result and optimize the multi-task joint detection method to output the multi-task detection result.

[0066] The fused data is obtained by fusing the monitoring data, audio data, and environmental data from the preprocessed data. The fused data is normalized and then concatenated on a time-scale and channel-wise basis to form the fused data, enhancing the feature representation capabilities of each data type. An improved YOLOv5 backbone network algorithm is employed through convolutional layers (CSP-Darknet53) and SPP modules to capture the spatial and semantic features of the image. This generates a shared feature map composed of feature vectors, which is shared across all detection tasks, avoiding repeated computation and reducing computational redundancy. A multi-task output head algorithm is employed to detect and analyze the feature vectors. The output heads for the detection results include a flame / smoke detection head, a temperature anomaly detection head, and a shared localization head. The flame / smoke detection head uses independent convolutional layers to predict bounding boxes and confidence scores for target localization and classification. The temperature anomaly detection head uses independent convolutional layers to output temperature anomaly probabilities for classification. The shared localization head allows all tasks to share bounding box regression parameters, reducing computational complexity.

[0067] A multi-task loss function is used to adjust the detection results and optimize the multi-task joint detection method. The process of outputting multi-task detection results involves: simultaneously optimizing the localization and classification tasks through a weighted time function to balance multi-task learning; and adjusting the loss contributions of different tasks to alleviate the imbalance between tasks. Finally, non-maximum suppression is used to filter out overlapping redundant boxes, retaining the boxes with the highest confidence. The bounding boxes of flames and smoke are then combined with the temperature anomaly areas to form the final output for safety decision-making.

[0068] In one embodiment of the present invention, the improved YOLOv5 backbone network algorithm is used to extract features from the fused data to obtain a feature vector, including:

[0069] The convolutional layer and spatial pyramid pooling method are used to process the fused data to capture the spatial features and semantic information of the image and generate a shared feature map containing feature vectors.

[0070] The specific process is: in multiple convolutional layers, the data processing formula of the convolutional layer is:

[0071] X l+1 =ReLU(W l *X l +b l ),

[0072] Among them, X l+1 is the input feature map of the l+1th layer, ReLU is the activation function, and in this embodiment, it is the maximum value function, W l is the convolution kernel weight, X lis the input feature map of the lth layer, b l is the bias term.

[0073] Then the output feature map is pooled and a fixed-length feature vector is output; the processing formula is: X spp =Concat(Pool(X,k=1),Pool(X,k=3),Pool(X,k=5)),

[0074] Among them, X spp is the concatenated multi-scale feature vector, Concat is the concatenation function, Pool is the pooling function, X is the input feature map, and k is the pooling scale.

[0075] Through pooling operations with different k values, global, medium and local features are captured, and then the global, medium and local features are spliced ​​to obtain a feature vector of fixed length as the input of subsequent tasks, achieving the feature of invariance to input size.

[0076] In one embodiment of the present invention, the use of a multi-task output head algorithm to detect and analyze feature vectors to obtain detection results includes: using a Sigmoid function to analyze and process the corresponding feature vectors to obtain flame confidence and smoke confidence; using a Softmax activation function to analyze and process the corresponding feature vectors to obtain temperature anomaly distribution information; using a preset shared positioning formula to analyze and process the corresponding feature vectors to obtain bounding box coordinate information; and integrating the flame confidence, smoke confidence, temperature anomaly distribution information, and bounding box coordinate information to obtain detection results.

[0077] The specific process is: use the Sigmoid function to analyze and process the corresponding eigenvectors, and obtain the formulas for flame confidence and smoke confidence respectively:

[0078] Flame confidence: Y flame =Sigmoid(W flame ·X feat +b flame ),

[0079] Among them, Y flame is the scalar probability value predicted by the flame bounding box, Sigmoid is the activation function, W flame is the flame weight matrix, X feat is the feature map tensor, i.e., the multi-scale feature vector, b flame is the flame bias vector.

[0080] Smoke confidence: Y smoke =Sigmoid(W smoke ·X feat +bsmoke ),

[0081] Among them, Y smoke is the scalar probability value of the smoke bounding box prediction, W smoke is the smoke weight matrix, b smoke is the smoke bias vector.

[0082] Temperature anomaly distribution information: Y temp =Soft max(W temp ·X feat +b temp ),

[0083] Among them, Y temp is the vector probability distribution of temperature anomaly, W temp is the temperature weight matrix, b temp is the temperature bias vector.

[0084] Bounding box coordinate information: B = W loc ·X feat +b loc ,

[0085] Where B is the bounding box coordinate, W loc is the weight matrix of the shared localization head, b loc is the bias vector of the shared positioning head.

[0086] The detection results are obtained by calculating and integrating the flame confidence, smoke confidence, temperature anomaly distribution information, and bounding box coordinate information. This embodiment can simultaneously complete multi-task classification (flame, temperature anomaly, smoke) and shared positioning, taking into account data processing efficiency and multi-task relevance.

[0087] In one embodiment of the present invention, the use of a multi-task loss function to adjust the detection results and optimize the multi-task joint detection method to output the multi-task detection results includes: using each preset loss function to analyze the detection results and the true label to determine the loss information; adjusting the corresponding detection results according to the loss information; and determining the model optimization parameters according to the parameter information corresponding to the minimized total loss value of back propagation.

[0088] The specific process is: the total loss L total The calculation formula is: L total =λ1L loc +λ2L flame +λ3L smoke +λ4L temp ,

[0089] Among them, L loc is the shared positioning loss, L flame is the flame classification loss, Lsmoke is the smoke classification loss, L temp is the temperature anomaly classification loss, λ1, λ2, λ3, and λ4 are weight coefficients used to adjust the contribution of each task loss. Minimize L through back propagation total , while optimizing positioning and classification services.

[0090] The calculation formula of shared positioning loss is:

[0091]

[0092] in, is the predicted bounding box of the i-th target, is the ground-truth bounding box of the i-th target, SmoothL1 is the smooth L1 loss, which is more robust to small errors, and SmoothL1(x) is the calculation formula of SmoothL1. The shared positioning loss is used to optimize target positioning accuracy and reduce the deviation between the predicted box and the ground-truth box.

[0093] Flame classification loss L flame The calculation formula is:

[0094] in, is the true label of the flame (the value is 0 or 1), is the flame probability predicted by the model (the output of Sigmoid, in the range [0, 1][0, 1]); the flame classification loss is used to optimize the classification confidence of flame detection.

[0095] Smoke classification loss L smoke The calculation formula is:

[0096]

[0097] in, is the true label of the flame (the value is 0 or 1), is the smoke probability predicted by the model (the output of Sigmoid, in the range [0, 1][0, 1]); the smoke classification loss is used to optimize the classification confidence of smoke detection.

[0098] Temperature anomaly classification loss L temp The calculation formula is:

[0099]

[0100] Where C is the temperature category number (normal, high temperature, low temperature), c is one of the temperature categories, is the true label of the c-th temperature, The probability of the c-th temperature predicted by the model (Softmax output result, range [0, 1]); the temperature anomaly classification loss is used to optimize the confidence of the temperature anomaly classification.

[0101] The optimization goal of the total loss is to minimize L through back propagation total , while optimizing localization and classification tasks.

[0102] Through the setting method of this embodiment, the present invention uses a multi-task loss function to adjust the detection results and optimize the multi-task joint detection method, thereby improving the accuracy of multi-task classification while ensuring positioning accuracy.

[0103] In one embodiment of the present invention, the predictive analysis based on monitoring data, safety detection results and task order information to obtain safety warning information includes: using a digital twin warning model constructed based on digital twins to process the monitoring data, safety detection results and task order information, simulating dust diffusion information and calculating dust concentration and duration; determining a processing strategy based on the matching relationship between dust concentration, duration and a preset strategy library; and generating safety warning information based on the dust concentration and processing strategy.

[0104] The digital twin early warning model is a dust movement simulation model based on digital twin technology and the laws of physical motion.

[0105] The preset strategy library records the corresponding treatment strategies for different dust concentrations and durations. For example, if the dust concentration is greater than 200mg / m 3 If the dust concentration is greater than 500mg / m 3 If the duration is greater than 10s, the associated equipment will be stopped urgently.

[0106] Through the setting method of this embodiment, the present invention can simulate the movement state of dust in the digital twin warning model based on monitoring data, safety detection results and task order information, and calculate the dust concentration and duration. When the warning conditions are triggered, the corresponding safety protection measures are activated and warning information is sent to the supervisor.

[0107] Furthermore, when the resolution of the monitoring image obtained by the monitoring device is less than a preset value, a dust removal operation is triggered or an early warning message is sent to the management personnel so that personnel can be arranged to perform cleaning operations in time to ensure the normal operation of the monitoring device.

[0108] In one embodiment of the present invention, as shown in the attached Figure 2As shown, the present invention also provides a security monitoring system for an industrial park, which includes a data acquisition module, a preprocessing module, a security detection module, and a security warning module; the security monitoring system implements any of the security monitoring methods described in the above embodiments through the mutual cooperation between the modules. Specifically:

[0109] The data acquisition module is used to acquire monitoring data, which includes monitoring data, environmental data and audio data.

[0110] The preprocessing module is used to preprocess the monitoring data to obtain preprocessed data.

[0111] The security detection module uses a preset multi-task joint detection method to identify and analyze the preprocessed data to obtain security detection results.

[0112] The safety warning module performs predictive analysis based on monitoring data, safety test results and task order information to obtain safety warning information.

[0113] The present invention pre-processes the monitoring data through a dynamic dust suppression method, a light reflection compensation method and a motion area enhancement method to obtain pre-processed data that has undergone dust suppression processing, light reflection compensation and motion area enhancement, which is convenient for subsequent analysis and processing and improves the accuracy of data processing; then a preset multi-task joint detection method is used to identify and analyze the pre-processed data to obtain a multi-task integrated safety detection result; finally, a predictive analysis is performed based on the monitoring data, safety detection results and task order information to obtain safety warning information, so that supervisors can carry out targeted processing according to the safety warning information and avoid the occurrence of safety accidents in advance. At the same time, it helps to solve the problem that the existing AI model training data is mostly targeted at general scenarios, and is not optimized for the unique interference of powder plant areas in chemical plant areas (such as high concentration of dust, light reflection). When applied to powder plant areas, the general safety monitoring method has a high false alarm rate.

[0114] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system comprising at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0115] It should be noted that, in the description of the present invention, the terms "first", "second" and "third" are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0116] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0117] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0118] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0119] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0120] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0121] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A security monitoring method for an industrial park, characterized in that: include: Acquiring monitoring data, wherein the monitoring data includes monitoring data, environmental data, and audio data; Preprocessing the monitoring data to obtain preprocessed data; Use the preset multi-task joint detection method to identify and analyze the pre-processed data to obtain security detection results; Based on monitoring data, safety inspection results and task order information, predictive analysis is carried out to obtain safety warning information.

2. The security monitoring method according to claim 1, characterized in that: The preprocessing of the monitoring data to obtain preprocessed data includes: processing the monitoring data using a preset dynamic dust suppression method to suppress the optical interference of dust; processing the monitoring data using a transmitted light model to compensate for the powder shielding effect; and processing the audio data using wavelet frequency domain filtering to suppress high-frequency noise. The formula of the transmitted light compensation model is: Where C is the dust concentration, T is the transmittance, and a and b are on-site calibration coefficients.

3. The security monitoring method according to claim 1, characterized in that: The preprocessing of the monitoring data to obtain preprocessed data includes: using a preset light reflection compensation method to process the monitoring data in the monitoring data to achieve light reflection compensation; specifically including: using a linear polarizer group to construct a Stokes parameter matrix to solve the surface reflection component; based on the physical BRDF model, inferring the real surface texture to obtain corrected monitoring data; wherein the polarization filter formula is: Among them, S is the image after surface reflection suppression, I 0° , I 45° , I 90° is the image intensity at different polarization angles; The inverse rendering modeling formula is: v=κ·cosθ·(1+α·N·H n ), Among them, κ is the material refractive index, θ is the incident angle, N and H are the surface normal and the unit vector of the illumination direction, α and n are empirical parameters obtained through sample training, and v is the reflection intensity.

4. The security monitoring method according to claim 1, characterized in that: The preprocessing of the monitoring data to obtain preprocessed data includes: using a sparse optical flow method to process the monitoring data in the monitoring data to distinguish dynamic areas from static areas; using a saliency measurement method to identify dynamic areas to mark significant motion areas; performing weight processing on significant motion areas to suppress static backgrounds; and performing morphological correction and time domain filtering on significant motion areas to obtain enhanced images.

5. The security monitoring method according to claim 1, characterized in that: The method adopts a preset multi-task joint detection method to identify and analyze the preprocessed data to obtain a security detection result, including: fusing the monitoring data, audio data and environmental data in the preprocessed data to obtain fused data; using an improved YOLOv5 backbone network algorithm to extract features from the fused data to obtain a feature vector; using a multi-task output head algorithm to detect and analyze the feature vector to obtain a detection result; using a multi-task loss function to adjust the detection result and optimize the multi-task joint detection method to output the multi-task detection result.

6. The security monitoring method according to claim 5, characterized in that: The improved YOLOv5 backbone network algorithm is used to extract features from the fused data to obtain a feature vector, including: using a convolutional layer and a spatial pyramid pooling method to process the fused data, capturing the spatial features and semantic information of the image, and generating a shared feature map containing the feature vector.

7. The security monitoring method according to claim 5, characterized in that: The method of using a multi-task output head algorithm to detect and analyze feature vectors to obtain detection results includes: using a Sigmoid function to analyze and process corresponding feature vectors to obtain flame confidence and smoke confidence; using a Softmax activation function to analyze and process corresponding feature vectors to obtain temperature anomaly distribution information; using a preset shared positioning formula to analyze and process corresponding feature vectors to obtain bounding box coordinate information; and integrating the flame confidence, smoke confidence, temperature anomaly distribution information, and bounding box coordinate information to obtain detection results.

8. The security monitoring method according to claim 5, characterized in that: The described method of adjusting the detection results using a multi-task loss function and optimizing the multi-task joint detection method to output the multi-task detection results includes: using each preset loss function to analyze the detection results and the true label to determine the loss information; adjusting the corresponding detection results according to the loss information; and determining the model optimization parameters according to the parameter information corresponding to the minimized total loss value of back propagation.

9. The security monitoring method according to claim 1, characterized in that: The predictive analysis based on monitoring data, safety detection results and task order information to obtain safety warning information includes: using a digital twin warning model constructed based on digital twins to process monitoring data, safety detection results and task order information, simulating dust diffusion information and calculating dust concentration and duration; determining a processing strategy based on the matching relationship between dust concentration, duration and a preset strategy library; and generating safety warning information based on dust concentration and processing strategy.

10. A security monitoring system for an industrial park, characterized in that: The security monitoring system includes a data acquisition module, a preprocessing module, a security detection module and a security warning module; the security monitoring system implements the security monitoring method according to any one of the above claims through the mutual cooperation between the modules.

Images