Steel rail damage judgment method and device based on large model

Abstract

The invention relates to a steel rail damage judgment method and device based on a large model, and belongs to the technical field of artificial intelligence. The method comprises the steps that firstly, a steel rail damage result is judged according to ultrasonic signal features, and then further judgment is conducted according to a steel rail damage judgment large model trained through historical data; and finally, clutter detection correction is carried out on clutter results in large model judgment results, and a final steel rail damage judgment result is obtained. According to the method and the device, the steel rail damage judgment efficiency and accuracy can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to a method and device for judging rail damage based on a large model. Background Art

[0002] With the expansion of the high-speed rail network and the increase in operating speed, the accurate identification of rail damage has become crucial. Traditional manual inspections and simple image processing technologies have problems such as low efficiency, limited accuracy, and inconsistent injury judgment criteria when faced with complex injury types such as nuclear injuries, bolt hole injuries, and abnormal rail bottoms. Summary of the Invention

[0003] In view of the above deficiencies of the prior art, the purpose of the invention is to provide a method, device, and storage medium for judging rail damage based on a large model. Using the method of the present invention, the injury judgment criteria can be unified, and the accuracy and efficiency of rail injury judgment can be improved.

[0004] In the first aspect of the present invention, a method for judging rail damage based on a large model is proposed, including:

[0005] Judging the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result;

[0006] Training a preset large model according to the historical rail injury judgment data to obtain a rail damage judgment large model, and inputting the first judgment result into the rail damage judgment large model to obtain a second judgment result;

[0007] Correcting the second judgment result to obtain a final judgment result.

[0008] Further, judging the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result includes:

[0009] Determining a rectangular injury judgment detection area and moving it at a fixed step size S;

[0010] Extracting the picture of the injury judgment detection area at each step of movement, and obtaining an initial injury judgment result judged according to the ultrasonic signal characteristics, where the initial injury judgment result includes no injury, nuclear injury, bolt hole injury, and abnormal rail bottom; wherein, the height of the rectangular injury judgment detection area is H and the width is W;

[0011] Obtaining an image of the damaged area;

[0012] The obtaining of the image of the damaged area includes:

[0013] When the initial injury judgment result is no injury, continue to detect the next detection area;

[0014] When the initial injury judgment result is a nuclear injury, determining the image of the damaged area according to the first method;

[0015] When the initial injury judgment result is a screw hole damage, determine the damage area image according to the second method;

[0016] When the initial injury judgment result is an abnormal rail bottom, determine the damage area image according to the third method;

[0017] Wherein, the first method, the second method and the third method are not completely the same.

[0018] In a second aspect of the present invention, a device for implementing a method for judging rail damage based on a large model is proposed, including:

[0019] An initial judgment module, configured to judge the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result;

[0020] A large model injury judgment module, configured to train a preset large model according to the historical rail injury judgment data to obtain a rail damage judgment large model, and input the first judgment result into the rail damage judgment large model to obtain a second judgment result;

[0021] A correction module, configured to correct the injury judgment result of the second judgment result to obtain a final injury judgment result.

[0022] In a third aspect of the present invention, a device for implementing a method for judging rail damage based on a large model is proposed, including: a memory, a processor and a user interface;

[0023] The memory is used to store a computer program;

[0024] The user interface is used to interact with the user;

[0025] The processor is used to read the computer program in the memory. When the processor executes the computer program, the above-mentioned method for judging rail damage based on a large model is implemented.

[0026] In a fourth aspect of the present invention, a processor-readable storage medium is proposed. The processor-readable storage medium stores a computer program, and when the processor executes the computer program, the above-mentioned method for judging rail damage based on a large model is implemented.

[0027] The beneficial effects of the present invention are as follows:

[0028] By combining deep learning with computer vision, rail damage judgment based on ultrasonic waves was carried out respectively, then rail damage judgment based on large models was carried out, and finally the judgment results were corrected, so as to accurately identify and classify various damages, unify the criteria for damage judgment, and improve the accuracy and efficiency of damage judgment. Using the method of the present invention, the development trend of rail damage can also be predicted, providing data support for maintenance decision-making, thereby improving railway safety and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings are only for the purpose of illustrating specific embodiments and are not considered as a limitation of the present invention. Throughout the drawings, the same reference signs denote the same components. Obviously, the drawings in the following description are only some embodiments described in the embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0030] Figure 1 It is a flowchart of a method for judging rail damage based on a large model according to an embodiment of the present invention;

[0031] Figure 2 It is a schematic diagram of the initial judgment process of rail damage according to an embodiment of the present invention;

[0032] Figure 3 It is a schematic diagram of the clutter correction process according to an embodiment of the present invention;

[0033] Figure 4 It is a schematic diagram of a device for judging rail damage based on a large model according to an embodiment of the present invention;

[0034] Figure 5 It is a schematic diagram of another device for judging rail damage based on a large model according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0036] In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts disclosed in the present invention.

[0037] In the description of the present invention, it should be noted that unless otherwise clearly specified and defined, the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation on the present invention. In addition, the terms "first", "second", "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance. The terms "installation", "connection", "connection" 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 directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0038] Damage judgment, which is the abbreviation of judging the damage situation, is used to judge the damage situation of the rail.

[0039] Damage result, that is, the result of rail damage, includes no damage, nuclear damage, bolt hole damage and abnormal rail bottom.

[0040] Damage type, that is, the type of rail damage, includes nuclear damage, bolt hole damage and abnormal rail bottom.

[0041] Nuclear damage refers to the metal fatigue damage generated on the contact surface of the rail head due to long-term wheel rolling. It is manifested as spalling, cracks or depressions on the rail surface, usually occurring at the center position of the rail head and distributed like a "core", so it is called nuclear damage.

[0042] Bolt hole damage refers to the cracks or damage around the bolt holes used to fix the fishplate connection on the track. This kind of damage is mainly caused by metal fatigue due to repeated stress around the bolt holes, or due to improper bolt installation or poor fastening.

[0043] Abnormal rail bottom refers to the damage or defect that appears at the bottom of the rail. It mainly includes rail bottom cracks, corrosion, wear, etc. The rail bottom bears the gravity and bending stress of the rail. Once an abnormality occurs, it will seriously affect the overall strength and safety of the track.

[0044] Damage judgment detection area, that is, the area for detecting whether the rail is damaged.

[0045] Damage area image, that is, when it is judged that there is damage, the image of the damage judgment detection area is obtained.

[0046] MiniCPM is a large language model.

[0047] In the present invention, unless otherwise specified, the unit of length is millimeter.

[0048] Here, the exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0049] A method for judging rail damage based on a large model provided by the present invention, specifically, as Figure 1 shown, the method includes steps S101 to S103:

[0050] S101. Judge the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result;

[0051] S102. Train a preset large model according to the historical rail damage judgment data to obtain a rail damage judgment large model, and input the first judgment result into the rail damage judgment large model to obtain a second judgment result;

[0052] S103. Correct the judgment result of the second judgment result to obtain a final judgment result.

[0053] In the present invention, the step of S101 is called initial damage judgment. The damage judgment detection area is a window of a rectangular picture, which moves from left to right along the width of the rail image. The height of the window is H, the width of the window is W, and it moves according to a fixed step length S. Each time it moves one step length, the picture of the damage judgment detection area is extracted, and the rail damage result judged according to the ultrasonic signal characteristics is obtained, which is called the initial damage judgment result. If the initial damage judgment result shows that there is damage, the image of the damage area is obtained for further judgment by the subsequent large model. Specifically, as Figure 2 shown, it includes:

[0054] S201. Determine the rectangular damage judgment detection area and move it according to the fixed step length S;

[0055] S202. Extract the picture of the damage judgment detection area every time it moves one step, and obtain the initial damage judgment result judged according to the ultrasonic signal characteristics. The initial damage judgment result includes no damage, nuclear damage, bolt hole damage and abnormal rail bottom.

[0056] S203. Obtain the image of the damage area according to the initial damage judgment result.

[0057] In this step, in the present invention, according to different types of initial injury judgment results, different methods are adopted to obtain the image of the damaged area. Among them, the first method, the second method, and the third method for obtaining the image of the damaged area are not completely the same.

[0058] S203-1: If the initial injury judgment result is no injury, there is no need to obtain the image of the damaged area, and continue to detect the next detection area. That is, move a step length S and detect the next area.

[0059] S203-2: If the initial injury judgment result is a nuclear injury, determine the image of the damaged area according to the first method. Specifically, it can be:

[0060] Obtain the height h corresponding to the nuclear injury i , and obtain the damaged area O that needs to be detected i :

[0061]

[0062] Where is the transverse intercept length corresponding to the nuclear injury, dh1 i = max(0, h i - Δh1) is the lower bound of the intercept corresponding to the nuclear injury, uh1 i = min(H, h i + Δh1) is the upper bound of the intercept corresponding to the nuclear injury, Δh1 is the longitudinal intercept length corresponding to the nuclear injury, w1 and Δh1 are multiples of 10, and the unit is millimeter;

[0063] When , add a black area with a width of W and a height of 2Δh1 - uh1 i + dh1 i above the intercepted area to form the image of the damaged area;

[0064] When , add a black area with a width of W and a height of 2Δh1 - uh1 i + dh1 i below the intercepted area to form the image of the damaged area.

[0065] Among them, i is the number of the rectangular injury judgment detection area, and the value starts from 0, and is incremented by 1 every time a step length is moved. For example, at the beginning of the detection, the number of the rectangular injury judgment detection area is 0. After the detection is completed and a step length S is moved, the number of the rectangular injury judgment detection area is 1, and so on.

[0066] S203-3: If the initial injury judgment result is a screw hole damage, determine the image of the damaged area according to the second method. Specifically, it can be:

[0067] Obtain the height h corresponding to the screw hole damage i, the damaged area O to be detected is obtained i ;

[0068]

[0069] Among them is the horizontal intercept length corresponding to the screw hole damage, dh2 i = max(0, h i - Δh2) is the lower bound of the intercept corresponding to the screw hole damage, uh2 i = min(H, h i + Δh2) is the upper bound of the intercept corresponding to the screw hole damage, Δh2 is the vertical intercept length corresponding to the screw hole damage, and w2, Δh2 are multiples of 10;

[0070] When , a black area with a width of W and a height of 2Δh2 - uh2 i + dh2 i is added above the intercepted area to form the damaged area image;

[0071] When , a black area with a width of W and a height of 2Δh2 - uh2 i + dh2 i is added below the intercepted area to form the damaged area image.

[0072] S203-3: If the initial injury judgment result is abnormal at the rail bottom, the damaged area image is determined according to the third method. Specifically, it can be:

[0073] Obtain the height h corresponding to the abnormal rail bottom i , and the damaged area O to be detected is obtained i ;

[0074]

[0075] Among them is the horizontal intercept length corresponding to the abnormal rail bottom, dh3 i = max(0, h i - Δh3) is the lower bound of the intercept corresponding to the abnormal rail bottom, uh3 i = min(H, h i + Δh3) is the upper bound of the intercept corresponding to the abnormal rail bottom, Δh3 is the vertical intercept length corresponding to the abnormal rail bottom, and w3, Δh3 are multiples of 10;

[0076] When , a black area with a width of W and a height of 2Δh3 - uh3 i + dh3 i is added above the intercepted area to form the damaged area image;

[0077] When it is, a black area with a width of W and a height of 2Δh3 - uh3 i + dh3 i is added below the intercepted area to form an image of the group damage area.

[0078] Preferably, in S102, the preset large model is trained according to the historical rail injury judgment data to obtain a rail damage judgment large model. Inputting the first judgment result into the rail damage judgment large model to obtain the second judgment result includes:

[0079] S102-1: Model training: Establish an injury judgment large model based on the MiniCPM model. The training data set includes injury images and an injury description data set. The injury description data set includes injury types and whether it is clutter. Use the training data set to train the MiniCPM model, evaluate the performance of the model, and optimize the model according to the evaluation results to obtain the injury judgment large model.

[0080] S102-2: Model injury judgment: The preset large model is trained according to the historical rail injury judgment data to obtain a rail damage judgment large model. Input the first judgment result into the rail damage judgment large model to obtain the second judgment result. That is, input the injury area image and the initial injury judgment result obtained in step 101 into the model to obtain the second judgment result. The second judgment result includes whether the injury corresponding to the injury area image is clutter.

[0081] That is to say, the second judgment result includes a specific injury type or clutter, that is, the second judgment result is one of the following: no injury, nuclear injury, bolt hole injury, abnormal rail bottom or clutter;

[0082] When the second judgment result is clutter, perform clutter detection and correction to obtain the final injury judgment result.

[0083] Preferably, in S103, correcting the second judgment result to obtain the final injury judgment result may include:

[0084] When the second judgment result is clutter, perform clutter detection and correction, specifically as Figure 3 shown, including S301, S302 and S303:

[0085] S301. Injury feature extraction.

[0086] Convert the injury area image into a grayscale image. First, equally divide the width of the area into L parts, and then equally divide the height of the area into L parts to obtain L*L rectangular areas of the same size, where L is an integer greater than or equal to 5. For example, it can be to first equally divide the width of the area into 10 parts, and then equally divide the height of the area into 10 parts to obtain 100 rectangular areas of the same size;

[0087] The average value of the pixels in each rectangular area is calculated as g j,k , where j is the serial number of the rectangular area counted from left to right horizontally, and k is the serial number of the rectangular area counted from top to bottom vertically, obtaining the feature set {g j,k} of the damaged area, and the values of j and k are both integers from 0 to L - 1.

[0088] Construct a feature matrix where ts1 is the set first judgment threshold, i is the serial number of the damaged area image and also the number of the rectangular damage detection area.

[0089] For example, when L = 10, the feature matrix of the i-th damaged area image is:

[0090]

[0091] S302. Damage clustering.

[0092] Cluster the feature matrix to obtain m clusters, and extract the clustering center of each cluster and denote it as q = 1,..., m, dp j,k is the corresponding element of the center matrix, and the values of j and k are both integers from 0 to L - 1. For example, when L = 10, the clustering center of each cluster is denoted as:

[0093]

[0094] Compare the feature matrix D i with the clustering center ΔD q of each cluster to calculate the similarity to obtain the classification feature score gd q :

[0096]

[0097] For example, when L = 10, the expression of gd q is:

[0098]

[0099] Select the cluster with the highest score gd q as the classification of the detection area;

[0100] S303. Damage correction;

[0101] Obtain the historical verification times gt q and the historical misjudgment number ge q of the damaged area image;

[0102] When When obtaining the historical injury judgment record n of the detection area i and the historical clutter record n1 i When it is determined that the detection area image is clutter, and the detection area image is added to the label of this cluster for storage; when if manual review is performed and the review result is clutter, the historical verification times gt q is incremented by 1, and the historical misjudgment count ge q remains unchanged. If the review result is damage, the historical verification times gt q is incremented by 1, and the historical misjudgment count ge q is incremented by 1; where the label of this cluster includes the damage type; where a cluster is a set of detection areas of a specific damage type, and the specific damage types include nuclear damage, screw hole damage, or rail bottom abnormality. For example, all detection areas with the damage type of nuclear damage are used as the first cluster, all detection areas with the damage type of screw hole damage are used as the second cluster, and all detection areas with the damage type of rail bottom abnormality are used as the third cluster.

[0103] When it is sent to the designated person for manual review. If the review result is clutter, the historical verification times gt q is incremented by 1, and the historical misjudgment count ge q remains unchanged. If the review result is damage, the historical verification times gt q is incremented by 1, and the historical misjudgment count ge q is incremented by 1;

[0104] where ts2 is the set second judgment threshold, and ts3 is the set third judgment threshold;

[0105] After clutter correction through S301, S302, and S303, the final rail damage judgment result is obtained, including nuclear damage, screw hole damage, rail bottom abnormality, or clutter.

[0106] Using the method of the present invention, first, the rail damage result is judged according to the ultrasonic signal characteristics, then it is further judged according to the rail damage judgment large model trained with historical data, and finally, the clutter detection and correction are performed on the results judged as clutter in the results judged by the large model to obtain the final rail damage judgment result. The method of the present invention combines deep learning and computer vision, thereby accurately identifying and classifying various types of damage, unifying the criteria for damage judgment, and improving the accuracy and efficiency of damage judgment. Using the method of the present invention, the development trend of rail damage can also be predicted, providing data support for maintenance decision-making, thereby improving railway safety and maintenance efficiency.

[0107] Another specific embodiment of the present invention discloses a rail damage judgment device based on a large model, as Figure 4 shown, including.

[0108] The preliminary judgment module 401 is configured to judge the damage of the rail according to the ultrasonic signal characteristics, and obtain a first judgment result;

[0109] The large model damage judgment module 402 is configured to train a preset large model according to the historical rail damage judgment data to obtain a rail damage judgment large model, and input the first judgment result into the rail damage judgment large model to obtain a second judgment result;

[0110] The calibration module 403 is configured to calibrate the damage judgment result of the second judgment result to obtain a final damage judgment result.

[0111] Preferably, the preliminary judgment module 401 is further configured to:

[0112] Determine a rectangular damage judgment detection area and move it at a fixed step size S;

[0113] Extract the picture of the damage judgment detection area for each step of movement, and obtain an initial damage judgment result judged according to the ultrasonic signal characteristics. The initial damage judgment result includes no damage, nuclear damage, screw hole damage and abnormal rail bottom; wherein, the height of the rectangular damage judgment detection area is H and the width is W;

[0114] Obtain an image of the damaged area according to the initial damage judgment result;

[0115] The obtaining of the damaged area image according to the initial damage judgment result includes:

[0116] When the initial damage judgment result is no damage, continue to detect the next detection area;

[0117] When the initial damage judgment result is nuclear damage, determine the damaged area image according to the first method;

[0118] When the initial damage judgment result is screw hole damage, determine the damaged area image according to the second method;

[0119] When the initial damage judgment result is abnormal rail bottom, determine the damaged area image according to the third method;

[0120] Among them, the first method, the second method and the third method are not completely the same.

[0121] Preferably, the determining of the damaged area image according to the first method includes:

[0122] Obtain the height h corresponding to the nuclear damage i , and obtain the damaged area O that needs to be detected i :

[0123]

[0124] Among them is the transverse intercept length corresponding to the nuclear damage, dh1 i = max(0, h i - Δh1) is the lower bound of the intercept corresponding to the nuclear damage, uh1 i = min(H, h i + Δh1) is the upper bound of the intercept corresponding to the nuclear damage, Δh1 is the longitudinal intercept length corresponding to the nuclear damage, w1, Δh1 are multiples of 10;

[0125] When , a black area with a width of W and a height of 2Δh1 - uh1 i + dh1 i is added above the intercepted area to form the damaged area image;

[0126] When , a black area with a width of W and a height of 2Δh1 - uh1 i + dh1 i is added below the intercepted area to form the damaged area image;

[0127] Among them, i is the number of the rectangular damage detection area.

[0128] Preferably, determining the damaged area image according to the second method includes:

[0129] Obtain the height h corresponding to the screw hole damage i , and obtain the damaged area O that needs to be detected i ;

[0130]

[0131] Among them is the transverse intercept length corresponding to the screw hole damage, dh2 i = max(0, h i - Δh2) is the lower bound of the intercept corresponding to the screw hole damage, uh2 i = min(H, h i + Δh2) is the upper bound of the intercept corresponding to the screw hole damage, Δh2 is the longitudinal intercept length corresponding to the screw hole damage, w2, Δh2 are multiples of 10;

[0132] When , a black area with a width of W and a height of 2Δh2 - uh2 i + dh2 i is added above the intercepted area to form the damaged area image;

[0133] When , a black area with a width of W and a height of 2Δh2 - uh2 i + dh2 iThe black areas form an image of the damage area;

[0134] where i is the number of the rectangular damage detection area.

[0135] Preferably, determining the damage area image according to the third method includes:

[0136] Obtaining the height h corresponding to the bottom anomaly of the rail i , obtaining the damage area O to be detected i ;

[0137]

[0138] where is the horizontal intercept length corresponding to the bottom anomaly of the rail, dh3 i = max(0, h i - Δh3) is the lower bound of the intercept corresponding to the bottom anomaly of the rail, uh3 i = min(H, h i + Δh3) is the upper bound of the intercept corresponding to the bottom anomaly of the rail, Δh3 is the longitudinal intercept length corresponding to the bottom anomaly of the rail, and w3, Δh3 are multiples of 10;

[0139] When , add a black area with a width of W and a height of 2Δh3 - uh3 i + dh3 i above the intercepted area to form an image of the damage area;

[0140] When , add a black area with a width of W and a height of 2Δh3 - uh3 i + dh3 i below the intercepted area to form an image of the damage area;

[0141] where i is the number of the rectangular damage detection area.

[0142] Preferably, the calibration module 403 is further configured to: when the second judgment result is clutter, perform clutter detection and calibration to obtain the final damage judgment result.

[0143] Preferably, the performing clutter detection and calibration includes:

[0144] Damage feature extraction, damage clustering, and damage calibration;

[0145] The damage feature extraction includes:

[0146] Convert the damage area image into a grayscale image, first equally divide it into L parts according to the width of the area, and then equally divide it into L parts according to the height of the area to obtain L*L rectangular areas of the same size, where L is an integer greater than or equal to 10;

[0147] The average value of pixels in each rectangular area is calculated as g j,k , where j is the serial number of the rectangular area counted from left to right horizontally, and k is the serial number of the rectangular area counted from top to bottom vertically, obtaining the feature set {g j,k}} of the damage area, and the values of j and k are both integers from 0 to L-1;

[0148] Build a feature matrix where ts1 is the set first judgment threshold;

[0149] The damage clustering includes:

[0150] Cluster the feature matrix to obtain m clusters, and extract the clustering center of each cluster and denote it as q = 1,..., m, dp j,k is the corresponding element of the center matrix;

[0151] Compare the feature matrix D i with the clustering center ΔD of each cluster q to calculate the similarity to obtain the classification feature score gd q :

[0152]

[0153]

[0154] Select the cluster with the highest score gd q as the classification of the detection area;

[0155] The damage correction includes:

[0156] Obtain the historical verification times gt q of the damage area image and the historical misjudgment number ge q ;

[0157] When , obtain the historical injury judgment record n i of the detection area and the historical clutter record n1 i . When , determine that the detection area image is clutter, and add the detection area image to the label of this cluster for storage; when , if manual review is performed, if the review result is clutter, the historical verification times gt q increase by 1, and the historical misjudgment number ge q remains unchanged. If the review result is injury, the historical verification times gt q increase by 1, and the historical misjudgment number ge q increases by 1; where the label of this cluster includes the injury type; the cluster is a set of detection areas of a specific injury type;

[0158] When occurs, the image is sent to the designated person for manual review. If the review result is clutter, the historical verification count gt q is incremented by 1, and the historical misjudgment count ge q remains unchanged. If the review result is damage, the historical verification count gt q is incremented by 1, and the historical misjudgment count ge q is incremented by 1;

[0159] where ts2 is the set second judgment threshold and ts3 is the set third judgment threshold.

[0160] It should be noted that the preliminary judgment module 401 can implement all the method steps included in S101 in the above method embodiment, and the same parts will not be described in detail;

[0161] It should be noted that the large model injury judgment module 402 can implement all the method steps included in S102 in the above method embodiment, and the same parts will not be described in detail;

[0162] It should be noted that the calibration module 403 can implement all the method steps included in S103 in the above method embodiment, and the same parts will not be described in detail;

[0163] It should be noted that the device embodiment and the above method embodiment belong to the same inventive concept, solve the same technical problems, and achieve the same technical effects, and the same parts will not be described in detail.

[0164] In a third aspect, the present invention proposes an electronic device, as Figure 5 shown, including: a memory and one or more processors.

[0165] One or more application programs are stored in the memory, and the one or more application programs are adapted to be executed by the one or more processors to implement the method for judging rail damage based on a large model described in the first aspect:

[0166] Judge the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result;

[0167] Train a large model for rail damage judgment by using the historical rail injury judgment data, and input the first judgment result into the large model for rail damage judgment to obtain a second judgment result;

[0168] Calibrate the second judgment result to obtain a final judgment result of the injury.

[0169] As Figure 5 shown, the electronic device includes: a processor 501 and a memory 502. Among them, the processor 501 and the memory 502 are connected, such as through a bus interface.

[0170] The structure of the electronic device does not constitute a limitation on the embodiments of the present invention.

[0171] The processor 501 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logical blocks, modules, and circuits described in connection with the disclosure of the present invention. The processor 501 may also be a combination that implements a computing function, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0172] The bus interface may include a path for transmitting information between the above components. The bus interface may be a PCI bus or an EISA bus, etc. The bus interface can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 5 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0173] The memory 502 may be a ROM or other type of static storage device that can store static information and instructions, a RAM, or other type of dynamic storage device that can store information and instructions, or it may also be an EEPROM, a CD-ROM, or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0174] It should be noted that in the embodiments of the present device, the one or more application programs are adapted to be executed by the one or more processors to implement all the large model-based rail damage judgment methods described in the first aspect. The same parts will not be described in detail.

[0175] It should be noted that the embodiments of the present device and the above method embodiments belong to the same inventive concept, solve the same technical problems, and achieve the same technical effects. The same parts will not be described in detail.

[0176] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and the computer program can be loaded and executed by a processor to implement the large model-based rail damage judgment method described in the first aspect.

[0177] The applicant of the present invention has made a detailed description and illustration of the embodiments of the present invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are only the preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, rather than a limitation on the protection scope of the present invention. On the contrary, any improvement or modification made based on the spirit of the present invention should fall within the protection scope of the present invention.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for judging rail damage based on a large model, characterized in that Including: Judging the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result; Training a preset large model based on the historical rail damage judgment data to obtain a rail damage judgment large model, and inputting the first judgment result into the rail damage judgment large model to obtain a second judgment result; Correcting the injury judgment result of the second judgment result to obtain a final injury judgment result.

2. The rail damage judgment method based on a large model according to claim 1, wherein Judging the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result includes: Determining a rectangular injury judgment detection area and moving it at a fixed step size S; Extracting the picture of the injury judgment detection area at each step of movement, and obtaining an initial injury judgment result judged according to the ultrasonic signal characteristics, where the initial injury judgment result includes no damage, nuclear injury, screw hole damage and bottom anomaly of the rail; wherein, the height of the rectangular injury judgment detection area is H and the width is W; Obtaining an injury area image according to the initial injury judgment result; The obtaining an injury area image according to the initial injury judgment result includes: When the initial injury judgment result is no damage, continue to detect the next detection area; When the initial injury judgment result is a nuclear injury, determining the injury area image according to the first method; When the initial injury judgment result is screw hole damage, determining the injury area image according to the second method; When the initial injury judgment result is a bottom anomaly of the rail, determining the injury area image according to the third method; Among them, the first method, the second method and the third method are not completely the same.

3. The method for judging rail damage based on a large model according to claim 2, characterized in that The determining the injury area image according to the first method includes: Obtain the height h corresponding to the nuclear damage i , and obtain the damage area O that needs to be detected i : Among them is the transverse intercept length corresponding to the nuclear injury, dh1 i = max(0, h i - Δh1) is the lower bound of the intercept corresponding to the nuclear injury, uh1 i = min(H, h i + Δh1) is the upper bound of the intercept corresponding to the nuclear injury, Δh1 is the longitudinal intercept length corresponding to the nuclear injury, w1 and Δh1 are multiples of 10; When occurs, a black area with a width of W and a height of 2Δh1 - uh1 i + dh1 i is added above the intercepted area to form an image of the damaged area; When , a black area with a width of W and a height of 2Δh1 - uh1 i + dh1 i is added below the intercepted area to form an image of the damaged area; Where i is the number of the rectangular injury judgment detection area.

4. A method for judging rail damage based on a large model according to claim 2, characterized in that, The determining the injury area image according to the second method includes: Obtain the height h corresponding to the damage of the screw hole i , and obtain the damaged area O to be detected i ; Among them is the transverse intercept length corresponding to the screw hole damage, dh2 i = max(0, h i - Δh2) is the lower bound of the intercept corresponding to the screw hole damage, uh2 i = min(H, h i + Δh2) is the upper bound of the intercept corresponding to the screw hole damage, Δh2 is the longitudinal intercept length corresponding to the screw hole damage, w2 and Δh2 are multiples of 10; When occurs, a black area with a width of W and a height of 2Δh2 - uh2 i + dh2 i is added above the intercepted area to form an image of the damaged area; When occurs, a black area with a width of W and a height of 2Δh2 - uh2 i + dh2 i is added below the intercepted area to form an image of the damaged area; Where i is the number of the rectangular injury judgment detection area.

5. The method for judging rail damage based on a large model according to claim 2, wherein The determining the injury area image according to the third method includes: Obtain the height h corresponding to the abnormality at the rail bottom i to obtain the damaged area O that needs to be detected i ; Among them is the horizontal intercept length corresponding to the rail bottom anomaly, dh3 i = max(0, h i - Δh3) is the lower bound of the intercept corresponding to the rail bottom anomaly, uh3 i = min(H, h i + Δh3) is the upper bound of the intercept corresponding to the rail bottom anomaly, Δh3 is the longitudinal intercept length corresponding to the rail bottom anomaly, w3 and Δh3 are multiples of 10; When occurs, a black area with a width of W and a height of 2Δh3 - uh3 i + dh3 i is added above the intercepted area to form an image of the group damage area; When occurs, a black area with a width of W and a height of 2Δh3 - uh3 i + dh3 i is added below the intercepted area to form an image of the group damage area; Where i is the number of the rectangular injury judgment detection area.

6. The method for judging rail damage based on a large model according to claim 1, wherein, The correcting the injury judgment result of the second judgment result to obtain a final injury judgment result includes: The second judgment result includes no damage, nuclear injury, screw hole damage, bottom anomaly of the rail or clutter; When the second judgment result is clutter, performing clutter detection correction to obtain a final injury judgment result.

7. The rail damage judgment method based on a large model according to claim 6, characterized in that, The performing clutter detection correction includes: Injury feature extraction, injury clustering and injury correction; The injury feature extraction includes: Converting the injury area image into a grayscale image, equally dividing it into L parts according to the width of the area first, and then equally dividing it into L parts according to the height of the area to obtain L*L rectangular areas of the same size, where L is an integer greater than or equal to 10; The average value of pixels in each rectangular area is calculated to be g j,k , where j is the serial number of the rectangular area counted from left to right horizontally, and k is the serial number of the rectangular area counted from top to bottom vertically, obtaining the feature set {g j,k} of the damage area. The values of j and k are both integers from 0 to L - 1; Build a feature matrix where ts1 is the set first judgment threshold; The injury clustering includes: Cluster the feature matrix to obtain m clusters, and extract the cluster center of each cluster, denoted as dp j,k is the element corresponding to the center matrix; The feature matrix D i is calculated for similarity with each cluster center ΔD q to obtain a classification feature score gd q : Select the score gd q Classify the highest cluster as the detection area; The injury correction includes: Obtain the historical verification times gt of the image of the damaged area q And the historical misjudgment number ge q ; When obtain the historical injury judgment record n of the detection area i and the historical clutter record n1 i When judge that the detection area image is clutter, and add the detection area image to the label of this cluster for storage; when if manual review is carried out, if the review result is clutter, the historical verification times gt q increase by 1, and the historical misjudgment number ge q remains unchanged. If the review result is injury, the historical verification times gt q increase by 1, and the historical misjudgment number ge q increase by 1; among them, the label of this cluster includes the injury type; a cluster is a set of detection areas of a specific injury type; When occurs, the image is sent to the designated personnel for manual review. If the review result is clutter, the historical verification count gt q is incremented by 1, and the historical misjudgment count ge q remains unchanged. If the review result is damage, the historical verification count gt q is incremented by 1, and the historical misjudgment count ge q is incremented by 1; Where ts2 is a set second judgment threshold and ts3 is a set third judgment threshold.

8. A rail damage judgment device based on a large model, characterized in that, For implementing the large model-based rail injury judgment method according to any one of claims 1 to 7, the device includes: An initial judgment module configured to judge the damage of the rail according to the ultrasonic signal characteristics to obtain a first judgment result; A large model injury judgment module configured to train a preset large model based on the historical rail injury judgment data to obtain a rail injury judgment large model, and input the first judgment result into the rail injury judgment large model to obtain a second judgment result; A correction module configured to correct the injury judgment result of the second judgment result to obtain a final injury judgment result.

9. A rail damage judgment device based on a large model, characterized in that, It includes a memory, a processor, and a user interface; The memory is used to store computer programs; The user interface is used to interact with the user; The processor is used to read the computer programs in the memory. When the processor executes the computer programs, it implements the method for judging rail damage based on a large model as described in any one of claims 1 to 7.

10. A processor-readable storage medium, characterized in that, The processor-readable storage medium stores computer programs. When the processor executes the computer programs, it implements the method for judging rail damage based on a large model as described in any one of claims 1 to 7.

Images