Intelligent elevator steel wire rope defect detection method based on deep learning
By employing a deep learning-based intelligent elevator wire rope defect detection method, which utilizes multi-angle visible light images and an improved YOLOv8 model, the problem of traditional detection methods being unable to reflect the real-time status of elevator wire ropes is solved. This achieves efficient and real-time defect detection, improving the safety and operational efficiency of elevators.
Patent Information
- Application Number
- CN202510626409.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-15
- Publication Date
- 2025-12-12
AI Technical Summary
Existing technologies cannot reflect the actual condition of elevator wire ropes in real time, have a low detection frequency, and are prone to accumulating hidden dangers. Furthermore, traditional methods affect the normal operation of elevators, have limited detection accuracy, and are difficult to meet the needs of high frequency and rapid response.
A deep learning-based intelligent elevator wire rope defect detection method is adopted. By acquiring multi-angle visible light images, an improved YOLOv8 model is used for defect detection. Combined with a feature enhancement module and ghost convolution, automated defect recognition is achieved.
It enables dynamic monitoring of the elevator wire rope status, preventing elevator shutdowns, improving detection efficiency, and ensuring elevator safety and operational efficiency.
Smart Images
Figure CN121120477A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of image processing, and in particular to an intelligent elevator steel wire rope defect detection method based on deep learning. BACKGROUND
[0002] As an important vertical transportation tool with high frequency of use, the operation safety of an elevator is directly related to the personal safety of passengers and social public safety. As an important component for bearing and traction, the health status of an elevator steel wire rope directly affects the operation reliability and overall safety of the elevator. Therefore, regular and accurate defect detection of the elevator steel wire rope, such as detection of broken wires, wear, exposed steel wires and the like, becomes an indispensable important link in the elevator maintenance work.
[0003] At present, the traditional steel wire rope detection methods mainly include manual visual inspection and handheld magnetic induction detector detection and the like. Although these methods can find obvious surface or local defects such as wear and broken wires to some extent, they have many limitations. First, the traditional methods highly depend on manual regular inspection, the detection frequency is low, the actual state of the steel wire rope cannot be reflected in real time, potential risks are likely to be accumulated, and potential risks cannot be warned in time. Secondly, in the detection process, it is usually necessary to ensure that the elevator is shut down and strict safety protection measures are implemented, which leads to long-time shutdown of the elevator and affects the normal operation efficiency. In addition, manual detection is limited by complex shaft environment, insufficient light and personnel experience level, the detection quality is unstable, and misjudgment and missed detection often occur; and the magnetic induction detection device is complicated to operate, the detection speed is slow, and the detection precision is limited, which is difficult to meet the actual demand of high frequency and fast response. SUMMARY
[0004] The purpose of the present application is to solve the problem of not being able to reflect the actual state of the steel wire rope in real time mentioned in the background technology, and to propose an intelligent elevator steel wire rope defect detection method based on deep learning.
[0005] In the first aspect of the present application, an intelligent elevator steel wire rope defect detection method based on deep learning is provided, which comprises:
[0006] obtaining a visible light image set containing an elevator steel wire rope; the visible light image set includes visible light images of the steel wire rope at various positions and from multiple angles;
[0007] preprocessing any visible light image to obtain a target image;
[0008] inputting the target image into a pre-trained defect detection model to obtain a defect detection result; the defect detection model is a model improved based on a YOLOv8 model;
[0009] According to the overall defect detection result, a defect detection report is generated.
[0010] Optionally, the acquiring the visible light image containing the elevator steel wire rope comprises:
[0011] A plurality of cameras are used to capture a multi-angle video of the steel wire rope;
[0012] According to the timestamp of the video frame rate, the position of the steel wire rope corresponding to each video frame is determined;
[0013] The steel wire rope is divided into multiple segments in units of the width of the field of view of the camera, and adjacent segments have an overlapping area greater than a first preset threshold;
[0014] The video frames corresponding to each segment of the steel wire rope are extracted from the video to form a set of visible light images.
[0015] Optionally, the pre-processing of the target visible light image to obtain a target image comprises:
[0016] The target visible light image is subjected to contrast enhancement to obtain an enhanced image; the target visible light image is any visible light image;
[0017] The enhanced image is subjected to steel wire rope identification to obtain a binary mask matrix of the steel wire rope image;
[0018] The foreground image of the steel wire rope is extracted from the enhanced image through the binary mask matrix;
[0019] The foreground image is subjected to pure color background filling to obtain a target image of a preset size.
[0020] Optionally, the contrast enhancement of the target visible light image to obtain an enhanced image comprises:
[0021] The target visible light image is converted to HSV space to obtain a hue component, a saturation component and a brightness component;
[0022] The brightness component is subjected to histogram equalization to obtain an enhanced brightness component;
[0023] The enhanced brightness component and the hue component and the saturation component are converted back to an RGB image to obtain an enhanced image.
[0024] Optionally, the defect detection model is improved based on a YOLOv8 model, and the specific improvements include:
[0025] A feature enhancement module is added between the skip connection of the backbone network and the neck network, the attention mechanism is used to enhance the feature map output by the backbone network, and then the enhanced feature map is fused with the feature map in the upsampling process of the neck network;
[0026] The operation process of the feature enhancement module includes: performing convolution on the input feature map through different receptive fields to obtain a plurality of feature maps; fusing the plurality of feature maps to convert into an attention weight map; and performing weighting on the input feature map through the attention weight map to obtain an enhanced feature map.
[0027] Optionally, the feature enhancement module includes a first scale branch, a second scale branch, a third scale branch, a fusion layer and an attention layer; wherein:
[0028] The first scale branch is configured to extract features of the input feature map by using a 3x3 convolution layer and a separable dilated convolution layer with a kernel size of 3x3 and a dilated rate of 3 to obtain a first feature map;
[0029] The second scale branch is configured to extract features of the input feature map by using a 1x1 convolution layer and a separable dilated convolution layer with a kernel size of 3x3 and a dilated rate of 1 to obtain a second feature map;
[0030] The third scale branch is configured to extract features of the input feature map by using two 3x3 convolution layers and a separable dilated convolution layer with a kernel size of 3x3 and a dilated rate of 5 to obtain a third feature map;
[0031] The fusion layer is configured to perform weighted summation on the first feature map, the second feature map and the third feature map to obtain a fourth feature map;
[0032] The attention layer is configured to splice two feature maps obtained by performing maximum pooling and average pooling on the fourth feature map along the channel dimension to obtain a fifth feature map; convert the fifth feature map into a first weight map by using a 7x7 convolution layer and a sigmoid function; divide the first weight map into a second weight map and a third weight map along the channel dimension; perform weighting on the first feature map and the third feature map by using the second weight map and the third weight map respectively, add the two weighted feature maps to obtain a sixth feature map; convert the sixth feature map into a fourth weight map by using a 1x1 convolution layer; and output the input feature map after weighting by using the fourth weight map.
[0033] Optionally, the specific improvement further includes: using a ghost convolution to replace a traditional convolution in the neck network; specifically:
[0034] Replacing a bottleneck structure of a C2f module in the neck network with a ghost bottleneck structure; the ghost bottleneck structure performs convolution operation on the input feature map by using two cascaded ghost convolution layers;
[0035] Replacing a traditional convolution module for down-sampling in the neck network with a ghost convolution module.
[0036] Optionally, in the training process of the defect detection model, the WIoU loss is used as the regression loss of the detection frame.
[0037] Advantages of the present application:
[0038] The present application provides an intelligent elevator steel wire rope defect detection method based on deep learning, which comprises the following steps: acquiring a visible light image set containing an elevator steel wire rope; the visible light image set comprises visible light images of the steel wire rope at different positions and angles; preprocessing any visible light image to obtain a target image; inputting the target image into a pre-trained defect detection model to obtain a defect detection result; the defect detection model is a model improved based on a YOLOv8 model; and generating a defect detection report according to all defect detection results.
[0039] By collecting visible light images of the elevator steel wire rope at different angles to ensure coverage of the steel wire rope surface, and using a deep learning model for automatic defect identification, defects can be detected in a timely manner, and dynamic monitoring of the state of the steel wire rope is realized. The method can detect during normal operation of the elevator, avoiding operation losses caused by shutdown, and improving detection efficiency. BRIEF DESCRIPTION OF DRAWINGS
[0040] Figure 1 A flowchart of an intelligent elevator steel wire rope defect detection method based on deep learning is provided for the embodiments of the present application;
[0041] Figure 2 A network architecture diagram of a defect detection model is provided for the embodiments of the present application;
[0042] Figure 3 A structural schematic diagram of a feature enhancement module is provided for the embodiments of the present application;
[0043] Figure 4 A structural schematic diagram of a ghost C2f module is provided for the embodiments of the present application. DETAILED DESCRIPTION
[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0045] The embodiments of the present application provide an intelligent elevator steel wire rope defect detection method based on deep learning. Referring to Figure 1 , Figure 1A flowchart of an intelligent elevator steel wire rope defect detection method based on deep learning is provided for an embodiment of the present application. The method comprises the following steps:
[0046] S101, a set of visible light images containing elevator steel wire ropes is obtained.
[0047] S102, any visible light image is preprocessed to obtain a target image.
[0048] S103, the target image is taken as the input of a pre-trained defect detection model to obtain a defect detection result.
[0049] S104, a defect detection report is generated according to all defect detection results.
[0050] The set of visible light images includes visible light images of the steel wire rope at various positions and multiple angles. The defect detection model is a model improved based on a YOLOv8 model.
[0051] Based on the intelligent elevator steel wire rope defect detection method based on deep learning provided by the embodiment of the present application, the multiple-angle visible light images of the elevator steel wire rope are collected to ensure that the steel wire rope circumferential surface is covered, and the deep learning model is used for automatic defect identification, which can timely find defects and realize dynamic monitoring of the state of the steel wire rope. The method can detect during normal operation of the elevator, avoid operation loss caused by shutdown, and improve detection efficiency.
[0052] In one embodiment, step S101 comprises:
[0053] Step one, a plurality of cameras are used to shoot to obtain multiple-angle videos of the steel wire rope. Specifically, three cameras can be used to shoot around the steel wire rope at an interval of 120 degrees.
[0054] Step two, according to the time stamp of the video frame, the position of the steel wire rope corresponding to each video frame is determined, and a position mark is added. Specifically, the steel wire rope position at the same time is matched with the video frame.
[0055] Step three, the steel wire rope is divided into multiple segments in units of the height of the camera viewfinder, and adjacent segments have an overlapping area greater than a first preset threshold. Specifically, the first preset threshold can be one-third of the unit length, that is, a window size of the unit length is taken, and a step of two-thirds of the unit length is used for sliding to intercept the steel wire rope segment of each window.
[0056] Step four, the video frame corresponding to each segment of the steel wire rope is extracted from the video to form a set of visible light images. Specifically, for any segment of the steel wire rope, the video frame closest to the position mark of the segment of the steel wire rope is extracted.
[0057] The embodiment adopts multiple cameras for multi-angle video shooting, which can comprehensively cover the surface of the steel wire rope and effectively avoid dead angles and missed detection problems. By determining the position of the steel wire rope based on the video frame timestamp and adding a position marker, accurate matching of the image and the actual position is realized, ensuring the spatiotemporal consistency of the data. At the same time, the steel wire rope is segmented according to the camera view height, and an overlapping area is set between adjacent segments, which not only improves the robustness of detection, but also enhances the integrity and fault tolerance of defect capture.
[0058] In one embodiment, step S102 comprises:
[0059] Step one, performing contrast enhancement on the target visible light image to obtain an enhanced image;
[0060] Step two, performing steel wire rope identification on the enhanced image to obtain a binary mask matrix of the steel wire rope image;
[0061] Step three, extracting the foreground image of the steel wire rope from the enhanced image through the binary mask matrix;
[0062] Step four, performing pure color background filling on the foreground image to obtain a target image of a preset size.
[0063] Wherein, the target visible light image is any visible light image.
[0064] The embodiment improves the effect of steel wire rope defect detection through a series of image processing steps. First, by performing contrast enhancement on the target visible light image, the details of the steel wire rope can be highlighted, and the contrast between the steel wire rope part and the background in the image is improved, which facilitates subsequent steel wire rope identification and defect positioning. Then, by performing steel wire rope identification on the enhanced image and generating a binary mask matrix, the steel wire rope can be accurately separated from the background, eliminating the interference of the background and providing a clear foreground image for subsequent feature extraction. Finally, by filling the pure color background and adjusting the target image to a preset size, the image size conforms to the unified standard, which facilitates model input, improves the processing efficiency and stability of the system.
[0065] In one implementation, performing contrast enhancement on the target visible light image to obtain an enhanced image comprises: converting the target visible light image to HSV space to obtain a hue component, a saturation component and a brightness component; performing histogram equalization on the brightness component to obtain an enhanced brightness component; converting the enhanced brightness component and the hue component and the saturation component back to an RGB image to obtain the enhanced image. This implementation enhances the brightness without changing the color characteristics, improves the clarity and contrast of the steel wire rope in the image, and thus provides more accurate images for subsequent steel wire rope identification and defect detection. Especially in low light areas, brightness enhancement can effectively improve image quality and reduce detection difficulty caused by insufficient light or shadows.
[0066] In an implementation manner, the steel wire rope recognition on the enhanced image can be implemented using a U-Net model to obtain a binary mask matrix of the steel wire rope image. The U-Net model is a classical image segmentation network, and the core of its structure is an encoder-decoder structure. Through multi-layer convolution operation, image features are extracted, and the spatial resolution of the image is gradually restored through upsampling, so that the target and the background can be accurately segmented. For the steel wire rope defect detection task, the U-Net can accurately identify the boundary and shape of the steel wire rope in the enhanced image, and generate an accurate binary mask matrix, that is, the position of the steel wire rope is represented by a mask of 0 and 1.
[0067] In an embodiment, the defect detection model is improved based on a YOLOv8 model. The YOLOv8 model includes a backbone network, a neck network, and a detection head. Among them, the backbone network is used for feature extraction of the input image; the neck network is used to receive the feature map output by the backbone network, and fuse at different stages to output feature maps of different sizes to the detection head; the detection head is used for target recognition and classification according to the feature map output by the neck network.
[0068] Referring to Figure 2 , Figure 2 A network architecture diagram of a defect detection model provided by an embodiment of the present application.
[0069] Compared with the original YOLOv8 model, the specific improvements include:
[0070] 1. A feature enhancement module (FEM) is added between the skip connection of the backbone network and the neck network, which enhances the feature map output by the backbone network using an attention mechanism, and then fuses the enhanced feature map with the feature map in the upsampling process of the neck network.
[0071] The operation process of the feature enhancement module includes: convolving the input feature map through different receptive fields to obtain a plurality of feature maps; fusing the plurality of feature maps to convert them into an attention weight map; and weighting the input feature map through the attention weight map to obtain an enhanced feature map. The attention mechanism can enable the model to adaptively focus on the most important regions or features in the image, thereby suppressing the influence of irrelevant regions. This process can make the model pay more attention to the key regions in the steel wire rope defect detection task through weighted feature maps, especially in the case of complex background or small target size, which can significantly improve the recognition accuracy of the target region.
[0072] In an implementation manner, referring to Figure 3 , Figure 3A structural diagram of a feature enhancement module is provided for an embodiment of the present application. The feature enhancement module comprises a first scale branch, a second scale branch, a third scale branch, a fusion layer and an attention layer; wherein:
[0073] The first scale branch is configured to extract features of an input feature map by using a 3x3 convolution layer (k=m*m represents a convolution kernel size of m*m) and a separable dilated convolution layer with a kernel size of 3x3 and a dilated rate of 3 (r=d represents a dilated rate of d), to obtain a first feature map.
[0074] The second scale branch is configured to extract features of the input feature map by using a 1x1 convolution layer and a separable dilated convolution layer with a kernel size of 3x3 and a dilated rate of 1, to obtain a second feature map.
[0075] The third scale branch is configured to extract features of the input feature map by using two 3x3 convolution layers and a separable dilated convolution layer with a kernel size of 3x3 and a dilated rate of 5, to obtain a third feature map.
[0076] The fusion layer is configured to perform weighted summation on the first feature map, the second feature map and the third feature map to obtain a fourth feature map. The weighted coefficients are model learnable parameters.
[0077] The attention layer is configured to splice two feature maps obtained by performing global max pooling (GMP) and global average pooling (GAP) on the fourth feature map along the channel dimension to obtain a fifth feature map; convert the fifth feature map into a first weight map by using a 7x7 convolution layer and a sigmoid function; divide the first weight map into a second weight map and a third weight map along the channel dimension; perform weighting on the first feature map and the third feature map by using the second weight map and the third weight map respectively, and add the two weighted feature maps to obtain a sixth feature map; convert the sixth feature map into a fourth weight map by using a 1x1 convolution layer; and perform weighting on the input feature map by using the fourth weight map and output.
[0078] The implementation extracts feature information of different scales through three convolution branches with different dilated rates, helps to capture various details of the steel wire rope, uses separable convolution to reduce the calculation overhead while maintaining a large receptive field. The fusion layer combines the feature maps of multiple scales by weighting, dynamically adjusts the contribution of each scale feature, and enhances the effectiveness of feature representation. At the same time, the attention mechanism is introduced to improve the focusing ability of the model on the defect area.
[0079] 2. The CBS (standard convolution Conv+batch normalization BatchNorm+activation function SiLU) in the original YOLOv8 model is referred to as a standard convolution, and the standard convolution in the neck network is replaced with a ghost convolution. Specifically:
[0080] The bottleneck structure of the C2f module in the neck network is replaced by a ghost bottleneck structure, and the obtained new C2f module is referred to as a ghost C2f module. Referring to Figure 4 , Figure 4 FIG. 1 is a structural schematic diagram of a ghost C2f module provided by an embodiment of the present application. Figure 4 (a) of FIG. 1 shows the structure of the ghost C2f module, Figure 4 (b) of FIG. 1 shows a ghost bottleneck structure, which adopts two cascaded ghost convolution layers to perform convolution operation on the input feature map.
[0081] The standard convolution module (CBS module) for down-sampling in the neck network is replaced by a ghost convolution module, that is, the convolution operation (Conv) in the standard convolution module is replaced by a ghost convolution operation (GhostConv).
[0082] The ghost convolution is a lightweight convolution method, which generates more features through a small amount of calculation, can significantly reduce the amount of calculation and the amount of parameters, and at the same time maintains a high feature extraction capability. Replacing the standard convolution in the neck network can reduce the calculation cost while maintaining the model performance, making the entire model more efficient and suitable for real-time defect detection tasks.
[0083] 3. In the training process of the defect detection model, the Wiou loss is used as the regression loss of the detection frame. The Wiou loss can suppress the over-optimization of simple samples through a dynamic weight distribution mechanism, so that the model can focus on difficult example samples, and the defect detection rate is significantly improved.
[0084] In one embodiment, the defect detection result includes the defect type, the defect position and the confidence. When no defect of any type is detected in the image, it is marked as normal. When any type of defect is identified in the image, the defect type and the defect position are labeled on the image. The image containing the defect label is packaged into a defect detection report and sent to the terminal of the operation and maintenance personnel, so that the operation and maintenance personnel can judge the risk degree. The defect detection report also includes the elevator number and the detection time.
[0085] It should be noted that in this document, terms such as "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or equipment.
[0086] The above detailed description of the embodiments of the present application is only preferred embodiments of the present application, and should not be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the patent coverage of the present application.
Claims
1. A deep learning-based intelligent elevator steel wire rope defect detection method, characterized in that, The method comprises: acquiring a visible light image set containing an elevator steel wire rope; the visible light image set comprises multi-angle visible light images of each position of the steel wire rope; preprocessing any visible light image to obtain a target image; inputting the target image into a pre-trained defect detection model to obtain a defect detection result; the defect detection model is a model improved based on a YOLOv8 model; generating a defect detection report according to all defect detection results. 2.The deep learning-based intelligent elevator steel wire rope defect detection method according to claim 1, characterized in that, The acquiring of the visible light image set containing the elevator steel wire rope comprises: shooting to obtain multi-angle videos of the steel wire rope by using multiple cameras; determining the position of the steel wire rope corresponding to each video frame according to the time stamp of the video frame rate; dividing the steel wire rope into multiple segments in units of the width of the field of view of the camera, and adjacent segments having an overlapping area greater than a first preset threshold; extracting the video frames corresponding to each segment of the steel wire rope from the video to form the visible light image set. 3.The deep learning-based intelligent elevator steel wire rope defect detection method according to claim 1, characterized in that, The preprocessing of any visible light image to obtain a target image comprises: performing contrast enhancement on the target visible light image to obtain an enhanced image; the target visible light image is any visible light image; performing steel wire rope identification on the enhanced image to obtain a binary mask matrix of the steel wire rope image; extracting a foreground image of the steel wire rope from the enhanced image through the binary mask matrix; performing pure color background filling on the foreground image to obtain a target image of a preset size.
4. The deep learning-based intelligent elevator steel wire rope defect detection method according to claim 3, characterized in that, The contrast enhancement on the target visible light image to obtain an enhanced image comprises: converting the target visible light image to HSV space to obtain a hue component, a saturation component and a brightness component; performing histogram equalization on the brightness component to obtain an enhanced brightness component; converting the enhanced brightness component and the hue component and the saturation component back to an RGB image to obtain an enhanced image.
5. The deep learning-based intelligent elevator steel wire rope defect detection method according to claim 1, characterized in that, The defect detection model is improved based on the YOLOv8 model, and the specific improvements comprise: adding a feature enhancement module between the skip connection of the backbone network and the neck network, enhancing the feature map output by the backbone network through the attention mechanism, and then fusing the enhanced feature map with the feature map in the upsampling process of the neck network; the operation process of the feature enhancement module comprises: convolving the input feature map through different receptive fields to obtain multiple feature maps; fusing the multiple feature maps to obtain an attention weight map; weighting the input feature map through the attention weight map to obtain an enhanced feature map.
6. The deep learning-based intelligent elevator steel wire rope defect detection method according to claim 5, characterized in that, The feature enhancement module comprises a first scale branch, a second scale branch, a third scale branch, a fusion layer and an attention layer; wherein: the first scale branch is configured to extract features of the input feature map by using a 3x3 convolution layer and a separable dilated convolution layer with a kernel size of 3x3 and an expansion rate of 3 to obtain a first feature map; the second scale branch is configured to extract features of the input feature map by using a 1x1 convolution layer and a separable dilated convolution layer with a kernel size of 3x3 and an expansion rate of 1 to obtain a second feature map; The third scale branch is configured to perform feature extraction on the input feature map by using two 3x3 convolution layers and a separable dilated convolution layer with a kernel size of 3x3 and a dilated rate of 5, to obtain a third feature map; The fusion layer is configured to perform weighted summation on the first feature map, the second feature map and the third feature map, to obtain a fourth feature map; The attention layer is configured to perform maximum pooling and average pooling on the fourth feature map along the channel dimension, to obtain two feature maps; the attention layer is further configured to concatenate the two feature maps, to obtain a fifth feature map; the attention layer is further configured to convert the fifth feature map into a first weight map by using a 7x7 convolution layer and a sigmoid function; the attention layer is further configured to divide the first weight map into a second weight map and a third weight map along the channel dimension; the attention layer is further configured to perform weighting on the first feature map and the third feature map by using the second weight map and the third weight map respectively, to obtain two weighted feature maps; the attention layer is further configured to add the two weighted feature maps, to obtain a sixth feature map; the attention layer is further configured to convert the sixth feature map into a fourth weight map by using a 1x1 convolution layer; and the attention layer is further configured to perform weighting on the input feature map by using the fourth weight map, and output the weighted feature map.
7. The method according to claim 5, wherein the method is characterized by, Specific improvements further include: using a ghost convolution to replace a traditional convolution in the neck network; in particular: A bottleneck structure of a C2f module in the neck network is replaced by a ghost bottleneck structure; the ghost bottleneck structure is configured to perform convolution operation on the input feature map by using two cascaded ghost convolution layers; A traditional convolution module for down-sampling in the neck network is replaced by a ghost convolution module. 8.The deep learning-based intelligent elevator steel wire rope defect detection method according to claim 5, characterized in that, In the training process of the defect detection model, a WIoU loss is used as a regression loss of the detection frame.