Intelligent unmanned maintenance vehicle and intelligent unmanned maintenance vehicle control method

Abstract

The invention relates to an intelligent unmanned maintenance vehicle and an intelligent unmanned maintenance vehicle control method, which are applied to the technical field of track detection.The intelligent unmanned maintenance vehicle comprises a vehicle main body, vehicle driving mechanisms used for driving the vehicle main body to advance and retreat are installed on the two sides of the vehicle main body, and the vehicle driving mechanisms are used for driving the vehicle main body to advance and retreat. The vehicle body is further provided with a cooperative detection mechanism, the cooperative detection mechanism comprises a cooperative mechanical arm and an execution collection device installed on the cooperative mechanical arm, and the cooperative mechanical arm is electrically connected with the execution collection device. And the power supply identification control system is installed on the vehicle body, and the vehicle body and the cooperation detection mechanism are both electrically connected with the power supply identification control system. The method has the effect of efficiently and accurately overhauling the train and the rail.

Description

Technical Field

[0001] The present application relates to the technical field of track detection, and in particular to an intelligent unmanned maintenance vehicle and an intelligent unmanned maintenance vehicle control method. Background Art

[0002] By the end of 2016, there were 2,856 standard train sets with an operating mileage of 24,000 kilometers. A large number of trains are running every day, which will cause wear and tear on both the trains themselves and the tracks. In the process of operation, foreign objects may enter the tracks and cause damage to the trains. In order to ensure the safe use of tracks and trains, regular maintenance of tracks and trains is required.

[0003] At present, manual maintenance is mainly used for maintenance. However, due to the large number of trains, long operating mileage and track location that is not suitable for manual operation, the maintenance tasks are becoming more and more difficult and the work pressure is increasing. This has a great impact on the sense of responsibility, mental and physical state of the maintenance personnel, making it impossible to guarantee the quality of the work, thus affecting the normal operation of rail transit. There is an urgent need for a technology that can efficiently and accurately maintain trains and tracks. Summary of the invention

[0004] In order to efficiently and accurately inspect trains and tracks, the present application provides an intelligent unmanned inspection vehicle and an intelligent unmanned inspection vehicle control method.

[0005] In the first aspect, the present application provides an intelligent unmanned maintenance vehicle, which adopts the following technical solution:

[0006] An intelligent unmanned maintenance vehicle, comprising:

[0007] A vehicle body, wherein a vehicle driving mechanism for driving the vehicle body forward and backward is installed on both sides of the vehicle body, and a cooperative detection mechanism is also provided on the vehicle body, wherein the cooperative detection mechanism includes a cooperative mechanical arm and an execution collection device installed on the cooperative mechanical arm, and the cooperative mechanical arm is electrically connected to the execution collection device;

[0008] A power supply identification control system is installed on the vehicle body, and the vehicle body and the cooperative detection mechanism are both electrically connected to the power supply identification control system.

[0009] By adopting the above technical solution, a vehicle driving mechanism is installed on the vehicle body, and the vehicle driving mechanism can drive the vehicle body to move its position. When the position needs to be changed, it automatically moves. When maintenance is needed, the cooperative detection mechanism installed on the vehicle body automatically performs maintenance work. The power supply identification system is set up to supply power and automatically control the vehicle body, thereby realizing efficient and accurate maintenance of trains and tracks.

[0010] Optionally, the vehicle body is provided with a left-right telescopic mechanism for driving the vehicle driving mechanism to extend and retract left-right, and the vehicle body is also provided with a lifting mechanism for driving the vehicle body and the vehicle driving mechanism to rise and fall, the left-right telescopic mechanism includes a first telescopic component and a second telescopic component, the first telescopic component and the second telescopic component are both connected to the vehicle driving mechanism, the lifting mechanism includes a supporting base and a lifting drive device, the lifting drive device is fixedly connected to the supporting base, and the lifting drive device is arranged on the vehicle body.

[0011] By adopting the above technical solution, when inspecting trains and tracks, there may be differences in the width of the tracks or the placement heights of the trains under different inspection conditions. In order to adapt to the above differences, left and right telescopic mechanisms and lifting mechanisms are provided to change the distance between the wheels and the vehicle body to adapt to driving environments of various widths. The lifting mechanism changes the overall height of the vehicle body to adapt to inspection environments of various heights, thereby improving the convenience of using intelligent unmanned inspection vehicles.

[0012] Optionally, the vehicle body is also provided with a linear motor for controlling the lifting and lowering of the collaborative robot arm. The linear motor is arranged on the same side as the vehicle driving mechanism. A tray device is arranged on the linear motor. The tray device is slidably connected to the linear motor. The fixed end of the collaborative robot arm is installed on the side of the tray device away from the ground.

[0013] By adopting the above technical solution, the linear motor controls the collaborative robot arm to rise and fall along the height direction of the vehicle body. When performing maintenance, the maintenance height can be further adjusted to adapt to more maintenance environments at different heights, further improving the convenience of using the intelligent unmanned maintenance vehicle.

[0014] Optionally, the power supply identification and control system includes a road surface identification module, a vehicle power supply module and a vehicle control module, the road surface identification module and the vehicle power supply module are both electrically connected to the vehicle control module, and the vehicle identification module, the vehicle power supply module and the vehicle control module are all arranged on the vehicle body.

[0015] By adopting the above technical scheme, the vehicle control module controls the vehicle body and the vehicle drive mechanism and collaborative robotic arm installed on the vehicle body respectively to realize automatic maintenance of the vehicle. The road surface recognition module collects and identifies information about the train, track and the environment. The vehicle power supply module is used to power the road surface recognition module, the vehicle control module and all electrical devices carried on the vehicle body, while the vehicle control module controls the overall maintenance without human intervention. When the road surface recognition module collects and identifies abnormal conditions that require maintenance, the vehicle control module controls the intelligent unmanned maintenance vehicle to perform automatic maintenance, further improving the convenience of using the intelligent unmanned maintenance vehicle.

[0016] In the second aspect, the present application provides an intelligent unmanned maintenance vehicle control method, which adopts the following technical solution:

[0017] A control method for an intelligent unmanned maintenance vehicle is applied to the intelligent unmanned maintenance vehicle as described in any one of the first aspects, and the method comprises:

[0018] Obtain the on-site image dataset collected by the intelligent unmanned maintenance vehicle and the model to be trained;

[0019] Training the model to be trained based on the on-site image data set to generate a target detection model;

[0020] Obtaining the image to be processed and image data;

[0021] Performing target detection on the image to be processed based on the target detection model to generate a target detection result;

[0022] Performing target recognition processing on the target detection result based on preset recognition rules to generate a target recognition result;

[0023] A maintenance control plan is generated based on the target recognition result and the image data.

[0024] By adopting the above technical solution, a neural network model is used to perform target detection on the image to be processed. After the detection is successful, a target detection result is generated. According to the obtained target detection result, the target is identified and processed, and the specific information of the target is identified to facilitate subsequent maintenance processing. The specific information of the target is used as the target recognition result, and the maintenance personnel control plan is set through the target recognition structure. The intelligent unmanned maintenance vehicle automatically performs maintenance according to the generated maintenance control plan, thereby realizing efficient and accurate maintenance of trains and tracks.

[0025] Optionally, the training the model to be trained based on the on-site image dataset to generate a target detection model includes:

[0026] Performing image preprocessing on the on-site image data set to generate an enhanced grayscale data set;

[0027] The enhanced grayscale data set enters the encoder, pyramid pooling module, generative adversarial network, boundary guided network and skip connection and decoder of DenseNet121 to train the model to be trained;

[0028] Acquire a training end criterion, train the model to be trained based on the training end criterion, and generate a target detection model.

[0029] Optionally, performing image preprocessing on the on-site image dataset to generate an enhanced grayscale dataset includes:

[0030] selecting a key area of ​​the on-site image data;

[0031] Performing enhancement processing on the key area to generate an enhanced data set;

[0032] Performing image grayscale processing on the enhanced data set to obtain an enhanced grayscale data set.

[0033] Optionally, performing target recognition processing on the target detection result based on a preset recognition rule to generate a target recognition result includes:

[0034] Get the abnormal feature template;

[0035] Performing feature comparison on the target detection result based on the abnormal feature template to generate a feature similarity value;

[0036] Obtaining a valid judgment threshold;

[0037] Determining whether the feature similarity value is greater than the validity determination threshold;

[0038] If the feature similarity value is greater than the effective determination threshold, a target recognition result is generated based on the abnormal feature template and the target detection result;

[0039] If the feature similarity value is not greater than the effective determination threshold, performing correction processing on the image to be processed to generate a corrected image;

[0040] The corrected image is subjected to target recognition processing based on the abnormal feature template to generate a target recognition result.

[0041] Optionally, generating a maintenance control scheme based on the target recognition result and the image data includes:

[0042] Obtain maintenance strategies;

[0043] Determining whether automatic maintenance is possible based on the target recognition result and the maintenance strategy;

[0044] If automatic maintenance is possible, determining the relative coordinate information of the target recognition result from the intelligent unmanned maintenance vehicle based on the image data;

[0045] Generate a maintenance control plan based on the relative coordinate information and the target recognition result;

[0046] If automatic inspection is not possible, determining the abnormal coordinates of the target recognition result based on the image data;

[0047] A maintenance control plan is generated based on the abnormal coordinates.

[0048] Optionally, after generating the maintenance control scheme based on the target recognition result and the image data, the method further includes:

[0049] Obtaining the content of the maintenance control plan;

[0050] Generate a report record based on the solution content, the target recognition result and the image data;

[0051] The report record is sent to a preset control center.

[0052] In summary, the present application includes at least one of the following beneficial technical effects:

[0053] 1. A vehicle driving mechanism is installed on the vehicle body, which can drive the vehicle body to move. When the position needs to be changed, it will automatically move. When the cooperative detection mechanism installed on the vehicle body needs to be repaired, it will automatically carry out the repair work. The power supply identification system is set to supply power and automatically control the vehicle body, so as to realize efficient and accurate repair of trains and tracks;

[0054] 2. The vehicle control module controls the vehicle body and the vehicle drive mechanism and cooperative robot arm installed on the vehicle body respectively to realize automatic maintenance of the vehicle. The road surface recognition module collects and recognizes information about the train, track and the environment. The vehicle power supply module is used to supply power to the road surface recognition module, the vehicle control module and all electrical devices carried on the vehicle body, while the vehicle control module controls the overall maintenance without manual intervention. When the road surface recognition module collects and recognizes abnormal conditions that require maintenance, the vehicle control module controls the intelligent unmanned maintenance vehicle to perform automatic maintenance, further improving the convenience of using the intelligent unmanned maintenance vehicle;

[0055] 3. Use a neural network model to perform target detection on the image to be processed. After the detection is successful, generate a target detection result. According to the target detection result, identify the target and identify the specific information of the target to facilitate subsequent maintenance. Use the specific information of the target as the target recognition result, and set the maintenance personnel control plan through the target recognition structure. The intelligent unmanned maintenance vehicle automatically performs maintenance according to the generated maintenance control plan, thereby realizing efficient and accurate maintenance of trains and tracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the overall structure of the intelligent unmanned maintenance vehicle in the embodiment of the present application.

[0057] Figure 2 It is a structural diagram of the intelligent unmanned maintenance vehicle cooperation mechanism in the embodiment of the present application.

[0058] Figure 3 It is a schematic diagram of the structure inside the intelligent unmanned maintenance vehicle in the embodiment of the present application.

[0059] Figure 4 It is a structural schematic diagram of the lifting drive device in the embodiment of the present application.

[0060] Figure 5 It is a structural schematic diagram of the bottom of the intelligent unmanned maintenance vehicle in the embodiment of the present application.

[0061] Figure 6 It is a structural block diagram of the vehicle control relationship provided in an embodiment of the present application.

[0062] Figure 7 It is a flow chart of an intelligent unmanned maintenance vehicle control method provided in an embodiment of the present application.

[0063] Figure 8 It is a structural block diagram of the target detection model provided in the embodiment of the present application.

[0064] Explanation of the reference numerals: 1. vehicle body; 2. vehicle driving mechanism; 21. wheel; 22. protective shell; 23. connecting rod; 24. first gear; 25. second gear; 26. driving main rod; 27. driving motor; 28. mounting platform; 3. cooperative detection mechanism; 31. cooperative robot arm; 32. execution acquisition device; 33. linear motor; 34. tray device; 35. binocular camera; 36. end effector; 37. cooperative controller; 4. power supply identification control system; 41. road surface identification module; 42. vehicle power supply module; 43. vehicle control module; 5. left and right telescopic mechanism; 51. first telescopic component; 52. second telescopic component; 53. electric push rod; 54. push rod bearing; 55. fixing block; 56. transmission motor; 57. transmission belt; 6. lifting mechanism; 61. support base; 62. lifting drive device; 63. lifting drive motor; 64. lifting drive gear; 65. threaded rod; 7. storage slot; 8. protective fence. DETAILED DESCRIPTION

[0065] The present application is further described in detail below with reference to the accompanying drawings.

[0066] The present application embodiment discloses an intelligent unmanned maintenance vehicle, referring to Figure 1 and Figure 2 The intelligent unmanned maintenance vehicle includes a vehicle body 1, which is arranged as a rectangular shell, and a vehicle driving mechanism 2 is arranged on the narrower sides of the rectangular body, and a cooperative detection mechanism 3 and a power supply identification control system 4 are arranged on the side of the vehicle body 1 away from the ground, and the vehicle driving mechanism 2 and the cooperative detection mechanism 3 are both electrically connected to the power supply identification control system 4. The intelligent unmanned maintenance vehicle is mainly used to inspect the tracks and the bottom of the train, that is, according to actual practical needs, the intelligent unmanned maintenance vehicle is placed on the rails to inspect the tracks, or the intelligent unmanned maintenance vehicle is placed in the maintenance workshop to inspect the bottom of the train.

[0067] Reference Figure 2 and Figure 3The vehicle driving mechanism 2 includes two wheels 21 on one side and a protective shell 22 for protecting the two wheels 21 on the same side. A connecting rod 23 is fixedly connected to the side of the vehicle close to the vehicle body 1. A first gear 24 is arranged on the connecting rod 23. The first gear 24 is meshed with a second gear 25. A driving main rod 26 is arranged. The second gear 25 is arranged at both ends of the driving main rod 26. A driving motor 27 for driving the driving main rod 26 to rotate is arranged on the driving main rod 26. Two mounting platforms 28 are placed in the vehicle body 1. The two mounting platforms 28 are both abutted against the inner wall of the lower side of the vehicle body 1. The mounting platforms 28 are rectangular plate-shaped structures and are respectively located on one side close to the vehicle driving mechanism 2. The driving motor 27 is mounted on the mounting platform 28. The other end of the connecting rod 23 is fixedly connected to the mounting platform 28. When in use, the driving motor 27 drives the main rod 26 to drive the second gear 25 to rotate forward or reversely, so that the first gear 24 drives the wheel 21 to rotate forward or backward, thereby driving the vehicle body 1 forward or backward. Furthermore, the wheel 21 is detachable, and a tire for protecting the wheel 21 can be mounted on the outside of the wheel 21 according to actual needs. The detachable setting of the wheel 21 can facilitate timely replacement of the wheel 21 when the wheel 21 is damaged, and due to different inspection environments, wheels 21 of different forms can be provided. When inspecting the track, the wheel 21 without an additional tire can be used. When inspecting the bottom of the train, in order to reduce damage to the wheel 21 and the ground, the wheel 21 with an additional tire can be replaced. It should be noted that the wheel 21 with an additional tire can use the tire of an ordinary car.

[0068] In this embodiment, the vehicle drive mechanism 2 also includes a wheel odometer, which is used to measure the distance from the starting point to the current position of the vehicle. When the vehicle is running, the wheel odometer detects the number of rotations of the wheel. By knowing the wheel diameter, the running distance of the vehicle from the starting point can be obtained, thereby calculating and determining the position of the vehicle, which is convenient for subsequent calculations.

[0069] Reference Figure 3The left and right telescopic mechanisms 5 include a first telescopic component 51 and a second telescopic component 52. When in use, the first telescopic component 51 and the second telescopic component 52 work simultaneously to control the vehicle drive mechanism 2 to approach the vehicle body 1 at the same time, or to control the vehicle drive mechanism 2 to move away from the vehicle at the same time. The first telescopic components 51 are arranged in four groups, and each group of the first telescopic components 51 includes an electric push rod 53. The telescopic end of the electric push rod 53 passes through the side wall of the vehicle body 1 and is fixedly connected to the protective shell 22. The driving end of the electric push rod 53 is connected to a push rod bearing 54, and the push rod bearing 54 is fixedly connected to the upper inner wall of the vehicle body. The second telescopic assembly 52 is set into two groups, and the second telescopic assembly 52 is installed on the mounting platform 28. The second telescopic assembly 52 includes two fixed blocks 55, two transmission motors 56 and a transmission belt 57. The two transmission motors 56 are relatively arranged on the mounting platform 28, and are both located at one end of the mounting platform 28 away from the vehicle drive mechanism 2. The two mounting blocks are fixed on the mounting platform 28 on opposite sides and are arranged by the transmission motor 56. The transmission belt 57 is fixedly connected to the two mounting blocks. When the transmission motor 56 rotates forward or reversely, the driving belt drives the two mounting blocks to approach or move away. Since the mounting blocks are fixedly connected to the mounting platform 28, the two mounting platforms 28 are driven to approach or move away from each other, thereby driving the vehicle drive mechanisms 2 on both sides to approach or move away from the vehicle body 1, and then the driving width of the vehicle intelligent unmanned maintenance vehicle is adjusted according to the actual train or track width when in use.

[0070] In this embodiment, the left and right telescopic mechanisms 5 also include a laser ranging sensor, which is electrically connected to the power supply identification control system 4. When the vehicle needs to be telescoped to the left and right, the power supply identification control system 4 sends a signal to the left and right telescopic mechanisms 5. When reaching the specified position, the laser sensor measures the telescopic distance of the left and right telescopic mechanisms 5, and feeds back the measured telescopic distance to the power supply identification control system 4. The power supply identification control system 4 compares the telescopic distance with the set value. If it is within the error range, the next process is executed. If it exceeds the error range, the telescopic distance and the set value are adjusted, and the adjusted value is sent to the left and right telescopic mechanisms 5 again. The left and right telescopic mechanisms 5 complete the adjustment of the telescopic distance.

[0071] Reference Figure 2 and Figure 4In order to ensure that sufficient lifting power can be achieved, the lifting mechanism 6 is set to at least four groups. The lifting mechanism 6 includes a support base 61 and a lifting drive device 62. The support base 61 is a rectangular plate-shaped base. Two support bases 61 are set, and the length direction is consistent with the traveling direction of the vehicle. The support base 61 is set close to the wheel 21. Among them, the lifting drive device 62 includes a lifting drive motor 63, a lifting drive gear 64 and a threaded rod 65. One end of the threaded rod 65 is fixedly connected to the support base 61, and the lifting drive gear 64 is meshed with the threaded rod 65. The output end of the lifting drive motor 63 is connected to the lifting drive gear 64 for driving the lifting drive gear 64 to rotate. The motor end of the lifting drive motor 63 is installed on the vehicle body 1. When in use, the lifting drive motor 63 drives the lifting drive gear 64 to rotate forward or reverse, thereby driving the vehicle body 1 to move upward or downward to complete the lifting operation.

[0072] Reference Figure 5 Two storage grooves 7 are provided on the outer wall of the vehicle body 1 close to the ground, and the storage grooves 7 cooperate with the support base 61. When the position of the vehicle body 1 needs to be lowered to the lowest, the support base 61 enters the storage grooves 7, thereby avoiding the thickness of the support base 61 from affecting the use height.

[0073] Reference Figure 2 A collaborative robot arm 31 is provided on the vehicle body 1. An execution and collection device 32 is provided at one end of the collaborative robot arm 31 away from the ground. A binocular camera 35 is provided on the execution and collection device 32. When the collaborative robot arm 31 is working, the binocular camera 35 can record the working process. A pallet device 34 is installed at the fixed end of the collaborative robot arm 31. The pallet device 34 is installed on the linear motor 33 and is slidably connected to the linear motor 33. A photoelectric sensor is also provided on the vehicle body 1. The photoelectric sensor is electrically connected to the power supply identification system 4. The photoelectric sensor is used to determine whether the pallet device has reached the bottom / top. When the pallet device reaches the bottom / top and the linear motor 33 is started, the power supply identification system 4 controls the pallet device 34 through the signal of the photoelectric sensor to drive the execution collaborative robot arm 31 to rise or fall. Refer to Figure 1 A protective fence 8 for protecting the power supply identification control system 4, the collaborative robot arm 31 and other devices is also provided on the vehicle body 1. During the driving of the vehicle, the power supply identification control system 4, the collaborative robot arm 31 and other devices are protected to reduce the possibility of damage by foreign objects and improve the safety of use.

[0074] Reference Figure 2 and Figure 6The vehicle body 1 is also provided with a power supply identification control system 4 including a road surface identification module 41, a vehicle power supply module 42 and a vehicle control module 43. The road surface identification module 41 and the vehicle power supply module 42 are both electrically connected to the vehicle control module 43. The vehicle power supply module 42 is mainly used to power all motors arranged in the vehicle body as well as the road surface identification module 41 and the vehicle control module 43. The road surface identification module 41 includes a track detection device such as a linear array camera arranged on both sides of the vehicle body 1 for image acquisition, detection and identification of the vehicle track. It also includes a road condition identification device arranged at the front of the vehicle body 1 for collecting the position of the vehicle body 1, so that the vehicle control module 43 can judge whether the intelligent unmanned maintenance vehicle enters the track. The vehicle control module 43 includes a main control device, an end effector 36, a cooperative controller 37, etc. The end effector 36 is installed at the execution end of the execution robot arm and is electrically connected to the execution robot arm. The cooperative controller 37 is electrically connected to the end effector 36 and the main control device. The staff can control the cooperative controller 37 through the main control device to control the end effector 36 to execute the cooperative instruction, so that the cooperative robot arm 31 performs maintenance work.

[0075] In this embodiment, two sets of programs are provided in the vehicle control module 43, which are respectively used to detect the track and the bottom of the train. The differences between the two sets of programs include that when the track is inspected, the collaborative robot arm 31 is retracted into the protective fence 8, and a protective cover that automatically opens and closes can be added to the top of the protective fence 8 according to actual needs. At the same time, the unmanned maintenance vehicle can only move along the track with the steering function turned off. When the bottom of the train is inspected, the collaborative robot arm 31 is extended out of the protective fence according to actual maintenance needs. At the same time, the steering function of the unmanned maintenance vehicle is turned on to facilitate inspection of all directions of the bottom of the train.

[0076] The embodiment of the present application provides an intelligent unmanned maintenance vehicle control method, which can be executed by an electronic device, which can be a server or a terminal device, wherein the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smart phone, a tablet computer, a desktop computer, etc., but is not limited thereto.

[0077] Figure 7 A flow chart of an intelligent unmanned maintenance vehicle control method provided in an embodiment of the present application.

[0078] like Figure 7 As shown, the main process of the method is described as follows (steps S101 to S106):

[0079] Step S101, obtaining a field image dataset collected by an intelligent unmanned maintenance vehicle and a model to be trained.

[0080] In this embodiment, before the intelligent unmanned maintenance vehicle is used, it is necessary to train the neural network model carried in the vehicle control module set in the vehicle, so as to achieve the accuracy of abnormal target recognition and detection. The scene image data set collected by the intelligent unmanned maintenance vehicle is to place the intelligent unmanned maintenance vehicle in the actual detection environment. The intelligent unmanned maintenance vehicle collects images of the scene environment, and uses all the collected scene images as the scene image data set. The scene environment images include trains and tracks, but the two are independent, that is, the scene environment images are divided into two types, one is an image that must contain tracks in the picture, and the other is an image that must contain trains in the picture. However, the models to be trained and the subsequent training methods used for the two scene environment image training are the same. Here, it is only to illustrate that abnormal situations can be recognized and collected for two situations. Among them, the image containing the train is an image containing the bottom of the train. This application is mainly for the inspection of the bottom of the train, and the image containing the track is an image containing two tracks instead of a single track. In the subsequent description, the track is used for further example. The processing method of the train is consistent with the track, and will not be further elaborated.

[0081] Step S102: training the model to be trained based on the scene image data set to generate a target detection model.

[0082] For step S102, the on-site image data set is image preprocessed to generate an enhanced grayscale data set; the enhanced grayscale data set enters the encoder, pyramid pooling module, generative adversarial network, boundary guided network and jump connection and decoder of the added DenseNet121 to train the model to be trained; the training end standard is obtained, and the model to be trained is trained based on the training end standard to generate a target detection model.

[0083] In this embodiment, the model to be trained is an improved TransUNet model. After the on-site image data set is preprocessed, it enters the model to be trained, and the training model is trained. The grayscale data set is strengthened to enter the encoder, pyramid pooling module, generative adversarial network, boundary guided network and skip connection and decoder of the added DenseNet121 to train the model to be trained. For details, refer to Figure 8, strengthen the grayscale dataset encoder, through the DenseNet121 backbone network, use the parameter sharing and feature reuse characteristics of DenseNet to solve the model overfitting problem caused by the small target image dataset, add the boundary guide (BG) network and jump connection, respectively train the boundary features in the image of parts that may be occluded, and solve the problem of occluded target boundaries; after the encoder processing, the hidden layer feature information of the image is obtained; a newly designed deformable spatial pyramid pooling module (DASPPM) is introduced between the encoder and the decoder, that is, the pyramid pooling module, to enhance the extraction of complex shapes and improve the detection accuracy of small targets; it will receive the feature map output by the encoder as input, and then generate new feature representations through a series of pooling operations and upsampling operations. These new feature representations will be sent to GAN for further processing, and the feature map will be gradually restored to the spatial dimension of the original image. After feature processing by the GAN generator and the GAN discriminator, until the preset stop condition is reached or converges to a certain state, after GAN processing, the quality of the segmentation results generated by the model is further improved, the segmentation edges are clearer, and the blurred areas are reduced. In the decoder, the proposed model uses a newly designed parallel channel and spatial attention modules (PCSAMs) to better capture the basic features in the image by simultaneously considering channel information and spatial information, and the boundary attention feature fusion module (BAFFMs), which combines the multi-scale channel attention mechanism, obtains attention-based fusion features, and alleviates the semantic differences between the shallow encoder and decoder. In addition, in each BAFFM, the boundary features extracted by BG are integrated.

[0084] The model to be trained is continuously trained according to the training end criterion until the training end criterion is reached and the training is stopped to generate a target detection model. The training end criterion can be a fixed number of training rounds or a loss function value reaching a certain value. The specific training end criterion needs to be set according to actual needs and is not specifically limited here.

[0085] Furthermore, a key area of ​​the on-site image data is selected; the key area is enhanced to generate an enhanced data set; and the enhanced data set is subjected to image grayscale processing to obtain an enhanced grayscale data set.

[0086] Take the railroad as an example. In the scene image, the railroad extends linearly in the image. When selecting the key area, it is necessary to ensure that the two rails are within the selected range, that is, the two rails are divided into key areas. Considering the interference and influence of the background at the end of the railroad on target recognition, it is necessary to strengthen the straight line features of the two rails, and take the straight line area as a benchmark, select ±5% as the rail adjacent area, that is, the enhanced area around the railroad, and shield the outer area as the unconsidered area. The image is processed in black to avoid erroneous recognition caused by the surrounding background, etc., thereby completing the enhancement processing, and performing image grayscale processing on the enhanced data to obtain enhanced grayscale data. All image data in the scene image data set are subjected to the above-mentioned enhancement and grayscale processing to obtain an enhanced grayscale data set.

[0087] Step S103, obtaining the image to be processed and the image data.

[0088] In this embodiment, after obtaining the target detection model, it is necessary to use the target detection model to detect the target in the image to obtain the image to be processed and the image data. The image to be processed is the image that needs to perform target detection, and the image data is the relevant data of the image to be processed, including but not limited to the image acquisition time, image acquisition position, and image acquisition coordinates of the image to be processed.

[0089] Step S104: performing target detection on the image to be processed based on the target detection model to generate a target detection result.

[0090] In this embodiment, the image to be processed is preprocessed, that is, the two rails are divided into key areas, the straight line area is used as a reference, ±5% is selected as the rail adjacent area, the image is processed to black, and enhanced processing is performed. The image to be processed after the enhanced processing is performed grayscale processing to obtain an enhanced grayscale image to be processed, and the image to be processed after the enhancement and grayscale processing is input into the target detection model for target detection. The target detection model outputs the target detection result, and the target detection result includes target data such as abnormal objects, abnormal cracks, and abnormal parts.

[0091] Step S105, performing target recognition processing on the target detection result based on preset recognition rules to generate a target recognition result.

[0092] For step S104, obtain an abnormal feature template; perform feature comparison on the target detection result based on the abnormal feature template to generate a feature similarity value; obtain a valid judgment threshold; determine whether the feature similarity value is greater than the valid judgment threshold; if the feature similarity value is greater than the valid judgment threshold, generate a target recognition result based on the abnormal feature template and the target detection result; if the feature similarity value is not greater than the valid judgment threshold, perform correction processing on the image to be processed to generate a corrected image; perform target recognition processing on the corrected image based on the abnormal feature template to generate a target recognition result.

[0093] In this embodiment, only an abnormal feature template is provided, and the feature data of all abnormal targets are recorded in the abnormal feature template, and the abnormal feature template is divided into a plurality of different types of abnormal feature templates according to the type of abnormal target. When performing feature comparison, the corresponding features are searched in the area where the target detection result is located based on the template features, including the pixel information of each orientation of the image of the problematic target detection result, which is used as the image element, and the obtained image element is compared again with the feature element previously recorded in the abnormal feature template, and the difference of the statistical contour pixels and the relevant data of the template edge contour are scanned from left to right to obtain the similarity probability P of the classification summary, that is, the feature similarity value, and the feature similarity value is compared with the effective judgment threshold. If the feature similarity value is greater than the effective judgment threshold, the abnormal feature corresponding to the abnormal feature template for abnormal comparison is used as the target recognition result. If the feature similarity value is less than or equal to the effective judgment threshold, the abnormal feature template used will be replaced, and in order to avoid misjudgment, a secondary correction template will be performed. By changing the selection of the feature points of the target detection result, the new average scaling ratio K and the average rotation angle detected in the previous stage will be obtained, and the new image of the target detection result after the secondary correction will be obtained. The template matching will be performed again. If it is still less than or equal to the effective judgment threshold, the result will be reported and wait for the staff to handle the instructions. If it is greater than the effective judgment threshold, the abnormal feature corresponding to the abnormal feature template for abnormal comparison will also be used as the target recognition result. It should be noted that the target recognition result for the track is mainly foreign objects on the track, mainly including plastic products, metal products, domestic garbage, etc. blown off by strong winds and other weather on the track, or maintenance tools accidentally left behind by maintenance personnel, etc., defects on the track such as cracks on the high-reflective surface of the track, deeply worn lines, corroded surfaces, track joints and other trackside connectors, etc. Therefore, there are many types of abnormal feature templates.

[0094] Step S106, generating a maintenance control plan based on the target recognition result and the image data.

[0095] For step S106, the maintenance strategy is obtained; based on the target recognition result and the maintenance strategy, it is determined whether automatic maintenance is possible; if automatic maintenance is possible, the relative coordinate information of the target recognition result from the intelligent unmanned maintenance vehicle is determined based on the image data; a maintenance control plan is generated based on the relative coordinate information and the target recognition result; if automatic maintenance is not possible, the abnormal coordinates of the target recognition result are determined based on the image data; and a maintenance control plan is generated based on the abnormal coordinates.

[0096] In this embodiment, the target recognition result is the specific target content and whether the target is successfully recognized. When the target recognition result is that the target is not successfully recognized, the intelligent unmanned maintenance vehicle cannot be repaired. The position coordinates of the target recognition result that failed to successfully recognize the target content will be determined based on the image data, and it will be used as abnormal coordinates. A maintenance control plan is generated based on the abnormal coordinates. The maintenance control plan here mainly indicates that maintenance cannot be performed and abnormal coordinates. When the target recognition result is a specific target content, it is determined that the target is successfully recognized, and the intelligent unmanned maintenance vehicle can be repaired. The image data determines the relative coordinate information of the target recognition result from the intelligent unmanned maintenance vehicle, and a maintenance control plan is generated based on the relative coordinate information and the target recognition result, and the maintenance process is performed through the maintenance control plan.

[0097] In this embodiment, the solution content of the maintenance control solution is obtained; a reporting record is generated based on the solution content, the target recognition result and the image data; and the reporting record is sent to a preset control center.

[0098] In order to ensure subsequent review and understanding of the actual situation, it is necessary to report the content of the maintenance plan, target recognition results and image data to the preset control center for recording when the maintenance plan is generated. In order to facilitate the preset control center to receive and record, the plan content, target recognition results and image data are packaged and processed to generate a reporting record, which is sent to the preset control center. This can reduce the computer's processing pressure and the possibility of data loss to a certain extent while facilitating storage at the preset control center.

[0099] The terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus.

[0100] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of application involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned application concept. For example, the above features are replaced with (but not limited to) technical features with similar functions applied in the present application.

Claims

1. An intelligent unmanned maintenance vehicle, characterized in that: include: A vehicle body (1), wherein vehicle drive mechanisms (2) for driving the vehicle body (1) forward and backward are installed on both sides of the vehicle body (1), and a cooperative detection mechanism (3) is also arranged on the vehicle body (1), and the cooperative detection mechanism (3) comprises a cooperative mechanical arm (31) and an execution collection device (32) installed on the cooperative mechanical arm (31), and the cooperative mechanical arm (31) is electrically connected to the execution collection device (32); A power supply identification control system (4), wherein the power supply identification control system (4) is installed on the vehicle body (1), and the vehicle body (1) and the cooperative detection mechanism (3) are both electrically connected to the power supply identification control system (4).

2. The intelligent unmanned maintenance vehicle according to claim 1, characterized in that: The vehicle body (1) is provided with a left-right telescopic mechanism (5) for driving the vehicle driving mechanism (2) to telescope left-right. The vehicle body (1) is also provided with a lifting mechanism (6) for driving the vehicle body (1) and the vehicle driving mechanism (2) to rise and fall. The left-right telescopic mechanism (5) comprises a first telescopic component (51) and a second telescopic component (52). The first telescopic component (51) and the second telescopic component (52) are both connected to the vehicle driving mechanism (2). The lifting mechanism (6) comprises a supporting base (61) and a lifting driving device (62). The lifting driving device (62) is fixedly connected to the supporting base (61). The lifting driving device (62) is provided on the vehicle body (1).

3. The method according to claim 1, wherein: The vehicle body (1) is also provided with a linear motor (33) for controlling the lifting and lowering of the collaborative robot arm (31); the linear motor (33) is arranged on the same side as the vehicle driving mechanism (2); a tray device (34) is arranged on the linear motor (33); the tray device (34) is slidably connected to the linear motor (33); and the fixed end of the collaborative robot arm (31) is installed on the side of the tray device (34) away from the ground.

4. The intelligent unmanned maintenance vehicle according to claim 1, characterized in that: The power supply identification control system (4) comprises a road surface identification module (41), a vehicle power supply module (42) and a vehicle control module (43); the road surface identification module (41) and the vehicle power supply module (42) are both electrically connected to the vehicle control module (43); and the vehicle identification module, the vehicle power supply module (42) and the vehicle control module (43) are all arranged on the vehicle body (1).

5. An intelligent unmanned maintenance vehicle control method, characterized in that: Applied to the intelligent unmanned maintenance vehicle according to any one of claims 1 to 4, the method comprises: Obtain the on-site image dataset collected by the intelligent unmanned maintenance vehicle and the model to be trained; Training the model to be trained based on the on-site image data set to generate a target detection model; Obtaining the image to be processed and image data; Performing target detection on the image to be processed based on the target detection model to generate a target detection result; Performing target recognition processing on the target detection result based on preset recognition rules to generate a target recognition result; A maintenance control plan is generated based on the target recognition result and the image data.

6. The method according to claim 5, characterized in that The training of the model to be trained based on the on-site image data set to generate a target detection model comprises: Performing image preprocessing on the on-site image data set to generate an enhanced grayscale data set; The enhanced grayscale data set enters the encoder, pyramid pooling module, generative adversarial network, boundary guided network and skip connection and decoder of DenseNet121 to train the model to be trained; Acquire a training end criterion, train the model to be trained based on the training end criterion, and generate a target detection model.

7. The method according to claim 6, characterized in that The performing image preprocessing on the on-site image data set to generate an enhanced grayscale data set comprises: selecting a key area of ​​the on-site image data; Performing enhancement processing on the key area to generate an enhanced data set; Performing image grayscale processing on the enhanced data set to obtain an enhanced grayscale data set.

8. The method according to claim 5, characterized in that The performing target recognition processing on the target detection result based on the preset recognition rule to generate the target recognition result comprises: Get the abnormal feature template; Performing feature comparison on the target detection result based on the abnormal feature template to generate a feature similarity value; Obtaining a valid judgment threshold; Determining whether the feature similarity value is greater than the validity determination threshold; If the feature similarity value is greater than the effective determination threshold, a target recognition result is generated based on the abnormal feature template and the target detection result; If the feature similarity value is not greater than the effective determination threshold, performing correction processing on the image to be processed to generate a corrected image; The corrected image is subjected to target recognition processing based on the abnormal feature template to generate a target recognition result.

9. The method according to claim 5, characterized in that The generating of the maintenance control scheme based on the target recognition result and the image data comprises: Obtain maintenance strategies; Determining whether automatic maintenance is possible based on the target recognition result and the maintenance strategy; If automatic maintenance is possible, determining the relative coordinate information of the target recognition result from the intelligent unmanned maintenance vehicle based on the image data; Generate a maintenance control plan based on the relative coordinate information and the target recognition result; If automatic inspection is not possible, determining the abnormal coordinates of the target recognition result based on the image data; A maintenance control plan is generated based on the abnormal coordinates.

10. The method according to claim 9, characterized in that After generating the maintenance control scheme based on the target recognition result and the image data, the method further includes: Obtaining the content of the maintenance control plan; Generate a report record based on the solution content, the target recognition result and the image data; The report record is sent to a preset control center.

Images