Device and method for detecting broken area of tunnel surrounding rock

Abstract

The invention belongs to the technical field of tunnel face image recognition, and provides a tunnel surrounding rock broken area detection device and method, and the method is a deep tunnel surrounding rock broken area recognition method based on a Unt-Transform technology, and comprises the steps: inputting a tunnel face image shot at a construction site into a trained U-net neural network; a broken area image is extracted through model calculation, and other equipment facilities contained in the image and a non-broken area image are removed. According to the method, the pictures collected during construction are fully utilized, the broken area of the tunnel face is rapidly recognized, technical support is provided for engineering personnel to take engineering measures in time, and the method has high popularization and guiding significance.

Description

Technical Field

[0001] The present invention belongs to the technical field of tunnel face image recognition, and in particular relates to a tunnel surrounding rock crushing zone detection device and method. Background Art

[0002] During the construction of highway and railway tunnels, hydraulic tunnels, underground mines, etc., the face is the frontier of tunnel excavation, and its stability is directly related to the safety and progress of the entire project. The face crushing area is the most vulnerable part of tunnel construction, which is easily affected by geological stress and construction disturbance, causing safety accidents such as landslides and mud bursts. By accurately identifying this unstable area, timely measures can be taken to prevent dangers such as landslides and mud bursts, ensure the safety of construction personnel, reasonably arrange construction plans, and avoid construction delays. At present, the determination of face stability mainly relies on experienced engineering personnel, which is highly subjective. Therefore, there is an urgent need for a method that can objectively, quickly and accurately identify the face crushing area. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a tunnel surrounding rock crushing zone detection device and method to solve the problems in the prior art. The technical solution adopted by the present invention is: A method for detecting a tunnel surrounding rock crushing area comprises the following steps: Step 1, tunnel face image acquisition: multiple local images of the tunnel face including the broken rock mass area are acquired by an image acquisition device of the detection device, and the multiple local images are synthesized into a panoramic image; Step 2, image annotation: clearly annotate the image to distinguish the broken rock area from the non-target area; Step 3, data preparation: Divide the labeled image data into two parts: Training set: used for model training; Test set: used to test the model effect; Step 4, model training: use a neural network combining U-Net and Transformer for image segmentation training; Step 5, model verification: input the broken area images of the test set into the trained model to evaluate the actual performance of the model; Step 6, save the optimal model: save the model that has been evaluated and confirmed as the optimal one; Step 7, inputting a newly shot tunnel face image containing a broken area: inputting a newly shot tunnel face image containing a broken rock mass area at the construction site into the saved optimal model; Step 8: Broken area extraction: Using the computing power of the model, the broken rock area is identified from the input image and other construction equipment and facilities in the image are excluded; finally, a pure broken area image is output.

[0004] Further, the step 4 comprises: Step 4-1, data input: input the labeled training set images into the model combining U-Net and Transformer. The U-Net part is used to extract local features of the image, and the Transformer part is used to capture the global context information of the image. Step 4-2, encoder processing: extract the features of the image of the broken rock mass area of ​​the tunnel face through several convolution layers and pooling layers. The convolution layer extracts the local features of the broken area, and the pooling layer reduces the spatial resolution of the image to generate a feature map. The convolution operation uses the following formula: , in, I represents the tunnel face image acquired under the input complex working conditions, K represents the convolution kernel, represents the convolution operation, and Both represent the position index of the palm face feature map output after the convolution operation, and m and n both represent the position index of the convolution kernel; Use the nonlinear activation function ReLU: , The pooling operation is used to reduce the spatial resolution of the feature map: , in, represents the pooled face feature map, Represents the position index of the face feature map output after the pooling operation, and Represents the position index of the pooling window, Indicates the size of the pooling window; Step 4-3, bottleneck layer processing: The deepest convolutional layer further abstracts the broken area features here to generate high-level feature representation; Step 4-4, decoder processing: The spatial resolution of the image is gradually restored through deconvolution layers and upsampling layers. Each deconvolution layer is combined with the feature map of the corresponding layer of the encoder to retain high-resolution features. The deconvolution operation uses the following formula: , in, represents the feature map of the palm face after the deconvolution operation, represents the deconvolution kernel, represents the deconvolution operation, and represents the position index of the face feature map output after the deconvolution operation, and Both represent the position index of the deconvolution kernel; Step 4-5, jump connection: During the decoding process, the jump connection directly passes the encoder's high-resolution fragmentation area feature map to the corresponding layer of the decoder, enhancing the model's ability to recognize the boundary of the target area. The feature map cascade formula is: , in, It represents the characteristic diagram of the tunnel face after cascading. represents the face feature map from the encoder, represents the palm face feature map from the decoder; Steps 4-6, loss calculation and optimization: The cross entropy loss function is used to calculate the difference between the model's prediction results for the broken rock area and the actual annotation, and the model parameters are adjusted through the optimization algorithm. The cross entropy loss function formula is: , in, L represents the cross entropy loss function, N represents the number of samples of the face image, Indicates palm face images, Indicates The true labels of the palm face images, Indicates The predicted probability of a tunnel face image.

[0005] A tunnel surrounding rock crushing zone detection device, comprising a mounting plate, a housing, a rotating column, an image acquisition device, a rotating plate, a supporting column, a motor, a connecting mechanism and an adjusting mechanism; The housing is mounted on the mounting plate, the bottom of the rotating column is rotatably arranged in the housing, the top of the rotating column passes through the housing, the top of the rotating column is connected to the image acquisition device, the bottom of the rotating column is connected to the rotating plate, and the bottom of the rotating plate is rotatably connected to the top of the supporting column; The motor is connected to the rotating plate through the connecting mechanism, and is used to drive the rotating plate to reciprocate or rotate with interval pauses; The rotating plate is connected to the rotating column through the adjusting mechanism, and is used to adjust the rotation angle of the rotating column.

[0006] Furthermore, the mounting plate is rotatably connected to a drive ring, the drive ring is connected to the motor, and the bottom of the rotating plate is fixedly connected to a gear ring; The connecting mechanism comprises a rotating shaft, a bevel gear, an arc-shaped tooth portion and a disc; the rotating shaft is rotatably arranged, the end of the rotating shaft is fixedly connected to the disc, and the circumference of the disc is fixedly connected to the arc-shaped tooth portion; the rotating shaft is fixedly connected to the bevel gear, and the driving ring is used to drive the bevel gear to drive the rotating shaft to rotate; Two connecting mechanisms are symmetrically arranged, one of the arc-shaped toothed portions on the two connecting mechanisms is meshed with the gear ring, and the other arc-shaped toothed portion is separated from the gear ring; the bevel gears on the two connecting mechanisms are respectively connected to the drive ring; The driving ring is used to drive the bevel gear to rotate, thereby rotating the rotating shafts on the two connecting mechanisms, so that the arc-shaped teeth on the two connecting mechanisms alternately mesh with the gear ring, and the rotating plate reciprocates.

[0007] Furthermore, the connecting mechanism also includes a linear drive device, which is installed on the housing. The output end of the linear drive device is rotatably connected to the rotating shaft. The linear drive device is used to push the rotating shaft to move linearly, so that one of the bevel gears is separated from the drive ring.

[0008] Furthermore, the bottom of the mounting plate is fixedly connected to the bottom shell, and the motor is installed in the bottom shell; The driving ring includes a hollow portion, a connecting portion and a bevel gear portion; the hollow portion is rotatably mounted on the support column, the bottom of the support column is fixedly connected to the bottom shell, the hollow portion is fixedly connected to the bevel gear portion through the connecting portion, the bevel gear portion is provided with a tooth portion meshing with the bevel gear, and the output end of the motor is connected to the hollow portion for driving the hollow portion to rotate.

[0009] Furthermore, the connection mechanism also includes a support block, the support block is fixedly connected to the mounting plate, and the rotating shaft can rotatably pass through the support block.

[0010] Further, the adjustment mechanism includes a slide plate, a fixed plate, a gear set and an intermediate column; The top of the housing is fixedly connected to the fixed plate, the fixed plate is slidably connected to the slide plate, the rotating column is rotatably connected to the slide plate, the bottom of the rotating column passes through the slide plate, the bottom of the rotating column is fixedly connected to the gear set, and the gear set includes a plurality of gears with different diameters; The top of the rotating plate is fixedly connected to the middle column, and a driving gear is sleeved on the middle column. The driving gear meshes with one of the gears in the gear set. The slide plate is used to slide to adjust the position so that different gears in the gear set mesh with the driving gear respectively.

[0011] Furthermore, the bottom of the mounting plate is fixedly connected to a walking device.

[0012] Furthermore, the top of the rotating column is fixedly connected to a telescopic rod, and the top of the telescopic rod is connected to the image acquisition device.

[0013] The present invention has the following beneficial effects: the present invention makes full use of the photos collected during the construction, quickly identifies the broken area of ​​the tunnel face, and improves the efficiency and accuracy of image recognition, so that the determination of the broken area of ​​the tunnel face is more timely, and provides technical support for engineering personnel to take engineering measures in a timely manner, which has strong promotion and guiding significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flow chart of the detection method of the present invention; Figure 2 This is an example of the original tunnel face image and its broken area annotation; Figure 3 This is an example of a broken area image after processing with the Unet-transformer technique; Figure 4 is a schematic diagram of a detection device of the present invention; Figure 5 is a schematic diagram of the connection mechanism; Figure 6 It is a schematic diagram of one of the bevel gears disengaging from the drive ring; Figure 7 is a schematic diagram of an arc-shaped tooth portion. DETAILED DESCRIPTION

[0015] The following will be combined with the embodiments of the present invention Figure 1-Figure 7 , the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0016] The detection method of the present invention is a method for identifying the broken rock area of ​​a deep buried tunnel based on the Unet-transformer technology. The method uses deep learning technology to train the U-net neural network model to process the images of the tunnel face containing broken rock mass taken at the construction site, thereby automatically identifying and extracting the broken rock mass area, while eliminating the interference of other equipment and facilities on site. Through this method, the photos collected during the construction can be fully utilized to quickly identify the broken area of ​​the tunnel face, and the efficiency and accuracy of image recognition can be improved, so that the determination of the broken area of ​​the tunnel face is more timely, providing technical support for engineering personnel to take engineering measures in a timely manner, and having a strong promotion and guiding significance.

[0017] like Figure 1-Figure 3, a method for detecting a tunnel surrounding rock crushing area, comprising the following steps: Step 1: Tunnel face image acquisition At the construction site, the image acquisition device 5 of the detection device acquires a local image of the tunnel face including the broken rock mass area, and synthesizes multiple local images into a panoramic image. These images should be as clear as possible and cover scenes in various construction environments as much as possible.

[0018] Step 2: Image Annotation Use Labelme, an image annotation tool, to clearly annotate the acquired images. During the annotation process, it is necessary to clearly distinguish between the broken rock areas and the non-target areas. After the annotation is completed, these images are divided into training sets and test sets for subsequent model training and performance evaluation.

[0019] Step 3: Data Preparation The labeled image data is divided into two parts, one is the training set for model training, and the other is the test set for testing the model effect.

[0020] Step 4: Model training A neural network combining U-Net and Transformer is used for image segmentation training. This model integrates the precise positioning capability of U-Net and the global context capture capability of Transformer. The U-net network structure consists of an encoder (compression path) and a decoder (expansion path). The encoder is responsible for extracting image features through continuous convolution and pooling operations, while the decoder restores the spatial details of the image through continuous deconvolution and upsampling and generates the final segmentation result. The Transformer model is a deep learning model based on the self-attention mechanism, which can process all elements in the sequence in parallel, which greatly improves the training efficiency of the model.

[0021] Step 4-1: Data input: The labeled training set images are input into the U-Net and Transformer model. The U-Net part is responsible for extracting local features of the image, while the Transformer part is used to capture the global context information of the image. Step 4-2, encoder processing: The features of the image of the fractured rock mass area of ​​the tunnel face are extracted through several convolutional layers and pooling layers. The convolutional layer extracts the local features of the fractured area, and the pooling layer reduces the spatial resolution of the image to generate a feature map. At the same time, the Transformer encoder further extracts global features through the self-attention mechanism; The convolution operation uses the following formula: , in, I represents the tunnel face image acquired under the input complex working conditions, K represents the convolution kernel, represents the convolution operation, and Both represent the position index of the palm face feature map output after the convolution operation, and m and n both represent the position index of the convolution kernel; Use the nonlinear activation function ReLU: , The pooling operation is used to reduce the spatial resolution of the feature map: , in, represents the pooled face feature map, Represents the position index of the face feature map output after the pooling operation, and Represents the position index of the pooling window, Indicates the size of the pooling window; Step 4-3, bottleneck layer processing: The deepest convolutional layer further abstracts the broken area features here to generate high-level feature representation; Step 4-4, decoder processing: The spatial resolution of the image is gradually restored through deconvolution layers and upsampling layers. Each deconvolution layer is combined with the feature map of the corresponding layer of the encoder to retain high-resolution features. The Transformer decoder further refines the segmentation results through self-attention layers and cross-attention layers to enhance the model's ability to recognize the boundaries of the target area; The deconvolution operation uses the following formula: , in, represents the feature map of the palm face after the deconvolution operation, represents the deconvolution kernel, represents the deconvolution operation, and represents the position index of the face feature map output after the deconvolution operation, and Both represent the position index of the deconvolution kernel; Step 4-5, jump connection: During the decoding process, the jump connection directly passes the encoder's high-resolution fragmentation area feature map to the corresponding layer of the decoder, enhancing the model's ability to recognize the boundary of the target area. The feature map cascade formula is: , in, It represents the characteristic diagram of the tunnel face after cascading. represents the face feature map from the encoder, represents the palm face feature map from the decoder; Steps 4-6, loss calculation and optimization: The cross entropy loss function is used to calculate the difference between the model's prediction results for the broken rock area and the actual annotation, and the model parameters are adjusted through the optimization algorithm (Adam optimization algorithm) to reduce the loss. The formula of the cross entropy loss function is: , in, L represents the cross entropy loss function, N represents the number of samples of the face image, Indicates palm face images, Indicates The true labels of the palm face images, Indicates The predicted probability of a tunnel face image.

[0022] Step 5: Model Validation The broken area images of the test set are input into the trained model to evaluate the actual performance of the model. The accuracy is mainly used to measure the effect of the model on the test set, and the model with the best performance is selected as the final preferred model.

[0023] Step 6: Save the best model The model that has been evaluated and confirmed as the best will be saved for subsequent reuse.

[0024] Step 7: Newly captured tunnel face image with broken area input New images of the tunnel face containing the broken rock mass area taken at the construction site are input into the saved optimal model. The model will automatically process these images to identify and extract the broken areas.

[0025] Step 8: Extraction of the broken area By using the computing power of the model, the broken rock mass area is identified from the input image, and other construction equipment and facilities in the image are excluded. Finally, a pure broken area image is output, providing a basis for the 3D reconstruction of the tunnel face.

[0026] like Figure 4-Figure 7 , a tunnel surrounding rock crushing zone detection device, comprising a mounting plate 1, a housing 2, a rotating column 3, an image acquisition device 5, a rotating plate 6, a supporting column 7, a motor 9, a connecting mechanism 11 and an adjusting mechanism; The housing 2 is mounted on the mounting plate 1, the bottom of the rotating column 3 is rotatably arranged in the housing 2, the top of the rotating column 3 passes through the housing 2, the top of the rotating column 3 is connected to the image acquisition device 5, the bottom of the rotating column 3 is connected to the rotating plate 6, and the bottom of the rotating plate 6 is rotatably connected to the top of the supporting column 7; The motor 9 is connected to the rotating plate 6 via the connecting mechanism 11, and is used to drive the rotating plate 6 to reciprocate or rotate with interval pauses; The rotating plate 6 is connected to the rotating column 3 via the adjusting mechanism, and is used to adjust the rotation angle of the rotating column 3 .

[0027] Specifically, the image acquisition device 5 is a device of the prior art, such as a video camera, a camera, etc., which can realize image acquisition. The housing 2 is a structure with an opening facing downward, and can be mounted on the mounting plate 1 by bolts. The rotating column 3 and the supporting column 7 are vertically arranged and coaxial, the supporting column 7 plays a supporting function, and the rotating column 3 plays a function of driving the image acquisition device 5 to rotate.

[0028] In specific implementation, the entire detection device is placed or moved to the target position so that the lens direction of the image acquisition device 5 faces the palm face. Then the motor 9 is started, and the rotating plate 6 rotates through the connecting mechanism 11, so that the image acquisition device 5 generates two kinds of movement modes according to the needs: Reciprocating rotation: At this time, the image acquisition device 5 performs reciprocating rotation in a fan-shaped range, that is, reciprocating swing, and the swing direction is the left and right direction of the tunnel face. In this motion mode, the image acquisition device 5 collects video. The collected video is a full-view video of the tunnel face and its surrounding rock wall.

[0029] Interval pause rotation: At this time, the image acquisition device 5 rotates a fixed angle in sequence within a sector range. Each time the image acquisition device 5 rotates a fixed angle, it pauses slightly and then rotates the fixed angle again, and reciprocates within the sector range. In this motion mode, the image acquisition device 5 captures a local image of the palm face when pausing. The local image includes a portion of the palm face, and multiple local images of the palm face can be combined into a panoramic image in the left and right directions.

[0030] The multiple local images collected by the image acquisition device 5 can be synthesized into a panoramic image by using software such as Photoshop and post-stack synthesis processing. The panoramic image has the characteristics of high resolution and includes the entire and adjacent range of the tunnel face, which is convenient for subsequent identification of broken rock areas. The video collected by the image acquisition device 5 can be used as a reference. With reference to the construction environment of the tunnel face, the panoramic image can be compared through the video to promptly discover defects such as missing, distortion, and deformation of the panoramic image, and the panoramic image can be modified in a timely manner.

[0031] The present invention can realize unified collection of images and videos of the tunnel face. The image acquisition device 5 can realize the functions of intermittent rotation and reciprocating rotation through the transmission of the connecting mechanism 11. It can take multiple local images of the tunnel face to synthesize a high-resolution panoramic image, which is convenient for subsequent identification of broken rock areas. It can take reference videos to record the construction environment where the images are taken and the surrounding rock walls.

[0032] The rotation angle, that is, the rotation amplitude, of the image acquisition device 5 is adjusted by the adjustment mechanism to adapt to tunnels of different sizes and their faces.

[0033] In addition, the bottom of the mounting plate 1 is fixedly connected to a walking device, which is a prior art and can be four Mecanum wheels or walking wheels to realize automatic walking of the entire device. A corresponding camera can be installed on the mounting plate 1 to remotely control the walking of the entire device. Remotely controlling the walking device to walk is a prior art, and a manual method can also be used, in which the walking device is a roller, to manually move the entire device.

[0034] Furthermore, the mounting plate 1 is rotatably connected to a driving ring 8, the driving ring 8 is connected to the motor 9, and the bottom of the rotating plate 6 is fixedly connected to a gear ring 12; The connecting mechanism 11 includes a rotating shaft 112, a bevel gear 114, an arc-shaped tooth portion 115 and a disc 116; the rotating shaft 112 is rotatably arranged, the end of the rotating shaft 112 is fixedly connected to the disc 116, and the circumference of the disc 116 is fixedly connected to the arc-shaped tooth portion 115; the rotating shaft 112 is fixedly connected to the bevel gear 114, and the driving ring 8 is used to drive the bevel gear 114 to drive the rotating shaft 112 to rotate; Two connecting mechanisms 11 are symmetrically arranged, one of the arc-shaped toothed portions 115 on the two connecting mechanisms 11 is meshed with the gear ring 12, and the other arc-shaped toothed portion 115 is separated from the gear ring 12; the bevel gears 114 on the two connecting mechanisms 11 are respectively connected to the drive ring 8; The driving ring 8 is used to drive the bevel gear 114 to rotate, thereby rotating the shafts 112 on the two connecting mechanisms 11, and making the arc-shaped teeth 115 on the two connecting mechanisms 11 alternately mesh with the gear ring 12, so that the rotating plate 6 reciprocates.

[0035] Furthermore, the connecting mechanism 11 also includes a linear drive device 111, which is installed on the housing 2. The output end of the linear drive device 111 is rotatably connected to the rotating shaft 112. The linear drive device 111 is used to push the rotating shaft 112 to move linearly, so that one of the bevel gears 114 is separated from the drive ring 8.

[0036] Specifically, the rotating shaft 112 is arranged horizontally, and the rotating shaft 112, the bevel gear 114 and the disc 116 are coaxially distributed. The opposite ends of the two rotating shafts 112 are respectively fixedly connected to the corresponding discs 116. Two arc-shaped tooth portions 115 are arranged one above and one below, and only one arc-shaped tooth portion 115 is engaged with the ring gear 12 at the same time. The ring gear 12 is annular, and its teeth face downward. The linear drive device 111 is a prior art, such as a cylinder, an electric cylinder, etc.

[0037] When the image acquisition device 5 reciprocates: the two linear drive devices 111 are in the initial state, the two bevel gears 114 are both engaged with the drive ring 8, when the motor 9 is started, the drive ring 8 rotates, and then the two corresponding rotating shafts 112 are rotated through the two bevel gears 114, so that the disk 116 and the arc-shaped tooth portion 115 rotate. When one of the arc-shaped tooth portions 115 engages with the ring gear 12, the ring gear 12 and the rotating plate 6 rotate forward by a certain amplitude. When the arc-shaped tooth portion 115 moves to disengage from the ring gear 12, the other arc-shaped tooth portion 115 engages with the ring gear 12, so that the ring gear 12 rotates in the opposite direction. Thus, through the alternating engagement of the two arc-shaped tooth portions 115, the ring gear 12 and the rotating plate 6 can be reciprocated. It should be noted that the two arc-shaped teeth 115 are located on both sides of the gear ring 12, and the rotation directions of the two arc-shaped teeth 115 are always the same, that is, the rotation directions of the two rotating shafts 112 are always the same, so that the two arc-shaped teeth 115 push the gear ring 12 to rotate in opposite directions. The arc-shaped teeth 115 are distributed in an arc shape, and their arc length is less than 180°.

[0038] like Figure 6 , when the image acquisition device 5 stops rotating at intervals: for the convenience of explanation, Figure 6 Taking the action of the linear drive device 111 on the right side as an example, the linear drive device 111 on the right side is actuated to disengage the bevel gear 114 on the right side from the drive ring 8, so that only the bevel gear 114 on the left side rotates. Since there is a gap between the disc 116 and the ring gear 12, when the arc-shaped tooth portion 115 on the left side drives the ring gear 12 to rotate, the ring gear 12 rotates a fixed angle, and when the arc-shaped tooth portion 115 moves to the bottom of the ring gear 12, the ring gear 12 is stationary, thereby forming a function of interval pause rotation, each rotation is a fixed angle, and after rotating to a fixed sector range, the linear drive device 111 on the right side returns to its initial state, and the linear drive device 111 on the left side is actuated to make the arc-shaped tooth portion 115 on the left side drive the ring gear 12, thereby forming a reciprocating rotation function.

[0039] The rotation range of the gear ring 12 and the rotating plate 6 is a sector-shaped range, which is less than 180°.

[0040] Furthermore, the bottom of the mounting plate 1 is fixedly connected to the bottom housing, and the motor 9 is installed in the bottom housing; The driving ring 8 includes a hollow portion 801, a connecting portion 802 and a bevel gear portion 803; the hollow portion 801 is rotatably mounted on the support column 7, the bottom of the support column 7 is fixedly connected to the bottom shell, the hollow portion 801 is fixedly connected to the bevel gear portion 803 through the connecting portion 802, the bevel gear portion 803 is provided with a tooth portion meshing with the bevel gear 114, and the output end of the motor 9 is connected to the hollow portion 801 for driving the hollow portion 801 to rotate.

[0041] Specifically, the driving ring 8 is an annular structure, and the bottom of the hollow part 801 can rotatably pass through the mounting plate 1. Gears can be set on the hollow part 801 to connect to the motor 9 through the meshing of the gears, or they can be connected through belts and pulleys.

[0042] It should be noted that the bevel tooth portion 803 is located at the top of the connecting portion 802, and the connecting portion 802 forms an annular groove structure. The bevel tooth portion 803 is distributed on the outer top surface and the inner top surface of the groove structure. The bevel tooth portions 803 on the outer top surface and the inner top surface respectively engage with two bevel gears 114, thereby realizing the function that the two bevel gears 114, that is, the two rotating shafts 112 always rotate in the same direction, so that the gear ring 12 reciprocates through the transmission of the arc-shaped tooth portion 115.

[0043] In addition, the hollow portion 801 , the connecting portion 802 , and the support column 7 are coaxially distributed, and the two rotating shafts 112 are located on both sides of the support column 7 and in the radial direction of the support column 7 .

[0044] Furthermore, the connection mechanism 11 further comprises a support block 113, the support block 113 is fixedly connected to the mounting plate 1, and the rotating shaft 112 can rotatably pass through the support block 113. The support block 113 plays a supporting function, so that the rotation of the rotating shaft 112 is more stable.

[0045] Furthermore, the adjustment mechanism includes a slide plate 13, a fixing plate 14, a gear set 15 and an intermediate column 16; The top of the housing 2 is fixedly connected to the fixed plate 14, the fixed plate 14 is slidably connected to the slide plate 13, the rotating column 3 is rotatably connected to the slide plate 13, the bottom of the rotating column 3 passes through the slide plate 13, and the bottom of the rotating column 3 is fixedly connected to the gear set 15, and the gear set 15 includes a plurality of gears with different diameters; The top of the rotating plate 6 is fixedly connected to the middle column 16, and a driving gear is sleeved on the middle column 16. The driving gear meshes with one of the gears in the gear set 15. The slide plate 13 is used to slide to adjust the position so that different gears in the gear set 15 mesh with the driving gear respectively.

[0046] Specifically, the slide plate 13 can be connected to the fixed plate 14 by means of an electric slide rail or a slide groove, and the position of the slide plate 13 can be locked by bolts. The gears in the gear set 15 are arranged side by side up and down along the axial direction of the rotating column 3. A square through hole is opened on the top of the housing 2, and the rotating column 3 passes through the square through hole, which facilitates the front and rear adjustment of the rotating column 3.

[0047] The gears of different diameters in the gear set 15 mesh with the driving gear of the intermediate column 16 , thereby changing the transmission ratio to achieve different rotation amplitudes of the rotating column 3 , and ultimately achieving different rotation amplitudes of the image acquisition device 5 .

[0048] Furthermore, the top of the rotating column 3 is fixedly connected to the telescopic rod 4, the top of the telescopic rod 4 is connected to the image acquisition device 5, the height of the image acquisition device 5 is adjusted by the telescopic rod 4, and the top of the telescopic rod 4 can be connected to the image acquisition device 5 through a hinge seat, thereby adjusting the pitch angle of the image acquisition device 5.

[0049] The embodiments described above are only descriptions of the preferred modes of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations, modifications, and substitutions made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for detecting a tunnel surrounding rock crushing area, characterized in that: The following steps are involved: Step 1, tunnel face image acquisition: a plurality of local images of the tunnel face including the broken rock mass area are acquired by an image acquisition device (5) of the detection device, and the plurality of local images are synthesized into a panoramic image; Step 2, image annotation: clearly annotate the image to distinguish the broken rock area from the non-target area; Step 3, data preparation: Divide the labeled image data into two parts: Training set: used for model training; Test set: used to test the model effect; Step 4, model training: use a neural network combining U-Net and Transformer for image segmentation training; Step 5, model verification: input the broken area images of the test set into the trained model to evaluate the actual performance of the model; Step 6, save the optimal model: save the model that has been evaluated and confirmed as the optimal one; Step 7, inputting a newly shot tunnel face image containing a broken area: inputting a newly shot tunnel face image containing a broken rock mass area at the construction site into the saved optimal model; Step 8: Broken area extraction: Using the computing power of the model, the broken rock area is identified from the input image and other construction equipment and facilities in the image are excluded; finally, a pure broken area image is output.

2. A method for detecting a broken area of ​​surrounding rock in a tunnel according to claim 1, characterized in that: The step 4 comprises: Step 4-1, data input: input the labeled training set images into the model combining U-Net and Transformer. The U-Net part is used to extract local features of the image, and the Transformer part is used to capture the global context information of the image. Step 4-2, encoder processing: extract the features of the image of the broken rock mass area of ​​the tunnel face through several convolution layers and pooling layers. The convolution layer extracts the local features of the broken area, and the pooling layer reduces the spatial resolution of the image to generate a feature map. The convolution operation uses the following formula: , in, I represents the tunnel face image acquired under the input complex working conditions, K represents the convolution kernel, represents the convolution operation, and Both represent the position index of the palm face feature map output after the convolution operation, and m and n both represent the position index of the convolution kernel; Use the nonlinear activation function ReLU: , The pooling operation is used to reduce the spatial resolution of the feature map: , in, represents the pooled face feature map, Represents the position index of the face feature map output after the pooling operation, and Represents the position index of the pooling window, Indicates the size of the pooling window; Step 4-3, bottleneck layer processing: The deepest convolutional layer further abstracts the broken area features here to generate high-level feature representation; Step 4-4, decoder processing: The spatial resolution of the image is gradually restored through deconvolution layers and upsampling layers. Each deconvolution layer is combined with the feature map of the corresponding layer of the encoder to retain high-resolution features. The deconvolution operation uses the following formula: , in, represents the feature map of the palm face after the deconvolution operation, represents the deconvolution kernel, represents the deconvolution operation, and represents the position index of the face feature map output after the deconvolution operation, and Both represent the position index of the deconvolution kernel; Step 4-5, jump connection: During the decoding process, the jump connection directly passes the encoder's high-resolution fragmentation area feature map to the corresponding layer of the decoder, enhancing the model's ability to recognize the boundary of the target area. The feature map cascade formula is: , in, It represents the characteristic diagram of the tunnel face after cascading. represents the face feature map from the encoder, represents the palm face feature map from the decoder; Steps 4-6, loss calculation and optimization: The cross entropy loss function is used to calculate the difference between the model's prediction results for the broken rock area and the actual annotation, and the model parameters are adjusted through the optimization algorithm. The cross entropy loss function formula is: , in, L represents the cross entropy loss function, N represents the number of samples of the face image, Indicates palm face images, Indicates The true labels of the palm face images, Indicates The predicted probability of a tunnel face image.

3. A tunnel surrounding rock crushing zone detection device, characterized in that: It comprises a mounting plate (1), a housing (2), a rotating column (3), an image acquisition device (5), a rotating plate (6), a supporting column (7), a motor (9), a connecting mechanism (11) and an adjusting mechanism; The housing (2) is mounted on the mounting plate (1), the bottom of the rotating column (3) is rotatably arranged in the housing (2), the top of the rotating column (3) passes through the housing (2), the top of the rotating column (3) is connected to the image acquisition device (5), the bottom of the rotating column (3) is connected to the rotating plate (6), and the bottom of the rotating plate (6) is rotatably connected to the top of the supporting column (7); The motor (9) is connected to the rotating plate (6) via the connecting mechanism (11), and is used to drive the rotating plate (6) to reciprocate or rotate with interval pauses; The rotating plate (6) is connected to the rotating column (3) via the adjusting mechanism, and is used to adjust the rotation angle of the rotating column (3).

4. A tunnel surrounding rock crushing zone detection device according to claim 3, characterized in that: The mounting plate (1) is rotatably connected to a drive ring (8), the drive ring (8) is connected to the motor (9), and the bottom of the rotating plate (6) is fixedly connected to a gear ring (12); The connecting mechanism (11) comprises a rotating shaft (112), a bevel gear (114), an arc-shaped tooth portion (115) and a disc (116); the rotating shaft (112) is rotatably arranged, the end of the rotating shaft (112) is fixedly connected to the disc (116), and the circumferential surface of the disc (116) is fixedly connected to the arc-shaped tooth portion (115); the rotating shaft (112) is fixedly connected to the bevel gear (114), and the driving ring (8) is used to drive the bevel gear (114) to drive the rotating shaft (112) to rotate; Two connecting mechanisms (11) are symmetrically arranged, one of the arc-shaped tooth portions (115) on the two connecting mechanisms (11) is meshed with the gear ring (12), and the other arc-shaped tooth portion (115) is separated from the gear ring (12); the bevel gears (114) on the two connecting mechanisms (11) are respectively connected to the drive ring (8); The driving ring (8) is used to drive the bevel gear (114) to rotate, thereby causing the rotating shafts (112) on the two connecting mechanisms (11) to rotate, causing the arc-shaped teeth (115) on the two connecting mechanisms (11) to alternately mesh with the gear ring (12), thereby causing the rotating plate (6) to reciprocate.

5. A tunnel surrounding rock crushing zone detection device according to claim 4, characterized in that: The connecting mechanism (11) further comprises a linear drive device (111), wherein the linear drive device (111) is mounted on the housing (2), wherein an output end of the linear drive device (111) is rotatably connected to the rotating shaft (112), and the linear drive device (111) is used to push the rotating shaft (112) to move linearly, so that one of the bevel gears (114) is separated from the drive ring (8).

6. A tunnel surrounding rock crushing zone detection device according to claim 4, characterized in that: The bottom of the mounting plate (1) is fixedly connected to the bottom shell, and the motor (9) is mounted in the bottom shell; The driving ring (8) comprises a hollow portion (801), a connecting portion (802) and a bevel gear portion (803); the hollow portion (801) is rotatably sleeved on the supporting column (7); the bottom of the supporting column (7) is fixedly connected to the bottom shell; the hollow portion (801) is fixedly connected to the bevel gear portion (803) via the connecting portion (802); the bevel gear portion (803) is provided with a tooth portion meshing with the bevel gear (114); and the output end of the motor (9) is connected to the hollow portion (801) for driving the hollow portion (801) to rotate.

7. A tunnel surrounding rock crushing zone detection device according to claim 4, characterized in that: The connection mechanism (11) further comprises a support block (113), wherein the support block (113) is fixedly connected to the mounting plate (1), and the rotating shaft (112) can rotatably pass through the support block (113).

8. The tunnel surrounding rock crushing zone detection device according to claim 4, characterized in that: The adjustment mechanism comprises a slide plate (13), a fixing plate (14), a gear set (15) and an intermediate column (16); The top of the housing (2) is fixedly connected to the fixed plate (14), the fixed plate (14) is slidably connected to the slide plate (13), the rotating column (3) is rotatably connected to the slide plate (13), the bottom of the rotating column (3) passes through the slide plate (13), the bottom of the rotating column (3) is fixedly connected to the gear set (15), and the gear set (15) comprises a plurality of gears with different diameters; The top of the rotating plate (6) is fixedly connected to the intermediate column (16), and a driving gear is sleeved on the intermediate column (16). The driving gear meshes with one of the gears in the gear set (15), and the slide plate (13) is used to slide to adjust the position so that different gears in the gear set (15) respectively mesh with the driving gear.

9. The tunnel surrounding rock crushing zone detection device according to claim 1, characterized in that: The bottom of the mounting plate (1) is fixedly connected to a walking device.

10. The tunnel surrounding rock crushing zone detection device according to claim 1, characterized in that: The top of the rotating column (3) is fixedly connected to a telescopic rod (4), and the top of the telescopic rod (4) is connected to the image acquisition device (5).

Images