Safety production early warning system and method based on AI image recognition technology

Abstract

The invention belongs to but not limited to the technical field of artificial intelligence, and particularly relates to a safety production early warning method and system based on an AI image recognition technology, and the method comprises a camera module which is used for collecting a real-time image of a production field; the image processing and analyzing module is used for preprocessing and analyzing the image acquired by the camera so as to detect the safety helmet wearing condition of the worker; the safety monitoring and early warning module is used for analyzing and displaying a detection result in real time and giving an alarm when potential safety hazards are found; the data storage and management module is used for storing and managing detection data and early warning records for subsequent analysis and auditing; and the network and communication module is used for ensuring data transmission and communication among the modules. According to the invention, efficient and real-time safety production monitoring and early warning are realized through an innovative AI image recognition technology and an optimized system architecture.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the field of artificial intelligence technology, and in particular relates to a production safety early warning system and method based on AI image recognition technology. Background Art

[0002] Work safety has always been the focus of attention in various industries, especially in high-risk industries such as construction, manufacturing and mining. With the development of industrialization and urbanization, enterprises face various safety risks in the production process, such as mechanical injuries, chemical leakage and high-altitude falls. Therefore, enterprises and regulatory agencies are seeking effective ways to reduce accidents and protect the lives of workers and the property safety of enterprises.

[0003] With the advancement of information technology, especially the rapid development of artificial intelligence and computer vision technology, safety production monitoring systems based on image recognition are widely used. These systems capture on-site images through cameras and use AI algorithms to analyze image data. They can automatically identify safety hazards, such as workers not wearing safety equipment and abnormal behavior in the area. Compared with traditional manual monitoring, such systems have obvious advantages in recognition speed, accuracy and coverage.

[0004] In the construction and manufacturing industries, wearing hard hats and other protective gear is the most basic safety requirement. However, traditional monitoring and management methods, such as manual inspections and regular training, often cannot guarantee that all workers comply with safety regulations at all times. Therefore, automated image recognition systems are introduced into safety production management to monitor workers' wearing of protective gear in real time, improve safety supervision efficiency and protect workers' lives.

[0005] In existing technologies, there are already some systems based on image recognition that are applied to production safety monitoring. For example, some systems can detect whether workers are wearing helmets by capturing images through cameras. The typical method is to use YOLO (You Only Look Once) for image classification and target detection. However, these systems face some challenges in practical applications, including:

[0006] In a complex production environment, factors such as background clutter and lighting changes will directly affect the accuracy of detection.

[0007] For production safety scenarios that require real-time response, the system's detection speed is crucial.

[0008] Running complex deep learning models on resource-constrained devices such as edge devices may face performance bottlenecks.

[0009] In view of the above analysis, the technical problems that need to be solved urgently in the prior art are:

[0010] (1) Accuracy: In a complex production environment, factors such as background clutter and lighting changes will directly affect the accuracy of detection.

[0011] (2) Real-time: For production safety scenarios that require real-time response, the system’s detection speed is crucial.

[0012] (3) Device resource limitations: Running complex deep learning models on resource-constrained devices (such as edge devices) may face performance bottlenecks. Summary of the invention

[0013] In view of the problems existing in the prior art, the present invention provides a production safety early warning system and method based on AI image recognition technology.

[0014] The present invention is implemented as follows: a safety production early warning system based on AI image recognition technology, comprising:

[0015] The camera module is responsible for collecting real-time images of the production site.

[0016] The image processing and analysis module is used to pre-process and analyze the images collected by the camera to detect whether the workers are wearing safety helmets.

[0017] The security monitoring and early warning module is used to analyze and display the detection results in real time and issue an alarm when a security risk is found.

[0018] The data storage and management module is used to save and manage detection data and warning records for subsequent analysis and auditing.

[0019] The network and communication module is used to ensure data transmission and communication between modules.

[0020] Furthermore, the camera module specifically includes: a plurality of high-definition cameras installed at different locations of the production site to cover all key areas, and these cameras are capable of transmitting image data to the core processing unit of the system in real time.

[0021] Furthermore, the image processing and analysis module specifically includes:

[0022] Preprocessing unit: responsible for denoising, enhancing, and scaling the original image to ensure that the image quality is suitable for subsequent AI analysis;

[0023] AI model: MobileNet is used as the backbone network, a three-layer FPN (Feature Pyramid Network) structure is adopted for feature extraction, and FCOS (anchor-free object detection model) is used for helmet detection;

[0024] MobileNet: A lightweight convolutional neural network for efficiently extracting image features.

[0025] FCOS: An anchor-free object detection model for predicting the location and classification of helmets directly on feature maps.

[0026] Furthermore, the security monitoring and early warning module specifically includes:

[0027] Monitoring interface: displays real-time images and test results of the production site, and highlights personnel who are not wearing helmets;

[0028] Early warning system: including sound and light alarms. When it is detected that the mask is not worn or worn improperly, the system will sound an alarm.

[0029] Furthermore, the data storage and management module specifically includes:

[0030] Database: stores image data, detection results and warning records;

[0031] Management software: provides data retrieval, report generation and statistical analysis functions.

[0032] Furthermore, the network and communication module specifically includes:

[0033] Wired / wireless network: used to connect cameras, processing units and monitoring terminals, supporting fast data transmission;

[0034] Communication protocol: Standard network communication protocol (TCP / IP) is used to ensure system compatibility and data security.

[0035] Another object of the present invention is to provide a safety production early warning method based on AI image recognition technology, comprising:

[0036] S1, Image acquisition: The camera module collects image data from the production site in real time and transmits it to the image processing and analysis module through the network. The position and angle of the camera are designed to ensure that there are no blind spots and key areas are monitored;

[0037] S2, image preprocessing: In the image processing and analysis module, the image is first preprocessed, including: denoising: filtering out noise and interference in the image; enhancement: adjusting the brightness and contrast of the image to enhance important features; scaling: adjusting the image to a size suitable for model input;

[0038] S3, feature extraction and target detection: Using the MobileNet backbone network, a three-layer FPN (Feature Pyramid Network) structure is used to extract features from the preprocessed image to obtain high-level image features. Subsequently, the FCOS target detection model is used to directly detect and locate the helmets on the feature map, including: Feature extraction: MobileNet's convolutional layer performs multi-level feature extraction on the image to generate feature maps of different dimensions; Target detection: The FCOS target detection model predicts the location information and category of the target on feature maps of different dimensions, and identifies the location of workers and their helmets;

[0039] S4, test result analysis: Analyze the test results to determine whether each detected worker wears a safety helmet correctly. Generate corresponding safety reminder labels based on the test results and update the monitoring interface;

[0040] S5, real-time monitoring and warning: the safety reminder label is displayed on the monitoring interface. If the helmet is not worn or worn improperly, the system will highlight it on the interface and trigger the sound and light alarm for real-time warning;

[0041] S6, Data storage and management: All image data, detection results and warning records will be stored in the data storage and management module to facilitate subsequent auditing and analysis; the management software provides data retrieval and analysis tools to help security managers generate reports and statistical analysis.

[0042] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the safety production warning method based on AI image recognition technology.

[0043] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the safety production warning method based on AI image recognition technology.

[0044] Another object of the present invention is to provide an information data processing terminal, which includes the above-mentioned production safety early warning system based on AI image recognition technology.

[0045] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0046] First, the advantages of the present invention include:

[0047] 1. Application of lightweight models (MobileNet): Use lightweight convolutional neural networks such as MobileNet to extract image features.

[0048] High resource efficiency: MobileNet is designed as a lightweight model with fewer parameters and computation, which makes it perform well when running on devices with limited computing resources (such as edge devices and embedded systems). Compared with traditional deep learning models, MobileNet significantly reduces the demand for hardware resources while maintaining high accuracy.

[0049] Real-time performance: Due to its high computational efficiency, MobileNet can quickly process image data in real-time applications, meeting the real-time and responsiveness requirements in safe production environments.

[0050] Easy to deploy: The low computational requirements and small model size of lightweight networks make them easy to deploy and integrate on a variety of devices without the need for high-performance dedicated hardware.

[0051] 2. Anchor-free object detection model (FCOS): The anchor-free object detection model FCOS is introduced to replace the traditional anchor-based detection method to classify the extracted features.

[0052] Simplified model design: The anchor-free detection method eliminates the reliance on predefined anchor boxes, which simplifies the model design and training process. It reduces the complexity by directly predicting the target boundary and category on the feature map.

[0053] Improved detection accuracy: Anchor-free detection can dynamically adapt to various target sizes and shapes, avoiding the mismatch problem in anchor selection, thereby improving the accuracy of target detection, especially in application scenarios that require high precision such as helmet detection.

[0054] Higher versatility: Since it does not rely on predefined anchor boxes, the FCOS method can better adapt to different application scenarios and target types, making the system more adaptable in different safety production environments.

[0055] 3. Optimization of deep learning models: Improve detection capabilities in complex backgrounds by adjusting parameters and optimizing the structure and hyperparameters of deep learning models.

[0056] Enhanced robustness: The optimized model can better handle the complex background and lighting changes at the safety production site, and improve the accurate detection of workers wearing helmets. Especially in the case of cluttered background or poor lighting conditions, the optimized model shows higher robustness.

[0057] Reduce false alarm rate: Through model optimization, the occurrence of false alarms and missed alarms is reduced, making the system more reliable in practical applications and reducing interference with normal operations.

[0058] Better generalization ability: The optimized model performs well in different factory environments and production conditions, indicating that it has strong generalization ability and can adapt to various practical application scenarios.

[0059] 4. Diversity and richness of the dataset: During the preparation of the dataset, a large amount of image data in different scenarios was collected and annotated, including normal wearing, not wearing, and irregular wearing.

[0060] Improve model training results: Rich and diverse data sets enable the model to learn more features and patterns during training, improving the model's ability to recognize different situations.

[0061] Adaptable to different application environments: Diverse data sets ensure that the model can run effectively in a variety of production environments and conditions, and is not easily limited to a specific environment.

[0062] Reduce overfitting: By training with a dataset that contains a variety of situations, the model is better able to generalize to unseen scenarios and data, reducing the risk of overfitting in a specific scenario.

[0063] 5. Integration of real-time monitoring and early warning systems: Integrate the trained model into the actual monitoring system to provide real-time production safety monitoring and early warning.

[0064] Instant response: The system can analyze surveillance images in real time and issue an immediate alarm when it detects that the helmet is not worn properly, which helps to take corrective measures as soon as the problem occurs and avoid accidents.

[0065] Improve safety management efficiency: Through automated real-time monitoring, it reduces dependence on manpower, improves safety management efficiency, and reduces operating costs.

[0066] Easy to use and manage: Real-time monitoring systems are usually equipped with a user-friendly visual interface, allowing safety managers to easily view and manage the safety status of the production site.

[0067] 6. Seamless integration and flexible deployment: The system is designed to facilitate seamless integration with existing production and monitoring equipment and support flexible deployment.

[0068] Reduce integration costs: The system is designed with compatibility with existing equipment and systems in mind, reducing additional integration and configuration costs.

[0069] Flexible adaptation to different needs: The system can be customized and adjusted according to the specific needs and production conditions of different enterprises, and flexibly adapt to various application scenarios.

[0070] Simplified maintenance and upgrades: Due to the modular design of the system, maintenance and upgrades are made simpler, allowing for quick response to new safety requirements and technological advances.

[0071] 7. Powerful scalability: The system architecture is designed to support expansion and can integrate more security monitoring functions, such as detecting other types of security equipment or behaviors.

[0072] Support multiple safety checks: In addition to helmet detection, the system can be expanded to detect other protective equipment, such as protective glasses, gloves, etc., or monitor other safety-related behaviors, such as fire detection, personnel location tracking, etc.

[0073] Meeting future needs: The system is designed with the possibility of future expansion in mind, allowing it to be continuously upgraded and improved as new security requirements and technologies develop.

[0074] The safety production warning method based on AI image recognition proposed in this invention solves the shortcomings of the existing technology in detection accuracy, real-time performance and resource efficiency by introducing lightweight models, anchor-free detection methods and deep learning model optimization. At the same time, the system integrates real-time monitoring and warning functions, has flexible deployment and strong scalability, and significantly improves the efficiency and effectiveness of safety production management. These improvements have not only improved the system in terms of technical performance, but also demonstrated outstanding advantages in practical applications.

[0075] Second, the expected benefits and commercial value of the technical solution of the present invention after transformation are:

[0076] Expected Return:

[0077] 1. Improve safety production efficiency: Through real-time monitoring and intelligent early warning systems, effectively reduce the occurrence of safety accidents, reduce the company's investment in safety management and accident handling costs.

[0078] 2. Reduce labor costs: Replace traditional manual inspections and monitoring, significantly reduce the consumption of human resources, and at the same time improve the coverage and accuracy of monitoring.

[0079] 3. Enhance brand image: Use advanced AI technology to demonstrate the company's foresight in safe production and intelligent management, and enhance the company's technological image and market competitiveness.

[0080] Commercial value:

[0081] 1. Wide market demand: The present invention is applicable to multiple industries, including construction, chemical industry, manufacturing, mining, etc., and has a wide market demand for improving the level of safety production management.

[0082] 2. High return on investment: After a one-time investment, the investment can be recovered in a short time by reducing safety accidents and improving production efficiency, and it can bring sustainable economic benefits.

[0083] 3. Diversified business models: It can make profits through various means such as selling hardware equipment, providing software subscription services, data analysis and reporting, etc., and has high business expansion potential.

[0084] Third, the technical solution of the present invention fills the technical gap in the industry at home and abroad:

[0085] 1. Application of anchor-free target detection: In existing production safety monitoring systems at home and abroad, anchor-based target detection algorithms are generally used. This invention introduces the anchor-free FCOS target detection model in the field of production safety early warning for the first time, achieving higher detection accuracy and efficiency.

[0086] 2. Real-time warning system: Traditional systems are mostly post-analysis and processing. The present invention realizes the instant detection and processing of safety hazards through real-time image processing and warning mechanism, filling the technical gap of real-time intelligent warning system.

[0087] 3. Multi-scenario adaptability: The present invention realizes efficient application in different production scenarios and complex environments through feature pyramid network and multi-scale feature extraction, overcoming the limitations of traditional technology in specific scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 This is a flow chart of a safety production early warning method based on AI image recognition technology provided by an embodiment of the present invention;

[0089] Figure 2 It is a structural diagram of a production safety early warning system based on AI image recognition technology provided by an embodiment of the present invention;

[0090] Figure 3 This is a safety production warning flow chart based on AI image recognition technology provided by an embodiment of the present invention;

[0091] Figure 4 This is the recognition effect of the present invention in indoor scenes provided by the embodiment of the present invention;

[0092] Figure 5 This is the recognition effect of the present invention in an outdoor scene provided by an embodiment of the present invention;

[0093] Figure 6 This is a diagram of a portion of the data set provided by an embodiment of the present invention.

[0094] Figure 7 is a data set diagram of the present invention provided by an embodiment of the present invention;

[0095] Figure 8 is a map diagram provided by an embodiment of the present invention;

[0096] Fig. 9 : are loss maps provided by an embodiment of the present invention; a) qfl loss map, a) dfl loss map, a) bbox loss map. DETAILED DESCRIPTION

[0097] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0098] like Figure 1 As shown, the safety production early warning system based on AI image recognition technology provided by the embodiment of the present invention includes:

[0099] The camera module is responsible for collecting real-time images of the production site;

[0100] An image processing and analysis module is used to pre-process and analyze images collected by the camera to detect whether the worker is wearing a helmet;

[0101] Security monitoring and early warning module, used to analyze and display detection results in real time and issue alarms when potential safety hazards are found;

[0102] Data storage and management module, used to save and manage detection data and warning records for subsequent analysis and audit;

[0103] The network and communication module is used to ensure data transmission and communication between modules.

[0104] like Figure 2 As shown, the safety production early warning method based on AI image recognition technology provided by the embodiment of the present invention includes:

[0105] S1, Image acquisition: The camera module collects image data from the production site in real time and transmits it to the image processing and analysis module through the network. The position and angle of the camera are designed to ensure that key areas are monitored without blind spots;

[0106] S2, image preprocessing: In the image processing and analysis module, the image is first preprocessed, including: denoising: filtering out noise and interference in the image; enhancement: adjusting the brightness and contrast of the image to enhance important features; scaling: adjusting the image to a size suitable for model input;

[0107] S3, feature extraction and target detection: Use the MobileNet network to extract features from the preprocessed image to obtain high-level image features. Then, use the FCOS model to directly detect and locate the helmets on the feature map, including: Feature extraction: The convolutional layer of MobileNet performs multi-level feature extraction on the image to generate a feature map; Target detection: The FCOS model predicts the location information and category of the target on each pixel of the feature map, and identifies the location of the workers and their helmets;

[0108] S4, Detection result analysis: Analyze the detection results to determine whether each detected worker wears a safety helmet correctly. Generate corresponding safety labels based on the detection results and update the monitoring interface;

[0109] S5, real-time monitoring and early warning: the analysis results are displayed on the monitoring interface. If a helmet is not worn or worn improperly, the system will highlight it on the interface and trigger the sound and light alarm for real-time early warning;

[0110] S6, Data storage and management: All image data, detection results and warning records will be stored in the data storage and management module to facilitate subsequent auditing and analysis; the management software provides data retrieval and analysis tools to help security managers generate reports and statistical analysis.

[0111] This paper proposes a safety production early warning method based on AI image recognition technology, which aims to monitor the safety status of the production site in an automated and intelligent way, especially to detect whether the workers are wearing safety equipment (such as helmets) correctly. The system combines a lightweight deep learning model and an anchor-free target detection algorithm to achieve efficient and real-time safety monitoring and early warning.

[0112] 1. System structure and components

[0113] 1.1 Camera Module

[0114] Function: Responsible for collecting real-time images of the production site.

[0115] Specific structure: Multiple high-definition cameras are installed at different locations on the production site to cover all key areas. These cameras can transmit image data to the core processing unit of the system in real time.

[0116] 1.2 Image processing and analysis module

[0117] Function: Preprocess and analyze the images collected by the camera to detect whether the workers are wearing safety helmets.

[0118] Specific structure:

[0119] Preprocessing unit: responsible for denoising, enhancing, and scaling the original image to ensure that the image quality is suitable for subsequent AI analysis.

[0120] AI model: Use MobileNet for feature extraction and combine it with FCOS (anchor-free object detection algorithm) for helmet detection.

[0121] MobileNet: A lightweight convolutional neural network for efficiently extracting image features.

[0122] FCOS: An anchor-free object detection model for predicting the location and classification of helmets directly on feature maps.

[0123] 1.3 Security monitoring and early warning module

[0124] Function: Analyze and display detection results in real time, and issue an alarm when safety hazards are found.

[0125] Specific structure:

[0126] Monitoring interface: displays real-time images and test results of the production site, and highlights personnel who are not wearing helmets.

[0127] Early warning system: including sound and light alarms. When it is detected that the mask is not worn or worn improperly, the system will sound an alarm.

[0128] 1.4 Data Storage and Management Module

[0129] Function: Save and manage detection data and warning records for subsequent analysis and auditing.

[0130] Specific structure:

[0131] Database: stores image data, detection results and warning records.

[0132] Management software: provides data retrieval, report generation and statistical analysis functions.

[0133] 1.5 Network and Communication Module

[0134] Function: Ensure data transmission and communication between modules.

[0135] Specific structure:

[0136] Wired / wireless network: used to connect cameras, processing units and monitoring terminals, supporting fast data transmission.

[0137] Communication protocol: Standard network communication protocol (TCP / IP) is used to ensure system compatibility and data security.

[0138] 2. Operation process and working principle

[0139] 2.1 Image acquisition

[0140] The camera module collects image data from the production site in real time and transmits it to the image processing and analysis module through the network. The position and angle design of the camera ensures that key areas are monitored without blind spots.

[0141] 2.2 Image Preprocessing

[0142] In the image processing and analysis module, the image is first preprocessed, including:

[0143] Denoising: Filter out noise and interference in images.

[0144] Enhance: Adjust the brightness and contrast of an image to enhance important features.

[0145] Scaling: Resize the image to a suitable size for the model input.

[0146] 2.3 Feature extraction and object detection

[0147] Use the MobileNet network to extract features from the preprocessed image to obtain high-level image features. Then, use the FCOS model to directly detect and locate the helmet on the feature map. The specific steps are as follows:

[0148] Feature extraction: The convolutional layer of MobileNet performs multi-level feature extraction on the image to generate a feature map.

[0149] Object detection: The FCOS model predicts the location and category of the object at each pixel in the feature map, identifying the workers and their helmets.

[0150] 2.4 Analysis of test results

[0151] Analyze the test results to determine whether each detected worker is wearing a helmet correctly. Generate corresponding safety labels based on the test results and update the monitoring interface.

[0152] 2.5 Real-time monitoring and early warning

[0153] The analysis results are displayed on the monitoring interface. If a helmet is not worn or is worn improperly, the system will highlight it on the interface and trigger the sound and light alarm for real-time warning.

[0154] 2.6 Data Storage and Management

[0155] All image data, detection results and warning records will be stored in the data storage and management module for subsequent audit and analysis. The management software provides data retrieval and analysis tools to help security managers generate reports and statistical analysis.

[0156] 3. Implementation details and key technologies

[0157] 3.1 MobileNet Network

[0158] Structure: MobileNet uses depth-wise separable convolution to reduce the number of parameters and computation while maintaining efficient feature extraction capabilities.

[0159] Function: As a lightweight feature extractor, MobileNet can run efficiently on devices with low computing resources and is suitable for edge computing environments.

[0160] 3.2 FCOS Detection Model

[0161] Structure: FCOS is an anchor-free object detection model that simplifies the complexity of traditional object detection by predicting the category and bounding box of the object at each pixel position in the feature map.

[0162] Function: Improves the accuracy and speed of target detection, and is particularly suitable for real-time helmet detection tasks.

[0163] 3.3 System Integration

[0164] Network communication: Use standard network communication protocols and security measures to ensure stable and secure data transmission between modules.

[0165] Data management: An efficient data storage and management mechanism has been designed to support the storage and rapid retrieval of large amounts of images and detection results.

[0166] 3.4 Real-time warning mechanism

[0167] Real-time: The system can respond and process helmet detection results at the millisecond level, which is suitable for safe production scenarios that require immediate response.

[0168] Early warning sensitivity: Through flexible configuration options, the system can adjust the early warning sensitivity and alarm strategy according to different scenarios to meet the needs of various production environments.

[0169] Case 1: Safety helmet wearing monitoring at construction sites

[0170] 1. Background

[0171] A large construction site needs to monitor the wearing of workers' helmets. Traditional manual inspection methods are inefficient and have a high rate of missed inspections. To solve these problems, the project team decided to deploy a safety production early warning system based on AI image recognition.

[0172] 2. Deployment plan

[0173] Camera Installation: Multiple high-definition cameras are installed at key areas of the construction site (such as entrances and exits, work areas, overhead working areas, etc.) to ensure comprehensive coverage.

[0174] Device selection: We selected a smart camera that supports edge computing, with a built-in MobileNet network and FCOS target detection model, which can perform image feature extraction and preliminary analysis directly on the device side.

[0175] Data transmission: The camera is connected to the on-site monitoring center via a wireless network to transmit image data in real time.

[0176] Early warning system: An audible and visual alarm is installed in the monitoring center. When the system detects a worker who is not wearing a safety helmet, an alarm will be sounded immediately.

[0177] 3. Implementation steps

[0178] like Figure 3 As shown, the camera collects images in real time: the camera captures images of various areas of the construction site in real time.

[0179] Image preprocessing and feature extraction: The smart camera denoises and enhances the image and uses the MobileNet network to extract image features.

[0180] Object detection and analysis: Use the FCOS model to detect workers and their helmets in the image and determine whether they are worn correctly.

[0181] Real-time monitoring and early warning: The analysis results are displayed on the monitoring interface. For workers who are not wearing safety helmets, the system will highlight them and trigger the sound and light alarm.

[0182] Data storage and management: All test results and alarm records are automatically stored in the monitoring center's database for subsequent review and analysis.

[0183] 4. Effect evaluation

[0184] Detection accuracy: The system’s accuracy in detecting whether workers are wearing helmets exceeds 95%, which is much higher than the efficiency of manual inspection.

[0185] Response time: The entire process from image acquisition to early warning is completed within milliseconds, ensuring real-time performance.

[0186] Reduce accident rate: After deploying the system, the compliance rate of hard hat wearing on construction sites increased by more than 90%, significantly reducing safety accidents caused by not wearing hard hats.

[0187] Improved management efficiency: Through automated monitoring and early warning, managers can focus on other important security management tasks, improving overall management efficiency.

[0188] like Figure 4 The figure shows the recognition effect of the present invention in indoor scenes.

[0189] like Figure 5 The figure shows the recognition effect of the present invention in an outdoor scene.

[0190] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the method.

[0191] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of a safety production early warning method based on AI image recognition technology.

[0192] An application embodiment of the present invention provides an information data processing terminal, which includes a production safety early warning system based on AI image recognition technology.

[0193] like Figure 6 As shown, the present invention uses the SHWD dataset, which contains 7,800 images in total. In order to make full use of the dataset, this paper divides the dataset in a ratio of 7:3, where 70% of the dataset is used to train the model and 30% of the dataset is used to verify the model.

[0194] like Figure 7 As shown in the figure, the experimental platform is used to train this dataset. The image input size is 640×640, the batch size is set to 20, the optimizer uses SGD, the initial learning rate is 0.14, and the learning rate is reduced to 1 / 10 of the previous one at 130, 160, 175, and 185 epochs. A total of 300 epochs are trained. The training result of the algorithm is shown in the figure below.

[0195] like Figure 8 As shown in Figure 1, the average recall rate of the algorithm reached 91.6% when the IoU was 0.5, and the average precision of the algorithm reached 88.5%. This performance met the experimental requirements. The average detection accuracy of the algorithm training (IoU = 0.5) is shown in Figure 1. Fig. 9 shown.

[0196] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0197] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A safety production early warning system based on AI image recognition technology, characterized in that: include: The camera module is responsible for collecting real-time images of the production site; An image processing and analysis module is used to pre-process and analyze images collected by the camera to detect whether the worker is wearing a helmet; Security monitoring and early warning module, used to analyze and display detection results in real time and issue alarms when potential safety hazards are found; Data storage and management module, used to save and manage detection data and warning records for subsequent analysis and audit; The network and communication module is used to ensure data transmission and communication between modules.

2. The safety production early warning system based on AI image recognition technology according to claim 1, characterized in that: The camera module specifically includes: multiple high-definition cameras installed at different locations on the production site to cover all key areas. These cameras can transmit image data to the core processing unit of the system in real time.

3. The safety production early warning system based on AI image recognition technology according to claim 1, characterized in that: The image processing and analysis modules specifically include: Preprocessing unit: responsible for denoising, enhancing, and scaling the original image to ensure that the image quality is suitable for subsequent AI analysis; AI model: Use MobileNet for feature extraction and FCOS for helmet detection; MobileNet: A lightweight convolutional neural network for efficiently extracting image features. FCOS: An anchor-free object detection model for predicting the location and classification of helmets directly on feature maps.

4. The safety production early warning system based on AI image recognition technology according to claim 1, characterized in that: The security monitoring and early warning modules specifically include: Monitoring interface: displays real-time images and test results of the production site, and highlights personnel who are not wearing helmets; Early warning system: including sound and light alarms. When it is detected that the mask is not worn or worn improperly, the system will sound an alarm.

5. The safety production early warning system based on AI image recognition technology as claimed in claim 1, characterized in that: The security monitoring and early warning modules specifically include: Database: stores image data, detection results and warning records; Management software: provides data retrieval, report generation and statistical analysis functions.

6. The safety production early warning system based on AI image recognition technology according to claim 1, characterized in that: The network and communication modules specifically include: Wired / wireless network: used to connect cameras, processing units and monitoring terminals, supporting fast data transmission; Communication protocol: Use standard network communication protocol to ensure system compatibility and data security.

7. A safety production warning method based on AI image recognition technology using the safety production warning system based on AI image recognition technology as claimed in any one of claims 1 to 6, characterized in that: include: S1, Image acquisition: The camera module collects image data from the production site in real time and transmits it to the image processing and analysis module through the network. The position and angle of the camera are designed to ensure that key areas are monitored without blind spots; S2, image preprocessing: In the image processing and analysis module, the image is first preprocessed, including: denoising: filtering out noise and interference in the image; enhancement: adjusting the brightness and contrast of the image to enhance important features; scaling: adjusting the image to a size suitable for model input; S3, feature extraction and target detection: Use the MobileNet network to extract features from the preprocessed image to obtain high-level image features. Subsequently, the FCOS target detection model is used to directly detect and locate the helmets on the feature map, including: Feature extraction: The convolutional layer of MobileNet performs multi-level feature extraction on the image to generate a feature map; Target detection: The FCOS target detection model predicts the location information and category of the target on each pixel of the feature map, and identifies the location of the workers and their helmets; S4, test result analysis: analyze the test results, determine whether each detected worker wears a safety helmet correctly, generate corresponding safety labels according to the test results, and update the monitoring interface; S5, real-time monitoring and early warning: the analysis results are displayed on the monitoring interface. If a helmet is not worn or worn improperly, the system will highlight it on the interface and trigger the sound and light alarm for real-time early warning; S6, Data storage and management: All image data, detection results and warning records will be stored in the data storage and management module to facilitate subsequent auditing and analysis; the management software provides data retrieval and analysis tools to help security managers generate reports and statistical analysis.

8. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the safety production early warning method based on AI image recognition technology as described in claim 7.

9. A computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the safety production early warning method based on AI image recognition technology as described in claim 1.

10. An information data processing terminal, comprising the production safety early warning system based on AI image recognition technology as described in any one of claims 1 to 6.

Images