Construction risk early warning method based on multi-dimensional supervision information fusion

Abstract

The invention discloses a construction risk early warning method based on multi-dimensional supervision information fusion. The method comprises the steps of periodically obtaining a sensor data set and a construction image of each sub-region every preset time interval; generating a first construction feature vector and a second construction feature vector by using a preset construction feature extraction algorithm; based on each sub-region, inputting the first construction feature vector and the second construction feature vector into a preset dual-channel risk prediction model, and outputting a construction risk value corresponding to the sub-region; according to the construction risk value of each sub-region and a preset risk threshold interval, determining a construction risk level of each sub-region; and according to the construction risk level of each sub-region, executing a region differentiation early warning control strategy. Therefore, the accuracy of construction risk assessment is improved, and the probability of accidents is reduced.

Description

Technical Field

[0001] The present invention relates to the field of construction technologies, and in particular, to a construction risk early warning method for multi-dimensional supervision information fusion. Background Art

[0002] In the field of risk early warning, quantitative assessments that have been widely implemented at home and abroad are concentrated in the finance and financial industries. However, the risk management of construction project supervision in China is still in the initial development stage, and it is relatively difficult to practice quantitative risk assessment and risk early warning in the construction industry. At the same time, quantitative risk assessment outside the finance and financial industries is mostly in the trial stage, especially in the construction industry, where there are no existing research results. In the fields of data mining and online analytical processing, both at home and abroad have entered the mature stage. Foreign technologies started relatively earlier and have more mature tools that occupy a larger market share, such as Tableu, PowerBI, sales force, etc. Such analytical tools are mostly used for supply chain and sales data analysis, and the data models are more suitable for business data analysis and cannot be directly applied to the fields of risk quantitative assessment and risk early warning.

[0003] The traditional methods have the following problems: Data singularity: Traditional methods often rely only on single data and are difficult to comprehensively reflect the actual situation of the construction site; Static risk assessment: The risk assessment of traditional methods is often based on historical data or preset rules and is difficult to dynamically adjust according to real-time data; Lack of pertinence in risk control: The risk control strategies of traditional methods are often relatively general and are difficult to accurately control the risk situations of different sub-regions. Summary of the Invention

[0004] The present application provides a construction risk early warning method for multi-dimensional supervision information fusion, which improves the accuracy of construction risk assessment and reduces the probability of accidents.

[0005] The present application provides a construction risk early warning method for multi-dimensional supervision information fusion, including: S101, based on a construction site that is pre-divided into several sub-regions, using the deployed sensors and image acquisition devices, periodically obtaining the sensor data set and construction images of each sub-region at preset time intervals; S102, based on the sensor data set and construction images, using a preset construction feature extraction algorithm to generate a first construction feature vector and a second construction feature vector; S103, based on each sub-region, inputting the first construction feature vector and the second construction feature vector into a preset dual-channel risk prediction model to output the construction risk value corresponding to the sub-region; S104. Determine the construction risk level of each sub-region according to the construction risk value of each sub-region and the preset risk threshold range. S105. Implement the regional differentiation early warning control strategy according to the construction risk level of each sub-region.

[0006] Preferably, a number of sensors are deployed in each of the sub-regions to collect a sensor data set, which includes a dust pollution coefficient, temperature, humidity, wind speed, and wind direction; an image acquisition device is provided in each sub-region, and the image acquisition device is used to collect the construction image of the corresponding sub-region.

[0007] Preferably, the preset construction feature extraction algorithm specifically includes: A1. Combine the dust pollution coefficient, temperature, humidity, and wind speed to form a first construction feature vector. A2. Input the construction image into a pre-trained construction hidden danger identification model, and output a target image with the same resolution as the input image. The classification value of each pixel in the target image is used to represent the hidden danger object identifier to which the pixel in the construction image belongs. The hidden danger objects include dust, landslide, water accumulation, and personnel. A3. Perform pixel merging according to the identifiers of all pixel points in the target image to obtain at least one hidden danger domain, and replace the original target image with the merged target image. A4. Based on the target image, use the preset hidden danger domain analysis algorithm to generate a second construction feature vector, including dust distribution density, landslide area ratio, water accumulation area ratio, personnel density, and hidden danger coupling index.

[0008] Preferably, the preset hidden danger domain analysis algorithm includes: B1. Calculate the ratio of the total area of the hidden danger domain corresponding to each hidden danger object identifier in the target image to the area of the target image to obtain the dust distribution density, landslide area ratio, water accumulation area ratio, and personnel density. B2. For each hidden danger domain with the hidden danger object being personnel and each hidden danger domain with each other hidden danger object identifier , calculate the hidden danger coupling index:

[0009] where Q is the hidden danger coupling index, is the spatial proximity between the hidden danger domain and the hidden danger domain with the jth hidden danger object identifier , [[ID=3,9]] is the preset weight value corresponding to the hidden danger object identifier j, is the area value of the hidden danger domain ; B3. The dust distribution density, the proportion of the area of the collapse area, the proportion of the area of the water accumulation area, the personnel density, and the hidden danger coupling index are combined to form the second construction feature vector.

[0010] Preferably, the spatial proximity is determined according to the Euclidean distance value, specifically:

[0011] Among them, is the hidden danger domain and the hidden danger domain of the jth hidden danger object identifier The spatial proximity between them, K is the total number of hidden danger domains corresponding to the jth hidden danger object identifier, is the hidden danger domain The Euclidean distance value between and the center point of the kth hidden danger domain.

[0012] Preferably, the pre-set dual-channel risk prediction model includes a first-channel risk prediction model, a second-channel risk prediction model, and a dual-channel fusion layer. The first-channel risk prediction model is used to receive the first construction feature vector and output the first risk value; the second-channel risk prediction model is used to receive the second construction feature vector and output the second risk value; the output ends of the first-channel risk prediction model and the second-channel risk prediction model are connected to the input end of the dual-channel fusion layer, and the dual-channel fusion layer is used to perform weighted summation on the received first risk value and the second risk value and output the construction risk value.

[0013] Preferably, each of the sub-regions corresponds to a dual-channel risk prediction model. The training method of the dual-channel risk prediction model includes: C1. Collect the first construction feature vectors and the second construction feature vectors corresponding to a large number of historical construction images in this sub-region, and perform label annotation on the first construction feature vectors and the second construction feature vectors. The label annotation content is set as the first risk value and the second risk value; C2. Use the first construction feature vectors and the second construction feature vectors after label annotation of all historical construction images as the training data set, train the pre-selected neural network structure, and continuously optimize the model parameters to obtain the first-channel risk prediction model and the second-channel risk prediction model; C3. Connect the first-channel risk prediction model and the second-channel risk prediction model with the pre-set dual-channel fusion layer to generate the final dual-channel risk prediction model.

[0014] Preferably, before the S103, the method further includes: S201. Based on the construction plan, define the spatial adjacency relationship between sub-regions, obtain all sub-regions with spatial adjacency relationships as target sub-regions, and obtain the wind direction of all target sub-regions from the sensor data set; S202. Based on each target sub-region, determine whether its wind direction points to any other target sub-region. If so, use the other target sub-region pointed to as the transfer sub-region of this target sub-region; S203. Based on each target sub-region and its corresponding transfer sub-region, introduce a dynamic adjacency weight value , representing the dust transfer probability from target sub-region x to its transfer sub-region y:

[0015] where, is the dust transfer probability, is the dynamic adjacency weight value from target sub-region x to its transfer sub-region y, is the trend weight value, is the wind speed weight value, is the distance weight value; , is the transfer angle, the wind speed weight value is the normalized value of the wind speed, is the value after normalizing the Euclidean distance between the centers of the two sub-regions; S204. Use the transfer sub-region with a dust transfer probability greater than the preset transfer threshold as the hidden danger sub-region. Based on each hidden danger sub-region, adjust the preset risk threshold interval of the hidden danger sub-region according to the difference between the dust transfer probability corresponding to the hidden danger sub-region and the preset transfer threshold.

[0016] Preferably, the obtaining method of the transfer angle is as follows: D1. Obtain all the hidden danger areas in the target sub-region with the hidden danger object identifier being dust. Calculate the average value of all pixel values in each hidden danger area. Use the hidden danger areas corresponding to the maximum and minimum average values as the first hidden danger area and the second hidden danger area respectively. Make a first ray from the center of the first hidden danger area to the center of the second hidden danger area; D2. Make a second ray from the center of the first hidden danger area to the center of the transfer sub-region corresponding to the target sub-region. Determine the angle formed by the first ray and the second ray as .

[0017] Preferably, after the S105, the method further includes: S301. Periodically execute steps S101 to S105 to obtain the target sub-regions corresponding to the dust transfer probability greater than the preset transfer threshold within the current period; S302. According to the image acquisition device, the first hidden danger area and the second hidden danger area in the target sub-region, adjust the shooting angle of the image acquisition device in the target sub-region in the next period.

[0018] One or more technical solutions provided in this application have at least the following technical effects or advantages: By fusing sensors and image data sources, a multi-dimensional construction feature vector containing environmental parameters (such as dust pollution coefficient, temperature, humidity, wind speed, wind direction) and image features (such as dust distribution density, landslide area, water accumulation area, personnel density) is generated. This fusion method can more comprehensively reflect the actual situation of the construction site and provide a rich data basis for subsequent risk assessment; introducing the relative distribution of construction site personnel and various potential hazard objects can comprehensively consider the construction safety hazards in this sub-region and improve the personal safety of construction workers; Using a dual-channel risk prediction model, it can process sensor data and image data separately, and combine the two through a dual-channel fusion layer to output a more accurate construction risk value. It can make full use of the advantages of different data sources to improve the accuracy of risk assessment; according to the construction risk value, the construction risk level of each sub-region can be determined in real time, and a differential early warning control strategy can be executed. The dynamic management method can timely discover and handle potential safety hazards, reduce the probability of accidents, and improve construction efficiency; By introducing the spatial adjacency relationship and dynamic adjacency weight values between sub-regions, the dynamic transfer of hazards such as dust between sub-regions is considered, making the risk assessment more comprehensive and accurate; based on the dust transfer probability, potential hazard sub-regions can be mined, and the risk threshold interval of potential hazard sub-regions can be dynamically adjusted, avoiding the problem of unreasonable risk threshold setting caused by the lack of dynamic correlation analysis between sub-regions in traditional methods; By more accurately assessing the risk transfer between sub-regions, potential safety hazards can be discovered in advance, improving the accuracy and timeliness of early warning; by dynamically adjusting the risk threshold and early warning strategy, the probability of accidents can be more effectively reduced, and the safety of the construction site can be improved. Brief Description of the Drawings

[0019] Figure 1 It is a schematic flow chart of the construction risk early warning method for multi-dimensional supervision information fusion according to an embodiment of the present invention. Detailed Embodiments

[0020] To facilitate the understanding of the present invention, the present application will be described more comprehensively with reference to the relevant drawings; the preferred embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, these embodiments are provided to make the disclosure of the present invention more thorough and comprehensive.

[0021] It should be noted that the terms "vertical", "horizontal", "upper", "lower", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only embodiments.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this invention belongs; the terms used in the specification of this invention are only for the purpose of describing specific embodiments and are not intended to limit this invention; the term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0023] Embodiment 1: Figure 1 It is a schematic flowchart of a construction risk early warning method for multi-dimensional supervision information fusion according to an embodiment of the present invention.

[0024] As Figure 1 shown, a construction risk early warning method for multi-dimensional supervision information fusion includes the following steps: S101, based on a construction site that is pre-divided into several sub-regions, using deployed sensors and image acquisition devices, periodically obtain a sensor data set and construction images for each sub-region at preset time intervals.

[0025] Specifically, several sensors are deployed in each sub-region for collecting a sensor data set, and the sensor data set includes but is not limited to dust pollution coefficient, temperature, humidity, wind speed, and wind direction; an image acquisition device is provided in each sub-region, and the image acquisition device is used to collect construction images of the corresponding sub-region.

[0026] S102, based on the sensor data set and construction images, use a pre-set construction feature extraction algorithm to generate a first construction feature vector and a second construction feature vector.

[0027] Specifically, the first construction feature vector is composed of dust pollution coefficient, temperature, humidity, and wind speed, and the second construction feature vector is composed of dust distribution density, area value of the collapse area, area value of the water accumulation area, and personnel density. For example, the first construction feature vector of the i-th sub-region is expressed as , is the dust pollution coefficient, is the temperature, is the humidity, v is the wind speed; the second construction feature vector of the i-th sub-region is expressed as , D is the dust distribution density, A is the area value of the collapse area, W is the area value of the water accumulation area, P is the personnel density, and Q is the hidden danger coupling index.

[0028] In some embodiments, the pre-set construction feature extraction algorithm specifically includes: A1. Combine the dust pollution coefficient, temperature, humidity, and wind speed to form the first construction feature vector.

[0029] Exemplarily, the sensors deployed in the sub-region directly measure the dust concentration in the air through technologies such as laser scattering and optical sensing, and convert it into a dust pollution coefficient according to a preset standard or algorithm. This coefficient reflects the severity of dust pollution in the current region; in addition, temperature, humidity, and wind speed can all be obtained by collecting and processing through corresponding sensors, which will not be elaborated in this invention and can refer to relevant existing technologies.

[0030] A2. Input the construction image into the pre-trained construction hazard identification model, and output a target image with the same resolution as the input image. The classification value of each pixel in the target image is used to represent the hazard object identifier to which the pixel in the construction image belongs. The hazard objects include dust, landslide, water accumulation, and personnel.

[0031] Among them, for the training of the construction hazard identification model, a large number of historical construction images in this region are used for annotation (hazard object identifier, dust, landslide, water accumulation, personnel) to obtain a training data set, and the pre-set neural network structure is trained using the training data set to obtain the final construction hazard identification model.

[0032] A3. Perform pixel point merging according to the identifiers of all pixel points in the target image to obtain at least one hazard domain, and replace the original target image with the target image after the merging process.

[0033] Specifically, the conditions for pixel point merging are set as follows: Pixel points with the same hazard object identifier and a dispersion degree less than the dispersion threshold are merged. The dispersion degree can be set as the Euclidean distance value between the centers of two pixel points, and the dispersion threshold can be set according to the actual situation and actual needs.

[0034] A4. Based on the target image, use the pre-set hazard domain analysis algorithm to generate a second construction feature vector, including dust distribution density, landslide area ratio, water accumulation area ratio, personnel density, and hazard coupling index.

[0035] Specifically, the pre-set hazard domain analysis algorithm includes: B1. Calculate the ratio of the total area of the hazard domain corresponding to each hazard object identifier in the target image to the area of the target image to obtain the dust distribution density, landslide area ratio, water accumulation area ratio, and personnel density.

[0036] It should be noted that the area of the target image is set as the total number of all pixels, and the total area of the hazard domain corresponding to each hazard object identifier is set as the total number of pixels in all hazard domains corresponding to this identifier; if a certain hazard object identifier does not appear in the target image, the total area of its hazard domain can be recorded as 0.

[0037] B2. For each hazard domain with the hazard object being personnel The hazard domain identified with each other hazard object (which can be any one of the hazard object identifications of dust, landslide, and water accumulation, where j is the number of the type of hazard object identification, and the value range of j is [1, 3]), calculate the spatial coupling index between the person and the hazard domain, that is, the hazard coupling index:

[0038] Among them, Q is the hazard coupling index, is the hazard domain and the hazard domain of the j-th type of hazard object identification between the spatial proximity, is the preset weight value corresponding to the hazard object identification j, is the hazard domain the area value of; the spatial proximity is determined according to the Euclidean distance value, specifically:

[0039] Among them, is the hazard domain and the hazard domain of the j-th type of hazard object identification between the spatial proximity, K is the total number of hazard domains corresponding to the j-th type of hazard object identification, is the hazard domain the Euclidean distance value from the center point of the k-th hazard domain.

[0040] It should be noted that the spatial proximity reflects the spatial proximity degree between the person's hazard domain and the hazard domains of other hazard objects. The larger the value, the closer or more overlapping the two hazard domains are, and the higher the risk; the weight value reflects the relative importance of different hazard object identifications to the safety of personnel, which is set according to the actual situation and expert experience, and the range is between 0 and 1. For example, the landslide hazard object may pose a greater threat to personnel safety than the water accumulation hazard object, so a higher weight can be assigned to the landslide hazard.

[0041] Therefore, the hazard coupling index is defined as: a quantitative value of the spatial overlap degree or proximity degree between the hazard domain corresponding to the person and the hazard domains corresponding to other hazard object identifications, considering the distribution position relationship between the hazard domain corresponding to the person and the hazard domains corresponding to other hazard objects (such as dust, landslide, water accumulation), so as to more accurately evaluate the risk environment where the person is located. The higher the value, the higher the coupling degree between the person and the dangerous environment, and the greater the risk.

[0042] B3. Combine the dust distribution density, the proportion of the landslide area, the proportion of the water accumulation area, the personnel density, and the hazard coupling index to form the second construction feature vector.

[0043] S103. Based on each sub-region, input the first construction feature vector and the second construction feature vector into a pre-set dual-channel risk prediction model, and output the construction risk value corresponding to this sub-region.

[0044] In some embodiments, the pre-set dual-channel risk prediction model includes a first-channel risk prediction model, a second-channel risk prediction model, and a dual-channel fusion layer. The first-channel risk prediction model is used to receive the first construction feature vector and output a first risk value; the second-channel risk prediction model is used to receive the second construction feature vector and output a second risk value; the output ends of the first-channel risk prediction model and the second-channel risk prediction model are connected to the input end of the dual-channel fusion layer, and the dual-channel fusion layer is used to perform weighted summation on the received first risk value and second risk value and output the construction risk value.

[0045] In some embodiments, each sub-region corresponds to a dual-channel risk prediction model. The training method of the dual-channel risk prediction model includes: C1. Collect the first construction feature vector and the second construction feature vector corresponding to a large number of historical construction images in this sub-region, and perform label annotation on the first construction feature vector and the second construction feature vector. The label annotation content is set as the first risk value and the second risk value.

[0046] Specifically, for the first construction feature vector and the second construction feature vector of each historical construction image, the risk experience judgment of the construction image and the importance degree of different parameters in the feature vector for risk assessment (determined through expert experience and actual requirements) can be combined with expert experience to determine the first risk value of the first construction feature vector and the second risk value of the second construction feature vector. The annotation range is set from 0 to 1. The larger the value, the greater the risk. The present invention does not elaborate and make specific limitations on this.

[0047] In other examples, for the first construction feature vector, importance degree factors corresponding to the dust pollution coefficient, temperature, humidity, and wind speed are respectively assigned (the range is between 0 and 1, and the larger the value, the more important). All parameters in the first construction feature vector are normalized and weighted summed to determine the first risk value. The risk processing method for construction natural environment parameters can also refer to relevant existing technologies, and the present invention does not elaborate on this; it can be understood that the annotation of the second construction feature vector is the same and will not be elaborated.

[0048] C2. Use the labeled first construction feature vector and the second construction feature vector of all historical construction images as the training data set to train a pre-selected neural network structure, continuously optimize the model parameters, and obtain the first-channel risk prediction model and the second-channel risk prediction model.

[0049] C3. Connect the first-channel risk prediction model and the second-channel risk prediction model to a preset dual-channel fusion layer to generate a final dual-channel risk prediction model.

[0050] S104. Determine the construction risk level of each sub-region according to the construction risk value of each sub-region and a preset risk threshold range.

[0051] Specifically, the preset risk threshold is set to [a, b], where 0 < a < b and a < b < 1; the specific values of a and b are determined according to the types of construction work in the sub-region. The more important the type of construction work, the lower the values of a and b.

[0052] Specifically, step S104 includes: When the construction risk value is less than the lower limit of the risk threshold range, determine that the construction risk level of this sub-region is low; When the construction risk value is greater than the upper limit of the risk threshold range, determine that the construction risk level of this sub-region is high; Otherwise, determine that the construction risk level of this sub-region is medium.

[0053] S105. Execute a regional differential early warning control strategy according to the construction risk level of each sub-region.

[0054] Specifically, when the construction risk level is high, automatically trigger an audible and visual alarm and suspend the construction in this sub-region; when the construction risk level is medium, push adjustment suggestions (such as adding dust removal equipment, evacuating personnel, etc.); when the construction risk level is low, mark it as a safe area.

[0055] The technical solutions in the embodiments of the present application at least have the following technical effects or advantages: By fusing sensor and image data sources, a multi-dimensional construction feature vector containing environmental parameters (such as dust pollution coefficient, temperature, humidity, wind speed, wind direction) and image features (such as dust distribution density, landslide area, water accumulation area, personnel density) is generated. This fusion method can more comprehensively reflect the actual situation of the construction site and provides a rich data basis for subsequent risk assessment; introducing the relative distribution of construction site personnel and various potential hazard objects can comprehensively consider the construction safety hazards in this sub-region and improve the personal safety of construction workers; Using the dual-channel risk prediction model, sensor data and image data can be processed separately, and the two can be combined through the dual-channel fusion layer to output a more accurate construction risk value. It can make full use of the advantages of different data sources to improve the accuracy of risk assessment; according to the construction risk value, the construction risk level of each sub-region can be determined in real time, and a differential early warning control strategy can be executed. The dynamic management and control method can timely discover and handle potential safety hazards, reduce the probability of accidents, and improve construction efficiency.

[0056] Example 2: At large construction sites, there are dynamic correlations such as dust diffusion between sub-regions. Example 1 mainly focuses on the risk assessment of individual sub-regions and lacks the analysis of dynamic correlations between sub-regions, resulting in unreasonable risk threshold settings and risk transmission blind spots.

[0057] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0058] In some embodiments, before step S103, the method further includes: S201. Based on the construction plan, define the spatial adjacency relationships (such as adjacent, diagonally adjacent) between sub-regions, obtain all sub-regions with spatial adjacency relationships as target sub-regions, and obtain the wind directions of all target sub-regions from the sensor dataset.

[0059] S202. Based on each target sub-region, determine whether its wind direction points to any other target sub-region. If so, use the other target sub-region pointed to as the transfer sub-region of this target sub-region.

[0060] S203. Based on each target sub-region and its corresponding transfer sub-region, introduce a dynamic adjacency weight value , representing the dust transfer probability from target sub-region x to its transfer sub-region y:

[0061] Among them, is the dust transfer probability, is the dynamic adjacency weight value from target sub-region x to its transfer sub-region y, is the trend weight value, is the wind speed weight value, is the distance weight value; specifically, , is the transfer angle, the wind speed weight value is the normalized value of the wind speed, is the value after normalizing the Euclidean distance between the centers of the two sub-regions.

[0062] It should be noted that adding to 1 is to avoid from being negative.

[0063] Specifically, the method for obtaining the transfer angle is as follows: D1. Obtain all the hazard regions in the target sub-region where the hazard object identification is dust. Calculate the average value of all pixel values in each hazard region. Obtain the hazard regions corresponding to the maximum and minimum average values as the first hazard region and the second hazard region respectively. Draw a first ray from the center of the first hazard region towards the center of the second hazard region; D2. Draw a second ray from the center of the first hazard region towards the center of the transfer sub-region corresponding to the target sub-region. Determine the angle formed by the first ray and the second ray as . Thus, the smaller the transfer angle, the greater the trend of dust cross-region transfer and the greater the transfer weight value.

[0064] S204. Take the transfer sub-regions with dust transfer probability greater than the preset transfer threshold (the preset transfer threshold is set according to historical experience of the actual situation. For example, the preset transfer threshold is set to 0.6) as the hazard sub-regions. Based on each hazard sub-region, adjust the preset risk threshold interval of the hazard sub-region according to the difference between the dust transfer probability corresponding to the hazard sub-region and the preset transfer threshold.

[0065] Specifically, take the difference between the dust transfer probability corresponding to the hazard sub-region and the preset transfer threshold as the first difference , multiply the ratio of the first difference to the preset transfer threshold by the preset adjustment factor to obtain the first adjustment factor. Take the difference between 1 and the first adjustment factor as the target adjustment factor, and multiply the upper limit value and the lower limit value in the preset risk threshold interval of the hazard sub-region by the target adjustment factor respectively to obtain the updated risk threshold interval.

[0066] Specifically, the calculation formula of the target adjustment factor is specifically:

[0067] Wherein, is the target adjustment factor, is the first difference, is the preset adjustment factor, which is set according to the actual situation and historical experience, and its value range is from 0 to 1, and the value is as large as possible and close to 1.

[0068] The technical solutions in the embodiments of the present application above have at least the following technical effects or advantages: By introducing the spatial adjacency relationship and dynamic adjacency weight value between sub-regions, the dynamic transfer of hazards such as dust between sub-regions is considered, making the risk assessment more comprehensive and accurate; based on the dust transfer probability, the hazard sub-regions can be mined, and the risk threshold interval of the hazard sub-regions can be dynamically adjusted, avoiding the problem of unreasonable risk threshold setting caused by the lack of dynamic correlation analysis between sub-regions in the traditional method; By more accurately evaluating the risk transfer between sub-regions, potential safety hazards can be discovered in advance, and the accuracy and timeliness of early warnings can be improved; by dynamically adjusting the risk threshold and early warning strategy, the probability of accidents can be more effectively reduced, and the safety of the construction site can be enhanced.

[0069] Embodiment 3: In Embodiment 1 and Embodiment 2, the shooting angle of the image acquisition device is relatively fixed, and it may not be able to capture image information related to key safety hazards such as dust transfer in a timely manner, resulting in blind spots in the risk monitoring of the construction site. Due to the lack of pertinence in image acquisition, some potential safety hazards may not be discovered in a timely manner, thus affecting the timeliness of risk early warning and increasing the probability of accidents.

[0070] Therefore, the embodiments of the present application are optimized on the basis of the above embodiments.

[0071] In some embodiments, after step S105, the method further includes: S301, periodically execute steps S101 to S105 to obtain target sub-regions corresponding to the dust transfer probability greater than the preset transfer threshold within the current cycle.

[0072] Within each cycle, based on the construction site of several pre-divided sub-regions, use sensors and image acquisition devices to obtain data, generate construction feature vectors, output construction risk values through a dual-channel risk prediction model, determine the construction risk level, and execute the early warning control strategy. At the same time, considering the spatial adjacency relationship and dynamic adjacency weight values between sub-regions, analyze the dust transfer probability, excavate hidden danger sub-regions, and adjust the risk threshold interval; At the end of each cycle, record the dust transfer probability data of all target sub-regions, and screen out the target sub-regions corresponding to the dust transfer probability greater than the preset transfer threshold within the current cycle.

[0073] S302, according to the first hidden danger domain and the second hidden danger domain in the target sub-region, adjust the shooting angle of the image acquisition device in the target sub-region in the next cycle.

[0074] Specifically, obtain the angles formed by the connecting lines between the image acquisition device in the target sub-region and the center points of the first hidden danger domain and the second hidden danger domain respectively as the adjustment angle values, and move the image acquisition device in the direction of the second hidden danger domain by the adjustment angle values to obtain the new shooting angle of the image acquisition device, so as to be able to more accurately capture the image information related to dust transfer in the target sub-region in the next cycle.

[0075] It should be noted that by using the position information of the image acquisition device, the first hidden danger area center, and the second hidden danger area center in the actual scene, and combining the shooting parameters of the image acquisition device, the adjustment angle value can be determined. The specific principle of the conversion relationship between the acquired image and the shooting parameters can be referred to the relevant existing technologies, and the present invention will not elaborate on this.

[0076] The technical solutions in the embodiments of the present application at least have the following technical effects or advantages: By adjusting the shooting angle of the image acquisition device according to the first hidden danger area and the second hidden danger area in the target sub-region, the image acquisition device can be more focused on the area that may be related to dust transfer, improving the pertinence and effectiveness of the acquired construction images; more importantly, image acquisition helps to more accurately identify construction hidden dangers, especially those related to dust transfer, thus providing more reliable data support for subsequent risk assessment and early warning; by dynamically adjusting the shooting angle of the image acquisition device, more refined monitoring of the construction site is realized, which helps to improve the safety management level of the construction site and ensure the life safety of construction workers and the smooth progress of construction.

[0077] The above is only the preferred implementation manner of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Construction risk early warning method for multi-dimensional supervision information fusion, characterized in that Including: S101, based on a construction site that is pre-divided into several sub-regions, using deployed sensors and image acquisition devices, periodically obtaining a sensor data set and construction images for each sub-region at preset time intervals; S102, based on the sensor data set and construction images, using a pre-set construction feature extraction algorithm to generate a first construction feature vector and a second construction feature vector; S103, based on each sub-region, inputting the first construction feature vector and the second construction feature vector into a pre-set dual-channel risk prediction model, and outputting the construction risk value corresponding to the sub-region; S104, according to the construction risk value of each sub-region and a pre-set risk threshold interval, determining the construction risk level of each sub-region; S105, according to the construction risk level of each sub-region, implementing a regional differentiation early warning control strategy.

2. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 1, characterized in that A number of sensors are deployed in each of the sub-regions for collecting sensor data sets, and the sensor data sets include dust pollution coefficients, temperature, humidity, wind speed, and wind direction; an image acquisition device is provided in each sub-region, and the image acquisition device is used for collecting construction images of the corresponding sub-region.

3. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 2, wherein, The pre-set construction feature extraction algorithm specifically includes: A1. Composing a first construction feature vector from the dust pollution coefficient, temperature, humidity, and wind speed; A2. Inputting the construction image into a pre-trained construction hazard identification model, and outputting a target image with the same resolution as the input image. The classification value of each pixel in the target image is used to represent the hazard object identifier to which the pixel in the construction image belongs, and the hazard objects include dust, landslides, water accumulation, and personnel; A3. Performing pixel merging according to the identifiers of all pixel points in the target image to obtain at least one hazard domain, and replacing the original target image with the target image after the merging process; A4. Based on the target image, using a pre-set hazard domain analysis algorithm to generate a second construction feature vector, including dust distribution density, landslide area ratio, water accumulation area ratio, personnel density, and hazard coupling index.

4. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 3, characterized in that, The pre-set hazard domain analysis algorithm includes: B1. Calculating the ratio of the total area of the hazard domain corresponding to each hazard object identifier in the target image to the area of the target image to obtain the dust distribution density, landslide area ratio, water accumulation area ratio, and personnel density; B2. For each hazard domain where the hazard object is a person and each hazard domain identified by other hazard objects , calculate the hazard coupling index: Among them, Q is the hidden danger coupling index, is the hidden danger domain and the spatial proximity between the hidden danger domain and the hidden danger domain of the j-th hidden danger object identifier, is the preset weight value corresponding to the hidden danger object identifier j, is the hidden danger domain area value; B3. Composing a second construction feature vector from the dust distribution density, landslide area ratio, water accumulation area ratio, personnel density, and hazard coupling index.

5. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 4, characterized in that, The spatial proximity is determined according to the Euclidean distance value, specifically: Among them, is the hidden danger area and the spatial proximity between the hidden danger area of the j-th hidden danger object identifier, K is the total number of hidden danger areas corresponding to the j-th hidden danger object identifier, is the hidden danger area and the Euclidean distance value from the center point of the k-th hidden danger area.​ 6. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 1, characterized in that The pre-set dual-channel risk prediction model includes a first-channel risk prediction model, a second-channel risk prediction model, and a dual-channel fusion layer. The first-channel risk prediction model is used to receive the first construction feature vector and output a first risk value; the second-channel risk prediction model is used to receive the second construction feature vector and output a second risk value; the output ends of the first-channel risk prediction model and the second-channel risk prediction model are connected to the input end of the dual-channel fusion layer, and the dual-channel fusion layer is used to perform weighted summation on the received first risk value and second risk value and output the construction risk value.

7. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 6, wherein, Each of the sub-regions corresponds to a dual-channel risk prediction model. The training method of the dual-channel risk prediction model includes: C1. Collect the first construction feature vector and the second construction feature vector corresponding to a large number of historical construction images in this sub-region, and perform label annotation on the first construction feature vector and the second construction feature vector. The label annotation content is set as the first risk value and the second risk value; C2. Use the first construction feature vector and the second construction feature vector after label annotation of all historical construction images as the training data set, train the pre-selected neural network structure, and continuously optimize the model parameters to obtain the first-channel risk prediction model and the second-channel risk prediction model; C3. Connect the first-channel risk prediction model and the second-channel risk prediction model to a preset dual-channel fusion layer to generate the final dual-channel risk prediction model.

8. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 4, characterized in that Before the step S103, the method further includes: S201. Based on the construction plan, define the spatial adjacency relationship between sub-regions, obtain all sub-regions with spatial adjacency relationships as target sub-regions, and obtain the wind direction of all target sub-regions from the sensor data set; S202. Based on each target sub-region, determine whether its wind direction points to any other target sub-region. If so, use the other target sub-region pointed to as the transfer sub-region of this target sub-region; S203. Introduce dynamic adjacency weight values based on each target sub-region and its corresponding transfer sub-region , representing the dust transfer probability from the target sub-region x to its transfer sub-region y: Among them, is the dust transfer probability, is the dynamic adjacency weight value of the target sub-region x to its transfer sub-region y, is the trend weight value, is the wind speed weight value, is the distance weight value; , is the transmission included angle, and the wind speed weight value is the normalized value of the wind speed, is the value after normalizing the Euclidean distance between the centers of two sub-regions; S204. Use the transfer sub-region with a dust transfer probability greater than the preset transfer threshold as the hidden danger sub-region. Based on each hidden danger sub-region, adjust the preset risk threshold interval of the hidden danger sub-region according to the difference between the dust transfer probability corresponding to the hidden danger sub-region and the preset transfer threshold.

9. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 8, characterized in that, The acquisition method of the transmission included angle is as follows: D1. Obtain all hidden danger domains in the target sub-region with the hidden danger object identified as dust, calculate the average value of all pixel values in each hidden danger domain, and obtain the hidden danger domains corresponding to the maximum and minimum average values as the first hidden danger domain and the second hidden danger domain. Make a first ray from the center of the first hidden danger domain to the center of the second hidden danger domain; D2. Make a second ray from the center of the first hidden danger area towards the center of the transfer sub-area corresponding to the target sub-area, and determine the angle formed by the first ray and the second ray as .

10. The construction risk early warning method for multi-dimensional supervision information fusion according to claim 9, characterized in that, After the step S105, the method further includes: S301. Periodically execute steps S101 to S105 to obtain the target sub-regions corresponding to the dust transfer probability greater than the preset transfer threshold within the current period; S302. According to the image acquisition device, the first hidden danger domain and the second hidden danger domain in the target sub-region, adjust the shooting angle of the image acquisition device in the target sub-region in the next period.

Images