Labor efficiency quantitative description method considering multi-factor influence

Abstract

The invention discloses a labor efficiency quantitative description method considering multi-factor influence. The method comprises the following steps: collecting physical state data, environmental state data and construction progress data of workers; labeling the collected data according to label information; the workload of workers is identified based on the collected construction progress data, the labor efficiency is calculated, and the label information of the construction progress data and the labor efficiency data are synchronously labeled; performing data alignment on the data with the same label information according to the label information, taking the label information as an entity, taking'worker physical state data, environmental state data-labor efficiency data 'as attributes, and establishing an'influence factor-labor efficiency' knowledge graph; based on an influence factor-labor efficiency knowledge graph, deep learning is utilized to establish a mapping relationship between the physical state of a worker and the environmental factors and the labor efficiency. According to the method, the incidence relation between the influence factors and the labor efficiency can be scientifically evaluated, and the influence of the influence factors on the fluctuation of the labor efficiency can be quantitatively described.

Description

Technical Field

[0001] The present invention relates to the field of engineering construction, and specifically to a method for quantitatively characterizing labor efficiency considering the influence of multiple factors. Background Technique

[0002] Engineering construction includes three main stages: survey, design, and construction. Among them, the construction stage is the stage with the highest investment and the most important in engineering construction. Therefore, carrying out construction operations safely and efficiently determines whether the engineering construction can be completed safely and with high quality. As the core key to ensuring the smooth progress of construction operations, scientific and reasonable construction scheduling is crucial.

[0003] Construction scheduling mainly includes the coordination of resources such as personnel, materials, and machinery. Currently, when carrying out construction scheduling operations at the construction site, managers usually calculate the number of construction personnel planned to be invested based on the total construction period plan of the engineering construction, the size of the construction volume, and in combination with the labor efficiency quota. However, labor efficiency is not fixed in actual production and is extremely prone to fluctuations due to factors such as environmental factors and the physical condition of workers. Therefore, predicting the number of personnel in the form of a quota cannot accurately adapt to the changes in labor efficiency affected by various factors. This is bound to lead to the occurrence of two scenarios: one is overestimating the expected labor efficiency, resulting in too low a labor input and causing delays in the construction period; the other is underestimating the expected labor efficiency, resulting in too high a labor input and causing waste of resources. Therefore, the existing form of using a quota will make the planned labor force number specified according to the quota in construction organization unable to match the actual demand, resulting in overstaffing and understaffing phenomena in actual construction, and ultimately resulting in construction period lag or resource waste.

[0004] In existing related research, the determination of the correlation between influencing factors and labor efficiency usually only includes a single factor variable or some factor variables, and there are few studies that combine the physical condition of construction workers with construction environmental factors. In addition, an integrated determination method has not been formed yet, so that in the process of research, it is necessary to separately collect environmental data and labor efficiency data, consuming a large amount of manpower and material resources. Summary of the Invention

[0005] The present invention provides a method for quantitatively characterizing labor efficiency considering the influence of multiple factors, which can scientifically evaluate the correlation between influencing factors and labor efficiency, and quantitatively characterize the influence of influencing factors on the fluctuation of labor efficiency, break through the existing extensive construction scheduling, and improve the industry level of construction scheduling.

[0006] A method for quantitatively characterizing labor efficiency considering the influence of multiple factors includes the following steps:

[0007] Step 1. Data collection: Collect data on the physical condition of workers, environmental condition data, and construction progress data;

[0008] Step 2: Label the collected worker physical state data, environmental state data, and construction progress data according to the label information, where the label information includes the collection time, collection location, and construction type;

[0009] Step 3: Identify the workload of workers and calculate the labor efficiency based on the collected construction progress data, and synchronously label the label information of the construction progress data with the labor efficiency data;

[0010] Step 4: Align the worker physical state data, environmental state data, and labor efficiency data with the same label information according to the label information. Use the label information as the entity and "worker physical state data, environmental state data - labor efficiency data" as the attribute to establish an "influence factor - labor efficiency" knowledge graph;

[0011] Step 5: Based on the "influence factor - labor efficiency" knowledge graph, use deep learning to establish a mapping relationship between the worker physical state, environmental factors, and labor efficiency.

[0012] Furthermore, the worker physical state data is collected through an electronic file and an integrated electronic bracelet; the environmental state data is collected through an integrated sensing platform; the construction progress data is obtained from the video images collected by a mobile wireless motion camera.

[0013] Furthermore, the worker physical state data includes basic state data, physical sign state data, sleep quality data, and motion state data. The basic state data of workers is collected using an electronic file, and the basic state data includes the height, weight, gender, age, job type, proficiency, working years, and health status of the workers; the physical sign state data of workers is collected using an integrated electronic bracelet, and the physical sign state data includes the body temperature, blood oxygen, blood pressure, and heart rate of the workers; the sleep quality data of the previous night is collected before the workers start working, and the motion state data is continuously measured during the construction process. The motion state data includes the cumulative working hours of the workers, the energy consumption per unit time, and the motion amplitude of the construction operations.

[0014] Furthermore, the worker physical state data includes basic state data, physical sign state data, sleep quality data, and motion state data. The basic state data of workers is collected using an electronic file, and the basic state data includes the height, weight, gender, age, job type, proficiency, working years, and health status of the workers; the physical sign state data of workers is collected using an integrated electronic bracelet, and the physical sign state data includes the body temperature, blood oxygen, blood pressure, and heart rate of the workers; the sleep quality data of the previous night is collected before the workers start working, and the motion state data is continuously measured during the construction process. The motion state data includes the cumulative working hours of the workers, the energy consumption per unit time, and the motion amplitude of the construction operations.

[0015] Furthermore, in step three, identifying the workload of workers and calculating labor efficiency based on the collected construction progress data specifically includes:

[0016] Step 31: Using object detection to identify various construction activities in the video image;

[0017] Step 32: For the identified construction activities, through image segmentation, achieving progress estimation at the pixel level of the image;

[0018] Step 33: By comparing the progress at different times, solving the workload of workers per unit time and calculating labor efficiency.

[0019] Furthermore, step 31 uses object detection to identify various construction activities in the video image, specifically including:

[0020] ① Establishing a dedicated dataset for construction activities

[0021] For various construction activities in the video image, select a sufficient number of images, use manual annotation, and through the partition of construction activity areas, label the activity types to construct a dedicated dataset for construction activities to meet the training of the YOLO model;

[0022] ② Grid division

[0023] First, divide the input image into a finite number of grids. Each grid is responsible for detecting a partial area of the object in the image. The size of each grid is:

[0024]

[0025] In the formula, CS represents the size of each grid, W and H respectively represent the length and width of the image, and S represents the number of grids in the two directions of W and H;

[0026] ③ Boundary prediction

[0027] For image data, the YOLO model can predict multiple bounding boxes and the corresponding object class probabilities for each grid. For each grid i, B bounding boxes are output, and the information of each bounding box includes:

[0028] x i ,y i : The coordinates of the center of the bounding box relative to the grid;

[0029] w i ,h i : The width and height of the bounding box, relative to the proportion of the entire image;

[0030] c i,j : The class probability, indicating the probability that there is an object in this grid;

[0031] For each bounding box, predict the following information:

[0032] P(Objective): The confidence that the grid contains an object;

[0033] P(Class k ): The probability that the object in the grid belongs to class k;

[0034] ④ Model training

[0035] Use the special dataset for construction activities to train the YOLO model, and use the loss function to measure the difference between the predicted values and the actual values of the model. Then, optimize the model parameters to reduce the value of the loss function. The loss function consists of three parts: classification loss, which is used to measure the difference between the predicted class and the actual class; localization loss, which is used to measure the difference between the center point and size of the predicted bounding box and the true values; and confidence loss, which is used to measure the difference between the predicted probability of the object's existence and the actual value.

[0036]

[0037] In the formula, L cls represents the classification loss, is the indicator function, which is 1 when the grid (i, j) contains an object and 0 otherwise; P(Class k ) and represent the predicted class probability and the actual class probability, respectively;

[0038]

[0039] In the formula, L coord represents the location loss, and λ coord is the weight coefficient used to balance the location loss;

[0040]

[0041] In the formula, L conf represents the confidence loss, P(Objective) and represent the predicted confidence that the object is contained and the actual confidence that the object is contained, respectively;

[0042] L = L cls + L coord + L conf Equation (5)

[0043] In the formula, L represents the total loss function;

[0044] ⑤ Non-maximum suppression

[0045] When performing object detection, there may be a phenomenon where multiple detection boxes overlap and duplicate. The non-maximum suppression algorithm is used to remove the duplicate detection boxes. By calculating the overlap degree between the bounding boxes, those boxes with a high overlap degree with other boxes are removed, and only the bounding box with the highest score is retained. Finally, the above steps are repeated until all detection boxes are processed;

[0046] ⑥Output result

[0047] The category, location, and confidence of the objects in each image will be output. Among them, the category is the category probability calculated by the Softmax function, and the bounding box is represented by x i , y i , w i , h i to describe the location of the object.

[0048] Furthermore, for the identified construction activities in step 32, through image segmentation, progress estimation based on the pixel level of the image is achieved, specifically including:

[0049] ①Image preprocessing

[0050] The images within the bounding box ranges corresponding to each construction activity identified by object detection are used as inputs. By preprocessing each input image, including adjusting the image size and normalizing, it is made to meet the input requirements of the network model;

[0051] ②Feature extraction

[0052] A pre-trained convolutional neural network is used as the backbone network to extract high-level features from the input image. For each target region, precise feature extraction is performed on it through RoI operations, including edges, textures, and shapes;

[0053] ③Object mask generation

[0054] Based on each target region, a pixel-level mask for the target is generated through a fully convolutional network. This mask represents the exact shape of the object. The area where the pixel value in the mask image is 1 corresponds to the object, and the area with 0 is the background;

[0055] ④Object segmentation and region area calculation

[0056] Through the mask output, the region area of each target is calculated. By counting the number of pixel points with a value of 1 in the mask, the segmentation region area of each object is obtained;

[0057]

[0058] where M k is the mask of object k, Mask(x, y) is the mask image, representing the object pixel region. A value of 1 indicates the object region, and 0 indicates the background;

[0059] Furthermore, based on the area of the segmented region, according to the calibration ratio of the size of the video image data corresponding to the type of construction activity to the actual size, the corresponding progress value is calculated:

[0060] W k = Area(M k )·I k Equation (7)

[0061] Where W k represents the progress value, that is, the current construction progress, and I k represents the calibration ratio of the size of the image data to the actual size, and I k is obtained through calibration when arranging video measurement points.

[0062] Further, in step 33, by comparing the progress at different times, the workload per unit time of the workers is solved and the labor efficiency is calculated, specifically including:

[0063] Based on the acquisition time, acquisition location, and construction type of the image, and based on the quantified construction progress, the labor efficiency corresponding to each activity is calculated:

[0064]

[0065] Where t s represents the start time of labor efficiency monitoring, and t f represents the end time of monitoring, and min respectively represent the progress values corresponding to the construction activity k at time t s and t f .

[0066] Further, step four to establish the mapping relationship between influencing factors - labor efficiency specifically includes:

[0067] Based on the "influencing factors - labor efficiency" knowledge graph, using deep learning, establish the mapping relationship between the physical state of the workers, the environmental state and the labor efficiency L k :

[0068] L k = f(E, R) Equation (9)

[0069] Where E represents the environmental state; R represents the physical state of the workers.

[0070] The present invention has the following beneficial effects:

[0071] (1) A wireless acquisition framework based on a bracket-mounted detector is proposed. This framework can collect internal-external influencing factors and labor efficiency data, and has the characteristics of being portable, highly adaptable, highly integrated, and remotely controllable. It eliminates the human consumption in the detection process and realizes the rapid acquisition and long-term monitoring of relevant original data at low cost.

[0072] (2) A labor efficiency identification method integrating machine vision and a method for constructing the "influence factor - labor efficiency" mapping relationship combining knowledge graph and deep learning are proposed. It can realize the intelligent analysis of a large amount of labor efficiency acquisition data and the accurate characterization of labor efficiency changes, so as to carry out construction scheduling more scientifically and avoid construction period delays and resource waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a schematic diagram of the modules of the acquisition framework of the present invention;

[0074] Figure 2 It is a schematic diagram of the wearing of the worker's body condition acquisition terminal of the present invention;

[0075] Figure 3 It is a schematic diagram of the arrangement of the sensor brackets of the present invention;

[0076] Figure 4 It is a schematic diagram of the arrangement of the acquisition framework of the present invention;

[0077] Figure 5 It is a flowchart of a method for quantitatively characterizing labor efficiency considering multiple factors of the present invention.

[0078] In the figure: 1 - integrated electronic bracelet; 2 - anemometer; 3 - wind pressure gauge; 4 - temperature and humidity meter; 5 - illuminometer; 6 - pyranometer; 7 - sound intensity meter; 8 - mobile smartphone; 9 - multi-channel Bluetooth transceiver; 10 - solar power panel; 11 - mobile wireless action camera; 12 - construction operation object. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0079] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0080] Please refer to Figures 1-5 , the embodiments of the present invention provide a method for quantitatively characterizing labor efficiency considering multiple factors, including the following steps:

[0081] Step 1, Data collection: Collect data on the physical state of workers, the environmental state, and the construction progress.

[0082] Data collection is mainly used to collect the influencing factors that affect the construction efficiency of workers during the construction process and the construction progress data containing information on the workload of workers. The influencing factors include internal influencing factors (data on the physical state of workers) and environmental influencing factors (data on the environmental state).

[0083] As Figure 1 shown, among which the data on the physical state of workers is collected through electronic files and integrated electronic bracelets; the data on the environmental state is collected through an integrated sensing platform; the construction progress data is obtained by collecting video images.

[0084] Specifically, the basic state data of workers is collected using electronic files, mainly recording information such as the height, weight, gender, age, job type, proficiency, working years, and health status of workers. The electronic file is recorded when the worker starts employment and is updated regularly according to the construction cycle.

[0085] As Figure 2 shown, the integrated electronic bracelet is used to collect the physical sign state data of workers, including real-time measurement of data such as the body temperature, blood oxygen, blood pressure, and heart rate of workers; and the sleep quality data of the previous night is collected before the worker starts working, and the motion state data such as the cumulative working hours, energy consumption per unit time, and motion amplitude of the construction operation of the worker is continuously measured during the construction process.

[0086] The integrated sensing platform is used to collect data on the environmental state. The integrated sensing platform includes a sensor bracket and several sensors. The sensor bracket is a telescopic bracket, and the sensors are fixed on the sensor bracket through buckles, and the number of sensor buckles can be adaptively adjusted according to different influencing factors of concern to realize the adjustment of the type and quantity of sensors.

[0087] As Figure 3 shown, based on the study of the sensitive influencing factors of labor efficiency, the sensors in this embodiment include an anemometer 2, a wind pressure gauge 3, a temperature and humidity gauge 4, an illuminance meter 5, a solar radiometer 6, and a sound intensity meter 7. The above sensors can all be connected via Bluetooth and come with independent batteries, and are connected to a smart mobile phone 8 through a multi-channel Bluetooth transceiver 9 fixed on the sensor bracket to achieve collaborative control through the smart mobile phone 8. In addition, the smart mobile phone 8 can be remotely controlled by multiple terminals. To ensure uninterrupted power supply for each sensor, a solar panel 10 is used as a backup power source. The data on the environmental state includes the wind speed, wind pressure, temperature, humidity, light intensity, solar radiation, noise, etc. collected by the sensors.

[0088] As Figure 4As shown in the figure, the construction progress data is collected by a number of mobile wireless motion cameras 11 installed at the construction site. The video images collected by the mobile wireless motion cameras 11 are transmitted to the intelligent mobile phone 8 in real time via Bluetooth. The installation positions and quantities of the mobile wireless motion cameras 11 should ensure that all workers in the construction scene are within the field of view formed by them.

[0089] Step 2: Label the collected worker physical state data, environmental state data, and construction progress data according to the collection time, collection location, and construction type, and store the data set in the database. Among them, the data is classified and stored according to structured data (worker physical state data, environmental state data) and unstructured image data (construction progress data).

[0090] Step 3: Identify the workload of workers based on the collected construction progress data and calculate the labor efficiency, and synchronously label the label information of the construction progress data with the labor efficiency data.

[0091] Specifically, based on the collected construction progress data, using object detection and progress estimation, identify the workload of workers and calculate the labor efficiency, as Figure 5 shown, the specific steps are as follows:

[0092] Step 31: Object detection: Use object detection to identify various construction activities in the video image; mainly including steps such as establishing a special data set for construction activities, grid division, boundary prediction, model training, duplicate elimination, and output of results.

[0093] Through manual annotation, various construction activities in the video image are annotated to establish a special data set for construction activities. The image to be recognized is partitioned through grid division, and the YOLO model is used to predict the boundaries of the object to be recognized. Considering that the general YOLO model is not applicable to the recognition of construction progress targets, the present invention uses a special data set for construction progress targets to train the YOLO model. After the training is completed, the non-maximum suppression algorithm is used to eliminate the phenomenon of overlapping detection frames, and finally the YOLO model can accurately output the construction progress target.

[0094] The embodiment of the present invention uses an end-to-end object detection algorithm, which can simultaneously complete the tasks of object classification, localization, and bounding box regression in the image through a single forward propagation. This algorithm transforms the object detection problem into a regression problem, and predicts the category and position information of the object through the YOLO model. Using object detection, various construction activities and corresponding operation ranges in the construction site can be recognized through video images. Step 21 specifically includes:

[0095] ① Establish a special data set for construction activities

[0096] For each construction activity in the video image, a sufficient number of images are selected, manually labeled, and the activity types are labeled through the division of construction activity zones to construct a dedicated dataset for construction activities to meet the training of the YOLO model.

[0097] ② Grid Division

[0098] First, the input image is divided into a finite number of grids, and each grid is responsible for detecting a partial area of the object in the image. The size of each grid is:

[0099]

[0100] In the formula, CS represents the size of each grid, W and H represent the length and width of the image respectively, and S represents the number of grids in the two directions of W and H.

[0101] ③ Boundary Prediction

[0102] For the image data, the YOLO model can predict multiple bounding boxes and the corresponding object class probabilities for each grid. For each grid i, the algorithm can output B bounding boxes, and the information of each bounding box includes:

[0103] x i ,y i : The coordinates of the center of the bounding box relative to the grid (range [0,1])

[0104] w i ,h i : The width and height of the bounding box, relative to the proportion of the entire image (range [0,1])

[0105] c i,j : Class probability, indicating the probability that there is an object in this grid.

[0106] For each bounding box, the algorithm will predict the following information:

[0107] P(Objective): The confidence that this grid contains an object (this value will be multiplied by the class probability to represent the confidence of the object).

[0108] P(Class k ): The probability that the object in this grid belongs to class k.

[0109] ④ Model Training

[0110] Using a special dataset for construction activities, the YOLO model is trained, and a loss function is used to measure the difference between the model's predicted values and the actual values, and the loss function value is reduced by optimizing the model parameters. The loss function consists of three parts: classification loss (the difference between the predicted class and the actual class); localization loss (the difference between the center point and size of the predicted bounding box and the true value); confidence loss (the difference between the predicted probability of object existence and the actual value).

[0111]

[0112] In the formula, L cls represents the classification loss, is the indicator function, which is 1 when the grid (i, j) contains an object and 0 otherwise. P(Class k ) and represent the predicted class probability and the actual class probability respectively.

[0113]

[0114] In the formula, L coord represents the location loss, and λ coord is the weight coefficient used to balance the location loss.

[0115]

[0116] In the formula, L conf represents the confidence loss, P(Objective) and represent the confidence of whether the predicted object is included and the actual confidence of object inclusion respectively.

[0117] L = L cls + L coord + L conf Equation (5)

[0118] In the formula, L represents the total loss function.

[0119] ⑤ Repeated elimination

[0120] When performing object detection, there may be a phenomenon of multiple detection boxes overlapping and repeating. This algorithm uses the non-maximum suppression algorithm to remove the repeated detection boxes. By calculating the overlap degree between the bounding boxes, those boxes with a higher overlap degree with other boxes are removed, and only the bounding box with the highest score is retained. It sorts all the detection results according to the predicted score. Then it selects the bounding box with the highest score. Subsequently, other boxes whose overlap degree with this box exceeds the threshold are deleted. Finally, the above steps are repeated until all the detection boxes are processed.

[0121] ⑥ Output result

[0122] This algorithm will output the category, location, and confidence of the objects in each image. The category is the category probability calculated by the Softmax function. The bounding box is represented by x i , y i , w i , h i to describe the location of the object.

[0123] Step 32: For the identified construction activities, through image segmentation, achieve progress estimation at the pixel level of the image.

[0124] The present invention utilizes a deep learning algorithm to achieve object segmentation at the pixel level on the basis of completing object detection. By taking the images within the bounding box ranges corresponding to the respective construction activities identified by object detection as inputs, extracting high-level features (such as edges, textures, shapes, etc.) in the images, and then precisely generating a pixel-level mask for each object, and finally performing progress estimation through the mask. Through this operation, the progress information of each construction activity in the video image can be obtained, providing data support for subsequent calculation of labor efficiency. Step 22 specifically includes:

[0125] ① Image preprocessing

[0126] Since the size of each input image may be different, by performing preprocessing on it, including adjusting the image size, normalization, etc., to make it meet the input requirements of the network model.

[0127] ② Feature extraction

[0128] Utilize a pre-trained convolutional neural network as the backbone network to extract high-level features from the input image. For each target region, perform precise feature extraction on it through RoI operation. RoI accurately calculates the values within each sub-region through bilinear interpolation, can avoid the quantization error caused by max pooling, and retains the detailed information of the target region.

[0129] ③ Object mask generation

[0130] Based on each target region, generate a pixel-level mask for the target through a fully convolutional network. This mask represents the precise shape of the object, and the regions with pixel values of 1 in the mask image correspond to the object, and the regions with 0 are the background.

[0131] ④ Object segmentation and region area calculation

[0132] Through the mask output, the region area of each target can be calculated. By counting the number of pixel points with a value of 1 in the mask, the segmented region area of each object can be obtained.

[0133]

[0134] where Mk is the mask of object k. Mask(x, y) is the mask image, representing the pixel area of the object. A value of 1 indicates the object area, and 0 indicates the background.

[0135] Furthermore, based on the area of the segmented region, according to the calibration ratio of the size of the video image data corresponding to the type of construction activity to the actual size, the corresponding progress value can be calculated:

[0136] W k = Area(M k )·I k Equation (7)

[0137] In the formula, W k represents the progress value, that is, the current construction progress. I k represents the calibration ratio of the size of the image data to the actual size, which is obtained through calibration when arranging the video measurement points.

[0138] Step 33: By comparing the progress at different times, solve the workload per unit time of the workers and calculate the labor efficiency.

[0139] Since the video image monitoring points are fixed, the difference between the progress values corresponding to different times represents the workload of the workers. Therefore, by comparing with time, the corresponding labor efficiency can be obtained. Based on the acquisition time, acquisition location, and construction type of the images, and based on the quantified construction progress, the labor efficiency corresponding to each activity can be calculated.

[0140]

[0141] In the formula, t s represents the start time of labor efficiency monitoring, and t f represents the end time of monitoring, and minutes respectively represent the progress values corresponding to construction activity k at time t s and t f .

[0142] Through the above process, the unstructured video images are converted into structured labor efficiency data, and the annotation information such as the acquisition time, acquisition location, and construction type of the original image video data (construction progress data) is synchronously annotated with the labor efficiency data.

[0143] Step Three: Align the data time scale and construct a "factor - labor efficiency" knowledge graph

[0144] Align the worker's physical condition data, environmental condition data, and labor efficiency data with the same annotation information according to the collection time, collection location, and construction type. Furthermore, using the label information as entities and "worker's physical condition data, environmental condition data - labor efficiency data" as attributes, establish a "factor - labor efficiency" knowledge graph.

[0145] Step 4. Establish the mapping relationship between factors and labor efficiency

[0146] Based on the "factor - labor efficiency" knowledge graph, use deep learning to establish the mapping relationship between the worker's physical condition, environmental condition, and labor efficiency L k among them.

[0147] L k = f(E, R) Equation (9)

[0148] In the formula, E represents the environmental condition, specifically including wind speed, wind pressure, temperature, humidity, light intensity, solar radiation, and noise; R represents the worker's physical condition, specifically including basic condition data, physical sign condition data, sleep quality data, and exercise state data.

[0149] Through establishing a set of labor efficiency measurement processes, the present invention forms a labor efficiency measurement method considering the influence of the worker's physical condition and external environment, proposes an integrated multi-factor data collection framework, realizes the synchronous collection of labor efficiency and factor data, and proposes a method for depicting the mapping relationship between factors and labor efficiency based on the combination of knowledge graph and deep learning, providing a method support for scientifically quantifying the change of labor efficiency in actual construction.

[0150] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for quantitatively characterizing labor efficiency considering the influence of multiple factors, characterized in that: The steps include: Step 1: Data collection: Collect workers’ physical condition data, environmental status data, and construction progress data; Step 2: Label the collected worker's physical condition data, environmental condition data, and construction progress data according to label information, where the label information includes collection time, collection location, and construction type; Step 3: Identify the workload of workers and calculate labor efficiency based on the collected construction progress data, and synchronously mark the label information of the construction progress data and the labor efficiency data; Step 4: Align the worker's physical condition data, environmental condition data, and labor efficiency data with the same label information according to the label information, take the label information as the entity, take "worker's physical condition data, environmental condition data-labor efficiency data" as the attribute, and establish the "influencing factors-labor efficiency" knowledge graph; Step 5: Based on the "influencing factors-labor efficiency" knowledge graph, use deep learning to establish a mapping relationship between workers' physical condition, environmental factors and labor efficiency.

2. The labor efficiency quantitative characterization method considering the influence of multiple factors as claimed in claim 1 is characterized by: The workers' physical condition data is collected through electronic files and integrated electronic bracelets; the environmental status data is collected through an integrated sensing platform; and the construction progress data is obtained through video images collected by a mobile wireless motion camera.

3. The labor efficiency quantitative characterization method considering the influence of multiple factors as claimed in claim 2 is characterized by: The workers' physical condition data include basic condition data, vital signs condition data, sleep quality data and motion status data. The workers' basic condition data are collected using electronic files, and the basic condition data include the workers' height, weight, gender, age, type of work, proficiency, length of service and health status; the workers' vital signs condition data are collected using an integrated electronic bracelet, and the vital signs condition data include the workers' body temperature, blood oxygen, blood pressure and heart rate; the workers' sleep quality data is collected the night before they start working, and the motion status data is continuously measured during the construction process, and the motion status data includes the workers' cumulative working hours, energy consumption per unit time and the motion amplitude of the construction operation.

4. The labor efficiency quantitative characterization method considering multiple factors as claimed in claim 2, characterized in that: The integrated sensing platform includes a sensor bracket and several sensors. The sensor bracket is a telescopic bracket. The sensors are fixed to the sensor bracket by buckles. The sensors include an anemometer, a wind pressure gauge, a thermometer and a hygrometer, an illuminance meter, a solar radiometer, and a sound intensity meter. The sensors are connected to a smart mobile phone through a multi-channel Bluetooth transceiver fixed on the sensor bracket to achieve collaborative control through the smart mobile phone. The environmental status data includes wind speed, wind pressure, temperature, humidity, light intensity, solar radiation, and noise collected by the sensors.

5. The labor efficiency quantitative characterization method considering multiple factors as claimed in claim 1, characterized in that: In step 3, the workload of workers is identified and the labor efficiency is calculated based on the collected construction progress data, which specifically includes: Step 31: using target detection to identify various construction activities in the video image; Step 32: for the identified construction activities, image segmentation is used to achieve progress estimation based on the image pixel level; Step 33: By comparing the progress at different times, solve the worker's workload per unit time and calculate the labor efficiency.

6. The method for quantitatively characterizing labor efficiency considering the influence of multiple factors as claimed in claim 5, characterized in that: Step 31 uses target detection to identify various construction activities in the video image, including: ① Establish a dedicated dataset for construction activities For each construction activity in the video image, a sufficient number of images are selected, and manually labeled, and the activity types are divided into construction activity zones to construct a special data set for construction activities to meet the training requirements of the YOLO model. ② Grid division First, the input image is divided into a finite number of grids, each of which is responsible for detecting a portion of the object in the image. The size of each grid is: Where CS represents the size of each grid, W and H represent the length and width of the image respectively, and S represents the number of grids in the W and H directions; ③Boundary prediction For image data, the YOLO model can predict multiple bounding boxes and corresponding object category probabilities for each grid. For each grid i, it outputs B bounding boxes. The information of each bounding box includes: x i ,y i : The coordinates of the center of the bounding box relative to the grid; w i ,h i : The width and height of the bounding box, relative to the ratio of the entire image; c i,j : Category probability, indicating the probability of an object existing in the grid; For each bounding box, predict the following information: P (Objective): The confidence level of whether the grid contains an object; P(Class k ): The probability that the object in the grid belongs to category k; ④Model training The YOLO model is trained using a special dataset for construction activities. The loss function is used to measure the difference between the model's predicted value and the actual value, and the loss function value is reduced by optimizing the model parameters. The loss function consists of three parts: classification loss, which is used to measure the difference between the predicted category and the actual category; positioning loss, which is used to measure the difference between the center point and size of the predicted bounding box and the true value; confidence loss, which is used to measure the difference between the predicted probability of the existence of an object and the actual value; Where L cls represents the classification loss, is an indicator function, which is 1 when the grid (i, j) contains an object, otherwise it is 0; P(Class k )and Represent the predicted category probability and the actual category probability respectively; Where L coord represents the position loss, λ coord is the weight coefficient used to balance the positioning loss; Where L conf represents the confidence loss, P(Objective) and Respectively represent the confidence of whether the prediction contains an object and the confidence of the actual inclusion of the object; L=L cls +L coord +L conf Formula (5) Where L represents the total loss function; ⑤Duplicate elimination When performing target detection, there may be multiple detection frames that overlap and repeat. The non-maximum suppression algorithm is used to remove duplicate detection frames. By calculating the overlap between bounding boxes, those that have a high overlap with other boxes are removed, and only the bounding box with the highest score is retained. Finally, the above steps are repeated until all detection frames are processed. ⑥Output results The category, location, and confidence of the object in each image will be output, where the category probability is calculated by the Softmax function, and the bounding box is calculated by x i ,y i ,w i ,h i Describes the location of an object.

7. The method for quantitatively characterizing labor efficiency considering the influence of multiple factors as claimed in claim 5, characterized in that: Step 32 implements pixel-level progress estimation for the identified construction activities through image segmentation, specifically including: ① Image preprocessing The images of the bounding box range corresponding to each construction activity identified by target detection are taken as input, and each input image is preprocessed, including adjusting the image size and normalizing it, to adapt it to the input requirements of the network model; ②Feature extraction A pre-trained convolutional neural network is used as the backbone network to extract high-level features from the input image. For each target area, accurate feature extraction including edge, texture, and shape is performed through RoI operation. ③Object mask generation Based on each target area, a pixel-level mask of the target is generated through a fully convolutional network. The mask represents the precise shape of the object. The area with a pixel value of 1 in the mask image corresponds to the object, and the area with a pixel value of 0 corresponds to the background. ④Object segmentation and area calculation Through the mask output, the area of ​​each target is calculated, and the segmented area of ​​each object is obtained by counting the number of pixels with a value of 1 in the mask; Where M k is the mask of object k, Mask(x,y) is the mask image, which represents the pixel area of ​​the object. The value 1 represents the object area, and 0 represents the background. Then, based on the area of ​​the segmented region, the corresponding progress value is calculated according to the calibration ratio between the size of the video image data corresponding to the type of construction activity and the actual size: W k = Area(M k )·I k Equation (7) Where W k Indicates the progress value, that is, the current construction progress, I k It represents the calibration ratio between the image data size and the actual size, I k It is obtained through calibration when laying out video measurement points.

8. The method for quantitatively characterizing labor efficiency considering the influence of multiple factors according to claim 7, characterized in that: Step 33 compares the progress at different times to solve the workload per unit time of the workers and calculate the labor efficiency, which specifically includes: According to the image acquisition time, acquisition location, construction type, and based on the quantified construction progress, the labor efficiency corresponding to each activity is calculated: Where t s represents the start time of labor efficiency monitoring, then t f Indicates the end time of monitoring. and points represents the construction activity k at time t s and t f The corresponding progress value.

9. The labor efficiency quantitative characterization method considering multiple factors as claimed in claim 8, characterized in that: Step 4: Establishing the influencing factor-labor efficiency mapping relationship includes: Based on the knowledge graph of "influencing factors-labor efficiency", deep learning is used to establish the relationship between workers' physical condition, environmental condition and labor efficiency L k The mapping relationship between them is: L k =f(E,R) Formula (9) Where E represents the environmental state; R represents the physical condition of the worker.

Images