Robot-assisted hydropower station safety monitoring and pipeline detection method and related equipment

Abstract

The invention discloses a robot-assisted hydropower station safety monitoring and pipeline detection method and related equipment, belongs to the technical field of hydropower station safety detection, adopts a YOLO V11 architecture, can perform real-time detection, and remarkably improves the detection precision at the same time. Therefore, the target can be recognized more accurately while high-speed detection is kept, especially in application scenes needing quick response, such as safety monitoring and pipeline leakage detection. A Transform layer is introduced into a YOLOV11 architecture, so that the capturing capability of the model on long-distance dependence is enhanced, which is particularly outstanding in the aspect of multi-scale feature fusion. The innovation not only improves the detection precision of the model, but also avoids the transition of computing resources through reasonable design.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydropower station safety detection, and specifically relates to a robot-assisted hydropower station safety monitoring and pipeline detection method and related equipment. Background Art

[0002] Hydropower station safety inspections require quasi-real-time data monitoring and analysis. However, due to the delays in existing systems when processing large-scale high-resolution images and multiple video streams, especially in the transmission environment of 5G+WiFi networks, the need to discover safety hazards in real time is difficult to be fully met, affecting the efficiency and timeliness of inspections.

[0003] Intelligent inspection robots need to perform multiple object detections in complex environments (such as helmet wear, water leakage, oil leakage, etc.). Traditional detection models such as early versions of YOLO perform poorly in multi-scale or small object detection, are prone to false detection or missed detection, and cannot effectively support accurate detection of multiple targets, which may result in missing key safety information during inspections.

[0004] Current detection systems often require high computing resources to improve detection accuracy. It is difficult to run high-energy consumption algorithms on resource-constrained devices such as edge devices or mobile devices, especially when deployed in power station environments for a long time, which may affect the system's energy efficiency management and continuous operation capabilities.

[0005] If the image recognition model training data set in the inspection system has problems such as inaccurate labeling and unbalanced samples, the adaptability of the detection model in the actual power station environment may be reduced. The lack of data quality directly leads to a decrease in the model detection accuracy and stability, affecting the accuracy of anomaly detection and the overall reliability of the system.

[0006] In a hydropower station, the environment is complex and changeable, and existing models often perform well under specific conditions, but have poor generalization capabilities when facing new scenes or object detection in different states. If the inspection system has difficulty maintaining performance when switching between multiple scenes and different detection tasks in the power station, it will increase the frequency of manual intervention and weaken the intelligence and autonomy of the system.

[0007] Existing inspection robots and fixed monitoring terminal devices may focus on a specific type of safety monitoring (such as limited to detecting the wearing of safety helmets), but lack the ability to simultaneously detect multiple abnormal situations (such as water leaks and oil leaks), resulting in the need for multiple independent systems to work together to complete comprehensive safety monitoring. Summary of the invention

[0008] The present invention provides a robot-assisted hydropower station safety monitoring and pipeline detection method and related equipment, which solves the problem that traditional detection models perform poorly in multi-scale or small object detection, are prone to false detection or missed detection, cannot effectively support accurate detection of multiple targets, and may lead to missing key safety information during inspections.

[0009] To achieve the above object, the present invention provides the following technical solutions: Robot-assisted hydropower station safety monitoring and pipeline inspection methods, including: Send data acquisition signals to the robot to obtain image data of the hydropower station; Perform image enhancement and preprocessing operations on image data; Input the image-enhanced and pre-processed data into the established hydropower station safety monitoring and pipeline inspection model, and output the inspection results; The hydropower station safety monitoring and pipeline detection model is based on the YOLO V11 architecture, and multiple layers of Transformer are added to the Neck part of the YOLO V11 architecture.

[0010] Preferably, a multi-layer GELU activation function is added to the YOLO V11 architecture, and the convolution is modified to a dilated convolution.

[0011] Preferably, the transformer adopts a Decoder-Only structure, uses RoPE relative position encoding instead of absolute position encoding, performs normalization on the input of each Transformer layer, uses the RMSNorm normalization function for normalization, and uses the SiLU activation function instead of the ReLU function.

[0012] Preferably, the data enhancement step is specifically as follows: Standard wavelet transform image enhancement, histogram equalization, grayscale change and image sharpening are performed in sequence.

[0013] Preferably, the data preprocessing step specifically includes normalization and resizing operations; Normalization adjusts the eigenvalues ​​of different dimensions to a similar range. The specific formula is as follows:

[0014] Among them, max and min represent the maximum and minimum values ​​of the image pixels respectively. , is the normalized result; The Resize operation reshapes the input data to the same size.

[0015] Preferably, the output of the hydropower station safety monitoring and pipeline detection model is:

[0016] in, Represents the center coordinates of the predicted i-th bounding box and the width and height of the bounding box, They respectively represent the confidence levels of improper wearing of the full cap, water leakage, and oil leakage.

[0017] Preferably, the hydropower station safety monitoring and pipeline detection model adopts a custom training loss function, and the custom training loss function is:

[0018] L 1 is the sum of the absolute values ​​of the coordinate deviations of the center point of the bounding box plus the sum of the absolute values ​​of the deviations of the width and height, L2 is the sum of the squares of the Euclidean distances between the classification confidence and the true label, is the average squared difference between the model predictions and the true values.

[0019] Robot-assisted hydropower station safety monitoring and pipeline inspection system, including Data acquisition module: used to send data acquisition signals to the robot to obtain image data of the hydropower station; Image data processing module: performs image enhancement and preprocessing operations on image data; Detection module: inputs the image enhancement and preprocessed data into the established hydropower station safety monitoring and pipeline detection model, and outputs the detection results; The hydropower station safety monitoring and pipeline detection model is based on the YOLO V11 architecture, and multiple layers of Transformer are added to the Neck part of the YOLO V11 architecture.

[0020] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a hydropower station safety monitoring and pipeline detection method is implemented.

[0021] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a hydropower station safety monitoring and pipeline detection method.

[0022] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a robot-assisted hydropower station safety monitoring and pipeline detection method and related equipment, which adopts the YOLOv11 architecture, can detect in real time and has significantly improved the detection accuracy. This enables the present invention to identify targets more accurately while maintaining high-speed detection, especially in application scenarios such as safety monitoring and pipeline leakage detection that require rapid response. The introduction of the Transformer layer in the YOLOv11 architecture enhances the model's ability to capture long-distance dependencies, which is particularly prominent in multi-scale feature fusion. This innovation not only improves the detection accuracy of the model, but also avoids excessive consumption of computing resources through reasonable design, achieving a good balance between performance and efficiency.

[0023] Furthermore, through the application of multiple image enhancement technologies, the dataset created by the present invention has high robustness and can work stably in a variety of environments, and can maintain good detection effects even in the face of lighting changes or complex backgrounds.

[0024] Furthermore, the GELU activation function is combined with the dilated convolution to expand the receptive field of the model, improve the feature extraction effect, and meet the needs of multi-scale feature fusion. Furthermore, the Decoder-Only structure is adopted, combined with RoPE relative position encoding and RMSNorm normalization strategy to improve the model performance; Furthermore, the SiLU activation function is applied to the model to replace the traditional ReLU in order to obtain better nonlinear mapping effect; Furthermore, a custom loss function that considers the position, classification accuracy, and prediction error is adopted to improve the model’s ability to distinguish specific categories (e.g., improper wearing of helmets, water leakage, and oil leakage). BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flow chart of a robot-assisted hydropower station safety monitoring and pipeline inspection method; Figure 2 This is a schematic diagram of the hydropower station safety monitoring and pipeline inspection model; Figure 3 Schematic diagram of the Transformer layer; Figure 4 Flowchart for model training; Figure 5 This is the model detection flow chart; Figure 6 This is a block diagram of a robot-assisted hydropower station safety monitoring and pipeline inspection system. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions 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 part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0027] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0029] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0030] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0031] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0032] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0033] like Figure 1 As shown, the present invention provides a robot-assisted hydropower station safety monitoring and pipeline detection method, comprising: S101 sends a data acquisition signal to the robot to acquire image data of the hydropower station; S102 performs image enhancement and preprocessing operations on the image data; S103 inputs the image enhanced and preprocessed data into the established hydropower station safety monitoring and pipeline detection model, and outputs the detection results; The hydropower station safety monitoring and pipeline detection model is based on the YOLO V11 architecture, and multiple layers of Transformer are added to the Neck part of the YOLO V11 architecture.

[0034] The hydropower station safety monitoring and pipeline inspection model architecture uses the Atten-YOLO model; The Atten-YOLO model design phase is: Introducing the Transformer layer in the YOLO (You Only Look Once) model can improve the model's ability to capture long-distance dependencies, thereby using the Transformer to capture more complex feature relationships while maintaining speed to maximize detection accuracy. This is why the present invention combines Yolo V11 and transformer.

[0035] YOLO V11 consists of three parts: Backbone, Neck, and Head. The Backbone is usually responsible for extracting image features, and the Neck is usually used to connect the Backbone and the Head, which handles the task of multi-scale feature fusion.

[0036] The present invention adds the Transformer layer to the Neck part for the following reasons: Multi-scale feature fusion: One of the main tasks of the Neck part is to perform multi-scale feature fusion. Transformer can help the model better understand the relationship between features of different scales, which is especially important for object detection tasks because objects may appear at any scale in the image.

[0037] Computational efficiency: Compared with Backbone, the feature map size of the Neck part is smaller, so using the Transformer layer here can reduce the computational burden while obtaining the benefits of the Transformer.

[0038] In order to avoid introducing too many parameters and computational complexity, in the experiment of the present invention, the number of Transformer layers was finally set to 4 based on the performance of speed and accuracy.

[0039] In the overall model, the present invention makes the following improvements: Figure 2 As shown: 1. Add the Transformer module to the YOLO model.

[0040] 2. Fine-tune the YOLO V11 model structure: Specifically, several layers of GELU activation functions are added, and the original convolution is modified to a dilated convolution, so that the convolution operation can obtain a larger receptive field and better extract features.

[0041] 3 For the added Transformer structure, the present invention also makes some structural modifications, such as Figure 3 As shown: The present invention adopts a decoder-only structured transfomer and uses RoPE relative position coding instead of absolute position coding to achieve better contextual relationship capture.

[0042] Furthermore, the present invention performs normalization on the input of each Transformer sublayer, rather than on the output. Specifically, the RMSNorm normalization function is used.

[0043] Finally, in order to improve performance, the present invention replaces ReLU with SiLU activation function. Model training phase: After the model design is completed, a large amount of data training is required to obtain the weights. The process of this step is as follows Figure 4 As shown: Generally speaking, the higher the training data instruction, the better the detection effect in actual application. In order to achieve better actual application effect, the present invention collects real data on site to construct its own collected data set, and performs data enhancement to ensure the robustness of the data set.

[0044] 1. Image Enhancement Since different image enhancement technologies have different advantages and disadvantages, the present invention uses a variety of image enhancement algorithms to improve the contrast, clarity and dynamic range of the image, as described below: 1) Standard wavelet transform image enhancement.

[0045] Wavelet transform is a mathematical method to represent or approximate an observed signal using a family of functions, which is called a wavelet function system. In addition to being an efficient and highly intuitive framework for representing and storing multi-resolution images, wavelet transform also provides powerful insights into the spatial and frequency characteristics of images. The steps of standard wavelet transform image enhancement are as follows: First, select a suitable wavelet basis. After the observed sample is transformed by wavelet, 4 coefficients are obtained, which are approximate coefficients containing low-frequency information and the speed of color change, horizontal detail coefficients containing high-frequency information and high-pass horizontally and low-pass vertically, vertical detail coefficients containing high-frequency information and low-pass horizontally and high-pass vertically, and diagonal detail coefficients containing high-frequency information and high-pass in both directions. Next, nonlinear enhancement is performed on the wavelet coefficients, and the functional relationship is as follows:

[0046] In the formula, G is the wavelet coefficient enhancement multiple, T is the wavelet coefficient threshold, is the wavelet coefficient after image decomposition, is the enhanced wavelet coefficient.

[0047] Finally, the enhanced wavelet coefficients are subjected to inverse wavelet transform to obtain the reconstructed enhanced image. The wavelet transform image enhancement algorithm can effectively highlight the details in the image by enhancing the wavelet coefficients, but attention should be paid to maintaining the image brightness while highlighting the image details.

[0048] 2) Histogram Equalization (HE). This is a technique that uses the histogram of an image to adjust contrast. This technique uses the Cumulative Distribution Function (CDF) to remap the histogram of the scene to a histogram with a probability density function that is close to a uniform distribution, and maps the specified input grayscale level to the output grayscale level, so that the output grayscale level has a probability density function that is approximately uniformly distributed, thereby achieving the purpose of improving image contrast and expanding the dynamic range of the image. If the histogram of any image has many peaks and valleys, then the equalized histogram will have peaks and valleys, but the peaks and valleys will shift. This technique improves contrast, and histogram equalization can redistribute the intensity distribution, with the goal of obtaining a uniform histogram. If the output of an observed image with a total grayscale level of L is Y after image enhancement, the mathematical expression of the histogram equalization process is as follows:

[0049] Represents the grayscale cumulative distribution function of the observed image, defined as:

[0050]

[0051] It represents the number of pixels with gray level k, N is the total number of pixels, and p(k) represents the probability density function of the gray level of the observed image.

[0052] 3) Grayscale change. Grayscale transformation is the simplest type of image enhancement technology, and has the following four main functions: brightness adjustment (brightening or darkening the image), contrast stretching (increasing or reducing contrast), local contrast enhancement or reduction, and grayscale slicing (grayscale cutting). The present invention relates to the following grayscale transformation functions, where and Indicates the location The current input and output of the pixel: Logarithmic transformation, the expression is as follows, where is a constant:

[0053] The power transformation expression is as follows, where is a constant:

[0054] The commonly used linear grayscale transformation is the piecewise linear transformation, which highlights the grayscale interval of the information of interest and suppresses the grayscale interval of no interest. The mathematical expression is as follows, where is a constant:

[0055] 4) Image sharpening. Image sharpening is a basic digital image processing solution used to achieve better image visual quality. The sharpened image often has better characteristics than the original observed image and is more suitable for extracting image features. The principle of image sharpening is to find the gradient of the image. This patent uses the Laplace operator to find the gradient of the image, and the expression is as follows:

[0056] Random Erase: Randomly erase part of the image: randomly select the upper left corner coordinates (x1, y1) and the lower right corner coordinates (x2, y2) in the image, and set the pixel value in the area to 0. The mathematical expression is as follows: set I(x,y)=0,∀x1≤y≤x2,y1≤y≤y2; 2. Preprocessing: In order to eliminate the negative impact of sample values ​​of different magnitudes, the present invention then preprocesses the images of the data set, including normalization and resizing operations.

[0057] Normalization adjusts the eigenvalues ​​of different dimensions to a similar range. The specific formula is as follows:

[0058] Among them, max and min represent the maximum and minimum values ​​of the image pixels respectively. After all samples are normalized, the final sample set is obtained.

[0059] The essence of the resize operation is scaling, which is to reshape the input data to the same size, so that when it is sent to the model, the size of the input dimension is guaranteed. The present invention follows the original configuration of YoloV11, that is, resizing the input image to a size of (640, 640).

[0060] Assuming the original image size is (W, H) and the resized image size is (640, 640), the scaling ratio of width and height is

[0061]

[0062] 3. Model training Because the application fields of the present invention are helmet wearing detection and pipeline water and oil leakage detection, there are three categories in total (irregular helmet wearing, water leakage, oil leakage). Therefore, the output of the model is defined as (8400, 7), where 8400 represents the 8400 boxes predicted by YOLOV11, and 7 represents the border and confidence, as follows:

[0063] in, Represents the predicted center coordinates of the i-th (1<=i<=8400) bounding box and the width and height of the bounding box. They represent the confidence of three categories respectively (improper wearing of the full hat, water leakage, and oil leakage). The one with a higher confidence is the classification result.

[0064] In order to achieve better training results, the present invention customizes a training loss function, which comprehensively considers the position accuracy, classification accuracy and prediction error. The loss function is defined as follows:

[0065] in, L 1 is used to improve the position accuracy of the detection box and ensure that the model has high accuracy when predicting the target position. Specifically, L 1 is defined as the sum of the absolute deviations of the bounding box center coordinates plus the sum of the absolute deviations of the width and height:

[0066] in, Represents the model's predicted bounding box center point and width and height values. and and It is the annotation data value of the center point, width and height of the real bounding box.

[0067] L 2 is used to enhance classification accuracy, optimize the model's ability to distinguish different categories, reduce classification errors, and enable the model to more accurately identify various types of targets. L2 can be the sum of the squares of the Euclidean distance between the classification confidence and the true label:

[0068] in , and is the true label of the corresponding category (one-hot encoding).

[0069] L3 uses M MSE (mean squared error) is used to optimize the prediction error of the model. Specifically, MSE It measures the average squared difference between the model's predicted value and the true value, and can effectively evaluate the prediction performance in regression tasks. In this application, MSE MSE The introduction of MSE helps improve the overall performance of the model in target positioning and classification tasks, and promotes the model's understanding and recognition of complex features. MSE It can be expressed as:

[0070] in is the output predicted by the model, is the actual label value.

[0071] The loss function defined in this way can balance the impact of position prediction, classification and overall prediction error, so that the model can better learn the task of target detection during training.

[0072] The experimental environment during training is as follows. This experiment is to save the weight with the smallest loss during the training process:

[0073] Step 3: Model Reasoning Application: Model application process Figure 5 As shown: The preprocessing steps are the normalization and resizing described above.

[0074] The purpose of the post-processing step is to remove the overlapping bounding boxes in the predicted 8400 boxes and keep only those boxes that are most likely to contain the target object. Specifically, for a set of bounding boxes (B1, B2), the NMS process can be simplified into the following steps: 1. Calculate IoU: First calculate the intersection over union (IoU) between each box and other boxes. The definition of IoU is as follows:

[0075] 2. Sorting: Sort the boxes in descending order based on the confidence score of each box.

[0076] 3. Select boxes: set an IOU threshold th, select the box with the highest score, and then remove all boxes that overlap with it by more than a certain th.

[0077] 4. Iteration: Repeat step 3 until there are no more boxes to remove.

[0078] Finally, we got the test results.

[0079] The present invention can not only be applied to pipeline leakage detection in specific fields, but can also be extended to other application scenarios that require real-time monitoring and target detection, such as vehicle detection in intelligent traffic management systems, abnormal behavior recognition in security monitoring, etc.

[0080] like Figure 6 As shown, another embodiment of the present invention provides a robot-assisted hydropower station safety monitoring and pipeline detection system, comprising Data acquisition module: used to send data acquisition signals to the robot to obtain image data of the hydropower station; Image data processing module: performs image enhancement and preprocessing operations on image data; Detection module: inputs the image enhancement and preprocessed data into the established hydropower station safety monitoring and pipeline detection model, and outputs the detection results; The hydropower station safety monitoring and pipeline detection model is based on the YOLO V11 architecture, and multiple layers of Transformer are added to the Neck part of the YOLO V11 architecture.

[0081] A terminal device is provided in one embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0082] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0083] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0084] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0085] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0086] If the module / unit integrated in the terminal device 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 present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0087] Although the embodiments of the present invention are described above in conjunction with the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments and application fields, and the above-mentioned specific embodiments are only illustrative and instructive, rather than restrictive. Under the guidance of the specification, a person skilled in the art can also make many forms without departing from the scope of protection of the claims of the present invention, all of which belong to the scope of protection of the present invention.

Claims

1. A robot-assisted hydropower station safety monitoring and pipeline detection method, characterized in that: include: Send data acquisition signals to the robot to obtain image data of the hydropower station; Perform image enhancement and preprocessing operations on image data; Input the image-enhanced and pre-processed data into the established hydropower station safety monitoring and pipeline inspection model, and output the inspection results; The hydropower station safety monitoring and pipeline detection model is based on the YOLO V11 architecture, and multiple layers of Transformer are added to the Neck part of the YOLO V11 architecture.

2. The robot-assisted hydropower station safety monitoring and pipeline detection method according to claim 1 is characterized in that: A multi-layer GELU activation function is added to the YOLO V11 architecture, and the convolution is modified to a dilated convolution.

3. The robot-assisted hydropower station safety monitoring and pipeline detection method according to claim 1 is characterized in that: The transformer adopts a Decoder-Only structure, uses RoPE relative position encoding instead of absolute position encoding, performs normalization on the input of each Transformer layer, uses the RMSNorm normalization function for normalization, and uses the SiLU activation function instead of the ReLU function.

4. The robot-assisted hydropower station safety monitoring and pipeline detection method according to claim 1 is characterized in that: The steps of data enhancement are specifically as follows: Standard wavelet transform image enhancement, histogram equalization, grayscale change and image sharpening are performed in sequence.

5. The robot-assisted hydropower station safety monitoring and pipeline detection method according to claim 1 is characterized in that: The specific steps of data preprocessing include normalization and resizing operations; Normalization adjusts the eigenvalues ​​of different dimensions to a similar range. The specific formula is as follows: Among them, max and min represent the maximum and minimum values ​​of the image pixels respectively. , is the normalized result; The Resize operation reshapes the input data to the same size.

6. The robot-assisted hydropower station safety monitoring and pipeline detection method according to claim 1 is characterized in that: The output of the hydropower station safety monitoring and pipeline detection model is: in, Represents the center coordinates of the predicted i-th bounding box and the width and height of the bounding box, They represent the confidence levels of improper full-hat wearing, water leakage, and oil leakage respectively.

7. The robot-assisted hydropower station safety monitoring and pipeline detection method according to claim 6 is characterized in that: The hydropower station safety monitoring and pipeline detection model uses a custom training loss function, which is: L 1 is the sum of the absolute values ​​of the coordinate deviations of the center point of the bounding box plus the sum of the absolute values ​​of the deviations of the width and height, L2 is the sum of the squares of the Euclidean distances between the classification confidence and the true label, is the average squared difference between the model predictions and the true values.

8. Robot-assisted hydropower station safety monitoring and pipeline inspection system, characterized in that: include Data acquisition module: used to send data acquisition signals to the robot to obtain image data of the hydropower station; Image data processing module: performs image enhancement and preprocessing operations on image data; Detection module: inputs the image enhancement and preprocessed data into the established hydropower station safety monitoring and pipeline detection model, and outputs the detection results; The hydropower station safety monitoring and pipeline detection model is based on the YOLO V11 architecture, and multiple layers of Transformer are added to the Neck part of the YOLO V11 architecture.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the hydropower station safety monitoring and pipeline detection method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, the steps of the hydropower station safety monitoring and pipeline detection method as claimed in any one of claims 1 to 7 are implemented.

Images