Coal mine tunnel point cloud segmentation method based on interactive grouping linear attention mechanism

Abstract

The application provides a coal mine tunnel point cloud segmentation method based on an interactive grouping linear attention mechanism, and relates to the technical field of computer vision, and the method comprises the following steps: according to a reference point in a neighborhood of an arbitrarily selected query point in coal mine tunnel point cloud data, a local coordinate system is constructed to describe three angle features between a normal vector of the query point and a normal vector of the reference point, and a distance feature of the query point and the reference point; high-dimensional feature coding is performed to construct a high-dimensional feature vector; an interactive grouping linear vector attention mechanism is introduced to map the high-dimensional feature vector into a value vector, channels of the value vector are divided into multiple groups, linear changes are applied in each group, and interaction between high-dimensional features between groups is performed to obtain a target feature vector; and a multilayer perception machine is used for semantic segmentation to obtain a segmentation category of the coal mine tunnel point cloud data. Therefore, without increasing equipment and computing capacity in the coal mine tunnel, efficient and accurate segmentation of point cloud data in a complex scene of the coal mine tunnel can be realized.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a method for segmenting point clouds in coal mine roadways based on an interactive grouped linear attention mechanism. Background Technology

[0002] The application of point cloud segmentation in coal mine development and management has gradually attracted attention, as coal mine point cloud segmentation methods can identify different targets inside the mine. In related technologies, point cloud segmentation methods for underground tunnel scenes often rely on rich attribute information, such as spatial distribution, color, normal vector, intensity, and density. However, there is currently a lack of semantic datasets for coal mine roadways, and the environment of coal mine roadways is complex and varied, containing objects and scenes of different shapes, sizes, and materials, such as dim lighting and messy pipelines. This makes it difficult to guarantee the accuracy of point cloud segmentation methods for coal mine roadways. Summary of the Invention

[0003] The present invention aims to at least partially solve one of the technical problems in the related art.

[0004] Therefore, the first objective of this invention is to propose a point cloud segmentation method for coal mine roadways based on an interactive grouped linear attention mechanism, which can effectively achieve efficient and accurate segmentation of point cloud data in complex coal mine roadway scenarios without increasing the equipment and computing power in the coal mine roadways.

[0005] The second objective of this invention is to propose a point cloud segmentation device for coal mine roadways based on an interactive grouped linear attention mechanism.

[0006] The third objective of this invention is to provide an electronic device.

[0007] The fourth objective of this invention is to provide a non-transitory computer-readable storage medium storing computer instructions.

[0008] To achieve the above objectives, a first aspect of the present invention proposes a method for segmenting point clouds in coal mine roadways based on an interactive grouped linear attention mechanism, the method comprising:

[0009] Based on an arbitrarily selected query point and a reference point within the neighborhood of the query point in the coal mine roadway point cloud data, a local coordinate system is constructed to describe the quaternary features between the query point and the reference point. These quaternary features include three angular features between the normal vector of the query point and the normal vector of the reference point. , , and the distance features between the query point and the reference point. ;

[0010] For the quaternary features ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors;

[0011] An interactive grouped linear vector attention mechanism is introduced to divide the channels that map the high-dimensional feature vector to the value vector into multiple groups, and apply linear transformation within each group. Furthermore, the interaction of high-dimensional features between groups is carried out through global attention weights to obtain the target feature vector of the interaction between linear transformation and high-dimensional features. In this case, the linear transformation weights within each group are shared when applying linear transformation within each group.

[0012] Multilayer perceptron is used to perform semantic segmentation on the target feature vector to obtain the segmentation categories of coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines.

[0013] To achieve the above objectives, a second aspect of the present invention provides a coal mine roadway point cloud segmentation device based on an interactive grouped linear attention mechanism, the device comprising:

[0014] The construction module is used to construct a local coordinate system based on an arbitrarily selected query point and a reference point within the neighborhood of the query point in the coal mine roadway point cloud data. This system describes the quaternary features between the query point and the reference point, whereby the quaternary features include three angular features between the normal vector of the query point and the normal vector of the reference point. , , and the distance features between the query point and the reference point. ;

[0015] Encoding module, used for encoding the four features ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors;

[0016] The interaction module is used to introduce an interactive grouped linear vector attention mechanism to divide the channels of the high-dimensional feature vector into value vectors into multiple groups, apply linear transformations within each group, and interact high-dimensional features between groups through global attention weights to obtain the target feature vector of the interaction between linear transformations and high-dimensional features. The linear transformation weights within each group are shared when applying linear transformations within each group.

[0017] The semantic segmentation module is used to perform semantic segmentation on the target feature vector using a multilayer perceptron to obtain the segmentation categories after segmenting the coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines.

[0018] This invention discloses a coal mine roadway point cloud segmentation method, apparatus, electronic device, and storage medium based on an interactive grouped linear attention mechanism. The method constructs a local coordinate system based on an arbitrarily selected query point and a reference point within its neighborhood in the coal mine roadway point cloud data. This system describes the three angular features between the normal vector of the query point and the normal vector of the reference point, as well as the distance features between the query point and the reference point. High-dimensional feature vectors are constructed through high-dimensional feature encoding. An interactive grouped linear vector attention mechanism is introduced to divide the channels that map the high-dimensional feature vectors into value vectors into multiple groups. Linear transformations are applied within each group, and high-dimensional features interact between groups to obtain the target feature vector. A multilayer perceptron is used for semantic segmentation to obtain the segmentation category of the coal mine roadway point cloud data. Therefore, without increasing the equipment and computing power within the coal mine roadway, efficient and accurate segmentation of point cloud data in complex coal mine roadway scenarios is effectively achieved.

[0019] To achieve the above objectives, a third aspect of the present invention provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0020] To achieve the above objectives, a fourth aspect of the present invention provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method described in the first aspect.

[0021] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0023] Figure 1 This is a flowchart illustrating a point cloud segmentation method for coal mine roadways based on an interactive grouped linear attention mechanism, provided in an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram illustrating the segmentation categories of point cloud data for coal mine roadways provided in an embodiment of the present invention.

[0025] Figure 3 This is a flowchart illustrating the technical process of a coal mine roadway point cloud segmentation method based on an interactive grouped linear attention mechanism, as provided in an embodiment of the present invention.

[0026] Figure 4This is a schematic diagram of a point cloud segmentation device for coal mine roadways based on an interactive grouped linear attention mechanism, provided in an embodiment of the present invention. Detailed Implementation

[0027] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0028] It should be noted that the acquisition, storage, use, and processing of data in the technical solution of this invention all comply with the relevant provisions of relevant laws and regulations.

[0029] The following describes, with reference to the accompanying drawings, a method and apparatus for segmenting coal mine roadway point clouds based on an interactive grouped linear attention mechanism.

[0030] Figure 1 This is a flowchart illustrating a point cloud segmentation method for coal mine roadways based on an interactive grouped linear attention mechanism, as provided in an embodiment of the present invention.

[0031] like Figure 1 As shown, the method includes the following steps:

[0032] Step 101: Based on an arbitrarily selected query point in the coal mine roadway point cloud data and a reference point within the query point's neighborhood, construct a local coordinate system to describe the quaternary features between the query point and the reference point. The quaternary features include three angular features between the normal vector of the query point and the normal vector of the reference point. , , and the distance features between the query point and the reference point. .

[0033] In some possible implementations, a local coordinate system is constructed to describe the quaternary features between the query point and the reference point, based on an arbitrarily selected query point in the coal mine roadway point cloud data and a reference point within the neighborhood of the query point. This includes: arbitrarily selecting a query point in the coal mine roadway point cloud data. and a reference point within the neighborhood of the query point. and according to and and their respective normal vectors and Construct a local coordinate system; calculate in the local coordinate system. normal vector and normal vector The relative deviation between them, i.e. normal vector and normal vector The three angular features between , , and the distance features between the query point and the reference point. To serve as a description of the query point With reference point Quaternary features between ( , , , );

[0034] in, ;

[0035] .

[0036] The coal mine roadway point cloud data was collected from a section of an underground roadway, for example, a total of 27,077,467 points. The coal mine roadway point cloud data can be acquired using a scanner. To ensure the reliability of segmentation training and evaluation, the coal mine roadway point cloud data is divided into a training set and a test set, with a random segmentation ratio of 8:2, but not limited to this.

[0037] Step 102, for the four-element features ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors.

[0038] In some possible implementations, for quaternary features ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors, including: representing the quaternary features of all query points in the coal mine roadway point cloud data as follows: Where i = 1, 2, ..., C represents the number of query points, and R represents the real number space; choose a number that follows a normal distribution. random projection B, Where m represents the dimension of the high-dimensional feature vector, and s represents the hyperparameter controlling the range of random number values; the high-dimensional feature vector after high-dimensional feature encoding of the query point's four-element features is represented as follows: ,in, .

[0039] Step 103 introduces an interactive grouped linear vector attention mechanism to divide the channels that map high-dimensional feature vectors to value vectors into multiple groups, and applies linear transformations within each group. Furthermore, the interaction of high-dimensional features between groups is achieved through global attention weights to obtain the target feature vector of the interaction between linear transformations and high-dimensional features. In this process, the linear transformation weights within each group are shared when applying linear transformations within each group.

[0040] In some possible implementations, an interactive grouped linear vector attention mechanism is introduced to divide the channels that map high-dimensional feature vectors to value vectors into multiple groups, and a linear transformation is applied within each group. Furthermore, the interaction of high-dimensional features between groups is achieved through global attention weights, resulting in a target feature vector from the interaction of linear transformation and high-dimensional features. This includes: dividing the channels that map high-dimensional feature vectors to value vectors into multiple groups using an interactive grouped linear vector attention mechanism, and applying a linear transformation within each group to obtain a linearly transformed value vector matrix; and employing a strip matrix. By learning the contextual information within each channel group and the information between channels, a global attention weight is constructed to interact with the high-dimensional features between groups on the value vector matrix, thereby obtaining the target feature vector; where, Involving There are two parameters, k represents different groups, and c represents the number of channels in the value vector.

[0041] .

[0042] Specifically, an interactive grouped linear vector attention mechanism is used to set up a high-dimensional feature vector... and the initial location information of the corresponding query point The initial points of the formation are gathered. The space is divided into several non-overlapping partitions using a grid strategy, and then different partitions are... Perform max pooling on the initial position information Average pooling is used for downsampling to obtain the point cloud. .

[0043] According to Dianyunji Calculate the vector attention value of the high-dimensional feature vector. And as a value vector, ,in, and They represent point clouds respectively. The high-dimensional feature vectors of the i-th and j-th query points. This represents the weighted feature of the j-th query point. This represents the Hadamard product. It is a relational function (e.g., subtraction). It is a sigmoid activation function. The location code represents the initial location information. .

[0044] Divide the value vector's channel count c into g groups, with each group containing the number of channels c. Within the same (attention) group, a linear transformation is applied within each group to obtain the linearly transformed value vector matrix. The value vectors share channels derived from the same linear transformation weights (attention weight parameters) used in the group attention. This approach maintains computational efficiency while capturing the interactions between features within a group. , involving Parameters, where,

[0045] .

[0046] Using a strip matrix By learning the contextual information within each channel group and the information between channels, a global attention weight is constructed to interact with the high-dimensional features between groups on the value vector matrix, thus obtaining the target feature vector.

[0047]

[0048] in, Involving There are two parameters, k represents different groups, and c represents the number of channels in the value vector.

[0049] At this point, the calculation of the global attention weights only needs to consider any one group. The interaction with the k neighboring groups is sufficient, and the global attention weight parameters are as follows: ,in, express The set of k adjacent groups, Learn the banded matrix of contextual information within each channel and information between channels.

[0050] In other possible implementations, the coverage of high-dimensional feature interactions between groups can be adjusted through cross-validation. Specifically, the coverage can be set to 6, which achieves the best segmentation performance. This not only reduces the complexity of channel calculations but also effectively captures group interactions between local channels, ensuring both efficiency and effectiveness.

[0051] Step 104: Use a multilayer perceptron to perform semantic segmentation on the target feature vector to obtain the segmentation categories of the coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines.

[0052] In some possible implementations, a multilayer perceptron is used to perform semantic segmentation on the target feature vector to obtain the segmentation categories of the coal mine roadway point cloud data. The segmentation categories include roof, floor, sidewall, ventilation ducts, and pipelines. This includes: extracting the point features of the target feature vector stored during the downsampling of the target feature vector by the multilayer perceptron for semantic segmentation, and mapping the point features to all points in the same set of points to obtain the segmentation categories of the coal mine roadway point cloud data. The segmentation categories include roof, floor, sidewall, ventilation ducts, and pipelines (pipes + cables).

[0053] Furthermore, such as Figure 2 As shown, Figure 2 This invention provides a schematic diagram of point cloud data segmentation categories for coal mine roadways. Figure 2 In the middle, 'a' represents the undivided point cloud data of coal mine roadways. Figure 2 The 'b' in the middle represents the segmentation category, which includes roof, floor, sidewall, ventilation duct, and pipe + cable (conduit), enabling accurate segmentation of point cloud data of coal mine roadways.

[0054] This invention discloses a coal mine roadway point cloud segmentation method based on an interactive grouped linear attention mechanism. The method constructs a local coordinate system based on an arbitrarily selected query point and a reference point within its neighborhood in the coal mine roadway point cloud data. This system describes the three angular features between the normal vector of the query point and the normal vector of the reference point, as well as the distance features between the query point and the reference point. High-dimensional feature vectors are constructed through high-dimensional feature encoding. An interactive grouped linear vector attention mechanism is introduced to map the high-dimensional feature vectors into value vectors, dividing the channels into multiple groups. Linear transformations are applied within each group, and high-dimensional features interact between groups to obtain the target feature vector. A multilayer perceptron is then used for semantic segmentation to determine the segmentation category of the coal mine roadway point cloud data. Therefore, without increasing equipment and computing power within the coal mine roadway, this method effectively achieves efficient and accurate segmentation of point cloud data in complex coal mine roadway scenarios.

[0055] In some embodiments, the present invention may use overall accuracy (OA), mean intersection between unions (mIoU), and mean accuracy within each class (mAcc) as evaluation metrics for class segmentation. In particular, mIoU provides the most reasonable evaluation for imbalanced datasets and can therefore be used as a priority measure. The calculation method for the evaluation metrics is as follows:

[0056]

[0057] Where K is the total number of classes and N is the total number of points. These represent the predicted values ​​for true positive, false positive, and false negative results, respectively.

[0058] Optionally, the coal mine roadway point cloud segmentation method based on the interactive grouped linear attention mechanism of the present invention can run on a high-performance computing cluster, using a graphics processor (48GB), a deep learning development environment, multiple deep learning frameworks supported by different programming languages, and a parallel computing platform and programming model. To mitigate overfitting, an optimizer (AdamW) can be used, and cross-entropy loss can be used as the loss function.

[0059] Therefore, the coal mine roadway point cloud segmentation method based on the interactive grouped linear attention mechanism of this invention was tested on a periodic coal mine roadway point cloud dataset. The overall accuracy (OA) reached 99.1%, the mean intersection (mIoU) between joints was 97.5%, and the mean accuracy (mAcc) was 97.5%. Among the five specific categories (roof, floor, face, ventilation ducts, and pipelines), the method of this invention achieved the highest Intersection over Union (IoU) score.

[0060] Furthermore, to better understand the present invention, in some possible embodiments, the present invention also provides a technical flowchart of a coal mine roadway point cloud segmentation method based on an interactive grouped linear attention mechanism, such as... Figure 3 As shown, the coal mine roadway point cloud segmentation method based on interactive grouped linear attention mechanism of the present invention mainly constructs a coal mine roadway point cloud segmentation network model based on interactive grouped linear attention mechanism. The core of this coal mine roadway point cloud segmentation network model is to construct an angle and distance feature encoding module and an interactive grouped linear attention module that represent local spatial relationships. Specifically, the angle and distance feature encoding module obtains the input coal mine roadway point cloud data (such as...) Figure 2 After (part a) of the above, a local coordinate system is constructed to describe the quaternary features between any selected query point and its neighboring points in the coal mine roadway point cloud data. The calculated quaternary features are then encoded using high-dimensional features to construct a high-dimensional feature vector for the query point. An interactive grouped linear attention module combines downsampling of the high-dimensional feature vector with point cloud partitioning, feature extraction, and interactive grouped linear attention calculation to output the target feature vector. Finally, a multilayer perceptron is used to perform semantic segmentation on the target feature vector to obtain the segmentation category of the coal mine roadway point cloud data, i.e., outputting the segmentation category of the coal mine roadway point cloud data (e.g., [missing information]). Figure 2 (Part b) of the model shows that the coal mine roadway point cloud segmentation network model is robust to the segmentation of complex shapes and noisy point clouds, and can provide strong support for subsequent digital modeling of coal mine roadway structures, safety monitoring and intelligent inspection.

[0061] The downsampling and point cloud partitioning process includes: dividing the space into several non-overlapping partitions using a grid strategy; then performing max pooling on the high-dimensional feature vectors of different partitions; and performing average pooling on the initial position information of the query points corresponding to the high-dimensional feature vectors to downsample and obtain a point cloud set. Feature extraction includes calculating the vector attention value of the high-dimensional feature vector based on the point cloud set and using it as the value vector. The number of channels c of the value vector is divided into g groups, and a linear transformation is applied within each group to obtain the linearly transformed value vector matrix. Interactive grouped linear attention calculation includes using a strip matrix. By learning the contextual information within each channel and the information between channels, a global attention weight is constructed to interact with the high-dimensional features between groups of the value vector matrix, thus obtaining the target feature vector.

[0062] In summary, the coal mine roadway point cloud segmentation method based on an interactive grouped linear attention mechanism of this invention achieves efficient and accurate segmentation of point cloud data in complex mining environments by comprehensively utilizing local coordinate system construction, high-dimensional feature encoding, grouped linear vector attention mechanism, and multilayer perceptron technology. This method is robust to complex shapes and noise, and can provide strong support for subsequent coal mine geological analysis and resource assessment.

[0063] To achieve the above embodiments, the present invention also proposes a coal mine roadway point cloud segmentation device based on an interactive grouped linear attention mechanism.

[0064] Figure 4 This is a schematic diagram of a point cloud segmentation device for coal mine roadways based on an interactive grouped linear attention mechanism, provided in an embodiment of the present invention.

[0065] like Figure 4 As shown, the coal mine roadway point cloud segmentation device 40 based on interactive grouped linear attention mechanism includes: a construction module 41, an encoding module 42, an interaction module 43, and a semantic segmentation module 44.

[0066] Construction module 41 is used to construct a local coordinate system based on an arbitrarily selected query point and a reference point within the neighborhood of the query point in the coal mine roadway point cloud data, to describe the four-element feature between the query point and the reference point. The four-element feature includes three angular features between the normal vector of the query point and the normal vector of the reference point. , , and the distance features between the query point and the reference point. ;

[0067] Encoding module 42 is used for encoding the four-element feature ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors;

[0068] Interaction module 43 is used to introduce an interactive grouped linear vector attention mechanism to divide the channels that map the high-dimensional feature vector into value vectors into multiple groups, apply linear transformations within each group, and interact high-dimensional features between groups through global attention weights to obtain the target feature vector of the interaction between linear transformations and high-dimensional features. The linear transformation weights within each group are shared when applying linear transformations within each group.

[0069] The semantic segmentation module 44 is used to perform semantic segmentation on the target feature vector using a multilayer perceptron to obtain the segmentation categories after segmentation of the coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines.

[0070] Furthermore, in one possible implementation of this invention, the construction module 41 is specifically used for:

[0071] Select any query point in the point cloud data of coal mine roadways. and a reference point within the neighborhood of the query point. and according to the and and their respective normal vectors and Construct a local coordinate system;

[0072] Calculate in the local coordinate system normal vector and normal vector The relative deviation between them, i.e. normal vector and normal vector The three angular features between , , and the distance features between the query point and the reference point. To serve as a description of the query point With reference point Quaternary features between ( , , , );

[0073] in, ;

[0074] .

[0075] Furthermore, in one possible implementation of this invention, the encoding module 42 is specifically used for:

[0076] The four-element feature of all query points in the coal mine roadway point cloud data is represented as follows: Where i = 1, 2, ..., C represents the number of query points, and R represents the real number space;

[0077] Choose one that follows a normal distribution random projection B, ;

[0078] Where m represents the dimension of the high-dimensional feature vector, and s represents the hyperparameter that controls the range of random number values;

[0079] The high-dimensional feature vector after high-dimensional feature encoding of the query point quadruple features is represented as follows: ,in, .

[0080] Furthermore, in one possible implementation of this invention, the interaction module 43 is specifically used for:

[0081] The high-dimensional feature vector is divided into multiple groups by an interactive grouped linear vector attention mechanism, which maps the channel to the value vector. A linear transformation is applied within each group to obtain the linearly transformed value vector matrix.

[0082] Using a strip matrix Learn the contextual information within each channel and the information between channels, construct global attention weights, and interact with the high-dimensional features between groups on the value vector matrix to obtain the target feature vector;

[0083] in, Involving There are two parameters, k represents different groups, and c represents the number of channels in the value vector.

[0084] .

[0085] Furthermore, in one possible implementation of this invention, the apparatus further includes:

[0086] The adjustment module is used to adjust the coverage of high-dimensional feature interactions between groups through cross-validation.

[0087] Furthermore, in one possible implementation of this embodiment of the invention, the semantic segmentation module 44 is specifically used for:

[0088] Extract the point features of the target feature vector stored during the downsampling of the target feature vector by the multilayer perceptron for semantic segmentation, and map the point features to all points in the same set to obtain the segmentation categories after segmentation of coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines.

[0089] This invention discloses a coal mine roadway point cloud segmentation device based on an interactive grouped linear attention mechanism. It constructs a local coordinate system based on an arbitrarily selected query point and a reference point within its neighborhood in the coal mine roadway point cloud data. This system describes the three angular features between the normal vector of the query point and the normal vector of the reference point, as well as the distance features between the query point and the reference point. High-dimensional feature encoding is then performed to construct a high-dimensional feature vector. An interactive grouped linear vector attention mechanism is introduced to map the high-dimensional feature vector into multiple groups of value vectors. Linear transformations are applied within each group, and high-dimensional features interact between groups to obtain the target feature vector. A multilayer perceptron is then used for semantic segmentation to determine the segmentation category of the coal mine roadway point cloud data. Therefore, without increasing equipment and computing power within the coal mine roadway, this device effectively achieves efficient and accurate segmentation of point cloud data in complex coal mine roadway scenarios.

[0090] To achieve the above embodiments, the present invention also proposes an electronic device, comprising:

[0091] At least one processor; and

[0092] A memory communicatively connected to the at least one processor; wherein,

[0093] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the aforementioned method.

[0094] To implement the above embodiments, the present invention also proposes a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the aforementioned method.

[0095] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0096] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0097] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0098] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0099] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0100] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0101] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0102] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A point cloud segmentation method for coal mine roadways based on an interactive grouped linear attention mechanism, characterized in that, The method includes: Based on an arbitrarily selected query point and a reference point within the neighborhood of the query point in the coal mine roadway point cloud data, a local coordinate system is constructed to describe the quaternary features between the query point and the reference point. These quaternary features include three angular features between the normal vector of the query point and the normal vector of the reference point. , , and the distance features between the query point and the reference point. ; For the quaternary features ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors; An interactive grouped linear vector attention mechanism is introduced to divide the channels that map the high-dimensional feature vector to the value vector into multiple groups, and apply linear transformation within each group. Furthermore, the interaction of high-dimensional features between groups is carried out through global attention weights to obtain the target feature vector of the interaction between linear transformation and high-dimensional features. In this case, the linear transformation weights within each group are shared when applying linear transformation within each group. Multilayer perceptron is used to perform semantic segmentation on the target feature vector to obtain the segmentation categories after segmentation of coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines. The step of constructing a local coordinate system based on an arbitrarily selected query point and a reference point within the neighborhood of the query point in the coal mine roadway point cloud data to describe the quaternary features between the query point and the reference point includes: Select any query point in the point cloud data of coal mine roadways. and a reference point within the neighborhood of the query point. and according to the and and their respective normal vectors and Construct a local coordinate system; Calculate in the local coordinate system normal vector and normal vector The relative deviation between them, i.e. normal vector and normal vector The three angular features between , , and the distance features between the query point and the reference point. To serve as a description of the query point With reference point Quaternary features between ( , , , ); in, ; ； The four-element feature ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors, including: The four-element feature of all query points in the coal mine roadway point cloud data is represented as follows: Where i = 1, 2, ..., C represents the number of query points, and R represents the real number space; Choose one that follows a normal distribution random projection B, ; Where m represents the dimension of the high-dimensional feature vector, and s represents the hyperparameter that controls the range of random number values; The high-dimensional feature vector after high-dimensional feature encoding of the query point quadruple features is represented as follows: ,in, ; The introduced interactive grouped linear vector attention mechanism divides the channels of the high-dimensional feature vector into multiple groups, maps them to value vectors, applies linear transformations within each group, and uses global attention weights to interact with the high-dimensional features between groups, thereby obtaining the target feature vector through the interaction of linear transformations and high-dimensional features. This includes: The high-dimensional feature vector is divided into multiple groups by an interactive grouped linear vector attention mechanism, which maps the channel to the value vector. A linear transformation is applied within each group to obtain the linearly transformed value vector matrix. Using a strip matrix Learn the contextual information within each channel and the information between channels, construct global attention weights, and interact with the high-dimensional features between groups on the value vector matrix to obtain the target feature vector; in, Involving There are two parameters, k represents different groups, and c represents the number of channels in the value vector. 。 2. The method according to claim 1, characterized in that, The method further includes adjusting the coverage of high-dimensional feature interactions between groups through cross-validation.

3. The method according to claim 1, characterized in that, The method utilizes a multilayer perceptron to perform semantic segmentation on the target feature vector to obtain segmentation categories from the segmented point cloud data of the coal mine roadway. These segmentation categories include roof, floor, face, ventilation ducts, and pipelines, among others. Extract the point features of the target feature vector stored during the downsampling of the target feature vector by the multilayer perceptron for semantic segmentation, and map the point features to all points in the same set to obtain the segmentation categories after segmentation of coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines.

4. A point cloud segmentation device for coal mine roadways based on an interactive grouped linear attention mechanism, characterized in that, The device includes: The construction module is used to construct a local coordinate system based on an arbitrarily selected query point and a reference point within the neighborhood of the query point in the coal mine roadway point cloud data. This system describes the quaternary features between the query point and the reference point, whereby the quaternary features include three angular features between the normal vector of the query point and the normal vector of the reference point. , , and the distance features between the query point and the reference point. ; Encoding module, used for encoding the four-element features ( , , , High-dimensional feature encoding is performed to construct high-dimensional feature vectors; The interaction module is used to introduce an interactive grouped linear vector attention mechanism to divide the channels of the high-dimensional feature vector into value vectors into multiple groups, apply linear transformations within each group, and interact high-dimensional features between groups through global attention weights to obtain the target feature vector of the interaction between linear transformations and high-dimensional features. The linear transformation weights within each group are shared when applying linear transformations within each group. The semantic segmentation module is used to perform semantic segmentation on the target feature vector using a multilayer perceptron to obtain the segmentation categories after segmenting the coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines. The building module is specifically used for: Select any query point in the point cloud data of coal mine roadways. and a reference point within the neighborhood of the query point. and according to the and and their respective normal vectors and Construct a local coordinate system; Calculate in the local coordinate system normal vector and normal vector The relative deviation between them, i.e. normal vector and normal vector The three angular features between , , and the distance features between the query point and the reference point. To serve as a description of the query point With reference point Quaternary features between ( , , , ); in, ; ； The encoding module is specifically used for: The four-element feature of all query points in the coal mine roadway point cloud data is represented as follows: Where i = 1, 2, ..., C represents the number of query points, and R represents the real number space; Choose one that follows a normal distribution random projection B, ; Where m represents the dimension of the high-dimensional feature vector, and s represents the hyperparameter that controls the range of random number values; The high-dimensional feature vector after high-dimensional feature encoding of the query point quadruple features is represented as follows: ,in, ; The interaction module is specifically used for: The high-dimensional feature vector is divided into multiple groups by an interactive grouped linear vector attention mechanism, which maps the channel to the value vector. A linear transformation is applied within each group to obtain the linearly transformed value vector matrix. Using a strip matrix Learn the contextual information within each channel and the information between channels, construct global attention weights, and interact with the high-dimensional features between groups on the value vector matrix to obtain the target feature vector; in, Involving There are two parameters, k represents different groups, and c represents the number of channels in the value vector. 。 5. The apparatus according to claim 4, characterized in that, The device further includes: The adjustment module is used to adjust the coverage of high-dimensional feature interactions between groups through cross-validation.

6. The apparatus according to claim 4, characterized in that, The semantic segmentation module is specifically used for: Extract the point features of the target feature vector stored during the downsampling of the target feature vector by the multilayer perceptron for semantic segmentation, and map the point features to all points in the same set to obtain the segmentation categories after segmentation of coal mine roadway point cloud data. The segmentation categories include roof, floor, face, ventilation ducts and pipelines.

7. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-3.

8. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-3.

Images