Safety rope wearing detection method and equipment for aerial operation
By improving the combination of deep learning network and LSKblock module, real-time monitoring of safety ropes in air operation environments is achieved, solving the problems of low efficiency and poor accuracy of traditional manual detection, and improving the accuracy and robustness of safety rope detection. Reduced the risk of accidents.
Patent Information
- Application Number
- CN202411041182.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-12-06
AI Technical Summary
In the air operation environment, traditional safety rope wear inspection relies on manual inspection, which has problems such as low efficiency, blind spots in monitoring and untimely response, resulting in frequent accidents and increased safety hazards.
Using an improved deep learning network, images are acquired through high-resolution cameras, and after preprocessing, the MobileViT network is built and the LSKblock module is introduced to train and detect models to realize real-time monitoring and identification of safety ropes.
It improves the accuracy and robustness of safety rope detection, ensures small target detection capabilities in complex environments, reduces accidents, and improves the safety guarantee of aerial operations.
Smart Images

Figure CN119091291A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of target detection, and in particular to a method and a device for detecting the wearing of a safety rope used for aerial operations. Background Art
[0002] With the development of the global construction industry and technological progress, aerial work has become more and more common in the construction and maintenance of high-rise buildings, bridges and power facilities. However, such work is often accompanied by high risks, such as falling and improper equipment operation, so effective safety management measures are particularly important. Safety ropes are one of the main safety equipment for aerial work, and their correct wearing and use are directly related to the life safety and work efficiency of workers.
[0003] In traditional work site management, supervision usually relies on manual inspection and observation, which has problems such as low efficiency, blind spots in monitoring, and untimely response. Especially in complex and changeable aerial work environments, it is difficult to monitor and accurately identify the safety ropes of workers in real time, resulting in frequent accidents and increased safety hazards. Summary of the invention
[0004] In view of this, an object of the present invention is to provide a method and device for detecting the wearing of a safety rope for aerial operations, so as to solve the problem that the detection of the safety rope usually relies on manual inspection and observation.
[0005] The present invention provides a method for detecting wearing of a safety rope for aerial work, comprising the steps of:
[0006] S1: Obtain an image set of workers wearing safety ropes in an aerial work area, preprocess the image set, and obtain a preprocessed image set;
[0007] S2: construct an improved deep learning network, train the improved deep learning network through the preprocessed image set, and obtain a trained improved deep learning network;
[0008] S3: Input the image to be detected in the aerial work area into the trained improved deep learning network to make a safety judgment on the aerial work.
[0009] Preferably, step S1 specifically comprises:
[0010] S11: In the aerial work area, a high-resolution camera is used to obtain an image set of the aerial work area;
[0011] S12: each image in the image set is sequentially subjected to median filtering, Gaussian filtering, histogram equalization, brightness adjustment and color correction to obtain a pre-processed image set.
[0012] Preferably, step S2 specifically comprises:
[0013] S21: Build an initial deep learning network, replace the Backbone part of the initial deep learning network with the MobileViT network, and introduce the LSKblock module into the initial deep learning network to obtain an improved deep learning network;
[0014] S22: Iteratively train the improved deep learning network through the preprocessed image set until the loss value of the improved deep learning network is less than a preset value, thereby obtaining a trained improved deep learning network.
[0015] Preferred:
[0016] Improved deep learning network includes: MobileViT network, Neck part and Head part;
[0017] The MobileViT network includes a first Conv module, a first MV2 module, a second MV2 module, a third MV2 module, a fourth MV2 module, a first MobileViT module, a fifth MV2 module, a second MobileViT module, a sixth MV2 module and a third MobileViT module connected in sequence;
[0018] The Neck part includes a first Upsample module, a first Concat module, a first C2f module, a second Upsample module, a second Concat module, a second C2f module, an LSKblock module, a second Conv module, a third Concat module, a third C2f module, a third Conv module, a fourth Concat module and a fourth C2f module, which are connected in sequence, and the first C2f module is connected to the third Concat module;
[0019] The Head part includes: the first Dedect module, the second Dedect module and the third Dedect module;
[0020] The first MobileViT module is connected to the second Upsample module, the second MobileViT module is connected to the first Concat module, and the third MobileViT module is connected to the first Upsample module and the fourth Concat module;
[0021] The LSKblock module is connected to the first Dedect module, the third C2f module is connected to the second Dedect module, and the fourth C2f module is connected to the third Dedect module.
[0022] Preferred:
[0023] Down-sampling layers are further added in the first Conv module, the second MV2 module, the fourth MV2 module, the fifth MV2 module and the sixth MV2 module.
[0024] Preferably, step S3 specifically comprises:
[0025] S31: Input the image to be detected into the trained improved deep learning network, and output the first recognition image, the second recognition image and the third recognition image after passing through the MobileViT network, the Neck part and the Head part in sequence;
[0026] S32: judging whether the worker is wearing a safety rope through the combined effect of the first recognition image, the second recognition image and the third recognition image; if the worker is wearing a safety rope, the aerial operation is judged to be safe; otherwise, the aerial operation is judged to be unsafe.
[0027] A storage medium stores instructions and data for implementing the safety rope wearing detection method for aerial work.
[0028] A safety rope wearing detection device for aerial work comprises: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the safety rope wearing detection method for aerial work.
[0029] The present invention has the following beneficial effects:
[0030] An improved deep learning network is provided for safety rope target detection. The Backbone part of the initial deep learning network is replaced with the MobileViT network. When extracting image features, the MobileViT module combines the processing of local information and global information to ensure the comprehensiveness and accuracy of feature extraction, so that the detection system can better understand and analyze the complex content in the image and improve the overall detection performance. The LSKblock module is introduced into the initial deep learning network. The LSKblock module effectively improves the detection capability of the small target of the safety rope by expanding the receptive field and multi-scale feature fusion. As a small and important target, the safety rope may be obscured or confused by other objects in a complex aerial working environment, so it is crucial to improve its detection accuracy. The introduction of the LSKblock module enables the detection system to more accurately identify and locate the safety rope, improving the robustness of the overall detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 A flow chart of a safety rope wearing detection method for aerial work;
[0032] Figure 2 To improve the deep learning network structure diagram;
[0033] Figure 3 Workflow diagram of the MobileViT network for improving deep learning networks;
[0034] Figure 4 This is the structure diagram of the MobileViT module;
[0035] Figure 5 This is a schematic diagram of the effect after being processed by the MobileViT module;
[0036] Figure 6 This is the structure diagram of the LSKNet module in the LSKblock module;
[0037] Figure 7 This is a structural diagram of the safety rope wearing detection equipment used for aerial operations;
[0038] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0039] 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.
[0040] Reference Figure 1 The present invention provides a safety rope wearing detection method for aerial operations. Through deep learning technology and embedded systems, real-time monitoring of the safety rope wearing status of operators is achieved to improve the safety of aerial operations. Aerial operations are highly dangerous, and falling accidents are prone to occur if the safety rope is not worn correctly.
[0041] Includes steps:
[0042] S1: Obtain an image set of workers wearing safety ropes in an aerial work area, preprocess the image set, and obtain a preprocessed image set;
[0043] Furthermore, step S1 is specifically as follows:
[0044] S11: In the aerial work area, a high-resolution camera is used to obtain an image set of the aerial work area;
[0045] Specifically, high-resolution cameras can capture subtle details, ensuring that safety rope detection can be performed under different lighting conditions. The position and angle of the camera are carefully designed to cover the entire work area to ensure there are no blind spots. Through multi-angle and multi-view image acquisition, the safety status of workers can be fully monitored. This step is not only the foundation of the system, but also the key to ensuring data quality and detection accuracy.
[0046] S12: each image in the image set is sequentially subjected to median filtering, Gaussian filtering, histogram equalization, brightness adjustment and color correction to obtain a pre-processed image set.
[0047] Specifically, the collected image data will undergo a series of enhancement processes to improve its clarity and quality. First, the image is denoised through techniques such as median filtering and Gaussian filtering to remove noise from the image and make the image clearer. Next, the image contrast is adjusted through methods such as histogram equalization to make the image details more obvious. Then, the image brightness is adjusted according to the actual light conditions to ensure that the safety rope can be clearly identified under different lighting environments. In addition, color correction is performed to correct the color deviation of the image so that the color of the image is more realistic and accurate. These enhancement processing steps significantly improve the quality of the image and provide a solid foundation for subsequent image analysis and feature extraction. Through these processes, the image is not only clearer, but also maintains consistent high-quality performance under various environmental conditions, providing reliable data support for the accurate detection and analysis of the safety rope.
[0048] The enhanced image data is transmitted to the embedded computing platform in real time through the wireless transmission module. The wireless transmission module uses high-bandwidth, low-latency communication technology to ensure that the image data can be transmitted to the computing platform in real time and stably. During the transmission process, the data will be encrypted to ensure the security and integrity of the data.
[0049] S2: construct an improved deep learning network, train the improved deep learning network through the preprocessed image set, and obtain a trained improved deep learning network;
[0050] Furthermore, step S2 is specifically as follows:
[0051] S21: Build an initial deep learning network, replace the Backbone part of the initial deep learning network with the MobileViT network, and introduce the LSKblock module into the initial deep learning network to obtain an improved deep learning network;
[0052] Furthermore, the improved deep learning network includes: MobileViT network, Neck part and Head part;
[0053] The MobileViT network includes a first Conv module, a first MV2 module, a second MV2 module, a third MV2 module, a fourth MV2 module, a first MobileViT module, a fifth MV2 module, a second MobileViT module, a sixth MV2 module and a third MobileViT module connected in sequence;
[0054] The Neck part includes a first Upsample module, a first Concat module, a first C2f module, a second Upsample module, a second Concat module, a second C2f module, an LSKblock module, a second Conv module, a third Concat module, a third C2f module, a third Conv module, a fourth Concat module and a fourth C2f module, which are connected in sequence, and the first C2f module is connected to the third Concat module;
[0055] The Head part includes: the first Dedect module, the second Dedect module and the third Dedect module;
[0056] The first MobileViT module is connected to the second Upsample module, the second MobileViT module is connected to the first Concat module, and the third MobileViT module is connected to the first Upsample module and the fourth Concat module;
[0057] The LSKblock module is connected to the first Dedect module, the third C2f module is connected to the second Dedect module, and the fourth C2f module is connected to the third Dedect module.
[0058] Furthermore, downsampling layers are added to the first Conv module, the second MV2 module, the fourth MV2 module, the fifth MV2 module and the sixth MV2 module.
[0059] Specifically, MobileViT (Mobile Vision Transformer) is introduced in the backbone of the deep learning framework to replace the traditional deep learning network. MobileViT significantly reduces the number of model parameters and computational complexity by introducing the ViT block (Vision Transformer Block), while maintaining efficient feature extraction capabilities. Figure 3 As shown in Figure 2, the MobileViT module aims to model local and global information in the input tensor using fewer parameters. Formally, for a given input tensor MobileViT applies n×n standard convolutional layers followed by point-wise (or 1×1) convolutional layers to produce The n×n convolutional layer encodes local spatial information, while the point-by-point convolution projects the tensor into a high-dimensional space (or d-dimensional, where d>C) by learning linear combinations of the input channels. The introduction of MobileViT not only improves the performance of the model, but also reduces the demand for hardware resources, allowing embedded devices to run deep learning models efficiently. In this way, the system has powerful feature extraction capabilities while being able to operate efficiently in resource-constrained environments, providing reliable protection for safety rope detection in aerial work areas.
[0060] With the support of the MobileViT module, a lightweight deep learning network is applied to extract features from the preprocessed images. In order to enable MobileViT to learn a global representation with spatially sensitive bias, we transform X L Expand into N non-overlapping flat patches where P = wh, N = HW / P is the number of patches, h≤n and w≤n are the height and width of the patch. For each p∈{1,…,P}, the relationship between patches is encoded by applying the transformation to obtain
[0061] X G (p) = Transformer(X U (p)),1≤p≤P(Formula 1)
[0062] Unlike VITS, which loses the spatial order of pixels, MobileViT neither loses the patch order nor the spatial order of pixels within each patch. Therefore, we can convert Fold to get Then X F Projected to a low C-dimensional space through point-wise convolution and merged with X through a cascade operation. Then, another n×n convolution layer is used to fuse these cascaded features. U (p) Use convolution to encode local information from n×n regions, while X G (p) Encode the global information on the P patches at the pth position, so X G Each pixel in can encode information from all pixels in X, such as Figure 5 Therefore, the overall effective receiving field of view of MobileViT is H×W.
[0063] In order to improve the detection capability of small targets (such as safety ropes), the LSKblock module is introduced in the deep learning network. The LSKblock module achieves high-precision detection of safety ropes by expanding the receptive field and fusing multi-scale features. The overall architecture of the LSKNet backbone is simply built on repeated LSKNet blocks. Figure 6The composition of the LSKNet block is illustrated. Each LSKNet block consists of two residual sub-blocks: a large kernel selection (LK selection) sub-block and a feed-forward network (FFN) sub-block. The LK selection sub-block dynamically adjusts the receptive field of the network as needed. The core LSK module is embedded in the LK selection sub-block. It consists of a series of large kernel convolutions and a spatial kernel selection mechanism. The FFN sub-block is used for channel mixing and feature refinement, which consists of a sequence of fully connected layers, depthwise convolutions, Gelu activations, and a second fully connected layer. This module constructs a larger kernel convolution by explicitly decomposing it into a sequence of depthwise convolutions with larger growing kernels and adding dilation. Specifically, for the i-th depthwise convolution, the kernel size k, dilation rate d, and the expansion of the receptive field RF are defined as follows:
[0064] k i-1 ≤k i ; d1=1,d i-1 <d i ≤RF i-1 (Formula 2)
[0065] RF1=k1,RF i =d i (k i -1)+RF i-1 (Formula 3)
[0066] The increase in kernel size and dilation rate ensures that the receptive field can expand fast enough. This module sets an upper bound on the dilation rate to guarantee that the dilated convolution does not introduce gaps between feature maps. For example, we can decompose a large kernel into 2 or 3 depthwise convolutions. The proposed design has two advantages. First, it explicitly produces multiple features with various large receptive fields, which makes later kernel selection easier. Second, the sequential decomposition is more effective than simply applying a single larger kernel. As shown in the tab. Under the same theoretical receptive field, our decomposition greatly reduces the number of parameters compared to the standard large convolution kernel. In order to obtain features with rich contextual information from different ranges of the input X, a series of depthwise convolutions with different receptive fields are applied:
[0067]
[0068] in is a core k i and expansion d i Suppose there are N decomposed kernels, each of which consists of a 1×1 convolution layer. Further processing:
[0069]
[0070] Channel mixing is allowed for each spatial feature vector. Then, a selection mechanism is proposed to dynamically select kernels for different objects based on the obtained multi-scale features.
[0071] The introduction of the LSKblock module enables the system to maintain efficient and accurate detection capabilities in complex environments.
[0072] S22: Iteratively train the improved deep learning network through the preprocessed image set until the loss value of the improved deep learning network is less than a preset value, thereby obtaining a trained improved deep learning network.
[0073] S3: Input the image to be detected in the aerial work area into the trained improved deep learning network to make a safety judgment on the aerial work.
[0074] Furthermore, step S3 is specifically as follows:
[0075] S31: Input the image to be detected into the trained improved deep learning network, and output the first recognition image, the second recognition image and the third recognition image after passing through the MobileViT network, the Neck part and the Head part in sequence;
[0076] S32: judging whether the worker is wearing a safety rope through the combined effect of the first recognition image, the second recognition image and the third recognition image; if the worker is wearing a safety rope, the aerial operation is judged to be safe; otherwise, the aerial operation is judged to be unsafe.
[0077] Specifically, the system uses a deep learning model to conduct a detailed analysis of the image data collected at the aerial work site to identify and determine whether the safety rope is worn. If it detects that the safety rope is not worn, the system will immediately issue a warning signal to alert the operator and management personnel. At the same time, the system will also store the detected image data and analysis results for subsequent review and analysis. This process ensures the safety of aerial operations, prevents potential safety hazards in a timely manner, and provides strong support for on-site management.
[0078] The user interface displays the detection data, test results and warning information in real time. The system provides user-friendly interactive functions, allowing users to intuitively view the real-time monitoring data and analysis results. Once the system detects an abnormal situation, such as not wearing a safety rope, the warning information will be displayed on the interface in a timely manner, and the user can respond quickly. In addition, the user interface allows the adjustment of monitoring parameters to adapt to different operating environments and requirements, and supports viewing historical data to help users conduct long-term safety management and analysis. Through these functions, the system not only improves the safety of aerial operations, but also improves the convenience and effectiveness of management.
[0079] See also Figure 7 , Figure 74 is a schematic diagram of the working of the hardware device of an embodiment of the present invention, wherein the hardware device specifically comprises: a safety rope wearing detection device 401 for aerial work, a processor 402 and a storage medium 403.
[0080] A safety rope wearing detection device 401 for aerial work: The safety rope wearing detection device 401 for aerial work implements the safety rope wearing detection method for aerial work.
[0081] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the safety rope wearing detection method for aerial work.
[0082] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the safety rope wearing detection method for aerial operations.
[0083] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.
[0084] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order and these words may be interpreted as identifiers.
[0085] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A safety rope wearing detection method for aerial work, characterized in that: Includes steps: S1: Obtain an image set of workers wearing safety ropes in an aerial work area, preprocess the image set, and obtain a preprocessed image set; S2: construct an improved deep learning network, train the improved deep learning network through the preprocessed image set, and obtain a trained improved deep learning network; S3: Input the image to be detected in the aerial work area into the trained improved deep learning network to make a safety judgment on the aerial work.
2. The safety rope wearing detection method for aerial work according to claim 1, characterized in that: Step S1 is specifically as follows: S11: In the aerial work area, a high-resolution camera is used to obtain an image set of the aerial work area; S12: each image in the image set is sequentially subjected to median filtering, Gaussian filtering, histogram equalization, brightness adjustment and color correction to obtain a pre-processed image set.
3. The safety rope wearing detection method for aerial work according to claim 1, characterized in that: Step S2 is specifically as follows: S21: Build an initial deep learning network, replace the Backbone part of the initial deep learning network with the MobileViT network, and introduce the LSKblock module into the initial deep learning network to obtain an improved deep learning network; S22: Iteratively train the improved deep learning network through the preprocessed image set until the loss value of the improved deep learning network is less than a preset value, thereby obtaining a trained improved deep learning network.
4. The safety rope wearing detection method for aerial work according to claim 3, characterized in that: Improved deep learning network includes: MobileViT network, Neck part and Head part; The MobileViT network includes a first Conv module, a first MV2 module, a second MV2 module, a third MV2 module, a fourth MV2 module, a first MobileViT module, a fifth MV2 module, a second MobileViT module, a sixth MV2 module and a third MobileViT module connected in sequence; The Neck part includes a first Upsample module, a first Concat module, a first C2f module, a second Upsample module, a second Concat module, a second C2f module, an LSKblock module, a second Conv module, a third Concat module, a third C2f module, a third Conv module, a fourth Concat module and a fourth C2f module, which are connected in sequence, and the first C2f module is connected to the third Concat module; The Head part includes: the first Dedect module, the second Dedect module and the third Dedect module; The first MobileViT module is connected to the second Upsample module, the second MobileViT module is connected to the first Concat module, and the third MobileViT module is connected to the first Upsample module and the fourth Concat module; The LSKblock module is connected to the first Dedect module, the third C2f module is connected to the second Dedect module, and the fourth C2f module is connected to the third Dedect module.
5. The safety rope wearing detection method for aerial work according to claim 4, characterized in that: Down-sampling layers are further added in the first Conv module, the second MV2 module, the fourth MV2 module, the fifth MV2 module and the sixth MV2 module.
6. The safety rope wearing detection method for aerial work according to claim 1, characterized in that: Step S3 is specifically as follows: S31: Input the image to be detected into the trained improved deep learning network, and output the first recognition image, the second recognition image and the third recognition image after passing through the MobileViT network, the Neck part and the Head part in sequence; S32: judging whether the worker is wearing a safety rope through the combined effect of the first recognition image, the second recognition image and the third recognition image; if the worker is wearing a safety rope, the aerial operation is judged to be safe; otherwise, the aerial operation is judged to be unsafe.
7. A storage medium, characterized in that: The storage medium stores instructions and data for implementing the safety rope wearing detection method for aerial operations as described in any one of claims 1 to 6.
8. A safety rope wearing detection device for aerial work, characterized in that: include: Processor and storage medium; the processor loads and executes instructions and data in the storage medium to implement the safety rope wearing detection method for aerial operations as described in any one of claims 1 to 6.
