A method, device, medium and equipment for obtaining a track bed foreign body detection model

Abstract

The present invention is a method, device, medium and equipment for acquiring a track bed foreign object detection model, and relates to the field of data processing technology, including: acquiring an initial track bed image list, acquiring a candidate track bed image list, processing the candidate track bed image list using a first recognition method, acquiring an intermediate track bed image list, processing the intermediate track bed image list using a second recognition method, acquiring a target foreign object image list, inputting all acquired target foreign object images as a new training sample set into a model corresponding to the first recognition method for training, and acquiring a target track bed foreign object detection model. It can be seen that the present invention can avoid the problem of detecting and covering a small number of foreign object types but with high accuracy or detecting and covering a large number of foreign object types but with low accuracy, improves the recall rate of detecting track bed foreign objects, enables detection and identification of more types of foreign objects, and at the same time improves the accuracy of detecting track bed foreign objects.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, medium and equipment for acquiring a track bed foreign body detection model. Background Art

[0002] Track bed foreign body detection is an important daily inspection link for track safety operation, which directly affects the operation safety of trains. Due to the diversity of foreign bodies on the track bed, it is difficult for a single detection method to conduct comprehensive detection of all types of foreign bodies on the track bed. Traditional detection is based on manual detection methods, which is time-consuming, labor-intensive and inefficient. Since the detection is generally carried out late at night when the train is stopped, the track light is poor at this time, which can easily cause missed detection due to line of sight problems. In the existing technology, unsupervised learning or supervised learning methods are used to detect track bed foreign bodies. Although the unsupervised learning method can theoretically cover more types of foreign bodies, the detection accuracy is low because the abnormal threshold cannot be accurately defined. The method based on supervised deep learning training needs to realize the collection, labeling and training of negative samples, which causes the detection recall rate to drop sharply with the new types of foreign bodies. Therefore, how to detect and identify the diversified types of foreign bodies while having high detection accuracy has become a difficult problem in track bed foreign body inspection. Summary of the invention

[0003] In view of the above technical problems, the technical solution adopted by the present invention is a method for obtaining a track bed foreign body detection model, the method comprising the following steps:

[0004] An initial track bed image list is obtained, wherein the initial track bed image list includes a plurality of initial track bed images, and the initial track bed images are images obtained by real-time photographing of the target track bed using 3D cameras installed on both sides of the target track vehicle.

[0005] According to the initial track bed image list, a candidate track bed image list is obtained, wherein the candidate track bed image list includes a plurality of candidate track bed images, and the candidate track bed images are images obtained after image preprocessing of the initial track bed images.

[0006] The candidate track bed image list is processed using the first recognition method to obtain an intermediate track bed image list.

[0007] The middle track ballast image list is processed by the second identification method to obtain a target foreign object image list, wherein the target foreign object image list includes a plurality of target foreign object images.

[0008] All target foreign object images in the acquired target foreign object image list are input as new training sample sets into the model corresponding to the first recognition method for training, so as to acquire the target track bed foreign object detection model.

[0009] The present invention also provides a device for acquiring a track ballast foreign body detection model, the device comprising:

[0010] The first execution module is used to obtain an initial track bed image list, wherein the initial track bed image list includes a plurality of initial track bed images, and the initial track bed images are images obtained by real-time shooting of the target track bed using 3D cameras installed on both sides of the target rail vehicle.

[0011] The second execution module is used to obtain a candidate track bed image list according to the initial track bed image list, wherein the candidate track bed image list includes a plurality of candidate track bed images, and the candidate track bed images are images obtained after image preprocessing of the initial track bed images.

[0012] The third execution module is used to process the candidate track bed image list by adopting the first recognition method to obtain the intermediate track bed image list.

[0013] The fourth execution module is used to process the intermediate track bed image list using the second recognition method to obtain a target foreign object image list, wherein the target foreign object image list includes a plurality of target foreign object images.

[0014] The fifth execution module is used to input all target foreign object images in the acquired target foreign object image list as a new training sample set into the model corresponding to the first recognition method for training, so as to obtain the target track bed foreign object detection model.

[0015] The present invention also provides a non-transitory computer-readable storage medium, in which at least one instruction or at least one program is stored, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-mentioned method for acquiring a track bed foreign object detection model.

[0016] The present invention also provides an electronic device, comprising a processor and the above-mentioned non-transitory computer-readable storage medium.

[0017] The present invention has at least the following beneficial effects: obtaining an initial track bed image list, obtaining a candidate track bed image list based on the initial track bed image list, processing the candidate track bed image list using a first recognition method, obtaining an intermediate track bed image list, processing the intermediate track bed image list using a second recognition method, obtaining a target foreign object image list, inputting all target foreign object images in the obtained target foreign object image list as a new training sample set into the model corresponding to the first recognition method for training, and obtaining a target track bed foreign object detection model. It can be seen that the present invention can avoid the problem of detecting and covering a small number of foreign object types but with high accuracy or detecting and covering a large number of foreign object types but with low accuracy, improves the recall rate of detecting track bed foreign objects, enables detection and identification of more types of foreign objects, and at the same time improves the accuracy of detecting track bed foreign objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of 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.

[0019] Figure 1 A flow chart of a method for acquiring a track bed foreign body detection model provided in Embodiment 1 of the present invention;

[0020] Figure 2 This is a flow chart of S400 provided in the first embodiment of the present invention;

[0021] Figure 3 This is a flow chart of S405 provided in the first embodiment of the present invention;

[0022] Figure 4 A schematic diagram of the structure of a track bed foreign body detection model acquisition device provided in the second embodiment of the present invention;

[0023] Figure 5 A schematic diagram of the structure of the fourth execution module 4 provided in the second embodiment of the present invention;

[0024] Figure 6 This is a schematic diagram of the structure of the target foreign body image list acquisition module 45 provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

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

[0027] Embodiment 1,

[0028] This embodiment 1 provides a method for obtaining a track bed foreign body detection model, the method comprising the following steps: Figure 1 As shown:

[0029] S100, obtaining an initial track bed image list, wherein the initial track bed image list includes a plurality of initial track bed images, and the initial track bed images are images obtained by real-time photographing of a target track bed using a 3D camera installed on both sides of a target rail vehicle.

[0030] Specifically, the target track bed is a track bed to be inspected.

[0031] Specifically, the target rail vehicle is a vehicle inspected on the target rail track bed.

[0032] S200, obtaining a candidate track bed image list according to the initial track bed image list, wherein the candidate track bed image list includes a plurality of candidate track bed images, and the candidate track bed images are images obtained after image preprocessing of the initial track bed images.

[0033] Specifically, the image preprocessing is to first filter, remove noise and grayscale the image, then superimpose two images of the track bed taken at the same position on the left and right sides by a preset length, cut the excess parts and then splice them.

[0034] Furthermore, the preset length ranges from 25 cm to 30 cm.

[0035] Furthermore, those skilled in the art know that any filtering, denoising, biochemical and grayscale process in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0036] S300: Process the candidate track bed image list using a first recognition method to obtain an intermediate track bed image list.

[0037] Specifically, the process of processing the candidate track bed image list in the first identification method is to input the candidate track bed image list into a specified neural network model for foreign object identification.

[0038] Furthermore, the designated neural network model is a model obtained by inputting several first key track bed images and several second key track bed images as sample training sets into an initial neural network model for model training, wherein the first key track bed image is an image of the track bed with initial foreign objects, and the second key track bed image is an image of the track bed without initial foreign objects.

[0039] Furthermore, the initial foreign matter may be understood as a common type of foreign matter, such as a cardboard box, a suitcase, a fallen rock, a wooden stick, or a branch.

[0040] Furthermore, the method for acquiring the first key track bed image is consistent with the method for acquiring the candidate track bed image, which can be understood as: placing initial foreign objects on the track bed in advance, and using the images captured by 3D cameras installed on both sides of the track bed where the initial foreign objects are placed to obtain the key track bed image by image preprocessing.

[0041] Furthermore, those skilled in the art know that any method for training a neural network model in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0042] Specifically, the intermediate track bed image list includes a plurality of intermediate track bed images, wherein the intermediate track bed images are candidate track bed images in which no initial foreign objects can be detected using a specified neural network model.

[0043] S400, using a second identification method to process the middle track ballast image list to obtain a target foreign object image list, wherein the target foreign object image list includes a plurality of target foreign object images.

[0044] Specifically, S400 also includes the following steps: Figure 2 As shown:

[0045] S401, obtaining a sample track bed image ID list, wherein the sample track bed image ID list includes a plurality of sample track bed image IDs, the sample track bed image ID is a unique identifier characterizing the sample track bed image, and the sample track bed image is an image of a track bed without initial foreign objects.

[0046] Specifically, the sample track bed image ID is obtained based on the coding method of the mileage ID, for example: "DK2+580" means 580 meters after the 2nd kilometer of the design mileage.

[0047] Furthermore, each sample track bed image ID corresponds to a sample track bed image.

[0048] S402, obtaining an intermediate track roadbed image ID list, wherein the intermediate track roadbed image ID list includes a plurality of intermediate track roadbed image IDs, and the intermediate track roadbed image ID is a unique identifier representing the intermediate track roadbed image.

[0049] Specifically, the method for acquiring the intermediate track bed image ID is consistent with the method for acquiring the sample track bed image ID.

[0050] S403, according to the sample track bed image ID list, obtain the target track bed image list corresponding to the intermediate track bed image list, wherein the target track bed image list includes a plurality of target track bed images, and the target track bed image is a sample track bed image corresponding to the sample track bed image ID consistent with the intermediate track bed image ID.

[0051] S404, according to the target track bed image, obtaining a target priority list corresponding to the intermediate track bed image list, wherein the target priority list includes a plurality of target priorities, and each intermediate track bed image corresponds to a target priority.

[0052] Specifically, the target priority is the ratio of the target point number to the number of pixel points of the intermediate track bed image, wherein the target point number is the number of inconsistencies between the pixel points of the intermediate track bed image and the pixel points of the target track bed image corresponding to the intermediate track bed image. It can be understood that: each pixel point corresponds to a pixel value, and inconsistent pixel points mean different pixel values.

[0053] S405, obtaining a target foreign object image list according to the target priority list, wherein the target foreign object image list includes a plurality of target foreign object images, and the target foreign object image is an image of a foreign object obtained by judging object contour pixel points in the intermediate track bed image screened based on the target priority.

[0054] Specifically, S405 also includes the following steps: Figure 3 As shown:

[0055] S4051, obtaining a target abnormality image list according to the target priority list, wherein the target abnormality image list includes a plurality of target abnormality images, and the target abnormality image is an intermediate track bed image corresponding to a target priority greater than a preset first priority threshold in the target priority list.

[0056] Specifically, the preset first priority threshold value ranges from 0.01 to 0.05.

[0057] S4052, according to the target abnormal image list, obtain a target object image list and a final priority list corresponding to the target object image list, wherein the target object image list includes a plurality of target object images, the target object image is an image of an object identified from all target abnormal images, and the final priority list includes a plurality of final priorities, the final priority is a ratio between an absolute value of a difference between a first value and a second value and the second value, the first value is a sum of pixel values ​​corresponding to the target object image in a target abnormal image including the target object image, and the second value is a sum of pixel values ​​corresponding to the target object image in a sample track bed image including the target object image.

[0058] S4053: When the final priority is greater than a preset second priority threshold, determine that the target object image corresponding to the final priority is a target foreign body image.

[0059] Specifically, each target object image corresponds to a final priority.

[0060] Specifically, the preset second priority threshold value ranges from 0.4 to 0.6.

[0061] S500: All target foreign object images in the acquired target foreign object image list are input as newly added training sample sets into the model corresponding to the first recognition method for training, so as to acquire a target track ballast foreign object detection model.

[0062] Specifically, S500 also includes the following steps:

[0063] S501, obtaining a designated image list, wherein the designated image list includes a plurality of designated images, and the designated images are a plurality of images randomly selected from images corresponding to all target foreign body images, and the number of designated images in the designated image list is consistent with the number of key track bed images.

[0064] S502, inputting a specified image list and several key track bed images as training sample sets into a specified neural network model for training to obtain a final priority, wherein the final priority is a value obtained by calculating using a cross entropy loss function.

[0065] Specifically, those skilled in the art know that any method of calculating using the cross entropy loss function in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0066] S503, continuously adjusting the specified neural network model parameters so that the final priority meets the preset conditions to obtain the target track ballast foreign body detection model.

[0067] Specifically, the preset condition is less than a preset priority threshold, wherein those skilled in the art know that the preset priority threshold can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be repeated here.

[0068] From the above content, it can be seen that the target foreign object image obtained by the second recognition method, that is, the newly added foreign object, adds the image corresponding to the target foreign object image to the old training sample as sample training data for model training to obtain the target track bed foreign object detection model, which can avoid the problem of detecting a small number of foreign object types but high accuracy or detecting a large number of foreign object types but low accuracy, improves the recall rate of detecting track bed foreign objects, enables more types of foreign objects to be detected and identified, and at the same time, improves the accuracy of detecting track bed foreign objects.

[0069] The method for acquiring a track bed foreign object detection model provided in the present embodiment includes the following steps: acquiring an initial track bed image list, acquiring a candidate track bed image list based on the initial track bed image list, processing the candidate track bed image list using a first recognition method to acquire an intermediate track bed image list, processing the intermediate track bed image list using a second recognition method to acquire a target foreign object image list, inputting all target foreign object images in the acquired target foreign object image list as a newly added training sample set into a model corresponding to the first recognition method for training, and acquiring a target track bed foreign object detection model. It can be seen that the present invention can avoid the problem of detecting and covering a small number of foreign object types but with high accuracy or detecting and covering a large number of foreign object types but with low accuracy, improves the recall rate of detecting track bed foreign objects, enables detection and identification of more types of foreign objects, and at the same time improves the accuracy of detecting track bed foreign objects.

[0070] Embodiment 2:

[0071] like Figure 4 As shown, this embodiment provides a device for acquiring a track bed foreign body detection model, the device comprising:

[0072] The first execution module 1 is used to obtain an initial track bed image list, wherein the initial track bed image list includes a plurality of initial track bed images, and the initial track bed images are images obtained by real-time shooting of the target track bed using 3D cameras installed on both sides of the target rail vehicle.

[0073] Specifically, the target track bed is a track bed to be inspected.

[0074] Specifically, the target rail vehicle is a vehicle inspected on the target rail track bed.

[0075] The second execution module 2 is used to obtain a candidate track bed image list according to the initial track bed image list, wherein the candidate track bed image list includes a plurality of candidate track bed images, and the candidate track bed images are images obtained after image preprocessing of the initial track bed images.

[0076] Specifically, the image preprocessing is to first filter, remove noise and grayscale the image, then superimpose two images of the track bed taken at the same position on the left and right sides by a preset length, cut the excess parts and then splice them.

[0077] Furthermore, the preset length ranges from 25 cm to 30 cm.

[0078] Furthermore, those skilled in the art know that any filtering, denoising, biochemical and grayscale process in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0079] The third execution module 3 is used to process the candidate track bed image list by adopting the first recognition method to obtain the intermediate track bed image list.

[0080] Specifically, the process of processing the candidate track bed image list in the first identification method is to input the candidate track bed image list into a specified neural network model for foreign object identification.

[0081] Furthermore, the designated neural network model is a model obtained by inputting several first key track bed images and several second key track bed images as sample training sets into an initial neural network model for model training, wherein the first key track bed image is an image of the track bed with initial foreign objects, and the second key track bed image is an image of the track bed without initial foreign objects.

[0082] Furthermore, the initial foreign matter may be understood as a common type of foreign matter, such as a cardboard box, a suitcase, a fallen rock, a wooden stick, or a branch.

[0083] Furthermore, the method for acquiring the first key track bed image is consistent with the method for acquiring the candidate track bed image, which can be understood as: placing initial foreign objects on the track bed in advance, and using the images captured by 3D cameras installed on both sides of the track bed where the initial foreign objects are placed to obtain the key track bed image by image preprocessing.

[0084] Furthermore, those skilled in the art know that any method for training a neural network model in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0085] Specifically, the intermediate track bed image list includes a plurality of intermediate track bed images, wherein the intermediate track bed images are candidate track bed images in which no initial foreign objects can be detected using a specified neural network model.

[0086] The fourth execution module 4 is used to process the intermediate track bed image list by adopting the second recognition method to obtain a target foreign object image list, wherein the target foreign object image list includes a plurality of target foreign object images.

[0087] Specifically, Figure 5 As shown, the fourth execution module 4 also includes:

[0088] The sample track bed image ID list acquisition module 41 is used to acquire a sample track bed image ID list, wherein the sample track bed image ID list includes a plurality of sample track bed image IDs, the sample track bed image ID is a unique identifier characterizing the sample track bed image, and the sample track bed image is an image of a track bed without initial foreign objects.

[0089] Specifically, the sample track bed image ID is obtained based on the coding method of the mileage ID, for example: "DK2+580" means 580 meters after the 2nd kilometer of the design mileage.

[0090] Furthermore, each sample track bed image ID corresponds to a sample track bed image.

[0091] The intermediate track roadbed image ID list acquisition module 42 is used to acquire the intermediate track roadbed image ID list, wherein the intermediate track roadbed image ID list includes a plurality of intermediate track roadbed image IDs, and the intermediate track roadbed image ID is a unique identifier representing the intermediate track roadbed image.

[0092] Specifically, the method for acquiring the intermediate track bed image ID is consistent with the method for acquiring the sample track bed image ID.

[0093] The target track bed image list acquisition module 43 is used to acquire the target track bed image list corresponding to the intermediate track bed image list according to the sample track bed image ID list, wherein the target track bed image list includes a plurality of target track bed images, and the target track bed image is a sample track bed image corresponding to the sample track bed image ID consistent with the intermediate track bed image ID.

[0094] The target priority list acquisition module 44 is used to acquire a target priority list corresponding to the intermediate track bed image list according to the target track bed image, wherein the target priority list includes a plurality of target priorities, and each intermediate track bed image corresponds to a target priority.

[0095] Specifically, the target priority is the ratio of the target point number to the number of pixel points of the intermediate track bed image, wherein the target point number is the number of inconsistencies between the pixel points of the intermediate track bed image and the pixel points of the target track bed image corresponding to the intermediate track bed image. It can be understood that: each pixel point corresponds to a pixel value, and inconsistent pixel points mean different pixel values.

[0096] The target foreign object image list acquisition module 45 is used to acquire a target foreign object image list according to a target priority list, wherein the target foreign object image list includes a plurality of target foreign object images, and the target foreign object image is an image of a foreign object acquired by judging the object contour pixel points in the intermediate track bed image screened based on the target priority.

[0097] Specifically, Figure 6 As shown, the target foreign body image list acquisition module 45 also includes:

[0098] The first acquisition module 451 is used to acquire a target abnormality image list according to a target priority list, wherein the target abnormality image list includes a plurality of target abnormality images, and the target abnormality image is an intermediate track bed image corresponding to a target priority greater than a preset first priority threshold in the target priority list.

[0099] Specifically, the preset first priority threshold value ranges from 0.01 to 0.05.

[0100] The second acquisition module 452 is used to acquire a target object image list and a final priority list corresponding to the target object image list according to the target abnormal image list, wherein the target object image list includes a plurality of target object images, and the target object image is an image of an object identified from all target abnormal images; the final priority list includes a plurality of final priorities, and the final priority is a ratio between an absolute value of a difference between a first value and a second value and the second value; the first value is a sum of pixel values ​​corresponding to the target object image in a target abnormal image including the target object image; and the second value is a sum of pixel values ​​corresponding to the target object image in a sample track bed image including the target object image.

[0101] The third acquisition module 453 is used to determine that the target object image corresponding to the final priority is a target foreign body image when the final priority is greater than a preset second priority threshold.

[0102] Specifically, each target object image corresponds to a final priority.

[0103] Specifically, the preset second priority threshold value ranges from 0.4 to 0.6.

[0104] The fifth execution module 5 is used to input all target foreign object images in the acquired target foreign object image list as a new training sample set into the model corresponding to the first recognition method for training, so as to acquire the target track bed foreign object detection model.

[0105] Specifically, the fifth execution module 5 further includes:

[0106] The designated image list acquisition module is used to obtain the designated image list, wherein the designated image list includes a plurality of designated images, and the designated images are a plurality of images randomly selected from the images corresponding to all target foreign body images, and the number of designated images in the designated image list is consistent with the number of key track bed images.

[0107] The final priority acquisition module is used to input the specified image list and several key track bed images as training sample sets into the specified neural network model for training to obtain the final priority, wherein the final priority is a value obtained by calculating the cross entropy loss function.

[0108] Specifically, those skilled in the art know that any method of calculating using the cross entropy loss function in the prior art falls within the protection scope of the present invention and will not be described in detail herein.

[0109] The target track bed foreign object detection model acquisition module is used to continuously adjust the specified neural network model parameters so that the final priority meets the preset conditions to obtain the target track bed foreign object detection model.

[0110] Specifically, the preset condition is less than a preset priority threshold, wherein those skilled in the art know that the preset priority threshold can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be repeated here.

[0111] From the above content, it can be seen that the target foreign object image obtained by the second recognition method, that is, the newly added foreign object, adds the image corresponding to the target foreign object image to the old training sample as sample training data for model training to obtain the target track bed foreign object detection model, which can avoid the problem of detecting a small number of foreign object types but high accuracy or detecting a large number of foreign object types but low accuracy, improves the recall rate of detecting track bed foreign objects, enables more types of foreign objects to be detected and identified, and at the same time, improves the accuracy of detecting track bed foreign objects.

[0112] Embodiment 3,

[0113] An embodiment of the present invention provides a non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the steps:

[0114] An initial track bed image list is obtained, wherein the initial track bed image list includes a plurality of initial track bed images, and the initial track bed images are images obtained by real-time photographing of the target track bed using 3D cameras installed on both sides of the target track vehicle.

[0115] According to the initial track bed image list, a candidate track bed image list is obtained, wherein the candidate track bed image list includes a plurality of candidate track bed images, and the candidate track bed images are images obtained after image preprocessing of the initial track bed images.

[0116] The candidate track bed image list is processed using the first recognition method to obtain an intermediate track bed image list.

[0117] The middle track ballast image list is processed by the second identification method to obtain a target foreign object image list, wherein the target foreign object image list includes a plurality of target foreign object images.

[0118] All target foreign object images in the acquired target foreign object image list are input as new training sample sets into the model corresponding to the first recognition method for training, so as to acquire the target track bed foreign object detection model.

[0119] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0120] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0121] Embodiment 4:

[0122] An embodiment of the present invention provides an electronic device, which includes a processor and the non-transitory computer-readable storage medium in Embodiment 3 of the present invention.

[0123] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for acquiring a track bed foreign body detection model, characterized in that: The method comprises the following steps: Acquire an initial track bed image list, wherein the initial track bed image list includes a plurality of initial track bed images, and the initial track bed images are images acquired by photographing the target track bed in real time using 3D cameras installed on both sides of the target track vehicle; According to the initial track bed image list, a candidate track bed image list is obtained, wherein the candidate track bed image list includes a plurality of candidate track bed images, and the candidate track bed images are images obtained after image preprocessing of the initial track bed images; The candidate track bed image list is processed by a first recognition method to obtain an intermediate track bed image list, wherein the process of processing the candidate track bed image list by the first recognition method is to input the candidate track bed image list into a specified neural network model for foreign body recognition; The middle track ballast image list is processed by the second recognition method to obtain a target foreign object image list, wherein the target foreign object image list includes a plurality of target foreign object images. The middle track ballast image list is processed by the second recognition method to obtain the target foreign object image list, including the following steps: Obtaining a sample track bed image ID list, wherein the sample track bed image ID list includes a plurality of sample track bed image IDs, the sample track bed image ID is a unique identifier characterizing the sample track bed image, and the sample track bed image is an image of a track bed without initial foreign matter; Obtaining a list of intermediate track roadbed image IDs, wherein the list of intermediate track roadbed image IDs includes a plurality of intermediate track roadbed image IDs, and the intermediate track roadbed image ID is a unique identifier representing the intermediate track roadbed image; According to the sample track bed image ID list, a target track bed image list corresponding to the intermediate track bed image list is obtained, wherein the target track bed image list includes a plurality of target track bed images, and the target track bed image is a sample track bed image corresponding to a sample track bed image ID consistent with the intermediate track bed image ID; According to the target track bed image, a target priority list corresponding to the intermediate track bed image list is obtained, wherein the target priority list includes a plurality of target priorities, and each intermediate track bed image corresponds to a target priority; According to the target priority list, a target foreign object image list is obtained, wherein the target foreign object image list includes a plurality of target foreign object images, and the target foreign object image is an image of a foreign object obtained by judging object contour pixel points in the middle track ballast image screened based on the target priority; All target foreign object images in the acquired target foreign object image list are input as new training sample sets into the model corresponding to the first recognition method for training, so as to obtain the target track bed foreign object detection model.

2. The method for acquiring a track bed foreign body detection model according to claim 1, characterized in that: The designated neural network model is a model obtained by inputting a number of first key track bed images and a number of second key track bed images as sample training sets into an initial neural network model for model training, wherein the first key track bed images are images of the track bed with initial foreign objects, and the second key track bed images are images of the track bed without initial foreign objects.

3. The method for acquiring a track bed foreign body detection model according to claim 1, characterized in that: According to the target priority list, obtaining the target foreign body image list includes the following steps: According to the target priority list, a target abnormal image list is obtained, wherein the target abnormal image list includes a plurality of target abnormal images, and the target abnormal image is an intermediate track bed image corresponding to a target priority greater than a preset first priority threshold in the target priority list; According to the target abnormal image list, a target object image list and a final priority list corresponding to the target object image list are obtained, wherein the target object image list includes a plurality of target object images, the target object image is an image of an object identified from all target abnormal images, the final priority list includes a plurality of final priorities, the final priority is a ratio between an absolute value of a difference between a first value and a second value and the second value, the first value is a sum of pixel values ​​corresponding to the target object image in a target abnormal image including the target object image, and the second value is a sum of pixel values ​​corresponding to the target object image in a sample track bed image including the target object image; When the final priority is greater than a preset second priority threshold, the target object image corresponding to the final priority is determined to be a target foreign body image.

4. A device for acquiring a track bed foreign body detection model, characterized in that: The device comprises: A first execution module is used to obtain an initial track bed image list, wherein the initial track bed image list includes a plurality of initial track bed images, and the initial track bed images are images obtained by real-time shooting of the target track bed using 3D cameras installed on both sides of the target rail vehicle; A second execution module is used to obtain a candidate track bed image list according to the initial track bed image list, wherein the candidate track bed image list includes a plurality of candidate track bed images, and the candidate track bed images are images obtained after image preprocessing of the initial track bed images; A third execution module is used to process the candidate track bed image list using the first recognition method to obtain an intermediate track bed image list, wherein the process of processing the candidate track bed image list using the first recognition method is to input the candidate track bed image list into a specified neural network model for foreign body recognition; The fourth execution module is used to process the intermediate track bed image list by using the second recognition method to obtain a target foreign object image list, wherein the target foreign object image list includes a plurality of target foreign object images, and the fourth execution module also includes: A sample track bed image ID list acquisition module is used to acquire a sample track bed image ID list, wherein the sample track bed image ID list includes a plurality of sample track bed image IDs, the sample track bed image ID is a unique identifier characterizing the sample track bed image, and the sample track bed image is an image of a track bed without initial foreign matter; An intermediate track roadbed image ID list acquisition module is used to acquire an intermediate track roadbed image ID list, wherein the intermediate track roadbed image ID list includes a plurality of intermediate track roadbed image IDs, and the intermediate track roadbed image ID is a unique identifier representing the intermediate track roadbed image; A target track bed image list acquisition module is used to acquire a target track bed image list corresponding to the intermediate track bed image list according to the sample track bed image ID list, wherein the target track bed image list includes a plurality of target track bed images, and the target track bed image is a sample track bed image corresponding to the sample track bed image ID consistent with the intermediate track bed image ID; A target priority list acquisition module is used to acquire a target priority list corresponding to the intermediate track bed image list according to the target track bed image, wherein the target priority list includes a plurality of target priorities, and each intermediate track bed image corresponds to a target priority; A target foreign object image list acquisition module is used to acquire a target foreign object image list according to a target priority list, wherein the target foreign object image list includes a plurality of target foreign object images, and the target foreign object image is an image of a foreign object acquired by judging object contour pixel points in an intermediate track ballast image screened based on the target priority; The fifth execution module is used to input all target foreign object images in the acquired target foreign object image list as a new training sample set into the model corresponding to the first recognition method for training, so as to obtain the target track bed foreign object detection model.

5. The track bed foreign body detection model acquisition device according to claim 4, characterized in that: The designated neural network model is a model obtained by inputting a number of first key track bed images and a number of second key track bed images as sample training sets into an initial neural network model for model training, wherein the first key track bed images are images of the track bed with initial foreign objects, and the second key track bed images are images of the track bed without initial foreign objects.

6. The track bed foreign body detection model acquisition device according to claim 4, characterized in that: The target foreign body image list acquisition module also includes: A first acquisition module is used to acquire a target abnormal image list according to a target priority list, wherein the target abnormal image list includes a plurality of target abnormal images, and the target abnormal image is an intermediate track bed image corresponding to a target priority greater than a preset first priority threshold in the target priority list; a second acquisition module, configured to acquire, according to the target abnormal image list, a target object image list and a final priority list corresponding to the target object image list, wherein the target object image list includes a plurality of target object images, the target object image is an image of an object identified from all target abnormal images, the final priority list includes a plurality of final priorities, the final priority is a ratio between an absolute value of a difference between a first value and a second value and the second value, the first value is a sum of pixel values ​​corresponding to the target object image in a target abnormal image including the target object image, and the second value is a sum of pixel values ​​corresponding to the target object image in a sample track bed image including the target object image; The third acquisition module is used to determine that the target object image corresponding to the final priority is a target foreign body image when the final priority is greater than a preset second priority threshold.

7. A non-transitory computer-readable storage medium, wherein at least one instruction or at least one program is stored in the non-transitory computer-readable storage medium, characterized in that: The at least one instruction or the at least one program is loaded and executed by the processor to implement the method for acquiring a track bed foreign object detection model as described in any one of claims 1 to 3.

8. An electronic device, characterized in that: The invention comprises a processor and the non-transitory computer-readable storage medium as claimed in claim 7.

Images