Model integration method and system for detecting inclined parts of power equipment

Abstract

The invention relates to the technical field of power equipment, and provides a model integration method and system for detection of inclined parts of power equipment. The method comprises the following steps: based on a power equipment image, respectively adopting at least two different inclined target detection models to obtain an inclined target frame of each model; calibrating the confidence coefficient of the inclined target frame detected by each model, converting the geometric features of the inclined target frame into Gaussian distribution, and fusing the Gaussian distribution of the corresponding inclined target frame by adopting Kalman filtering weighted by the calibrated confidence coefficient to obtain new Gaussian distribution; and converting the new Gaussian distribution into a fusion target frame, and outputting the fusion target frame as a detection result. The method aims at integrating two or more existing inclined target detection models, and the final detection result of the inclined part of the power equipment is obtained through a confidence coefficient calibration and detection result fusion method.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power equipment, and in particular to a model integration method and system for detecting tilted components of electric power equipment. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Automatic detection of power equipment components is the basis for automatic identification and analysis of power equipment faults, and plays an important role in the daily inspection, maintenance and overhaul of power systems. By automatically detecting equipment components and identifying potential equipment failures, it can help power departments to find problems in a timely manner and take active overhaul measures to prevent accidents, reduce power outages, maintain the healthy operation of power infrastructure, improve the overall performance of the power grid, and ensure the safe and efficient operation of the power system.

[0004] Due to the changing shooting angles during inspections, power equipment components are tilted in the images. Current methods for detecting tilted components of power equipment are mostly based on deep learning. Although a single deep model has achieved good performance in component detection, there is still a certain gap between performance and actual application requirements when the background is complex, the component is too small, or the tilt angle of the component is too large. In addition, a single deep model also has problems such as insufficient generalization ability and overfitting. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a model integration method and system for detecting tilted parts of power equipment. The present invention aims to integrate two or more existing tilted target detection models, and obtain the final detection result of tilted parts of power equipment through confidence calibration and detection result fusion methods.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A first aspect of the present invention provides a model integration method for detecting tilted components of electric power equipment.

[0008] A model integration method for detecting tilted components of power equipment, comprising:

[0009] Based on the power equipment image, at least two different inclined target detection models are respectively used to obtain an inclined target frame of each model;

[0010] The confidence of the tilted target frame detected by each model is calibrated, the geometric features of the tilted target frame are converted into Gaussian distribution, and the Gaussian distribution of the corresponding tilted target frame is fused using the calibrated confidence-weighted Kalman filter to obtain a new Gaussian distribution;

[0011] The new Gaussian distribution is converted into a fused target frame, and the fused target frame is output as the detection result.

[0012] Furthermore, the process of converting the geometric features of the tilted target frame into a Gaussian distribution includes: taking the center point of the tilted target frame as the expectation of the Gaussian distribution; calculating the rotation matrix of the tilt angle, and obtaining the variance of the Gaussian distribution according to the rotation matrix and the length and height of the tilted target frame, and using a calibrated confidence-weighted Kalman filter to fuse the expectation and variance of each Gaussian distribution to obtain the expectation and variance of a new Gaussian distribution.

[0013] Furthermore, the Kalman filter weighted by the calibrated confidence factor is used to fuse the expectation and variance of each Gaussian distribution, and is expressed by the following formula:

[0014] G kf (μ,Σ)=p1G1(μ1,Σ1)p2G2(μ2,Σ2)

[0015] μ=μ1+K(μ2-μ1)

[0016] Σ=p1Σ1-Kp2Σ2

[0017] K=Σ1(Σ1+Σ2) -1

[0018] Among them, p1 represents the confidence of the tilted target box detected by the first model, μ1 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the first model, Σ1 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the first model, p2 represents the confidence of the tilted target box detected by the second model, μ2 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the second model, Σ2 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the second model, μ represents the expectation of the new Gaussian distribution, and Σ represents the variance of the new Gaussian distribution.

[0019] Furthermore, the process of calibrating the confidence of each tilted target frame includes designing a calibrator based on temperature scaling, optimizing the parameters of the calibrator by data training, and calibrating the confidence of each tilted target frame according to the trained calibrator.

[0020] Furthermore, the geometric features include: a center point, length, width, height and tilt angle of the tilted target frame.

[0021] Furthermore, the geometric features of the tilted target frame predicted by each model are converted into a Gaussian distribution, wherein only the tilted target frames whose confidence after calibration is greater than a set threshold are retained.

[0022] Furthermore, the method for converting the new Gaussian distribution into a fused target frame includes: calculating matrix eigenvalues ​​and matrix eigenvectors based on the variance of the new Gaussian distribution; calculating the width and height of the fused target frame according to the matrix eigenvalues; calculating the inclination angle of the fused target frame according to the matrix eigenvector; and obtaining the fused target frame based on the center point of the inclined target frame, the width and height of the fused target frame, and the inclination angle.

[0023] A second aspect of the present invention provides a model integration system for detecting tilted components of electric power equipment.

[0024] A model integration system for detecting tilted components of power equipment, comprising:

[0025] The inclined target detection module is configured to: based on the power equipment image, respectively use at least two different inclined target detection models to obtain an inclined target frame of each model;

[0026] A confidence calibration module is configured as follows: calibrating the confidence of the tilted target frame detected by each model, converting the geometric features of the tilted target frame into a Gaussian distribution, and using a Kalman filter weighted by the calibrated confidence to fuse the Gaussian distribution of the corresponding tilted target frame to obtain a new Gaussian distribution;

[0027] The detection result fusion module is configured to: transform the new Gaussian distribution into a fused target frame, and output the fused target frame as the detection result.

[0028] Furthermore, the detection result fusion module is also configured to: use the center point of the tilted target frame as the expectation of the Gaussian distribution; calculate the rotation matrix of the tilt angle, and obtain the variance of the Gaussian distribution according to the rotation matrix and the length and height of the tilted target frame; and use the calibrated confidence-weighted Kalman filter to fuse the expectation and variance of each Gaussian distribution to obtain the expectation and variance of the new Gaussian distribution.

[0029] Furthermore, the Kalman filter weighted by the calibrated confidence factor is used to fuse the expectation and variance of each Gaussian distribution, and is expressed by the following formula:

[0030] G kf (μ,Σ)=p1G1(μ1,Σ1)p2G2(μ2,Σ2)

[0031] μ=μ1+K(μ2-μ1)

[0032] Σ=p1Σ1-Kp2Σ2

[0033] K=Σ1(Σ1+Σ2) -1

[0034] Among them, p1 represents the confidence of the tilted target box detected by the first model, μ1 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the first model, Σ1 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the first model, p2 represents the confidence of the tilted target box detected by the second model, μ2 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the second model, Σ2 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the second model, μ represents the expectation of the new Gaussian distribution, and Σ represents the variance of the new Gaussian distribution.

[0035] Furthermore, the confidence calibration module is also configured to: design a calibrator based on temperature scaling, optimize the parameters of the calibrator by data training, and calibrate the confidence of each tilted target frame according to the trained calibrator.

[0036] Furthermore, the geometric features include: a center point, length, width, height and tilt angle of the tilted target frame.

[0037] Furthermore, the geometric features of the tilted target frame predicted by each model are converted into a Gaussian distribution, wherein only the tilted target frames whose confidence after calibration is greater than a set threshold are retained.

[0038] Furthermore, the detection result fusion module is also configured to: calculate the matrix eigenvalues ​​and matrix eigenvectors based on the variance of the new Gaussian distribution; calculate the width and height of the fused target frame according to the matrix eigenvalues; calculate the inclination angle of the fused target frame according to the matrix eigenvector; obtain the fused target frame based on the center point of the inclined target frame, the width and height of the fused target frame and the inclination angle.

[0039] A third aspect of the present invention provides a computer-readable storage medium.

[0040] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the model integration method for detecting tilted components of electric equipment as described in the first aspect above.

[0041] A fourth aspect of the present invention provides a computer device.

[0042] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the model integration method for detecting tilted components of electric power equipment as described in the first aspect above are implemented.

[0043] A fifth aspect of the present invention provides a computer program product or a computer program.

[0044] The present invention provides a computer program product or a computer program, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the model integration method for detecting tilted components of power equipment as described in the first aspect above.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The present invention adopts a variety of tilted target detection models with different architectures and characteristics, and makes full use of the advantages of each model in different scenarios through reasonable configuration in the model selection and construction stages. This multi-model integration strategy can complement the weaknesses of each model and greatly improve the overall accuracy of tilted component detection of power equipment.

[0047] The present invention effectively reduces the difference in prediction confidence between different models by using a confidence calibration method based on temperature scaling, so that the output probability of each model more accurately reflects its confidence level in the detection result. This calibration technology ensures the consistency between different models and provides a more reliable data basis for subsequent result fusion.

[0048] In the detection result fusion stage, the present invention adopts a confidence-weighted Kalman filtering method to comprehensively consider the geometric features and confidence of the tilted target frame. Through this fusion method, not only the advantages of each model are retained, but also the incidence of false detection and missed detection can be effectively reduced, making the final detection results of tilted parts of power equipment more accurate and stable.

[0049] The model integration method of the present invention is specially optimized for the complex characteristics of tilted parts of power equipment, and can handle detection tasks in a variety of complex environments. By combining the fusion strategy of geometric features and confidence, the system has higher adaptability and robustness when facing different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings in the specification, which constitute a part of the present invention, 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 improper limitations on the present invention.

[0051] Figure 1 is a flow chart of a model integration method for detecting tilted components of electric power equipment shown in the present invention;

[0052] Figure 2 It is a framework diagram of a model integration method for detecting tilted components of electric power equipment shown in the present invention;

[0053] Figure 3 It is a schematic diagram of frame fusion shown in the present invention. DETAILED DESCRIPTION

[0054] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0055] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0056] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0057] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of a code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram can be implemented using a dedicated hardware-based system that performs a specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0058] Embodiment 1

[0059] like Figure 1As shown, this embodiment provides a model integration method based on the detection of tilted components of power equipment. This embodiment uses the method applied to a server as an example for illustration. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in this application. In this embodiment, the method includes the following steps:

[0060] Step (1) Model selection and construction: This method first selects two or more models with different architectures and characteristics from the existing deep models for tilted target detection, and adapts them to the categories and characteristics of power equipment components for subsequent integration optimization. These models have their own advantages in architecture design, feature extraction capabilities, and performance in handling complex scenarios. By selecting these models for integration, not only can the advantages of each model in a specific situation be fully utilized, but they can also complement each other's weaknesses, thereby effectively improving detection accuracy.

[0061] Step (2) Confidence calibration: The output of the tilted target detection model includes the location, size, category and confidence of the target. Among them, different models have different predicted confidence values ​​due to different architectures, different output heads and different loss functions. The present invention adopts a confidence calibration method based on temperature scaling to calibrate the output confidence of each model so that the probability value of its output can more accurately reflect the model's confidence in the detection result, so as to reduce the prediction deviation between models and ensure that the final fusion result is more reliable.

[0062] Step (3) fusion of detection results: After completing the confidence calibration, the present invention fuses the detection results of each model.

[0063] The present invention designs a confidence-weighted Kalman filtering method, which combines the geometric features and confidence of the tilted target frame to comprehensively process the output results of each model. Through this fusion method, the incidence of false detection and missed detection can be minimized while maintaining the advantages of each model, thereby obtaining a more accurate and robust detection result of the tilted components of the power equipment.

[0064] In step (1), this method focuses on the diversity of model architectures when selecting candidate models for integration. This diversity can not only enhance the generalization ability of the model, but also take advantage of the advantages of different models in feature extraction and target recognition. Therefore, in the specific implementation, the following two models are selected:

[0065] Step (1-1): Figure 2 As shown in the figure, the model Oriented R-CNN based on the convolutional neural network (CNN) architecture is good at handling directional target detection problems. By introducing a rotation box and an optimized regression mechanism, it can more accurately detect and locate tilted or rotated targets. The hierarchical structure of CNN helps to gradually extract features from low to high layers, so that it has strong adaptability to targets of different scales.

[0066] Step (1-2): Model AO2-DETR based on Transformer architecture. The Transformer architecture can effectively capture global information and model the long-distance dependencies of input data through the self-attention mechanism. AO2-DETR combines the self-attention mechanism with the special requirements of the target detection task, and can show strong performance in dealing with target detection in complex scenes, especially in the case of multiple targets and target occlusion.

[0067] By selecting these two models with different architectural characteristics and detection advantages, this method can fully utilize their complementarity in the integrated model and improve the adaptability and accuracy of the overall system in different application scenarios. At the same time, this diversified model selection strategy can also effectively reduce the bias problem caused by the single architecture of the model, thereby enhancing the robustness of the detection task.

[0068] In addition, the two models will adapt the classification head according to the category and characteristics of the power equipment components. The feature extraction module of the model will use the weights pre-trained on the general inclined target detection dataset as initialization, while the weights of the classification head will be randomly initialized. Subsequently, the two models will be fine-tuned on the power equipment component detection dataset.

[0069] The present invention adopts a variety of tilted target detection models with different architectures and characteristics, and makes full use of the advantages of each model in different scenarios through reasonable configuration in the model selection and construction stages. This multi-model integration strategy can complement the weaknesses of each model and greatly improve the overall accuracy of tilted component detection of power equipment.

[0070] In step (2), the present invention first extracts a portion of sample data from the model training or verification process as a calibration data set, which includes the model output (target box center position, length and width, tilt angle, category and confidence) and the true label. Ensure that the selected data set covers as many target types, different confidence ranges and various detection scenarios as possible to improve the generalization ability of the calibration process.

[0071] Then, for each tilted target detection model, an independent temperature-scaling-based confidence calibrator is trained. The output confidence and true label in the calibration dataset are used to optimize the parameters of the calibration model so that the calibrated confidence value better matches the actual classification and positioning accuracy. The specific design is as follows:

[0072] Given an image X, the output of the tilted object detection model is {b i ,c i ,p i}, where b i =(x i ,y i ,w i ,h i ,θ i ) is the center point of the target box (x i ,y i ), width w i , high h i and the tilt angle θ i , c i is the component category, p i is the confidence level. When a calibration data training set is given, the target detection results of all images in it are recorded as Where M is the total number of target boxes. We divide the confidence value into J intervals. For a certain category c, the verification error considering both positioning and classification is defined as:

[0073]

[0074] Among them, E c Indicates the calibration error, For all target detection boxes of category c, is the target box set with confidence in interval j, is the average confidence of all target detection boxes of category c in interval j, precision c is the classification accuracy of the detection, IoU c It is the intersection-over-union ratio of the detected target box and the true value.

[0075] The present invention employs a temperature scaling based calibrator defined as

[0076]

[0077] Among them, p is the probability distribution of confidence, is the calibrated average confidence of all target detection boxes of category c in interval j, T>0 is the temperature scaling parameter to be learned, and the calibration error is minimized by During the training process, the calibrator will gradually adjust the temperature scaling parameters according to the distribution of the input confidence values. After the model training is completed, the calibrator can map the original input confidence value to a more accurate probability value. This mapping relationship can effectively reduce the deviation between the confidence value and the actual classification and positioning accuracy, thereby making the confidence output of the model more credible in various detection scenarios. Ultimately, the calibrated confidence value can more accurately reflect the performance of the target detection model at different confidence levels, providing more reliable data support for the subsequent fusion of target detection results.

[0078] The present invention effectively reduces the difference in prediction confidence between different models by using a confidence calibration method based on temperature scaling, so that the output probability of each model more accurately reflects its confidence level in the detection result. This calibration technology ensures the consistency between different models and provides a more reliable data basis for subsequent result fusion.

[0079] In the step (3), the present invention designs a confidence-weighted Kalman filtering method, which combines the geometric features and confidence of the tilted target frame to fuse the detection results. Specifically, the method is as follows:

[0080] First, if Figure 3 As shown, the results detected by each model with a confidence greater than the set threshold are retained. If two models have detection results for the same power equipment component, the frame is converted to a Gaussian distribution, and then the confidence-weighted Kalman filter is used to obtain the fused Gaussian distribution, and the fused detection frame is restored.

[0081] The steps to convert the box into a Gaussian distribution are as follows: the center point (x, y) of the rectangular box is used as the expected μ of the two-dimensional Gaussian distribution:

[0082] μ=(x,y) T

[0083] The variance Σ is obtained from the rotation matrix R and the width w and height h of the detection box:

[0084] Σ=RΛR T

[0085] Among them, the rotation matrix R is:

[0086]

[0087] The diagonal matrix Λ is:

[0088]

[0089] The confidence-weighted Kalman filter combines the confidence p to obtain the expectation μ and variance Σ of the new Gaussian distribution:

[0090] G kf (μ,Σ)=p1G1(μ1,Σ1)p2G2(μ2,Σ2)

[0091] μ=μ2+K(μ2-μ1)

[0092] Σ=p1Σ1-Kp2Σ2

[0093] Among them, p1 represents the confidence of the tilted target box detected by the first model, μ1 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the first model, Σ1 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the first model, p2 represents the confidence of the tilted target box detected by the second model, μ2 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the second model, Σ2 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the second model, μ represents the expectation of the new Gaussian distribution, and Σ represents the variance of the new Gaussian distribution.

[0094] K is:

[0095] K=Σ1(Σ1+Σ2) -1

[0096] Finally, according to the fused Gaussian distribution μ and Σ, the width w, height h and rotation angle θ of the fused tilted frame are restored. The specific steps to restore the frame from the Gaussian distribution are as follows: First, the matrix eigenvalues ​​λ1, λ2 and eigenvector v are obtained by using the function eig() for finding eigenvalues ​​and eigenvectors:

[0097] λ1,λ2,v=eig(Σ)

[0098] According to the eigenvalue and confidence p, the major axis a and the minor axis b are obtained:

[0099]

[0100] The angle is obtained from the eigenvector value:

[0101] θ=arccos(v)

[0102] Finally, the coordinates of the four points of the frame are calculated by the tilt angle and the major and minor axes to generate a fused frame, and the target in the fused target frame is detected.

[0103] The present invention integrates different deep models, fully combines the respective advantages of different models, and improves the overall detection performance. In the detection result fusion stage, the present invention adopts the confidence-weighted Kalman filtering method to comprehensively consider the geometric features and confidence of the tilted target frame. Through this fusion method, not only the advantages of each model are retained, but also the incidence of false detection and missed detection can be effectively reduced, making the final detection results of tilted parts of power equipment more accurate and stable.

[0104] Embodiment 2

[0105] This embodiment provides a model integration system for detecting tilted components of power equipment.

[0106] A model integration system for detecting tilted components of power equipment, comprising:

[0107] The inclined target detection module is configured to: based on the power equipment image, respectively use at least two different inclined target detection models to obtain an inclined target frame of each model;

[0108] A confidence calibration module is configured as follows: calibrating the confidence of the tilted target frame detected by each model, converting the geometric features of the tilted target frame into a Gaussian distribution, and using a Kalman filter weighted by the calibrated confidence to fuse the Gaussian distribution of the corresponding tilted target frame to obtain a new Gaussian distribution;

[0109] The detection result fusion module is configured to: transform the new Gaussian distribution into a fused target frame, and output the fused target frame as the detection result.

[0110] Furthermore, the detection result fusion module is also configured to: use the center point of the tilted target frame as the expectation of the Gaussian distribution; calculate the rotation matrix of the tilt angle, and obtain the variance of the Gaussian distribution according to the rotation matrix and the length and height of the tilted target frame; and use the calibrated confidence-weighted Kalman filter to fuse the expectation and variance of each Gaussian distribution to obtain the expectation and variance of the new Gaussian distribution.

[0111] Furthermore, the Kalman filter weighted by the calibrated confidence factor is used to fuse the expectation and variance of each Gaussian distribution, and is expressed by the following formula:

[0112] G kf (μ,Σ)=p1G1(μ1,Σ1)p2G2(μ2,Σ2)

[0113] μ=μ1+K(μ2-μ1)

[0114] Σ=p1Σ1-Kp2Σ2

[0115] K=Σ1(Σ1+Σ2) -1

[0116] Among them, p1 represents the confidence of the tilted target box detected by the first model, μ1 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the first model, Σ1 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the first model, p2 represents the confidence of the tilted target box detected by the second model, μ2 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the second model, Σ2 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the second model, μ represents the expectation of the new Gaussian distribution, and Σ represents the variance of the new Gaussian distribution.

[0117] In some embodiments, the confidence calibration module is further configured to: design a calibrator based on temperature scaling, optimize the parameters of the calibrator using data training, and calibrate the confidence of each tilted target frame according to the trained calibrator.

[0118] In some embodiments, the geometric features include: a center point, length, width, height and tilt angle of the tilted target frame.

[0119] In some embodiments, the geometric features of the tilted target frame predicted by each model are converted into a Gaussian distribution, wherein only the tilted target frames whose confidence after calibration is greater than a set threshold are retained.

[0120] In some embodiments, the detection result fusion module is further configured to: calculate the matrix eigenvalues ​​and matrix eigenvectors based on the variance of the new Gaussian distribution; calculate the width and height of the fused target frame according to the matrix eigenvalues; calculate the inclination angle of the fused target frame according to the matrix eigenvector; obtain the fused target frame based on the center point of the inclined target frame, the width and height of the fused target frame and the inclination angle.

[0121] The present invention optimizes the design for the complex characteristics of tilted parts of power equipment and can handle detection tasks in a variety of complex environments. By combining the fusion strategy of geometric features and confidence, the system has higher adaptability and robustness when facing different scenarios.

[0122] Embodiment 3

[0123] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the model integration method for detecting tilted components of power equipment as described in the first embodiment above are implemented.

[0124] Embodiment 4

[0125] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the model integration method for detecting tilted components of power equipment as described in the first embodiment above are implemented.

[0126] Embodiment 5

[0127] This embodiment provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the above-mentioned model integration method for detecting tilted components of power equipment.

[0128] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0129] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0130] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0132] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A model integration method for detecting tilted components of power equipment, characterized in that: include: Based on the power equipment image, at least two different inclined target detection models are respectively used to obtain an inclined target frame of each model; The confidence of the tilted target frame detected by each model is calibrated, the geometric features of the tilted target frame are converted into Gaussian distribution, and the Gaussian distribution of the corresponding tilted target frame is fused using the calibrated confidence-weighted Kalman filter to obtain a new Gaussian distribution; The new Gaussian distribution is converted into a fused target frame, and the fused target frame is output as the detection result.

2. The model integration method for detecting tilted components of power equipment according to claim 1, characterized in that: The process of converting the geometric features of the tilted target frame into a Gaussian distribution includes: taking the center point of the tilted target frame as the expectation of the Gaussian distribution; calculating the rotation matrix of the tilt angle, obtaining the variance of the Gaussian distribution according to the rotation matrix and the length and height of the tilted target frame, and using a calibrated confidence-weighted Kalman filter to fuse the expectation and variance of each Gaussian distribution to obtain the expectation and variance of a new Gaussian distribution.

3. The model integration method for detecting tilted components of electric power equipment according to claim 2, characterized in that: The Kalman filter with confidence weighting after calibration is used to fuse the expectation and variance of each Gaussian distribution, which is expressed by the following formula: G kf (μ, ∑)=p1G1(μ1, ∑1)p2G2(μ2, ∑2) μ=μ1+K(μ2-μ1) ∑=p1∑1-Kp2∑2 K＝∑1(∑1+∑2) -1 Among them, p1 represents the confidence of the tilted target box detected by the first model, μ1 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the first model, ∑1 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the first model, p2 represents the confidence of the tilted target box detected by the second model, μ2 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the second model, Σ2 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the second model, μ represents the expectation of the new Gaussian distribution, and Σ represents the variance of the new Gaussian distribution.

4. The model integration method for detecting tilted components of power equipment according to claim 1, characterized in that: The process of calibrating the confidence of each tilted target frame includes designing a calibrator based on temperature scaling, optimizing parameters of the calibrator by using data training, and calibrating the confidence of each tilted target frame according to the trained calibrator; or, The geometric features include: the center point, length, width, height and tilt angle of the tilted target frame.

5. The model integration method for detecting tilted components of power equipment according to claim 1, characterized in that: The geometric features of the tilted target frame predicted by each model are converted into a Gaussian distribution, wherein only the tilted target frame whose confidence after calibration is greater than a set threshold is retained; or, The method for converting the new Gaussian distribution into a fused target frame includes: calculating matrix eigenvalues ​​and matrix eigenvectors based on the variance of the new Gaussian distribution; calculating the width and height of the fused target frame according to the matrix eigenvalues; calculating the inclination angle of the fused target frame according to the matrix eigenvector; and obtaining the fused target frame based on the center point of the inclined target frame, the width and height of the fused target frame, and the inclination angle.

6. A model integration system for detecting tilted components of power equipment, characterized in that: include: The inclined target detection module is configured to: based on the power equipment image, respectively use at least two different inclined target detection models to obtain an inclined target frame of each model; A confidence calibration module is configured as follows: calibrating the confidence of the tilted target frame detected by each model, converting the geometric features of the tilted target frame into a Gaussian distribution, and using a Kalman filter weighted by the calibrated confidence to fuse the Gaussian distribution of the corresponding tilted target frame to obtain a new Gaussian distribution; The detection result fusion module is configured to: transform the new Gaussian distribution into a fused target frame, and output the fused target frame as the detection result.

7. The model integration system for detecting tilted components of electric power equipment according to claim 6, characterized in that: The detection result fusion module is further configured to: take the center point of the tilted target frame as the expectation of the Gaussian distribution; calculate the rotation matrix of the tilt angle, obtain the variance of the Gaussian distribution according to the rotation matrix and the length and height of the tilted target frame, and use the calibrated confidence-weighted Kalman filter to fuse the expectation and variance of each Gaussian distribution to obtain the expectation and variance of the new Gaussian distribution; or, The Kalman filter with confidence weighting after calibration is used to fuse the expectation and variance of each Gaussian distribution, which is expressed by the following formula: G kf (μ, ∑)=p1G1(μ1, ∑1)p2G2(μ2, ∑2) μ=μ1+K(μ2-μ1) ∑=p1∑1-Kp2∑2 K＝∑1(∑1+∑2) -1 Among them, p1 represents the confidence of the tilted target box detected by the first model, μ1 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the first model, ∑1 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the first model, p2 represents the confidence of the tilted target box detected by the second model, μ2 represents the expectation of the Gaussian distribution corresponding to the tilted target box detected by the second model, ∑2 represents the variance of the Gaussian distribution corresponding to the tilted target box detected by the second model, μ represents the expectation of the new Gaussian distribution, and ∑ represents the variance of the new Gaussian distribution; or, The confidence calibration module is further configured to: design a calibrator based on temperature scaling, optimize the parameters of the calibrator by using data training, and calibrate the confidence of each tilted target frame according to the trained calibrator; or, The geometric features include: the center point, length, width, height and tilt angle of the tilted target frame; or, The geometric features of the tilted target frame predicted by each model are converted into a Gaussian distribution, wherein only the tilted target frame whose confidence after calibration is greater than a set threshold is retained; or, The detection result fusion module is also configured to: calculate the matrix eigenvalues ​​and matrix eigenvectors based on the variance of the new Gaussian distribution; calculate the width and height of the fused target frame according to the matrix eigenvalues; calculate the inclination angle of the fused target frame according to the matrix eigenvector; obtain the fused target frame based on the center point of the inclined target frame, the width and height of the fused target frame and the inclination angle.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the model integration method for detecting tilted components of electric equipment as described in any one of claims 1 to 5 are implemented.

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 program, the steps in the model integration method for detecting tilted components of electric equipment as described in any one of claims 1 to 5 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps in the model integration method for detecting tilted components of electric power equipment as described in any one of claims 1 to 5 are implemented.

Images