Method for identifying incorrect wearing of safety belt on scaffold through intelligent AI algorithm

Abstract

The invention discloses a method for recognizing incorrect wearing of a safety belt on a scaffold through an intelligent AI algorithm, and particularly relates to the technical field of AI vision. An acquisition alarm end comprises a camera module, a 5G communication module and a voice alarm broadcasting module; the safety belt wearing detection model comprises a safety belt identification module, a safety belt hooking mode detection module, a safety belt hook state detection module and a data storage module, the camera module is controlled by the acquisition alarm terminal to acquire images, and the images are sent to the safety belt wearing detection model through the 5G communication module; the safety belt recognition module recognizes an operator, a safety belt and a safety belt hook in the image, judges whether the safety belt is correctly worn or not, judges whether the hooking mode of the safety belt is correct or not and whether the state of the safety belt hook is normal or not based on the recognition result of the safety belt recognition module, and stores data in the data storage module; and playing a corresponding voice prompt by using the voice alarm broadcast module.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of AI vision, and more particularly, to a method for identifying incorrect wearing of safety belts by workers on scaffolds by using intelligent AI algorithms. BACKGROUND

[0002] In a scaffold work site, the wearing of safety belts is crucial for the safety of workers. In the prior art, real-time monitoring of workers is performed by using computer vision technology, and a target detection algorithm is used to identify whether a worker is wearing a safety belt. These technical methods usually rely on image data, and by performing target detection on sample images, a detection region is extracted and a safety belt is classified and judged.

[0003] However, due to the complex environment of a construction site and the frequent activities of workers, occlusion, changes in illumination, and rapid movement of workers in images can make it difficult to accurately identify the wearing of safety belts. In addition, the prior art fails to effectively fuse different types of data, which makes it impossible for the system to guarantee the stability and accuracy of identification under low light or occlusion conditions, especially when the dynamic behavior of workers is complex, the judgment of existing image recognition technology is prone to misjudgment, and it is impossible to comprehensively consider safety hazards and effectively warn of dangerous situations where safety belts are not worn correctly. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a method for identifying incorrect wearing of safety belts by workers on scaffolds by using intelligent AI algorithms, to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme, a method for identifying incorrect wearing of safety belts by workers on scaffolds by using intelligent AI algorithms, specifically comprising: S1, a safety belt wearing detection model and a collection alarm end are constructed, the collection alarm end includes a camera module, a 5G communication module and a voice alarm broadcast module; the safety belt wearing detection model includes a safety belt recognition module, a safety belt hanging method detection module, a safety belt hook state detection module and a data storage module; the collection alarm end is connected to the safety belt wearing detection model through a 5G wireless network; S2, the collection alarm end controls the camera module to collect images, and sends the images to the safety belt wearing detection model through the 5G communication module; S3, the safety belt recognition module in the safety belt wearing detection model uses a deep learning network to process the received images, identifies the workers, safety belts and safety belt hooks in the images, and judges whether the safety belts are correctly worn; S4, the safety belt hanging mode detection module in the safety belt wearing detection model, judges whether the safety belt hanging mode is correct based on the identification result of the safety belt identification module; S5: the safety belt hook state detection module in the safety belt wearing detection model, judges whether the safety belt hook state is normal based on the identification result of the safety belt identification module; S6: when the judgment result of any one of S3, S4 and S5 is incorrect, the incorrect data is stored to the data storage module, and the corresponding preset voice prompt is played by using the voice alarm broadcast module.

[0006] Preferably, as a preferred scheme of the method for identifying incorrect safety belt wearing on a scaffold by using an intelligent AI algorithm, the method comprises the steps of S2, the collection alarm end controls the camera module to collect images, and the images are sent to the safety belt wearing detection model through the 5G communication module, and specifically comprises the following steps: Various different angle, hanging fine wire, hanging inclined line, buckle not closed, buckle wire working image is collected, to ensure the comprehensiveness and diversity of data, and each collected picture is accurately labeled, including safety belt labeling, safety belt hook labeling; The safety belt labeling is to label the area of the safety belt in the image by using a rectangular frame, to ensure that the model can identify the existence of the safety belt, and the labeling content is the coordinate information of the rectangular frame; The safety belt hook labeling is to label the position and state of the safety belt hook in the image, and the labeling content is the position and state of the hook, whether closed or in a correct state.

[0007] Preferably, as a preferred scheme of the method for identifying incorrect safety belt wearing on a scaffold by using an intelligent AI algorithm, the method comprises the steps of S3, the safety belt identification module in the safety belt wearing detection model uses a deep learning network to process the received images, identifies the working personnel, the safety belt and the safety belt hook in the images, and judges whether the safety belt is correctly worn, and specifically comprises the following steps: The deep learning network is improved in network structure and loss function based on a pre-trained YOLOv5 model, wherein the network structure improvement comprises replacing the SPPF module in the YOLOv5 network with an SPPFCSPC module and replacing the original network coupling head with a decoupling head; when the deep learning network is improved in network structure based on the YOLOv5 model, the SPPFCSPC module comprises first and second convolution units in parallel; the decoupling head uses a 1x1 convolution to reduce the dimension of the input feature layer, uses two 3x3 convolutions in the classification and regression branches respectively, performs object classification and target frame coordinate regression tasks in parallel, and finally fuses the classification result Cls, the regression result Reg and the target existence score Obj to obtain the prediction information; The loss function optimization is to optimize the network using an SIOU loss function, and the angle loss is calculated according to the height difference , the center point distance of the real frame and the prediction frame center point , the specific formula is , the distance loss is calculated based on the angle loss , and the shape loss is , wherein , , , and h represent the width and height of the prediction frame, respectively and are constant factors related to the frame width and height, which are used to scale and adjust the width and height errors to help calculate the distance loss and shape loss, and the comprehensive loss value is calculated according to the distance loss and shape loss , and the network parameters are optimized according to the value; After identifying the workers, safety belts and safety belt hooks, the area where the workers are located , the area where the safety helmet is located , and the area where the safety belt and the safety belt hook are located are further obtained, and a threshold value is preset, and the discrimination value is calculated as , wherein when the discrimination value is greater than the preset threshold value, it is determined that the safety belt is correctly worn; when the discrimination value is less than the preset threshold value, it is determined that the safety belt is not correctly worn, and a warning output is performed.

[0008] Preferably, as a preferred scheme of the method for identifying incorrect wearing of safety belts on scaffolds by the intelligent AI algorithm, it comprises the safety belt wearing detection model in the S4, and the safety belt hooking mode detection module in the safety belt detection model judges whether the safety belt hooking mode is correct based on the identification result of the safety belt detection module, and specifically comprises: Obtaining an identification frame containing the safety belt and extracting a picture; Performing a binarization process on the extracted picture; Obtaining the center coordinates of the safety belt region by calculating the geometric center of all safety belt pixels in the binary image In the safety belt region, the pixel coordinates farthest from the center point are calculated as ; According to the upper and lower boundary vertical coordinates and of the safety belt identification frame, the discrimination interval T is calculated as , where T is the discrimination interval; when , it is determined that the high-low use is high, indicating that the safety belt is normally hung; when , it is determined that the low-high use is low, and a warning output is performed.

[0009] Preferably, as a preferred scheme of the method for identifying incorrect wearing of safety belts on scaffolds by an intelligent AI algorithm, it comprises the safety belt hook state detection module in the safety belt wearing detection model in S5, which judges whether the safety belt hook state is normal based on the identification result of the safety belt identification module, and specifically comprises: Five states of the safety belt hook are identified by transfer learning of the deep learning network, including normal, hanging thin wires, hanging inclined lines, buckle not closed, and buckle wire, and when the safety belt hook state is any of the above, it is determined that the hook state is abnormal, and a warning output is performed; When transfer learning of the deep learning network is performed, all convolutional layers of the pre-trained network are frozen, and only the top layer parameters are trained and fine-tuned, further comprising: The pre-trained deep learning network is used to transfer the knowledge of the network to the safety belt hook state identification task through transfer learning, and all convolutional layers of the pre-trained network are frozen during the transfer learning process; The top layer of the network is fine-tuned, and only the parameters of the top layer are trained, so that it can adapt to the new safety belt hook state classification task, which can significantly reduce the training time and improve the accuracy of the model on the specific task.

[0010] The training process includes collecting a labeled data set containing various hook states, and the state label of the hook should be labeled in each image, including normal, hanging thin wires, hanging inclined lines, buckle not closed, and buckle wire; The deep learning model is trained using the transfer learning method, the convolutional layers are frozen, only the top layer is trained, and the model can accurately classify the hook state by optimizing the loss function for fine-tuning; After the training is completed, the model is deployed into the safety belt wearing detection system to identify the state of the hook through real-time image analysis, and when the abnormal state of the hook is detected, a pre-warning output is immediately given to remind the relevant personnel to check and adjust the hook, thereby ensuring safety.

[0011] Preferably, as a preferred solution of the method for identifying incorrect safety belt wearing of workers on a scaffold by using an intelligent AI algorithm, the method comprises the following steps in S6: when the judgment result of any one of S3, S4 and S5 is incorrect, the incorrect data is stored into the data storage module, and the corresponding preset voice prompt is played by using the voice alarm broadcast module, and the specific steps comprise the following steps: The worker data with the incorrect judgment result is stored into the data storage module, including the ID, time, safety belt wearing state, hooking mode, hook state and identification result of the worker, when the abnormality is judged, different voice alarms are triggered according to different error types, and the specific preset voice prompts comprise the following: Incorrect safety belt wearing: when the safety belt is not worn, the voice broadcast is: "incorrect safety belt wearing, please wear"; Incorrect hooking mode: when the safety belt hooking mode is incorrect, the voice broadcast is: "incorrect safety belt hooking mode"; Abnormal hook state: when the hook is abnormal, the voice broadcast is: "abnormal hook state, please check"; When the worker does not take corrective measures within the safe time, the alarm sound is continuously played to prompt the worker to respond, and the error record is transmitted to the back-end management system, so that the management personnel can check the working environment, analyze whether there is a common safety hazard and take preventive measures.

[0012] On the other hand, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program is executed by the processor to implement the steps of the method for identifying incorrect safety belt wearing of workers on a scaffold by using an intelligent AI algorithm.

[0013] On the other hand, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method for identifying incorrect safety belt wearing of workers on a scaffold by using an intelligent AI algorithm.

[0014] In the above technical solution, the present application has the following technical effects and advantages: Through the construction of the collection alarm end and the safety belt wearing detection model, the real-time collection and high-speed transmission of images are realized, the AI algorithm is combined for rapid analysis, the identification accuracy of the safety belt hook is significantly improved, the background is effectively distinguished, when the safety belt is not correctly worn, the voice alarm broadcast module can be started in time to remind the scene, and the related data is stored, the safety belt wearing condition of the operation personnel is realized uninterrupted, all-weather real-time monitoring, the illegal behavior can be found and alarmed in the first time, the time from the illegal behavior to the discovery and correction is greatly shortened, and the safety accidents caused by illegal operation are effectively prevented. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0016] Figure 1 The structural block diagram of the collection alarm end and the safety belt wearing detection model in the present application.

[0017] Figure 2 The method flowchart of the present application.

[0018] Figure 3 The diagram for identifying low hanging high use in the present application.

[0019] Table 1 is a data recording table of the simulation experiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0021] In the description of the present application, the terms "first", "second" are only used for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0022] In the description of the present application, the term "for example" is used to mean "serving as an example, instance, or illustration." Any embodiment described as "for example" in this application is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the application. In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present application. It will be apparent, however, to one skilled in the art that the present application can be practiced without using these specific details. In other instances, well-known structures and processes are not elaborated upon in order to avoid unnecessary detail, which can obscure the description of the present application. Thus, the present application is not intended to be limited by the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0023] Embodiment 1 The present embodiment provides a method for identifying incorrect safety belt wearing of workers on a scaffold by an intelligent AI algorithm, comprising the following steps: Figure 1 The present embodiment provides a method for identifying incorrect safety belt wearing of workers on a scaffold by an intelligent AI algorithm, comprising the following steps: S1, a collection and alarm end and a safety belt wearing detection model are constructed, the collection and alarm end comprises a camera module, a 5G communication module and a voice alarm broadcast module; the safety belt wearing detection model comprises a safety belt recognition module, a safety belt hanging method detection module, a safety belt hook state detection module and a data storage module; the collection and alarm end is connected to the safety belt wearing detection model through a 5G wireless network; S2, the collection and alarm end controls the camera module to collect images, and sends the images to the safety belt wearing detection model through the 5G communication module; S3, the safety belt recognition module in the safety belt wearing detection model processes the received images using a deep learning network, identifies the workers, safety belts and safety belt hooks in the images, and judges whether the safety belts are correctly worn; S4, the safety belt hanging method detection module in the safety belt wearing detection model judges whether the hanging method of the safety belt is correct based on the identification result of the safety belt recognition module; S5, the safety belt hook state detection module in the safety belt wearing detection model judges whether the state of the safety belt hook is normal based on the identification result of the safety belt recognition module; S6, when the judgment result of any one of S3, S4 and S5 is incorrect, the incorrect data is stored in the data storage module, and the corresponding preset voice prompt is played using the voice alarm broadcast module.

[0024] Preferably, in S2, the collection and alarm end controls the camera module to collect images, and sends the images to the safety belt wearing detection model through the 5G communication module, specifically comprising: Collect images of various angles, hanging fine wires, hanging inclined lines, buckles not closed, and buckle stringing operation images to ensure the comprehensiveness and diversity of the data, and accurately label each collected image, including safety belt labeling and safety belt hook labeling; The safety belt labeling is to label the area where the safety belt is located in the image by using a rectangular frame, to ensure that the model can identify the existence of the safety belt, and the labeling content is the coordinate information of the rectangular frame; The safety belt hook labeling is to label the position and state of the safety belt hook in the image, and the labeling content is the position and state of the hook, whether it is closed or in the correct state.

[0025] Preferably, in the S3, the safety belt identification module in the safety belt wearing detection model uses a deep learning network to process the received image, identifies the work personnel, safety belt and safety belt hook in the image, and judges whether the safety belt is correctly worn, specifically including: The deep learning network improves the network structure and optimizes the loss function based on the pre-trained YOLOv5 model, wherein the network structure improvement includes replacing the SPPF module in the YOLOv5 network with an SPPFCSPC module, and replacing the original network coupling head with a decoupling head; when the deep learning network improves the network structure based on the YOLOv5 model, the SPPFCSPC module contains a first and a second convolution unit in parallel; the decoupling head uses a 1x1 convolution to reduce the dimension of the input feature layer, uses two 3x3 convolutions in the classification and regression branches respectively, and performs object classification and target frame coordinate regression tasks in parallel, and finally fuses the classification result Cls, the regression result Reg and the target existence score Obj three output values to obtain the prediction information; The loss function optimization is to optimize the network using the SIOU loss function, and the angle loss is calculated according to the height difference of the center points of the real frame and the predicted frame and the distance factor of the center points , and the specific formula is The distance loss is calculated based on the angle loss , and the shape loss is , wherein , , , and h represent the width and height of the predicted frame respectively and are constant factors related to the width and height of the frame, used to scale and adjust the width and height errors to help calculate the distance loss and the shape loss, and the comprehensive loss value is calculated according to the distance loss and the shape loss , and the network parameters are optimized according to the value; After recognizing the worker, the safety belt and the safety belt hook, further obtain the area where the worker is located , the area where the safety helmet is located , and the area where the safety belt and the safety belt hook are located , and preset a threshold value, calculate a discrimination value as , wherein, when the discrimination value is greater than the preset threshold value, it is determined that the safety belt is correctly worn; when the discrimination value is less than the preset threshold value, it is determined that the safety belt is not correctly worn, and a warning output is performed.

[0026] Preferably, in S4, the safety belt hanging mode detection module in the safety belt wearing detection model judges whether the hanging mode of the safety belt is correct based on the recognition result of the safety belt recognition module, and specifically includes: obtaining the recognition frame containing the safety belt and extracting the picture; performing binaryzation processing on the extracted picture; obtaining the center coordinates of the safety belt area by calculating the geometric center of all safety belt pixels in the binary image , calculating the pixel coordinates farthest from the center point in the safety belt area as , and calculating the discrimination interval according to the upper and lower boundary vertical coordinates of the safety belt recognition frame and . , wherein T is the discrimination interval; when , it is determined that the safety belt is hung high and used low, indicating that the safety belt is normally hung; when , it is determined that the safety belt is hung low and used high, and a warning output is performed.

[0027] Preferably, in S5, the safety belt hook state detection module in the safety belt wearing detection model judges whether the state of the safety belt hook is normal based on the recognition result of the safety belt recognition module, and specifically includes: recognize five states of the safety belt hook by migration learning of the deep learning network, which are normal, hanging thin wire, hanging inclined wire, buckle not closed and buckle wire, respectively; when the safety belt hook state is any of hanging thin wire, hanging inclined wire, buckle not closed or buckle wire, it is determined that the hook state is abnormal, and a warning output is performed; when migration learning is performed on the deep learning network, freeze all convolutional layers of the pre-trained network, and only train and fine-tune the top part of the parameters, further including: use the pre-trained deep learning network to migrate the knowledge of the network to the safety belt hook state recognition task by migration learning, and freeze all convolutional layers of the pre-trained network during the migration learning process; Fine-tuning the top layer of the network, only training the parameters of the top layer, so that it can adapt to the new safety belt hook state classification task, can significantly reduce the training time and improve the accuracy of the model on specific tasks.

[0028] The training process includes collecting a labeled data set containing various hook states, and each image should be labeled with the state label of the hook, including normal, hanging fine wire, hanging inclined wire, buckle not closed, and buckle line; Using the transfer learning method to train the deep learning model, freezing the convolutional layer, only training the top layer, and fine-tuning through optimizing the loss function, finally making the model accurately classify the hook state; After the training is completed, the model is deployed to the safety belt wearing detection system, and the state of the hook is identified through real-time image analysis. When an abnormal hook state is detected, an early warning output is immediately given to remind relevant personnel to check and adjust the hook to ensure safety.

[0029] Preferably, in S6, when the judgment result of any one of S3, S4 and S5 is incorrect, the incorrect data is stored in the data storage module, and the corresponding preset voice prompt is played by using the voice alarm broadcast module. Specifically, it includes: The judgment result of the incorrect data of the work personnel, including the ID, time, safety belt wearing state, hooking method, hook state and identification result, is stored in the data storage module, and different voice alarms are triggered according to different error types when an exception is detected. The specific preset voice prompt includes: Not wearing a safety belt: When the safety belt is not worn, the voice broadcast is: "Please wear a safety belt"; Incorrect hooking method: When the safety belt hooking method is incorrect, the voice broadcast is: "Safety belt hooking method is incorrect"; Abnormal hook state: When the hook is abnormal, the voice broadcast is: "Please check the hook state"; When the work personnel does not take corrective measures within the safety time, the alarm sound is continuously played to prompt the work personnel to respond, and the error record will be transmitted to the backend management system, so that the management personnel can check the work environment, analyze whether there are common safety hazards and take preventive measures.

[0030] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, a carrier network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0031] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the functional modules of the method for identifying incorrect wearing of safety belts by workers on a scaffold by using an intelligent AI algorithm according to the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0032] Embodiment 2 In another embodiment of the present application, a method for identifying incorrect wearing of safety belts by workers on a scaffold by using an intelligent AI algorithm is provided. In order to verify the beneficial effects of the present application, a simulation experiment is performed for scientific demonstration.

[0033] The experiment aims to verify the effectiveness of the scaffold operation safety belt wearing recognition method based on intelligent AI algorithm. Through data collection, image processing, deep learning network training, real-time detection and early warning technology, the real-time monitoring and dynamic early warning capability of the safety belt wearing situation of the operating personnel is improved. The experiment uses simulated and actual collected scaffold operation data, including image features under different safety belt wearing states, wearing accuracy, hanging method, hook state and multi-dimensional environmental information. The recognition results of the analysis model are compared with the actual wearing labels to verify the accuracy and robustness of the invention in recognizing the safety belt wearing situation of the operating personnel. The simulation experiment step is implemented according to the content of the method for identifying incorrect safety belt wearing of operating personnel on a scaffold provided in embodiment 1. The specific steps include, Collect image data in the scaffold operation environment, including operating personnel, scaffold structure, environmental changes, etc. The collection frequency is set to once every 5 minutes; Construct a safety belt wearing detection model, which includes the following modules: Safety belt recognition module: process the collected images through a deep learning network to recognize the operating personnel, safety belt and hook in the images; Safety belt hanging method detection module: determine whether the safety belt is correctly hung; Safety belt hook state detection module: check whether the hook state is normal; Data storage module: store all collected data and recognition results; Train the image data using a deep learning network. The trained model can identify whether the operating personnel correctly wear safety belts, and determine the hanging method and hook state of the safety belt. The trained model can update the recognition results in real time and provide a wearing state evaluation for each image. Apply the safety belt wearing detection model in the actual operation environment. Real-time image data of operating personnel is collected by a camera and transmitted to the cloud for processing. The safety belt wearing detection model will automatically determine whether there are operating personnel who do not correctly wear safety belts based on the image data, and take the following actions: Real-time recognition and prediction of the safety belt wearing state of operating personnel. When an operating personnel who does not correctly wear a safety belt is found, the incorrect data is immediately stored in the data storage module, and a preset warning message is played through the voice broadcast module; Evaluate and warn the safety belt wearing situation of the operating personnel. According to the recognition results, determine whether the safety belt is correctly worn, and prompt the personnel who incorrectly wear the safety belt; Real-time update of the detection results, and generation of a safety management report to provide real-time support for the safety management of the scaffold.

[0034] The specific data of the simulation experiment are as follows: Time / minute Number of workers Safety belt wearing state Hanging method Hook state Identification result Actual label Warning measure 0-5 2 Correct High hanging low use Closed Correct Correct None 5-10 3 Not worn - - Not worn Not worn Announce "Please wear safety belt" 10-15 1 Low hanging high use Low hanging high use Closed Low hanging high use Low hanging high use Announce "Safety belt hanging method is improper" 15-20 2 Correct High hanging low use Not closed Hook abnormality Hook abnormality Announce "Hook state is abnormal, please check" 20-25 4 Correct High hanging low use Closed Correct Correct None 25-30 2 Correct High hanging low use Closed Correct Correct None 30-35 3 Not worn - - Not worn Not worn Announce "Please wear safety belt" 35-40 1 Low hanging high use Low hanging high use Not closed Hook abnormality Hook abnormality Announce "Hook is not closed, please check" 40-45 2 Correct High hanging low use Closed Correct Correct None Table 1 Experimental analysis: Through the simulation experiment, the safety belt wearing state, the hanging mode and the hook state can be accurately identified, when the abnormality is found, the voice alarm is automatically triggered, the operation personnel are reminded to correct in time, and the safety of the operation personnel is ensured.

[0035] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that the described embodiments can be modified in various ways without departing from the spirit and scope of the present application for those skilled in the art. Therefore, the above drawings and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. A method for identifying non-compliance with safety harness use on a scaffold by an AI algorithm, characterized in that: Specifically comprising: S1, construct a collection alarm end and a safety belt wearing detection model, the collection alarm end includes a camera module, a 5G communication module and a voice alarm broadcast module; the safety belt wearing detection model includes a safety belt identification module, a safety belt hanging method detection module, a safety belt hook state detection module and a data storage module; the collection alarm end is connected to the safety belt wearing detection model through a 5G wireless network; S2, the collection alarm end controls the camera module to collect images, and sends the images to the safety belt wearing detection model through the 5G communication module; S3, the safety belt identification module in the safety belt wearing detection model processes the received images using a deep learning network, identifies the work personnel, safety belt and safety belt hook in the images, and judges whether the safety belt is correctly worn; S4, the safety belt hanging method detection module in the safety belt wearing detection model judges whether the safety belt hanging method is correct based on the identification result of the safety belt identification module; S5: the safety belt hook state detection module in the safety belt wearing detection model judges whether the safety belt hook state is normal based on the identification result of the safety belt identification module; S6: when the judgment result of any one of S3, S4 and S5 is incorrect, the incorrect data is stored in the data storage module, and the corresponding preset voice prompt is played using the voice alarm broadcast module.

2. A method of identifying non-compliance with safety harness use on a scaffold according to claim 1, wherein: In S2, the collection alarm end controls the camera module to collect images, and sends the images to the safety belt wearing detection model through the 5G communication module, specifically comprising: Collect various different angle, hanging fine wire, hanging oblique line, buckle not closed, buckle line work image, accurately label each collected picture, including safety belt label, safety belt hook label; The safety belt label is to label the area of the safety belt in the image using a rectangular frame, and the label content is the coordinate information of the rectangular frame; The safety belt hook label is to label the position and state of the safety belt hook in the image, and the label content is the position and state of the hook, whether closed or in the correct state.

3. A method of identifying non-compliance with safety harness use on a scaffold according to claim 1, wherein: In S3, the safety belt identification module in the safety belt wearing detection model processes the received images using a deep learning network, identifies the work personnel, safety belt and safety belt hook in the images, and judges whether the safety belt is correctly worn, specifically comprising: The deep learning network is improved in network structure and loss function based on a pre-trained YOLOv5 model, wherein the network structure improvement comprises replacing the SPPF module in the YOLOv5 network with an SPPFCSPC module and replacing the coupling head of the original network with a decoupling head; when the deep learning network is improved in network structure based on the YOLOv5 model, the SPPFCSPC module comprises first and second convolution units in parallel; the decoupling head uses a 1x1 convolution to reduce the dimension of an input feature layer, uses two 3x3 convolutions in the classification and regression branches respectively, performs object classification and target frame coordinate regression tasks in parallel, and finally fuses three output values of a classification result Cls, a regression result Reg and a target existence score Obj to obtain prediction information; After recognizing the worker, the safety belt and the safety belt hook, further acquire the area where the worker is located , the area where the safety helmet is located , and the area where the safety belt and the safety belt hook are located , and preset a threshold value, calculate a discrimination value as , wherein, when the discrimination value is greater than the preset threshold value, it is determined that the safety belt is correctly worn; when the discrimination value is less than the preset threshold value, it is determined that the safety belt is not correctly worn, and a warning output is performed.

4. The method of claim 3, wherein the method is a method of identifying non-compliance with safety harness use on a scaffold by an AI algorithm, characterized in that: The loss function optimization is to optimize the network by using an SIOU loss function, according to the height difference between the real frame and the center point of the predicted frame , the center point distance , the angle loss , the specific formula is , the distance loss is calculated based on the angle loss , and the shape loss is , wherein , , , and h represent the width and height of the predicted frame respectively and are constant factors related to the width and height of the frame, which are used to scale the width and height errors to help calculate the distance loss and shape loss, and the comprehensive loss value is calculated according to the distance loss and shape loss , and the network parameters are optimized according to the value.

5. A method of identifying non-compliance with safety harness use on a scaffold according to claim 1, wherein: In the S4, the safety belt hanging mode detection module in the safety belt wearing detection model judges whether the hanging mode of the safety belt is correct based on the identification result of the safety belt identification module, and specifically comprises: An identification frame containing the safety belt is obtained and a picture is extracted; the extracted picture is binarized; The center coordinates of the safety belt region are obtained by calculating the geometric center of all the safety belt pixels in the binary image In the safety belt region, the pixel coordinates farthest from the center point are calculated as ; according to the upper and lower boundary vertical coordinates of the safety belt identification frame and , the discrimination interval is calculated as , wherein T is the discrimination interval; when , it is determined that the high-low use is determined, indicating that the safety belt is normally hung; when , it is determined that the low-high use is determined, and a warning output is performed.

6. A method of identifying non-compliance with safety harness use on a scaffold according to claim 1, wherein: In the S5, the safety belt hook state detection module in the safety belt wearing detection model judges whether the state of the safety belt hook is normal based on the identification result of the safety belt identification module, and specifically comprises: Five states of the safety belt hook are identified through transfer learning of the deep learning network, which are normal, hanging thin wire, hanging inclined wire, buckle not closed and buckle wire, respectively; when the safety belt hook state is any of the hanging thin wire, hanging inclined wire, buckle not closed and buckle wire, the hook state is determined to be abnormal, and a warning output is performed; When the deep learning network is subjected to transfer learning, all convolution layers of the pre-trained network are frozen, and only the top part of the parameters is trained and fine-tuned; after the training is completed, the state of the hook is identified through real-time image analysis, and a warning output is immediately performed to remind relevant personnel to check and adjust the hook when the hook state is detected to be abnormal.

7. A method of identifying non-compliance with safety harness use on a scaffold according to claim 1, wherein: In the S6, when the judgment result of any one of S3, S4 and S5 is incorrect, the incorrect data is stored in the data storage module, and the corresponding preset voice prompt is played by using the voice alarm broadcast module, and specifically comprises: The judgment result of the incorrect data, including the ID, time, safety belt wearing state, hanging mode, hook state and identification result of the worker, is stored in the data storage module, and different voice alarms are triggered according to different error types when an abnormality is judged, and the specific preset voice prompt comprises: No safety belt is worn: when the safety belt is not worn, the voice broadcast is "Please wear a safety belt"; Incorrect hanging mode: when the safety belt hanging mode is incorrect, the voice broadcast is "Incorrect safety belt hanging mode"; Abnormal hook state: when the hook is abnormal, the voice broadcast is "Abnormal hook state, please check"; When the worker does not take corrective measures within a safe time, the alarm sound is continuously played to prompt the worker to respond.

Images