Anti-falling device installation point position identification method based on YOLO-Mamba fusion model

Abstract

The invention provides an anti-falling device installation point position identification method based on a YOLO-Mamba fusion model, and the method comprises the steps: obtaining a to-be-detected image through employing an unmanned plane, carrying out the preprocessing of the to-be-detected image, and obtaining the to-be-detected image after preprocessing, and the to-be-detected image comprises the image information of an anti-falling device installation point position in a target region; carrying out real-time target detection on the image information by adopting a deep learning model combining YOLO11 and Mamba, marking the identified mounting point of the anti-falling device in the image in real time to obtain an identification result, and recording the position information of the mounting point of the anti-falling device; comparing the recognition result with a preset target database to obtain a comparison result, and evaluating the recognition performance of the YOLO11 and Mamba combined deep learning model based on the comparison result; and according to a preset rule, automatically evaluating the compliance of the installation point location of the anti-falling device, and generating an evaluation report.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a method for identifying installation points of an anti-fall device based on a YOLO-Mamba fusion model. Background Art

[0002] In scenarios such as power tower inspection, construction, and aerial work, the identification and management of the installation points of anti-fall devices are key links in ensuring the safety of workers. The installation points of anti-fall devices refer to the specific locations of anti-fall devices installed in aerial work environments to ensure the safety of workers. These points usually include safety belt fixing points, safety rope anchor points, anti-fall net support points, etc. Correctly identifying and managing these points can ensure the effectiveness and reliability of anti-fall devices, thereby reducing the risk of falling in aerial work.

[0003] The traditional method of identifying the installation point of fall protection devices usually relies on manual inspections, which has problems such as low efficiency, high cost, and susceptibility to environmental influences. Manual inspections require a lot of human resources and time, especially in large areas and complex environments, where the inspection efficiency is low. In addition, the cost of manual inspections includes staff wages, equipment maintenance, etc., which is costly in the long run. Manual inspections are greatly affected by environmental factors such as weather, terrain, and lighting, and it is difficult to conduct effective inspections in harsh environments. Manual inspections rely on personal experience and judgment, are highly subjective, and are prone to missed inspections or false inspections.

[0004] Small target detection in drone images has important application prospects in the identification of anti-fall device installation points, but it also faces many technical difficulties. First, when the drone is flying at high altitude, the target object (such as anti-fall device) in the captured image is usually small and has a low contrast with the background environment, resulting in unclear target features. Secondly, during the flight of the drone, affected by factors such as wind and airflow, the image is prone to jitter and blur, which further increases the difficulty of small target detection. In addition, factors such as occlusion, lighting changes, and shadows in complex environments can also interfere with the accuracy of target detection. In the prior art, deep learning models such as YOLO and Faster R-CNN have made certain progress in small target detection, but there are still some limitations. For example, when these models deal with small targets, they are prone to missed detection or false detection due to the small number of target features. In addition, model training requires a large amount of labeled data, and the labeling of small targets in drone images is costly and difficult, which limits the generalization ability and detection accuracy of the model. Therefore, developing a technical method that can efficiently and accurately identify the installation points of anti-fall devices is of great significance for improving safety management efficiency and ensuring the safety of staff.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The main purpose of the present invention is to provide a method for identifying the installation point of an anti-fall device based on a YOLO-Mamba fusion model, so as to solve the problem of low efficiency and accuracy in identifying the installation point of an anti-fall device in a complex environment in the prior art.

[0007] In order to achieve the above-mentioned purpose, according to one aspect of the present invention, a method for identifying the installation point of an anti-fall device based on a YOLO-Mamba fusion model is provided, comprising: step S1: using a drone to obtain an image to be detected and preprocessing it to obtain the preprocessed image to be detected, wherein the image to be detected includes: image information of the installation point of the anti-fall device in the target area; step S2: using a deep learning model combining YOLO11 and Mamba to perform real-time target detection on the image information, marking the identified installation point of the anti-fall device in real time in the image, obtaining a recognition result, and recording the location information of the installation point of the anti-fall device; step S3: comparing the recognition result with a preset target database to obtain a comparison result, and evaluating the recognition performance of the deep learning model combining YOLO11 and Mamba based on the comparison result; step S4: according to preset rules, automatically evaluating the compliance of the installation point of the anti-fall device, and generating an evaluation report.

[0008] Optionally, step S2 includes: step S21: optimizing the backbone, inputting the preprocessed image to be detected, replacing the original C2f module in YOLOv8 with the improved C3K2 module, the C3K2 module outputs a first feature map with feature enhancement and expression capabilities by increasing the depth and width of the convolution layer, optimizing the neck, inputting the first feature map output by the C3K2 module into the neck structure, integrating the C2PSA module after the SPPF module, the C2PSA module combines spatial attention and channel attention mechanisms, outputting a second feature map with detail capture capabilities, optimizing the head, inputting the second feature map into the head structure, using depthwise separable convolution, reducing the number of parameters and the amount of calculation with lightweight structuring, and outputting the detection result.

[0009] Optionally, step S2 includes: step S22: the Mamba module integrates CNN and SSM, specifically for the needs of small target detection, and inputs the first feature map output by the C3K2 module. Through lightweight convolution structure processing, the structural design of multiple convolution layers and pooling layers is utilized to reduce the number of parameters and the amount of calculation, and outputs a local feature map with local patterns and edge information. The Mamba module utilizes a combination of multiple state space modules to capture long-distance dependencies in the image, and outputs a global feature map with contextual information. The Mamba module fuses local features and global features, identifies small target features that are far apart in the image, and outputs a small target detail Mamba fused feature map.

[0010] Optionally, step S2 also includes: step S23: using the YOLO-Mamba fusion model, the first feature map output by the C3K2 module is sent to the mamba module, the long-distance dependency in the first feature map is captured, the local features and global features in the first feature map are extracted using CNN and SSM, and the local features and global features are fused using the Mamba fusion feature module to obtain a small target detail mamba fusion feature map, and the mamba fusion feature map and the YOLO feature map with detail capture capability are fused to obtain a YOLO-Mamba fusion map.

[0011] Optionally, step S3 includes: step S31: collect image samples of the installation points of the anti-fall device, mark them, and build a target database, wherein the samples in the target database cover different lighting conditions, background complexity, and target sizes. The database is updated regularly, new samples are introduced, and a real-time target database is obtained.

[0012] Optionally, step S3 includes: step S32: testing under different environmental conditions, collecting recognition results and performance indicators, analyzing the performance of the model, and further optimizing according to the evaluation results, wherein the environmental conditions include at least: lighting, weather, and terrain.

[0013] Optionally, the preprocessing step includes: step S11: using the high-definition image of the noisy power tower area captured by the drone, applying the image denoising algorithm to obtain a clean image after removing the noise points; adjusting the pixel value distribution of the denoised image to increase the local contrast of the image, so as to obtain an image that highlights the background edge and texture of the power tower; inputting a high-contrast image, and adjusting the resolution of the image to a size suitable for the model input according to the requirements of the target detection model to obtain an optimized image map; annotating the image after the resolution is adjusted through manual or semi-automatic tools to mark the specific location of the installation point of the anti-fall device, and obtaining label information for training the deep learning model combining YOLO11 and Mamba.

[0014] Optionally, S4: according to preset standards, automatically evaluate the compliance of the installation points of the anti-fall device and generate an evaluation report, including: step S41: set the compliance rules of the installation points of the anti-fall device, wherein the compliance rules include the rules of the installation position, quantity requirements, and spacing requirements; step S42: automatically collect the location information of the installation points of the anti-fall device, and compare the collected location information with the preset compliance standards to determine whether each installation point meets the standard requirements; step S43: when it is determined that the installation point does not meet the standard requirements, conduct a detailed analysis of the installation points that do not meet the standards, and identify the specific non-compliance reasons, and the non-compliance reasons include at least: position deviation and quantity deviation; step S44: generate an evaluation report, the evaluation report includes the compliance evaluation results and the non-compliance evaluation results, wherein the non-compliance evaluation results include a list of installation points that do not meet the standards and the non-compliance reasons.

[0015] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the anti-fall device installation point identification method based on the YOLO-Mamba fusion model of the above embodiment.

[0016] According to another aspect of the present invention, a processor is provided, and the processor is used to run a program, wherein when the program is running, the method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model of the above embodiment is executed.

[0017] By applying the technical solution of the present invention and using a drone equipped with a high-resolution camera, high-definition images of the target area can be quickly acquired. The fast detection capability of the YOLO11 model ensures real-time performance, greatly accelerating the entire process from image acquisition to target detection. Compared with manual inspection, the recognition efficiency is greatly improved. The Mamba model focuses on feature extraction of small targets. Combined with the fast detection capability of YOLO11, the recognition accuracy of small targets such as the installation points of anti-fall devices is enhanced. Especially in complex environments, the YOLO-Mamba fusion model can better capture the detailed information of the target, thereby improving the accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0020] Figure 1 is a flow chart of a method for identifying installation points of a fall prevention device based on a YOLO-Mamba fusion model according to an embodiment of the present invention;

[0021] Figure 2 This is the network architecture diagram of the C3K2 module in YOLO11;

[0022] Figure 3 This is the network architecture diagram of the C2PSA module in YOLO11;

[0023] Figure 4 This is the network architecture diagram of the Mamba module;

[0024] Figure 5 It is a network architecture diagram of a method for identifying installation points of anti-fall devices based on a YOLO-Mamba fusion model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] According to an embodiment of the present invention, a method embodiment of a method for identifying installation points of a fall protection device based on a YOLO-Mamba fusion model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0028] It should be noted that the method for identifying the installation point of an anti-fall device based on the YOLO-Mamba fusion model in the present disclosure can be used in the field of unmanned aerial vehicle technology to perform real-time detection and identification of the installation points of an anti-fall device in a power pole tower environment, and can also be used in any field other than the field of unmanned aerial vehicle technology to perform real-time detection and identification of small targets in a complex environment. The present disclosure does not limit the application field of the method for identifying the installation point of an anti-fall device in a power pole tower environment.

[0029] The following embodiments of the present invention can be applied to the method for identifying the installation point of the anti-fall device in various complex environments. The present invention uses a high-resolution camera mounted on an unmanned aerial vehicle to obtain high-definition images of the target area, and uses a deep learning model to perform real-time target detection to achieve automatic identification and compliance assessment of the installation point of the anti-fall device.

[0030] The specific steps include:

[0031] Figure 1 is a method for identifying installation points of anti-fall devices based on a YOLO-Mamba fusion model according to an embodiment of the present invention, such as Figure 1 As shown, the method comprises the following steps:

[0032] Step S1: using a drone to acquire an image to be detected and preprocessing it to obtain the preprocessed image to be detected, wherein the image to be detected includes: image information of the installation point of the anti-fall device in the target area.

[0033] Step S2: Use a deep learning model that combines YOLO11 and Mamba to perform real-time target detection on image information, mark the identified installation points of the anti-fall device in real time in the image, obtain the recognition results, and record the location information of the installation points of the anti-fall device.

[0034] Step S3: Compare the recognition result with the preset target database to obtain a comparison result, and evaluate the recognition performance of the deep learning model combining YOLO11 and Mamba based on the comparison result.

[0035] Step S4: According to preset rules, the compliance of the installation points of the fall protection device is automatically evaluated and an evaluation report is generated.

[0036] Through the above steps, using drones equipped with high-resolution cameras, high-definition images of the target area can be quickly acquired. The fast detection capability of the YOLO11 model ensures real-time performance, greatly accelerating the entire process from image acquisition to target detection. Compared with manual inspections, the recognition efficiency is greatly improved. The Mamba model focuses on feature extraction of small targets. Combined with the fast detection capability of YOLO11, the recognition accuracy of small targets such as the installation points of anti-fall devices is enhanced. Especially in complex environments, the YOLO-Mamba fusion model can better capture the detailed information of the target, thereby improving the accuracy of detection.

[0037] Optionally, step S2 includes: step S21: optimizing the backbone, inputting the preprocessed image to be detected, replacing the original C2f module in YOLOv8 with the improved C3K2 module, the C3K2 module outputs a first feature map with feature enhancement and expression capabilities by increasing the depth and width of the convolution layer, optimizing the neck, inputting the first feature map output by the C3K2 module into the neck structure, integrating the C2PSA module after the SPPF module, the C2PSA module combines spatial attention and channel attention mechanisms, outputting a second feature map with detail capture capabilities, optimizing the head, inputting the second feature map into the head structure, using depthwise separable convolution, reducing the number of parameters and the amount of calculation with lightweight structuring, and outputting the detection result.

[0038] Specifically, YOLO11 adopts an improved backbone (Backbone) and neck (Neck) architecture, which enhances the model's ability to extract features from images. This is especially important for small target detection, because small targets may have weaker features in the image and require stronger feature extraction capabilities for accurate identification. In addition, YOLO11 achieves faster processing speeds while maintaining a balance between accuracy and performance through refined architecture design and optimized training processes. This is especially important for small target detection, because while maintaining high accuracy, images need to be processed quickly to achieve real-time or near real-time detection. The specific contents of YOLO11 include: YOLO11 changes the C2f module in YOLOv8 to the C3K2 module. The C3K2 module enhances the feature extraction capability by increasing the depth and width of the convolutional layer. The structure of the C3K2 module is as follows Figure 2 As shown in Figure 1, it specifically includes multiple convolutional layers and residual connections to improve the richness and expressiveness of features; a C2PSA module is added after the SPPF module. The C2PSA module further enhances the expressiveness of the feature map by combining the spatial attention mechanism and the channel attention mechanism. The structure of the C2PSA module is shown in Figure 1. Figure 3As shown in the figure, it specifically includes the spatial attention module and the channel attention module. Through the synergy of these two modules, the detail capture ability of the feature map is improved; the head of YOLO11 is consistent with the head idea of ​​YOLOv10, using a depth-separable method to reduce redundant calculations and improve efficiency. Specifically, the head of YOLO11 adopts a lightweight convolutional structure, which achieves faster reasoning speed by reducing the number of parameters and the amount of calculation.

[0039] Optionally, step S2 includes: step S22: the Mamba module integrates CNN and SSM, specifically for the needs of small target detection, and inputs the first feature map output by the C3K2 module. Through lightweight convolution structure processing, the structural design of multiple convolution layers and pooling layers is utilized to reduce the number of parameters and the amount of calculation, and outputs a local feature map with local patterns and edge information. The Mamba module utilizes a combination of multiple state space modules to capture long-distance dependencies in the image, and outputs a global feature map with contextual information. The Mamba module fuses local features and global features, identifies small target features that are far apart in the image, and outputs a small target detail Mamba fused feature map.

[0040] Specifically, the Mamba module breaks the limitation of traditional networks in processing long-distance dependencies by integrating the advantages of CNN (convolutional neural network) and SSM (state space model), which is especially important for small target detection because the features of small targets may be far apart in the image, and the model needs to be able to capture these scattered features and correctly identify the target. Figure 4 As shown in the figure, the specific contents of the Mamba module include: using a lightweight convolutional structure to achieve faster inference speed by reducing the number of parameters and the amount of calculation. Specifically, the CNN part of the Mamba module includes multiple convolutional layers and pooling layers, and the local features in the image are extracted through the combination of these layers; the Mamba module captures the long-distance dependencies in the image through the state space model. Specifically, the SSM part includes multiple state space modules, and the global features in the image are captured through the combination of these modules.

[0041] Optionally, step S2 also includes: step S23: using the YOLO-Mamba fusion model, the first feature map output by the C3K2 module is sent to the mamba module, the long-distance dependency in the first feature map is captured, the local features and global features in the first feature map are extracted using CNN and SSM, and the local features and global features are fused using the Mamba fusion feature module to obtain a small target detail mamba fusion feature map, and the mamba fusion feature map and the YOLO feature map with detail capture capability are fused to obtain a YOLO-Mamba fusion map.

[0042] Specifically, the YOLO-Mamba fusion model can fully utilize the Mamba module's ability to handle long-distance dependencies by adding the Mamba module after the C3k2 module. This is especially important for small target detection because the features of small targets may be more scattered in the image. This fusion strategy helps improve the model's ability to recognize small targets, especially in complex scenes, and can better capture the details of small targets, thereby improving detection accuracy. Specifically, the network architecture is as follows: Figure 5 As shown in the figure, the specific contents of the YOLO-Mamba fusion model include: the feature map output by the C3k2 module is used as the input of the Mamba module, so that the Mamba module can directly use the feature information extracted by the C3k2 module. Specifically, the C3k2 module extracts local features in the image through multiple convolutional layers and residual connections, and passes these feature maps to the Mamba module; the Mamba module further processes the feature map output by the C3k2 module, and enhances the feature representation through its unique mechanism, especially in capturing the subtle features of small targets. Specifically, the Mamba module extracts local features and global features in the image through the combination of the CNN part and the SSM part, and fuses these features through the feature fusion module, thereby improving the feature expression ability; the feature map output by the Mamba module is fused with other feature maps of YOLO11, and these features are fused through the feature fusion module, thereby improving the overall expression ability of the model. Specifically, the feature fusion module fuses the feature map output by the Mamba module with other feature maps through weighted fusion, thereby improving the detection accuracy of the model.

[0043] In an embodiment of the present invention, the YOLO-Mamba fusion model training process includes: dividing the data set into a training set and a test set in a 1:1 ratio. Before training, all input images are normalized and adjusted to a resolution of 256×256, and then input into the network. The model uses the Soft-IoU loss function to train the network, and uses the Adam method to optimize, while applying the CosineAnnealingLR scheduler. During the training process, the learning rate is set to 0.001, the batch size is set to 16, and the training is performed for 500 rounds. PyTorch is used to implement it on a computer equipped with an RTX 3090 GPU.

[0044] Optionally, step S3 includes: step S31: collect image samples of the installation points of the anti-fall device, mark them, and build a target database, wherein the samples in the target database cover different lighting conditions, background complexity, and target sizes. The database is updated regularly, new samples are introduced, and a real-time target database is obtained.

[0045] Optionally, step S3 includes: step S32: testing under different environmental conditions, collecting recognition results and performance indicators, analyzing the performance of the model, and further optimizing according to the evaluation results, wherein the environmental conditions include at least: lighting, weather, and terrain.

[0046] In an embodiment of the present invention, the target database includes pre-labeled small target samples of fall protection device installation points, which are used to train and verify the recognition ability of the model. The samples in the database cover the diversity of different lighting conditions, background complexity, target size, etc. The specific contents include: collecting high-definition image samples of fall protection device installation points, annotating them, and constructing a target database; the samples in the database cover the diversity of different lighting conditions, background complexity, target size, etc. to ensure the generalization ability of the model; regularly updating the database, introducing new samples, and maintaining the timeliness and comprehensiveness of the database.

[0047] In an embodiment of the present invention, the periodic evaluation step includes testing under different environmental conditions, collecting recognition results and performance indicators, analyzing the performance of the model, and further optimizing according to the evaluation results. The specific evaluation method includes: testing under different environmental conditions, such as different lighting, weather, terrain and other conditions, collecting recognition results and performance indicators; analyzing the performance of the model, evaluating the performance indicators of the model; and further optimizing according to the evaluation results to improve the adaptability and robustness of the model.

[0048] In the embodiment of the present invention, the performance evaluation indicators are precision and recall. Precision measures the proportion of actual positive samples among the samples predicted by the model as positive. A high precision means that the model's prediction of the positive class is reliable. The calculation formula of precision is as follows:

[0049]

[0050] Among them, TP (True Positive) represents the number of true positive examples, that is, the number of samples that are actually positive and the model predicts correctly. FP (False Positive) represents the number of false positive examples, that is, the number of samples that are actually negative but the model incorrectly predicts as positive. The recall rate measures the proportion of actual positive samples that the model successfully captures. A high recall rate means that the model has a strong ability to identify positive samples. The calculation formula for the recall rate is as follows:

[0051]

[0052] Among them, FN (False Negative) represents the number of false negative examples, that is, the number of samples that are actually positive but are mistakenly predicted as negative by the model.

[0053] Optionally, the preprocessing step includes: step S11: using the high-definition image of the noisy power tower area captured by the drone, applying the image denoising algorithm to obtain a clean image after removing the noise points; adjusting the pixel value distribution of the denoised image to increase the local contrast of the image, so as to obtain an image that highlights the background edge and texture of the power tower; inputting a high-contrast image, and adjusting the resolution of the image to a size suitable for the model input according to the requirements of the target detection model to obtain an optimized image map; annotating the image after the resolution is adjusted through manual or semi-automatic tools to mark the specific location of the installation point of the anti-fall device, and obtaining label information for training the deep learning model combining YOLO11 and Mamba.

[0054] Specifically, during the aerial photography process, the high-resolution camera carried by the drone can capture high-definition images of the power tower area to ensure the clarity and integrity of the image details. After acquiring the image, denoising is first performed, and advanced image processing algorithms such as wavelet transform or median filtering are used to effectively remove noise points in the image and improve the image quality. Next, the contrast of the image is enhanced, and the features of the power tower and its surrounding environment are made more obvious through histogram equalization or adaptive contrast enhancement technology, which is convenient for subsequent target detection. According to the specific needs of target detection, the resolution of the image is adjusted to ensure the clarity of the image details, while avoiding the waste of computing resources caused by too high resolution. After the preprocessing is completed, the image is labeled to mark the specific location of the installation point of the anti-fall device. This step is crucial and provides training data for subsequent target detection. The labeling process uses automated labeling tools combined with manual verification to ensure the accuracy and consistency of the labeling. The labeling information includes the location, size, angle, etc. of the anti-fall device, providing rich feature information for model training.

[0055] Through the above preprocessing steps, not only the quality and clarity of the image are improved, but also high-quality training data is provided for subsequent target detection, ensuring the recognition efficiency and accuracy of the model in complex environments.

[0056] Optionally, step S4 includes: including: step S41: setting compliance rules for the installation points of the anti-fall device, wherein the compliance rules include rules for the installation position, quantity requirements, and spacing requirements; step S42: automatically collecting the location information of the installation points of the anti-fall device, and comparing the collected location information with the preset compliance standards to determine whether each installation point meets the standard requirements; step S43: when it is determined that the installation point does not meet the standard requirements, performing a detailed analysis of the installation points that do not meet the standards, and identifying specific non-compliance reasons, and the non-compliance reasons include at least: position deviation and quantity deviation; step S44: generating an evaluation report, the evaluation report includes compliance evaluation results and non-compliance evaluation results, wherein the non-compliance evaluation results include a list of installation points that do not meet the standards and non-compliance reasons.

[0057] The embodiment of the present invention solves the technical problem of low efficiency and accuracy in identifying the installation point of the anti-fall device in a complex environment by adopting a deep learning model combining YOLO11 and Mamba. The method first uses a drone to obtain high-definition images and preprocesses them to ensure the image quality. Subsequently, a deep learning model combining YOLO11 and Mamba is used to perform real-time target detection, mark and record the location information of the installation point of the anti-fall device. YOLO11 adopts an improved backbone (Backbone) and neck (Neck) architecture, which significantly enhances the model's ability to extract features from images, especially in small target detection. The Mamba module breaks the limitations of traditional networks in processing long-distance dependencies by integrating the advantages of CNN (convolutional neural network) and SSM (state space model), and effectively improves the model's ability to capture scattered features in complex environments. The YOLO-Mamba fusion model makes full use of the Mamba module's ability to process long-distance dependencies, significantly improving detection accuracy and efficiency. By comparing and evaluating the model performance with the target database, automatically evaluating compliance and generating reports according to preset standards, intelligent safety management is achieved. In summary, the present invention improves the efficiency and accuracy of identifying the installation points of anti-fall devices by integrating the advantages of YOLO11 and Mamba models, thereby effectively solving the technical problem of low efficiency and accuracy in identifying the installation points of anti-fall devices in complex environments in the related arts.

[0058] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the anti-fall device installation point identification method based on the YOLO-Mamba fusion model of the above embodiment.

[0059] According to another aspect of the present invention, a processor is provided, and the processor is used to run a program, wherein when the program is running, the method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model of the above embodiment is executed.

[0060] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0061] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0062] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0063] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0064] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0065] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0066] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model, characterized in that: include: Step S1: using a drone to acquire an image to be detected and preprocessing it to obtain a preprocessed image to be detected, wherein the image to be detected includes: image information of the installation point of the anti-fall device in the target area; Step S2: using a deep learning model combining YOLO11 and Mamba to perform real-time target detection on the image information, marking the identified installation points of the anti-fall device in real time in the image to obtain a recognition result, and recording the location information of the installation points of the anti-fall device; Step S3: comparing the recognition result with a preset target database to obtain a comparison result, and evaluating the recognition performance of the deep learning model combining YOLO11 and Mamba based on the comparison result; Step S4: According to preset rules, the compliance of the installation points of the fall protection device is automatically evaluated and an evaluation report is generated.

2. The method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model according to claim 1 is characterized in that: The step S2 comprises: Step S21: Optimize the backbone, input the preprocessed image to be detected, replace the original C2f module in YOLOv8 with the improved C3K2 module, the C3K2 module outputs a first feature map with feature enhancement and expression capabilities by increasing the depth and width of the convolution layer, optimize the neck, input the first feature map output by the C3K2 module into the neck structure, integrate the C2PSA module after the SPPF module, the C2PSA module combines the spatial attention and channel attention mechanisms, outputs a second feature map with detail capture capabilities, optimize the head, input the second feature map into the head structure, use depthwise separable convolution, reduce the number of parameters and the amount of calculation with lightweight structuring, and output the detection result.

3. The method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model according to claim 2 is characterized in that: The step S2 comprises: Step S22: The Mamba module integrates CNN and SSM, specifically for the needs of small target detection, and takes as input the first feature map output by the C3K2 module. Through lightweight convolution structure processing and the structural design of multiple convolution layers and pooling layers, the number of parameters and the amount of calculation are reduced, and a local feature map with local pattern and edge information is output. The Mamba module uses a combination of multiple state space modules to capture long-distance dependencies in the image, and outputs a global feature map with contextual information. The Mamba module fuses local features with global features, identifies small target features that are far apart in the image, and outputs a small target detail Mamba fused feature map.

4. The method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model according to claim 2 is characterized in that: The step S2 further comprises: Step S23: Using the YOLO-Mamba fusion model, the first feature map output by the C3K2 module is sent to the mamba module to capture the long-distance dependency in the first feature map, and the local features and global features in the first feature map are extracted using CNN and SSM. The local features and the global features are fused using the Mamba fusion feature module to obtain a small target detail mamba fusion feature map, and the mamba fusion feature map is fused with the feature map of YOLO with detail capture capability to obtain a YOLO-Mamba fusion map.

5. The method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model according to claim 1, characterized in that: Step S3 includes: Step S31: Collect image samples of the installation points of the anti-fall device, annotate them, and build a target database. The samples in the target database cover different lighting conditions, background complexity, and target sizes. The database is updated regularly, new samples are introduced, and a real-time target database is obtained.

6. The method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model according to claim 5, characterized in that: Step S3 includes: Step S32: Conduct tests under different environmental conditions, collect recognition results and performance indicators, analyze the performance of the model, and further optimize according to the evaluation results, wherein the environmental conditions include at least: lighting, weather, and terrain.

7. The method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model according to claim 1, characterized in that: Step S1 includes: Step S11: using the high-definition image of the noisy power tower area captured by the drone, applying the image denoising algorithm to obtain a clean image after removing the noise points; adjusting the pixel value distribution of the clean image to increase the local contrast of the image, so as to obtain a contrast image that highlights the background edge and texture of the power tower; inputting the contrast image, adjusting the image resolution to a size suitable for the model input according to the requirements of the target detection model, and obtaining an optimized image map; marking the specific location of the anti-fall device installation point on the optimized image map, and obtaining label information for training the deep learning model combining YOLO11 and Mamba.

8. The method for identifying installation points of anti-fall devices based on the YOLO-Mamba fusion model according to claim 1, characterized in that: Step S4 includes: Step S41: setting compliance rules for installation points of fall prevention devices, wherein the compliance rules include rules for installation positions, quantity requirements, and spacing requirements; Step S42: automatically collecting the location information of the installation points of the fall prevention device, comparing the collected location information with the preset compliance standard, and determining whether each installation point meets the standard requirements; Step S43: when it is determined that the installation point does not meet the standard requirements, a detailed analysis is performed on the installation point that does not meet the standard to identify specific non-compliance reasons, where the non-compliance reasons include at least: position deviation and quantity deviation; Step S44: Generate an assessment report, which includes compliance assessment results and non-compliance assessment results, wherein the non-compliance assessment results include a list of installation points that do not meet the standards and the reasons for the non-compliance.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the method for identifying the installation point of a fall protection device based on the YOLO-Mamba fusion model as described in any one of claims 1 to 8.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the method for identifying installation points of fall protection devices based on the YOLO-Mamba fusion model as described in any one of claims 1 to 8.

Images