Anti-falling track defect detection algorithm based on improved YOLOv8

Abstract

The invention discloses an anti-falling track defect detection algorithm based on improved YOLOv8. The detection algorithm comprises the following steps: data preparation and preprocessing; constructing and training a model; a dynamic threshold adjustment mechanism; online learning and model updating; and integrating and applying the system. According to the method, high-precision detection and real-time adaptive adjustment of track defects are realized through dynamic threshold adjustment and an online learning and model updating mechanism, the detection precision is improved, and the false alarm rate and the missing report rate are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of anti-fall track defect detection, and in particular to an anti-fall track defect detection algorithm based on improved YOLOv8. Background Art

[0002] In the daily maintenance of transmission lines, the safety and reliability of the guided anti-fall rail is one of the key factors to ensure the normal operation of the transmission lines. The guided anti-fall rail is used to support and guide the transmission line equipment to ensure its stability and safety under various environmental conditions. Track defects, such as cracks, wear, deformation, etc., may cause the transmission equipment to derail or fall, thereby causing serious safety accidents.

[0003] The existing methods for detecting defects in anti-fall rails mainly rely on the following methods:

[0004] Manual detection: Manual detection relies on the experience and skills of maintenance personnel, and has problems such as strong subjectivity, low detection efficiency, high false alarm rate and missed alarm rate;

[0005] Ultrasonic testing: Ultrasonic testing technology can detect defects inside the track, but it is less effective in detecting surface defects;

[0006] Infrared detection: Infrared detection identifies defects by detecting temperature changes on the track surface, but it is greatly affected by factors such as ambient temperature and light, and is prone to false alarms;

[0007] Traditional image processing algorithms: Traditional image processing algorithms are widely used in guided anti-fall track defect detection. However, due to the complex and changeable track environment, these algorithms have limitations in detection accuracy and real-time performance, and are difficult to adapt to different track conditions and defect types.

[0008] In view of the above defects, it is necessary to design an anti-fall track defect detection algorithm based on improved YOLOv8 to solve the above problems. Summary of the invention

[0009] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0010] In view of the above-mentioned problems existing in the existing anti-fall track defect detection algorithm based on improved YOLOv8, the present invention is proposed.

[0011] Therefore, the purpose of the present invention is to provide an anti-fall track defect detection algorithm based on improved YOLOv8, which is suitable for solving the problem that these algorithms have limitations in detection accuracy and real-time performance due to the complex and changeable track environment, and are difficult to adapt to different track conditions and defect types.

[0012] In order to solve the above technical problems, the present invention provides the following technical solutions: an anti-fall track defect detection algorithm based on improved YOLOv8, the detection algorithm comprises the following steps:

[0013] Data preparation and preprocessing;

[0014] Model building and training;

[0015] Dynamic threshold adjustment mechanism;

[0016] Online learning and model updating;

[0017] System integration and application.

[0018] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the dynamic threshold calculation formula is as follows:

[0019]

[0020] Wherein, T is the dynamically calculated detection threshold, which is used to adjust the rail defect detection algorithm to improve the detection accuracy, X is the characteristic vector of the rail defect, Y is the environmental factor vector, Z is the real-time feedback data vector, a, β, γ, λ, μ are all weight adjustment parameters, which are used to adjust the weights of each part in the formula to adapt to different defect detection requirements and environments;

[0021] The value range of T is a positive real number, which represents a dynamic detection threshold adapted to different track conditions. This value is used to adjust the sensitivity of the detection algorithm and optimize the detection effect.

[0022] Exponential function exp(-β.||X|| 2 ) is used to indicate the influence of defect characteristics;

[0023] The logarithmic function lop(1+||Y||) is used to represent the influence of environmental factors;

[0024] The linear combination λ.||X||+μ.||Z|| is used to integrate the influence of defect characteristics and real-time feedback data.

[0025] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the preliminary threshold calculation formula of the feature vector and environmental factors is as follows:

[0026] T1=a.exp(-β.||X|| 2 )+γ.log(1+||Y||)

[0027] The preliminary threshold T1 mainly considers the influence of defect characteristics and environmental factors;

[0028] If T1>T y1 , then it is considered that the comprehensive judgment defect exists;

[0029] If T1≤T y1 , then the comprehensive judgment is considered to be flawless.

[0030] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the comprehensive calculation formula of the preliminary threshold and the real-time feedback data is as follows:

[0031]

[0032] T2 is the threshold of the comprehensive preliminary threshold and the real-time feedback data. On the basis of the preliminary judgment, the real-time feedback data is combined to further judge the existence of track defects;

[0033] If T2>T y2 , then it is considered that the comprehensive judgment defect exists;

[0034] If T2≤T y2 , then it is considered that the comprehensive judgment defect is not defective.

[0035] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the final calculation formula of the comprehensive threshold and dynamic adjustment is as follows:

[0036]

[0037] Among them, F is the newly added real-time feedback weight factor, which represents the influence of real-time feedback in the dynamic adjustment process, G is the environment change vector, and δ is the adjustment parameter, which controls the amplitude of dynamic adjustment.

[0038] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, wherein: T3 is the final calculated value of the comprehensive threshold and dynamic adjustment, which is used to determine the existence of track defects and dynamically adjust the detection threshold;

[0039] If T3>T y3 , then it is finally judged that the defect exists;

[0040] If T3≤T y3 , then the final judgment is that there is no defect.

[0041] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the value range of T1, T2 and T3 is a positive real number, representing the detection thresholds at different stages, and each threshold is used for detection judgment at different stages to ensure high-precision detection of track defects and real-time adaptive adjustment.

[0042] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the online learning formula is as follows:

[0043]

[0044] Among them, θ t is the model parameter vector of the tth iteration, θ t+1 is the model parameter vector of the t+1th iteration, η is the learning rate, Represents the loss function L with respect to the current model parameter θ t The gradient reflects the error of the current model, D t represents the training data set used in the tth iteration, and λ is the regularization parameter used to control the complexity of the model.

[0045] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the dynamic threshold adjustment mechanism starts from three aspects: defect type and size classification, dynamic threshold calculation and threshold adjustment.

[0046] As a preferred solution of the anti-fall track defect detection algorithm based on improved YOLOv8 described in the present invention, the online learning and model updating include:

[0047] Collect real-time feedback information on test results;

[0048] Continuously update and optimize the detection model based on factual feedback information;

[0049] Combined with the feedback data, the model is retrained.

[0050] The beneficial effects of the present invention are as follows: the detection threshold is automatically adjusted according to the type and size of the detected track defects to improve the detection accuracy, reduce the false alarm rate and the missed alarm rate, and the detection model is continuously updated and optimized through real-time feedback to adapt to changes in different environments and track conditions, thereby improving the real-time and accuracy of detection. Through dynamic threshold adjustment and online learning and model update mechanisms, high-precision detection and real-time adaptive adjustment of track defects are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0052] Figure 1 This is a schematic diagram of the overall steps of the anti-fall track defect detection algorithm based on improved YOLOv8 proposed in the present invention. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0056] Secondly, the present invention is described in detail with reference to the schematic diagram. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0057] Embodiment 1

[0058] Reference Figure 1 , is an embodiment of the present invention, and provides an anti-fall track defect detection algorithm based on improved YOLOv8, the detection algorithm includes the following steps:

[0059] Data preparation and preprocessing;

[0060] Data collection: Collect image data containing various rail defects to ensure that the dataset covers defects of different types and sizes;

[0061] Data annotation: Annotate the collected images to mark the type and location of defects;

[0062] Data enhancement: Use image data enhancement techniques, such as rotation, cropping, flipping, adding noise, etc., to increase the diversity of the data set and improve the generalization ability of the model;

[0063] Model building and training;

[0064] Improved YOLOv8 model: Based on the YOLOv8 model, multi-scale feature fusion and attention mechanism are introduced to enhance the feature extraction capability of the model;

[0065] Multi-scale feature fusion: In the feature extraction stage of YOLOv8, a multi-scale feature fusion module is added to enable the model to better detect track defects of different sizes and shapes;

[0066] Attention mechanism: Introduce self-attention mechanism and channel attention mechanism in the feature extraction module to enhance attention to track defect areas;

[0067] Dynamic threshold adjustment mechanism;

[0068] The dynamic threshold adjustment mechanism starts from three aspects: defect type and size classification, dynamic threshold calculation and threshold adjustment;

[0069] Online learning and model updating;

[0070] Online learning and model updating include:

[0071] Collect real-time feedback information on test results;

[0072] Continuously update and optimize the detection model based on factual feedback information;

[0073] Retrain the model based on feedback data;

[0074] System integration and application.

[0075] The dynamic threshold calculation formula is as follows:

[0076]

[0077] Wherein, T is the dynamically calculated detection threshold, which is used to adjust the rail defect detection algorithm to improve the detection accuracy, X is the characteristic vector of the rail defect, Y is the environmental factor vector, Z is the real-time feedback data vector, a, β, γ, λ, μ are all weight adjustment parameters, which are used to adjust the weights of each part in the formula to adapt to different defect detection requirements and environments;

[0078] The value range of T is a positive real number, which represents a dynamic detection threshold adapted to different track conditions. This value is used to adjust the sensitivity of the detection algorithm and optimize the detection effect.

[0079] Exponential function exp(-β.||X|| 2) is used to indicate the influence of defect characteristics;

[0080] The logarithmic function lop(1+||Y||) is used to represent the influence of environmental factors;

[0081] The linear combination λ.||X||+μ.||Z|| is used to integrate the influence of defect characteristics and real-time feedback data.

[0082] The preliminary threshold calculation formula for feature vectors and environmental factors is as follows:

[0083] T1=a.exp(-β.||X|| 2 )+γ.log(1+||Y||)

[0084] The preliminary threshold T1 mainly considers the influence of defect characteristics and environmental factors;

[0085] If T1>T y1 , then it is considered that the comprehensive judgment defect exists;

[0086] If T1≤T y1 , then the comprehensive judgment is considered to be flawless.

[0087] The comprehensive calculation formula of the preliminary threshold and real-time feedback data is as follows:

[0088]

[0089] T2 is the threshold of the comprehensive preliminary threshold and the real-time feedback data. On the basis of the preliminary judgment, the real-time feedback data is combined to further judge the existence of track defects;

[0090] If T2>T y2 , then it is considered that the comprehensive judgment defect exists;

[0091] If T2≤T y2 , then it is considered that the comprehensive judgment defect is not defective.

[0092] The final calculation formula for the comprehensive threshold and dynamic adjustment is as follows:

[0093]

[0094] Among them, F is the newly added real-time feedback weight factor, which represents the influence of real-time feedback in the dynamic adjustment process, G is the environment change vector, and δ is the adjustment parameter, which controls the amplitude of dynamic adjustment.

[0095] T3 is the final calculated value of the comprehensive threshold and dynamic adjustment, which is used to determine the existence of track defects and dynamically adjust the detection threshold;

[0096] If T3>T y3 , then it is finally judged that the defect exists;

[0097] If T3≤T y3 , then the final judgment is that there is no defect.

[0098] The value ranges of T1, T2 and T3 are positive real numbers, representing the detection thresholds at different stages. Each threshold is used for detection judgment at different stages to ensure high-precision detection and real-time adaptive adjustment of track defects. In the present invention, T y1 , T y2 and T y3 These are the threshold judgment values ​​for the three stages, which are determined based on historical data and statistical analysis and can be adjusted according to actual usage.

[0099] The online learning formula is as follows:

[0100]

[0101] Among them, θ t is the model parameter vector of the tth iteration, θ t+1 is the model parameter vector of the t+1th iteration, η is the learning rate, Represents the loss function L with respect to the current model parameter θ t The gradient reflects the error of the current model, D t represents the training data set used in the tth iteration, and λ is the regularization parameter used to control the complexity of the model.

[0102] Embodiment 2

[0103] The difference from the first embodiment is that the improved YOLOv8 model and the dynamic threshold adjustment mechanism are integrated into the actual track monitoring system for field testing and verification.

[0104] Defect Detection Comparison Table

[0105]

[0106]

[0107] By comparing the data samples in the table, we can see that the present invention has significant advantages in many aspects:

[0108] Processing time: The processing time of the algorithm of the present invention on different defect types and comprehensive data sets is lower than that of the prior art, showing significant time efficiency. Specifically, the processing time of the prior art is 11.0 seconds, while the processing time of the algorithm of the present invention is 8.5 seconds, which is a reduction of 22.7%;

[0109] Detection accuracy: The detection accuracy of the algorithm of the present invention for various defects is higher than that of the existing technology. For example, the crack detection accuracy is increased from 84.3% to 91.2%, an increase of 6.9 percentage points, which shows that the algorithm of the present invention has significantly improved the detection accuracy;

[0110] Missed detection rate: The algorithm of the present invention significantly reduces missed detection. For example, for pit type, the missed detection rate of the prior art is 14.3%, while that of the algorithm of the present invention is 6.8%, which is reduced by 52.4%. This shows that the present invention can detect various defects more comprehensively and reduce missed detection.

[0111] Detection false alarm rate: The algorithm of the present invention also significantly reduces the false alarm rate. For example, for the shedding type, the false alarm rate of the prior art is 12.5%, while that of the algorithm of the present invention is 5.1%, a decrease of 59.2%. This shows that the present invention performs well in reducing false alarms and improves the reliability of detection.

[0112] The detection threshold is automatically adjusted according to the type and size of the detected track defects to improve detection accuracy and reduce false alarm and missed alarm rates. The detection model is continuously updated and optimized through real-time feedback to adapt to changes in different environments and track conditions, improve the real-time and accuracy of detection, and achieve high-precision detection and real-time adaptive adjustment of track defects through dynamic threshold adjustment and online learning and model update mechanisms.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. The anti-fall track defect detection algorithm based on improved YOLOv8 is characterized by: The detection algorithm comprises the following steps: Data preparation and preprocessing; Model building and training; Dynamic threshold adjustment mechanism; Online learning and model updating; System integration and application.

2. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 1, characterized in that: The dynamic threshold calculation formula is as follows: Wherein, T is the dynamically calculated detection threshold, which is used to adjust the rail defect detection algorithm to improve the detection accuracy, X is the characteristic vector of the rail defect, Y is the environmental factor vector, Z is the real-time feedback data vector, a, β, γ, λ, μ are all weight adjustment parameters, which are used to adjust the weights of each part in the formula to adapt to different defect detection requirements and environments; The value range of T is a positive real number, which represents a dynamic detection threshold adapted to different track conditions. This value is used to adjust the sensitivity of the detection algorithm and optimize the detection effect. Exponential function exp(-β.||X|| 2 ) is used to indicate the influence of defect characteristics; The logarithmic function lop(1+||Y||) is used to represent the influence of environmental factors; The linear combination λ.||X||+μ.||Z|| is used to integrate the influence of defect characteristics and real-time feedback data.

3. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 2, characterized in that: The preliminary threshold calculation formula of the feature vector and environmental factors is as follows: T1=a.exp(-β.||X|| 2 )+γ.log(1+||Y||) The preliminary threshold T1 mainly considers the influence of defect characteristics and environmental factors; If T1>T y1 , then it is considered that the comprehensive judgment defect exists; If T1≤T y1 , then the comprehensive judgment is considered to be flawless.

4. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 3 is characterized in that: The comprehensive calculation formula of the preliminary threshold and real-time feedback data is as follows: T2 is the threshold of the comprehensive preliminary threshold and the real-time feedback data. On the basis of the preliminary judgment, the real-time feedback data is combined to further judge the existence of track defects; If T2>T y2 , then it is considered that the comprehensive judgment defect exists; If T2≤T y2 , then it is considered that the comprehensive judgment defect is not defective.

5. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 4 is characterized in that: The final calculation formula of the comprehensive threshold and dynamic adjustment is as follows: Among them, F is the newly added real-time feedback weight factor, which represents the influence of real-time feedback in the dynamic adjustment process, G is the environment change vector, and δ is the adjustment parameter, which controls the amplitude of dynamic adjustment.

6. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 5, characterized in that: T3 is the final calculated value of the comprehensive threshold and dynamic adjustment, which is used to determine the existence of track defects and dynamically adjust the detection threshold; If T3>T y3 , then it is finally judged that the defect exists; If T3≤T y3 , then the final judgment is that there is no defect.

7. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 5, characterized in that: The value range of T1, T2 and T3 is a positive real number, representing the detection threshold at different stages. Each threshold is used for detection judgment at different stages to ensure high-precision detection of track defects and real-time adaptive adjustment.

8. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 1, characterized in that: The online learning formula is as follows: Among them, θ t is the model parameter vector of the tth iteration, θ t+1 is the model parameter vector of the t+1th iteration, η is the learning rate, Represents the loss function L with respect to the current model parameter θ t The gradient reflects the error of the current model, D t represents the training data set used in the tth iteration, and λ is the regularization parameter used to control the complexity of the model.

9. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 1, characterized in that: The dynamic threshold adjustment mechanism starts from three aspects: defect type and size classification, dynamic threshold calculation and threshold adjustment.

10. The anti-fall track defect detection algorithm based on improved YOLOv8 according to claim 1, characterized in that: The online learning and model updating include: Collect real-time feedback information on test results; Continuously update and optimize the detection model based on factual feedback information; Combined with the feedback data, the model is retrained.

Images