Workshop safety intelligent identification and analysis system
By designing an intelligent identification and analysis system in the factory workshop, using real-time monitoring video and improved deep learning model, the problem of lack of real-time nature of traditional safety management is solved, and timely monitoring of workshop safety conditions and hidden danger identification is achieved, which improves the Response speed for security management.
Patent Information
- Application Number
- CN202411545949.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-01
- Publication Date
- 2025-02-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The traditional factory workshop safety management methods lack real-time performance and cannot timely monitor the production environment and personnel behavior, making it difficult to detect and deal with potential safety issues in a timely manner.
Design a workshop security intelligent identification and analysis system, obtain real-time monitoring video through the data acquisition module, preprocess video frames, use an improved deep learning model to perform object detection, and trigger an alarm when preset dangerous items are detected.
Real-time monitoring of the safety conditions of the workshop is realized, potential safety hazards are identified in a timely manner, response speed is improved, and accidents are avoided.
Smart Images
Figure CN119445484A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image analysis, and in particular to a workshop safety intelligent recognition and analysis system. Background Art
[0002] The hidden safety risks in modern factory workshops have always been an important issue that cannot be ignored. With the continuous improvement of the level of industrial automation, the operation of various mechanical equipment, the handling of raw materials and the frequent flow of staff have made the workshop environment more complicated and increased potential safety risks. These hidden dangers not only endanger the lives of employees, but may also have a negative impact on the production process and the economic benefits of the enterprise. Therefore, how to ensure the safety management of the factory workshop during the production process is an issue that every enterprise needs to focus on.
[0003] At present, most factory workshops still rely on traditional safety management methods, such as regular safety education and training for workers, installation of safety signs, implementation of necessary labor protection measures, and intermittent on-site inspections by safety managers. Although these measures have improved workers' safety awareness to a certain extent, their limitations are also obvious.
[0004] Traditional safety management is not very real-time. Safety training is usually held regularly, and inspections are mostly planned. This intermittent management makes it impossible for factory workshops to monitor the production environment and personnel behavior in real time, making it difficult to detect and deal with potential safety issues in a timely manner. For example, workers may not wear safety helmets or run around the workshop when there is no supervision, and these unsafe behaviors cannot be monitored and corrected immediately. This management method actually has blind spots, and once a safety accident occurs, it is often difficult to remedy. Summary of the invention
[0005] The purpose of the present invention is to solve the problem of lack of real-time performance mentioned in the above background technology, and to propose a workshop safety intelligent identification and analysis system.
[0006] In a first aspect of the present invention, a workshop safety intelligent identification and analysis system is provided, the system comprising:
[0007] A data acquisition module is used to acquire the real-time monitoring video of the workshop and capture the latest video frame as the first target image;
[0008] A preprocessing module, used for preprocessing the first target image to obtain a second target image;
[0009] A detection module, configured to use the second target image as an input of a pre-trained deep learning model to obtain a target detection result;
[0010] The alarm module is used to issue an alarm if the target detection result contains a preset dangerous item.
[0011] Optionally, the preprocessing module includes:
[0012] A scaling module, used for scaling the first target image to a preset size to obtain a third target image;
[0013] The normalization module is used to perform normalization processing on the third target image to obtain a second target image.
[0014] Optionally, the deep learning model is a detection model improved based on the YOLOv5 network architecture; the YOLOv5 network architecture includes a backbone network, a neck network and a head network; the backbone network includes a spatial pyramid pooling module; the specific improvements include:
[0015] Adding dilated convolution to the spatial pyramid pooling module to obtain a first improved module;
[0016] An attention module is added to the neck network; the attention module consists of a channel attention submodule and a spatial attention submodule.
[0017] Optionally, the first improved module includes a front 1×1 convolution layer, a first expansion branch, a second expansion branch, a third expansion branch, a fourth expansion branch, a channel splicing layer and a rear 1×1 convolution layer; the first expansion branch includes a cascaded 1×1 convolution layer, a batch normalization and an activation function layer; the second expansion branch includes a cascaded 3×3 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization and an activation function layer; the third expansion branch includes a cascaded 5×5 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization and an activation function layer; the fourth expansion branch includes a cascaded 7×7 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization and an activation function layer; wherein:
[0018] The front 1×1 convolution layer processes the input feature map using a 1×1 convolution kernel and outputs a first intermediate feature map;
[0019] The first expansion branch, the second expansion branch, the third expansion branch and the fourth expansion branch are used to process the first intermediate feature map respectively to obtain a second intermediate feature map, a third intermediate feature map, a fourth intermediate feature map and a fifth intermediate feature map;
[0020] The channel concatenation layer is used to concatenate the second intermediate feature map, the third intermediate feature map, the fourth intermediate feature map and the fifth intermediate feature map according to the channel dimension to obtain a sixth intermediate feature map;
[0021] The post-1×1 convolution layer uses a 1×1 convolution kernel to process the sixth intermediate feature map and outputs a feature map.
[0022] Optionally, the expansion rate of the first expansion branch is 1; the expansion rate of the second expansion branch is 2; the expansion rate of the third expansion branch is 5; and the expansion rate of the third expansion branch is 9.
[0023] Optionally, the neck network includes multiple C3 convolution modules; and adding an attention module to the neck network includes:
[0024] An attention module is added after the last three C3 convolutional modules of the neck network.
[0025] Optionally, the alarm module includes:
[0026] A secondary confirmation module, used to intercept the surveillance video of a preset period of time according to the timestamp of the first target image, and send it to the office terminal of the manager;
[0027] The reminder module is used to broadcast the corresponding safety rules on site in the workshop if confirmation information of dangerous items is received.
[0028] Beneficial effects of the present invention:
[0029] The present invention proposes a workshop safety intelligent identification and analysis system, which includes: a data acquisition module, used to obtain real-time monitoring video of the workshop and capture the latest video frame as a first target image; a preprocessing module, used to preprocess the first target image to obtain a second target image; a detection module, used to use the second target image as an input of a pre-trained deep learning model to obtain a target detection result; and an alarm module, used to issue an alarm if the target detection result contains a preset dangerous item.
[0030] Through video surveillance and image processing technology, the system can monitor the safety status of the workshop in real time, promptly identify potential safety hazards, improve response speed, and avoid accidents. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The present invention will be further described below in conjunction with the accompanying drawings.
[0032] Figure 1 An architecture diagram of a workshop safety intelligent identification and analysis system is provided for an embodiment of the present invention;
[0033] Figure 2 An original structure diagram of YOLOv5 is provided for an embodiment of the present invention;
[0034] Figure 3 A schematic diagram of the structure of a CBS module is provided for an embodiment of the present invention;
[0035] Figure 4 A schematic diagram of the structure of the C3 module is provided for an embodiment of the present invention;
[0036] Figure 5 A structural diagram of an SPP module is provided for an embodiment of the present invention;
[0037] Figure 6 An improved network structure diagram is provided for an embodiment of the present invention;
[0038] Figure 7 A structural schematic diagram of an improved module is provided for an embodiment of the present invention. DETAILED DESCRIPTION
[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] The embodiment of the present invention provides a workshop safety intelligent identification and analysis system. Figure 1 , Figure 1 This is an architecture diagram of a workshop safety intelligent identification and analysis system provided by an embodiment of the present invention. The system includes:
[0041] The data acquisition module is used to acquire the real-time monitoring video of the workshop and capture the latest video frame as the first target image.
[0042] The preprocessing module is used to preprocess the first target image to obtain a second target image.
[0043] The detection module is used to use the second target image as an input of the pre-trained deep learning model to obtain a target detection result.
[0044] The alarm module is used to issue an alarm if the target detection result contains preset dangerous items.
[0045] Based on a workshop safety intelligent identification and analysis system provided by an embodiment of the present invention, through video monitoring and image processing technology, the system can monitor the safety status of the workshop in real time, promptly identify potential safety hazards, improve response speed, and avoid accidents.
[0046] In one implementation, see Table 1, which lists dangerous items that may occur in a workshop.
[0047] Table 1
[0048] serial number Project Category 01 smoke 02 Not wearing a helmet 03 fall 04 run 05 Hand transport
[0049] These project behaviors may lead to safety accidents. According to the project categories in Table 1, the collected monitoring images are annotated and the deep learning model is trained. This allows the model to perceive the occurrence of dangerous behaviors.
[0050] In one embodiment, the pre-processing module includes:
[0051] The scaling module is used to scale the first target image to a preset size to obtain a third target image. The normalization module is used to perform normalization processing on the third target image to obtain a second target image.
[0052] In one implementation, the preset size may be 640×640×3.
[0053] In one implementation, normalization can standardize image pixel values to a uniform range, reducing differences in image brightness and contrast caused by factors such as lighting conditions and camera hardware differences, thereby improving the model's adaptability to different scenarios and enhancing the robustness of target detection.
[0054] In one embodiment, the deep learning model is a detection model improved based on the YOLOv5 network architecture.
[0055] First, the original YOLOv5 is introduced.
[0056] See also Figure 2 , Figure 2 The original structure diagram of a YOLOv5 provided by an embodiment of the present invention. The YOLOv5 network architecture includes a backbone network, a neck network and a head network; the backbone network is used for feature extraction; the neck network is used for feature enhancement; and the head network is used for target detection and classification.
[0057] See also Figure 3 , Figure 3 A schematic diagram of the structure of the CBS module provided in an embodiment of the present invention, Conv+BN+SiLU is a commonly used convolutional layer combination structure, which can effectively improve the learning ability and expression ability of the model.
[0058] See also Figure 4 , Figure 4 The schematic diagram of the structure of the C3 module provided in the embodiment of the present invention is that the C3 module adopts residual connection, that is, the input features are added to the features processed by the convolution layer, which can effectively alleviate the gradient vanishing problem in the deep network and promote the flow of information. In the backbone network, shortcut = True, use Figure 4 The C3 module shown in part (a) of Figure 2. In the neck network, shortcut = False, using Figure 4 The C3 module shown in part (b).
[0059] See also Figure 5 , Figure 5 The main function of the spatial pyramid pooling module (SPP) is to enrich the multi-scale information of the feature map through pooling operations of different scales, thereby improving the target detection performance.
[0060] The improvements proposed in the embodiments of the present invention specifically include:
[0061] See also Figure 6 , Figure 6 An improved network structure diagram provided for an embodiment of the present invention.
[0062] Add dilated convolution to the spatial pyramid pooling module SPP to obtain the first improved module (SPPA, Spatial Pyramid Pooling with Atrous Convolution).
[0063] Add an attention module (CBAM, Convolutional Block Attention Module) to the neck network.
[0064] Among them, the attention module consists of a channel attention sub-module and a spatial attention sub-module.
[0065] In one implementation, by adding dilated convolutions to the spatial pyramid pooling module, the model can expand the receptive field to better capture large-scale contextual information while retaining detail information. This helps improve the model's ability to detect objects of different sizes. Adding an attention module to the neck network allows the model to better focus on key areas and ignore irrelevant information. The channel attention submodule can dynamically adjust feature weights based on the importance of different channels to enhance attention to important features; while the spatial attention submodule can guide the model to more accurately locate the target area, improving the efficiency and effectiveness of detection. These improvements can improve the accuracy and robustness of detection while maintaining the original high-speed performance of YOLOv5, ensuring that the system can provide efficient and accurate feedback when monitoring workshop safety in real time.
[0066] In one embodiment, see Figure 7 , Figure 7A schematic diagram of the structure of an improved module provided for an embodiment of the present invention. The first improved module includes a front 1×1 convolution layer, a first expansion branch, a second expansion branch, a third expansion branch, a fourth expansion branch, a channel splicing layer, and a rear 1×1 convolution layer; the first expansion branch includes a cascaded 1×1 convolution layer, a batch normalization, and an activation function layer; the second expansion branch includes a cascaded 3×3 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization, and an activation function layer; the third expansion branch includes a cascaded 5×5 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization, and an activation function layer; the fourth expansion branch includes a cascaded 7×7 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization, and an activation function layer; wherein:
[0067] The front 1×1 convolution layer uses a 1×1 convolution kernel to process the input feature map and output the first intermediate feature map;
[0068] The first expansion branch, the second expansion branch, the third expansion branch and the fourth expansion branch are used to process the first intermediate feature map respectively to obtain the second intermediate feature map, the third intermediate feature map, the fourth intermediate feature map and the fifth intermediate feature map;
[0069] A channel concatenation layer is used to concatenate the second intermediate feature map, the third intermediate feature map, the fourth intermediate feature map, and the fifth intermediate feature map according to the channel dimension to obtain a sixth intermediate feature map;
[0070] A 1×1 convolution layer is placed afterwards, and the sixth intermediate feature map is processed using a 1×1 convolution kernel to output the feature map.
[0071] In one implementation, by introducing branches with different expansion rates, the system can simultaneously extract feature information under different receptive fields, thereby better coping with object detection of different scales. This enables the model to achieve better results when facing both large and small objects, and adapt to complex workshop scenarios.
[0072] In one implementation, GeLU (Gaussian Error Linear Unit) is used in the activation function layer, and its approximate form is used to simplify the calculation:
[0073]
[0074] In one implementation, the expansion rate of the first expansion branch is 1; the expansion rate of the second expansion branch is 2; the expansion rate of the third expansion branch is 5; and the expansion rate of the third expansion branch is 9. By setting different expansion rates, each branch can capture context information of different ranges. Smaller expansion rates (such as 1 and 2) are suitable for extracting local detail information, while larger expansion rates (such as 5 and 9) can capture a wider range of context features. In this way, both fine detection of small targets in the workshop and global detection of large targets can be taken into account.
[0075] In one embodiment, the neck network includes a plurality of C3 convolution modules; adding an attention module to the neck network includes:
[0076] An attention module is added after the last three C3 convolutional modules of the neck network.
[0077] In one implementation, the attention module can help the model automatically focus on more important features and weaken irrelevant or noisy information. Adding it to the late stage of the neck network (i.e. the last three C3 convolutional modules) can strengthen the focus on key features at the stage of extracting high-level features, especially in workshop safety monitoring, which helps to accurately identify dangerous behaviors.
[0078] In one implementation, the attention module is added to the last few C3 convolutional modules, avoiding the computational burden that may be caused by adding the attention module to the entire network. This can control the complexity and computation of the model while maintaining high detection performance, which is suitable for real-time monitoring applications.
[0079] In one embodiment, the alarm module includes:
[0080] The secondary confirmation module is used to intercept the monitoring video of a preset period of time according to the timestamp of the first target image and send it to the office terminal of the manager so that the manager can confirm the dangerous project.
[0081] The reminder module is used to broadcast the corresponding safety rules on site in the workshop if confirmation information of dangerous items is received.
[0082] In one implementation, the system sends surveillance video clips of detected potentially dangerous items to managers, allowing them to review and confirm the detection results. This process can effectively reduce false alarms in the system, avoid invalid alarms caused by misjudgment, and improve the accuracy and reliability of the system.
[0083] In one implementation, the on-site instant reminder function can make educational broadcasts when dangerous behaviors occur, reminding employees to comply with regulations. Compared with traditional regular training or inspections, this real-time feedback mechanism can educate employees more directly, help them form good safety habits, and thus reduce the occurrence of similar problems in the future.
[0084] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.
Claims
1. A workshop safety intelligent identification and analysis system, characterized in that: The system comprises: A data acquisition module is used to acquire the real-time monitoring video of the workshop and capture the latest video frame as the first target image; A preprocessing module, used for preprocessing the first target image to obtain a second target image; A detection module, configured to use the second target image as an input of a pre-trained deep learning model to obtain a target detection result; The alarm module is used to issue an alarm if the target detection result contains a preset dangerous item.
2. The workshop safety intelligent identification and analysis system according to claim 1 is characterized in that: The preprocessing module comprises: A scaling module, used for scaling the first target image to a preset size to obtain a third target image; The normalization module is used to perform normalization processing on the third target image to obtain a second target image.
3. The workshop safety intelligent identification and analysis system according to claim 1 is characterized in that: The deep learning model is a detection model improved based on the YOLOv5 network architecture; the YOLOv5 network architecture includes a backbone network, a neck network and a head network; the backbone network includes a spatial pyramid pooling module; Specific improvements include: Adding dilated convolution to the spatial pyramid pooling module to obtain a first improved module; An attention module is added to the neck network; the attention module consists of a channel attention submodule and a spatial attention submodule.
4. The workshop safety intelligent identification and analysis system according to claim 3 is characterized in that: The first improved module includes a front 1×1 convolution layer, a first expansion branch, a second expansion branch, a third expansion branch, a fourth expansion branch, a channel splicing layer and a rear 1×1 convolution layer; the first expansion branch includes a cascaded 1×1 convolution layer, a batch normalization and an activation function layer; the second expansion branch includes a cascaded 3×3 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization and an activation function layer; the third expansion branch includes a cascaded 5×5 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization and an activation function layer; the fourth expansion branch includes a cascaded 7×7 maximum pooling layer, a 3×3 expansion convolution layer, a batch normalization and an activation function layer; wherein: The front 1×1 convolution layer processes the input feature map using a 1×1 convolution kernel and outputs a first intermediate feature map; The first expansion branch, the second expansion branch, the third expansion branch and the fourth expansion branch are used to process the first intermediate feature map respectively to obtain a second intermediate feature map, a third intermediate feature map, a fourth intermediate feature map and a fifth intermediate feature map; The channel concatenation layer is used to concatenate the second intermediate feature map, the third intermediate feature map, the fourth intermediate feature map and the fifth intermediate feature map according to the channel dimension to obtain a sixth intermediate feature map; The post-1×1 convolution layer uses a 1×1 convolution kernel to process the sixth intermediate feature map and outputs a feature map.
5. The workshop safety intelligent identification and analysis system according to claim 4 is characterized in that: The expansion rate of the first expansion branch is 1; the expansion rate of the second expansion branch is 2; the expansion rate of the third expansion branch is 5; and the expansion rate of the third expansion branch is 9.
6. The workshop safety intelligent identification and analysis system according to claim 3 is characterized in that: The neck network includes a plurality of C3 convolution modules; and adding an attention module to the neck network includes: An attention module is added after the last three C3 convolutional modules of the neck network.
7. The workshop safety intelligent identification and analysis system according to claim 3 is characterized in that: The alarm module comprises: A secondary confirmation module, used to intercept the surveillance video of a preset period of time according to the timestamp of the first target image, and send it to the office terminal of the manager; The reminder module is used to broadcast the corresponding safety rules on site in the workshop if confirmation information of dangerous items is received.