A method for measuring the volume of a cement conveyor using deep learning point cloud segmentation

Abstract

This application provides a method for measuring the volume of a cement conveyor using deep learning point cloud segmentation. The method includes: acquiring 3D point cloud data using a laser line-scanning 3D camera and determining the RGB values ​​of the corresponding pixels in the 3D point cloud data to obtain a 3D point cloud with color features; inputting the 3D point cloud with color features into an improved Point-Net++ neural network for 3D point cloud segmentation, resulting in 3D point cloud data containing only the cement surface within the conveyor; constructing a triangulation network based on the irregular cement surface in the segmented 3D point cloud data using triangulation, and calculating the volume based on the constructed triangulation network; wherein the improved Point-Net++ neural network is built on the framework of the original Point-Net++ point cloud segmentation network, including: core point convolution, a context-aware module, and a Focal Loss loss function. This improves the performance of the point cloud segmentation algorithm, enabling accurate segmentation of the surface point cloud data of cement material within the conveyor for volume measurement even in complex situations.

Description

Technical Field

[0001] This invention relates to the field of machine vision volume measurement technology, and in particular to a method for measuring the volume of a cement conveyor using deep learning point cloud segmentation. Background Technology

[0002] With the development of technology and the need for refined management of the production process, it is now necessary to measure the volume of cement being transported on a chain bucket conveyor in order to accurately grasp the daily production data of the factory.

[0003] Due to the uniform motion of the bucket chain conveyor, data collection on the cement on the conveyor is often not accurate enough, leading to complex point cloud segmentation in the later stages.

[0004] For point cloud segmentation tasks in complex environments, traditional segmentation paradigms based on single laser point clouds face the dual challenges of lacking multimodal data fusion and limited computational efficiency, making it difficult to meet the accuracy requirements of practical engineering scenarios.

[0005] Therefore, there is a need to provide an improved technical solution that addresses the shortcomings of the existing technology. Summary of the Invention

[0006] The purpose of this invention is to provide a method for measuring the volume of a cement conveyor based on deep learning point cloud segmentation, so as to solve or alleviate the problems existing in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] This invention provides a method for measuring the volume of a cement conveyor based on deep learning point cloud segmentation. The method includes:

[0009] Step S1: Obtain 3D point cloud data using a laser line scanning 3D camera. Based on the 3D point cloud data, determine the RGB values ​​of the corresponding pixels to obtain a 3D point cloud with color features.

[0010] Step S2: Input the 3D point cloud with color features into the improved Point-Net++ neural network for 3D point cloud segmentation to obtain segmented 3D point cloud data; the segmented 3D point cloud data only contains 3D point cloud data of the surface of cement in the conveyor.

[0011] Step S3: Using the triangulation method, construct a triangular mesh based on the irregular cement surface in the segmented 3D point cloud data, and calculate the volume based on the constructed triangular mesh;

[0012] The improved Point-Net++ neural network is built on the framework of the original Point-Net++ point cloud segmentation network, including: core point convolution and context-aware module;

[0013] The core point convolution is used to replace the feature extraction structure in the SA abstraction layer of the original Point-Net++ point cloud segmentation network;

[0014] The context-aware module is located after the last FP decoding layer of the improved Point-Net++ point cloud segmentation network;

[0015] The improved Point-Net++ neural network further includes replacing the original loss function in the original Point-Net++ point cloud segmentation network with the Focal Loss loss function.

[0016] Preferably, in step S1, obtaining 3D point cloud data using a laser line scanning 3D camera includes:

[0017] A line laser system is used to perform three-dimensional reconstruction on the raw data collected by the laser line scanning 3D camera based on the principle of triangulation, thereby obtaining 3D point cloud data.

[0018] Preferably, the improved Point-Net++ neural network specifically performs the following steps:

[0019] Perform downsampling and global feature acquisition in the space;

[0020] Then, the original point cloud size is gradually restored through an upsampling layer;

[0021] Finally, the fully connected layer effectively integrates multi-scale contextual information to achieve deep representation learning of the global spatial features of each point cloud, thereby completing the accurate semantic classification task.

[0022] Preferably, before performing downsampling and acquiring local spatial features, the method further includes:

[0023] Color features are used as auxiliary features, and 3D point cloud data with color features are used in the input of the improved Point-Net++ neural network to replace the 3D point cloud data containing only position information in the original Point-Net++ point cloud segmentation network.

[0024] The downsampling and local feature acquisition in the space are specifically as follows:

[0025] Hierarchical downsampling is achieved through core point convolution, which effectively reduces the amount of point cloud data. At the same time, local geometric features are captured by core points with fixed preset positions, and weights are calculated for each core point based on the distance and position of neighboring points. Finally, the core points and their weights are used to perform convolution operations on the set of neighboring points to obtain new features.

[0026] Preferably, the triangulation method is the Delaunay triangulation method.

[0027] Preferably, in step S3, a triangular mesh is constructed on the irregular surface of the cement, and the specific steps are as follows:

[0028] S31. Connect the points on the irregular surface of the cement to form an initial triangle;

[0029] S32. Expand the initial triangle into a triangular mesh;

[0030] S33. Repeat step S32 until all points in the point set have been processed;

[0031] The point set is the set of all points in the segmented 3D point cloud data.

[0032] Preferably, the Delaunay triangulation method specifically involves dividing the cement pile into two dimensions to obtain triangular sections.

[0033] Preferably, the volume calculation in step S3 specifically includes:

[0034] Each vertex of the segmented triangle corresponds to a projection point in the point cloud model of the irregular cement surface. There is a one-to-one correspondence between each segmented triangle and a triangular facet of the irregular cement surface. The projected triangle and the triangular facet constitute a composite volume. Each composite volume consists of two tetrahedrons and one triangular prism. The volume of each composite volume is equal to the sum of the volumes of these three spatial volumes, calculated using the following formula:

[0035] V i =V ABC-A′B′C′ +V B-A′B′C +V A-A′BC′ ,

[0036] Among them, V i Let V be the volume of the i-th composite body. ABC-A′B′C′ V represents the volume of a triangular prism. B-A′B′C and V A-A′BC′ Let A, B, and C represent the volumes of two tetrahedrons, where A, B, and C are points in the bisecting triangle, and A', B', and C' are the corresponding projection points.

[0037] Adding the volumes of all the components together gives the total volume of cement in the conveyor.

[0038] This disclosure also proposes a cement conveyor volume measurement system based on deep learning point cloud segmentation, wherein the system uses the steps of the aforementioned deep learning point cloud segmentation cement conveyor volume measurement method.

[0039] Compared with the closest existing technology, the technical solution of the embodiments of the present invention has the following beneficial effects:

[0040] This invention uses a laser line-scanning 3D camera to perform three-dimensional reconstruction to obtain point cloud data when collecting data on cement on a bucket conveyor. This technique is reliable and stable, and also has the advantages of being easier to visualize and more accurate. This application uses an improved deep learning convolutional neural network to perform semantic segmentation on the point cloud data collected by the laser line-scanning 3D camera, which addresses the image information problem, improves the performance of point cloud segmentation, and obtains point cloud data that only contains the surface of cement in the conveyor. Even in complex situations, it can perform point cloud segmentation and volume measurement on the pile inside the cement conveyor. Attached Figure Description

[0041] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. Wherein:

[0042] Figure 1 This is a schematic diagram of the structure of the improved Point-Net++ neural network according to an embodiment of this disclosure.

[0043] Figure 2 This is a diagram showing the transportation of cement stockpiles according to an embodiment of this disclosure.

[0044] Figure 3 This is a schematic diagram illustrating volume calculation according to an embodiment of the present disclosure. Detailed Implementation

[0045] The present application will now be described in detail with reference to the accompanying drawings and embodiments. Various examples are provided by way of explanation and not by way of limitation. In fact, those skilled in the art will recognize that modifications and variations can be made to the present application without departing from the scope or spirit thereof. For example, a feature shown or described as part of one embodiment may be used in another embodiment to produce yet another embodiment. Therefore, it is desirable that the present application encompass such modifications and variations that fall within the scope of the appended claims and their equivalents.

[0046] In the following description, the terms "first / second / third" are used merely to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of this disclosure only and is not intended to limit this disclosure.

[0048] In the description of this application, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," and "bottom," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and do not require that this application be constructed and operated in a specific orientation, and therefore should not be construed as limiting this application. The terms "connected," "linked," and "set up" used in this application should be interpreted broadly. For example, they can refer to fixed connections or detachable connections; direct connections or indirect connections through intermediate components; wired connections, radio connections, or wireless communication signal connections. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0049] This application mainly relates to the field of machine vision volume measurement technology, specifically disclosing a deep learning point cloud segmentation method for measuring the volume of a cement conveyor. This embodiment specifically performs the following steps:

[0050] Step S1: Obtain 3D point cloud data using a laser line scanning 3D camera. Based on the 3D point cloud data, determine the RGB values ​​of the corresponding pixels to obtain a 3D point cloud with color features.

[0051] Step S2: Input the 3D point cloud with color features into the improved Point-Net++ neural network for 3D point cloud segmentation to obtain segmented 3D point cloud data; the segmented 3D point cloud data only contains 3D point cloud data of the surface of cement in the conveyor.

[0052] Step S3: Using the triangulation method, construct a triangular mesh based on the irregular cement surface in the segmented 3D point cloud data, and perform volume calculation based on the constructed triangular mesh.

[0053] For step S1, a laser line-scan 3D camera is used to collect data on the cement pile in the uniformly moving bucket conveyor. During image acquisition, the camera's intrinsic and extrinsic parameters must first be calibrated. After calibration, the pixel coordinates on the image acquired by the camera can be converted into world coordinates for measurement. Due to the imaging characteristics of a monocular camera, the depth information of the target object from the optical axis cannot be directly obtained from a two-dimensional image alone. Therefore, this embodiment introduces a line laser structured light system to calculate the depth information of the target surface points using the laser triangulation principle. The light plane projected by the line laser system constrains the target object, allowing the depth information of the target object to be solved. Finally, through the solution of the depth information of the light plane and the calibration of the camera's intrinsic parameters, the data is projected onto the camera's UV coordinate system to obtain the RGB values ​​of the corresponding pixels, thus obtaining 3D point cloud data with color features.

[0054] Specifically, by using a laser line scan 3D camera to collect data on the cement pile in a bucket conveyor moving at a constant speed, it can achieve a higher accuracy in reconstructing the three-dimensional shape of the object's surface compared to traditional methods. Furthermore, since the target object is moving at a constant speed, the uniform motion characteristics of the target object perfectly match the frame rate-motion speed synchronization constraint of the laser line scan camera, ensuring the spatial continuity and geometric accuracy of the collected data.

[0055] Figure 1 A schematic diagram of the structure of the improved Point-Net++ neural network according to an embodiment of the present invention is shown.

[0056] like Figure 1 The process shown involves the following steps: 3D point cloud data with color features obtained in step S1 is downsampled layer by layer by a multi-level SA module to extract local geometric features, then gradually upsampled by a FP module to restore the original point cloud size, and then fed into a context-aware module. Afterwards, the Focal Loss loss function is used to optimize the class imbalance problem. Finally, the 3D point cloud data with color features is segmented by an improved Point-Net++ neural network to obtain 3D point cloud data containing only the surface of the cement in the conveyor. Step S2 is executed as follows... Figure 1 The processing flow shown includes:

[0057] The improved Point-Net++ neural network is built upon the framework of the original Point-Net++ point cloud segmentation network, including: core point convolution and a context-aware module. The core point convolution technique replaces the feature extraction structure in the SA abstraction layer of the original Point-Net++ point cloud segmentation network. The SA abstraction layer specifically consists of three abstraction layers: SA1, SA2, and SA3. A context-aware module is deployed after the last FP decoding layer in the improved Point-Net++ point cloud segmentation network. Furthermore, the Focal Loss loss function replaces the original loss function in the original Point-Net++ point cloud segmentation network.

[0058] For step S2, in one embodiment, the 3D point cloud segmentation processing of the 3D point cloud data with color features to be processed includes:

[0059] The 3D point cloud data with color features to be processed ( Figure 1 The input (ColorPoint Cloud) is downsampled into the improved Point-Net++ neural network SA1 abstraction layer, which includes a sampling grouping module ( Figure 1 Sample Grouping and Core Point Convolution ( Figure 1 In the PAConv module, the sampling and grouping module samples and groups the incoming 3D point cloud data with color features. Sampling aims to reduce the amount of point cloud data, lower the computational burden, and retain important structural information. Grouping aims to construct local neighborhoods for extracting local features. The result is the generation of a set containing neighboring points for each center point, used in subsequent convolution or feature aggregation operations. The point cloud data output from the sampling and grouping module is then fed to the core point convolution module for local feature extraction. The core point convolution captures local geometric features using core points with fixed preset positions and calculates a weight for each core point based on the distance and position of neighboring points. Finally, the core points and their weights are used to perform a convolution operation on the neighboring point set to obtain new features. This achieves the effects of reducing the amount of point cloud data, removing redundant points, retaining key geometric features, and lowering the computational burden. Finally, pooling is performed on the local features processed by convolution at the core points to compress and abstract the local features. The input features are then aggregated through a sliding window to achieve dimensionality reduction (reducing the amount of data and computational complexity), feature stability (enhancing robustness to small translations and rotations of the input), and hierarchical abstraction (extracting more global semantic information layer by layer).

[0060] In practical applications, while traditional convolutional methods are simple and efficient, their ability to extract local features from point clouds is limited, resulting in low efficiency and unsuitability for complex environments. The core-point convolution method, by adding spatial information during feature extraction from point cloud data, enhances the ability to capture local geometric features, significantly improving point cloud processing performance. This allows the modified model to more accurately identify the geometric differences between the regular geometric shapes of conveyor components and the geometric patterns of cement piles, reducing misclassification and improving the model's feature extraction capability for cement conveyor point cloud data. Furthermore, because the scanned point cloud exhibits sparse points far from the camera and dense points near the camera, the fixed-radius neighborhood grouping of the original Point-Net++ is prone to failure in sparse regions. Replacing it with core-point convolution allows for dynamic adjustment of the kernel coverage. A larger kernel size captures context in sparse regions, effectively reducing missed detections of sparse cement point clouds at the far end of the conveyor, avoiding overfitting in high-density regions and underfitting in low-density regions. For example, at a distance of 5m, the recall rate of cement point clouds is significantly improved.

[0061] Furthermore, the SA1 abstraction layer upsamples the local features of the processed point cloud into the FP3 decoding layer.

[0062] Based on this, in one embodiment, the method may further include:

[0063] The SA1 abstraction layer transmits the processed point cloud local feature data to the next SA2 abstraction layer, and the SA2 abstraction layer transmits the processed point cloud local feature data to the next SA3 abstraction layer. After downsampling processing by the SA3 abstraction layer, the data is transmitted to the FP1 decoding layer for upsampling. This allows each SA abstraction layer to extract local features through core point convolution and then generate higher-level features through pooling operations. The receptive field expands step by step from SA1 to SA3, covering the entire conveyor cross section. The higher-level features are then aggregated at the highest layer to generate global features.

[0064] The FP decoding layer upsamples the data from the SA abstraction layer. The main objective of the improved Point-Net++ spatial decoding part is to obtain the point features of each original point in the point cloud set by mapping the aggregated local spatial information onto all point clouds. The FP decoding layer recovers the point cloud density step by step through upsampling and a three-level FP pattern. Specifically, this is achieved using inverse distance weighted interpolation (IRI). Figure 1 The color-weighted features of the points obtained from the IDW (In-IDW) are propagated, and then the obtained weighted features and the spatial local features obtained from the corresponding point set SA abstraction layer during the encoding process are connected through cross-layer skip connections. Figure 1 The concatenated features are then stitched together using a single PointNet layer structure and aggregated using a core point convolution method to obtain the final point features, which are then gradually restored to the original point cloud size.

[0065] Based on this, in one embodiment, the method may further include:

[0066] The FP decoding layer consists of three decoding layers: FP1, FP2, and FP3. In the FP1 decoding layer, the inverse distance weighted interpolation module processes the local point cloud features transmitted from the SA3 layer and concatenates them with the local point cloud features transmitted from the SA2 layer via a cross-layer skip connection. The concatenated combined features are then aggregated using a core-point convolution method within a single PointNet layer structure to obtain new point cloud features. Next, the point cloud features obtained from the FP1 layer are used as the interpolation part in the next layer, FP2, where inverse distance weighted interpolation is used to obtain features. This is then concatenated with the point cloud features transmitted from the SA1 layer to obtain new combined features. These new combined features are then aggregated using a core-point convolution method within a single PointNet layer structure to obtain yet another new point cloud feature. Finally, the point cloud features obtained from the FP2 layer are used as the interpolation part in the next layer, FP3, where inverse distance weighted interpolation is used to obtain features. This is then concatenated with the original point cloud coordinates output from the laser line-scanning 3D camera, and the core-point convolution method is used to obtain the final point cloud features for each point cloud.

[0067] A context-aware module is added during the feature propagation (FP3) to the fully connected layer (FC) to address two complex environmental scenarios: misjudgment due to insufficient local features when the conveyor support partially obscures the cement area; and misjudgment due to highly similar local geometric features when cement contacts the conveyor chain or rollers. The final point cloud features of each point cloud obtained from the FP3 layer are transmitted to the context-aware module for processing. A Focal Loss function is used to optimize the class imbalance problem, balancing the significant differences in the amount of point cloud data separated by barriers such as the cement surface and the conveyor side. This ensures faster convergence and allows for sufficient training of classes with smaller point cloud proportions.

[0068] In practical applications, the original PointNet++ neural network's FP decoding layer relies on skip connections to recover details during hierarchical downsampling, resulting in a lack of explicit modeling of global semantics. This leads to misjudgments when the conveyor support partially obscures cement areas due to insufficient local features. Furthermore, when cement contacts the conveyor chain or rollers, the local geometric features are highly similar, making it impossible to distinguish them based solely on local information. To address these complexities, a context-aware module is introduced during feature propagation to the fully connected layer. This module aims to learn the correlations between features of the same category and the differences between features of different categories, thereby enhancing the representational power and robustness of the reinforced features.

[0069] Furthermore, the formula for calculating the Focal Loss function used is as follows:

[0070] L FL (y)=-α y log(p y (1-p) y ) λ ,

[0071] In the formula: α y These are class weights, which can suppress the imbalance between positive and negative samples. The λ coefficient reduces the weight of easily classified samples, allowing the model to focus more on difficult-to-classify samples during training. y represents the true class label of the sample, and p... y This is the probability value predicted as y. In this application, the Focal Loss loss function with parameters α=1 and γ=2 is used to train the model.

[0072] Finally, the processed point cloud features are input into the fully connected layer. The fully connected layer performs point-by-point feature mapping on all the input point cloud features through an equivalent fully connected layer and performs classification processing to obtain the predicted semantic result for each point, that is, the point cloud data containing only the surface of cement in the conveyor is obtained.

[0073] By segmenting the improved Point-Net++ neural network, point cloud data containing only the cement surface inside the conveyor is obtained, thus providing high-precision input for volume measurement. Because the surface shape of the cement in the conveyor is irregular, such as... Figure 2 As shown, the cement pile has a rectangular base and an irregular cone-shaped top. When calculating this irregular volume, since the cement pile has no overlapping parts in space, the Delaunay triangulation method is used to perform two-dimensional subdivision of the cement pile, and the resulting subdivided triangles are used to construct a triangulation network.

[0074] To construct a triangulation network, first find the two closest points in the point set and connect them to form an edge. Then, according to the Delaunay empty circle criterion of the triangulation network, find the other endpoint of the triangle containing this edge. Process all newly generated edges in this way until all points have been processed. The basic operation steps of the triangulation network are as follows:

[0075] S31. Connect the points on the irregular surface of the cement to form an initial triangle;

[0076] S32. Expand the initial triangle into a triangular mesh;

[0077] S33. Repeat step S32 until all points in the point set have been processed;

[0078] The point set is the set of all points in the segmented 3D point cloud data.

[0079] After partitioning the cement pile point cloud using the two-dimensional partitioning method described above, a cement pile point cloud model is obtained. First, the point cloud is projected onto the XY plane. At this point, the projected points are scattered points in the two-dimensional plane. Delaunay triangulation is then performed on the projected scattered points. According to the Delaunay triangulation criteria, the resulting triangles are adjacent and do not contain each other. Each vertex of a triangular partition corresponds to a projected point in the pile point cloud model. Therefore, each triangular partition corresponds one-to-one with a triangular facet on the pile surface. The projected triangle and the triangular facet can form a composite body, such as... Figure 3 As shown, A, B, and C are points in the original point cloud data, and A', B', and C' are the corresponding projected points. Figure 3 It can be seen that each composite solid consists of two tetrahedrons and one triangular prism, and the volume of each composite solid is equal to the sum of the volumes of these three spatial solids. The formula for calculating the volume of a single composite solid is:

[0080] V i =V ABC-A′B′C′ +V B-A′B′C +V A-A′BC′ ,

[0081] Among them, V i Let V be the volume of the i-th composite body. ABC-A′B′C′ V represents the volume of a triangular prism. B-A′B′C and V A-A′BC′ Let A, B, and C represent the volumes of the two tetrahedrons, where A, B, and C are points in the bisecting triangles, and A', B', and C' are the corresponding projection points. By adding the volumes of all the combined volumes, we can obtain the total volume of cement in the conveyor.

[0082] In addition, this application also discloses a cement conveyor volume measurement system based on deep learning point cloud segmentation, which calculates the total volume of cement in the conveyor by using the steps of the above embodiments.

[0083] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for measuring the volume of a cement conveyor using deep learning point cloud segmentation, characterized in that, The measurement method includes: Step S1: Obtain 3D point cloud data using a laser line scanning 3D camera. Based on the 3D point cloud data, determine the RGB values ​​of the corresponding pixels to obtain a 3D point cloud with color features. Step S2: Input the 3D point cloud with color features into the improved Point-Net++ neural network for 3D point cloud segmentation to obtain segmented 3D point cloud data; the segmented 3D point cloud data only contains 3D point cloud data of the surface of cement in the conveyor. Step S3: Using the triangulation method, construct a triangular mesh based on the irregular cement surface in the segmented 3D point cloud data, and calculate the volume based on the constructed triangular mesh; The improved Point-Net++ neural network is built on the framework of the original Point-Net++ point cloud segmentation network, including: core point convolution and context-aware module; The core point convolution is used to replace the feature extraction structure in the SA abstraction layer of the original Point-Net++ point cloud segmentation network; The context-aware module is located after the last FP decoding layer of the improved Point-Net++ point cloud segmentation network; The improved Point-Net++ neural network further includes replacing the original loss function in the original Point-Net++ point cloud segmentation network with the Focal Loss loss function.

2. The method for measuring the volume of a cement conveyor using deep learning point cloud segmentation according to claim 1, characterized in that, In step S1, 3D point cloud data is obtained using a laser line scanning 3D camera, including: A line laser system is used to perform three-dimensional reconstruction on the raw data collected by the laser line scanning 3D camera based on the principle of triangulation, thereby obtaining 3D point cloud data.

3. The method for measuring the volume of a cement conveyor using deep learning point cloud segmentation according to claim 1, characterized in that, The improved Point-Net++ neural network specifically performs the following steps: Perform downsampling and acquire local spatial features; Then, the original point cloud size is gradually restored through an upsampling layer; Finally, the fully connected layer effectively integrates multi-scale contextual information to achieve deep representation learning of the global spatial features of each point cloud, thereby completing the accurate semantic classification task.

4. The method for measuring the volume of a cement conveyor using deep learning point cloud segmentation according to claim 3, characterized in that, Before performing downsampling and local feature acquisition in the space, the method also includes: Color features are used as auxiliary features, and 3D point cloud data with color features are used in the input of the improved Point-Net++ neural network to replace the 3D point cloud data containing only position information in the original Point-Net++ point cloud segmentation network. The downsampling and local feature acquisition in the space are specifically as follows: Hierarchical downsampling is achieved through core point convolution, which effectively reduces the amount of point cloud data. At the same time, local geometric features are captured by core points with fixed preset positions, and weights are calculated for each core point based on the distance and position of neighboring points. Finally, the core points and their weights are used to perform convolution operations on the set of neighboring points to obtain new features.

5. The method for measuring the volume of a cement conveyor based on deep learning point cloud segmentation according to claim 1, characterized in that, The triangulation method mentioned is the Delaunay triangulation method.

6. The method for measuring the volume of a cement conveyor using deep learning point cloud segmentation according to claim 1, characterized in that, In step S3, a triangular mesh is constructed on the irregular surface of the cement. The specific steps are as follows: S31. Connect the points on the irregular surface of the cement to form an initial triangle; S32. Expand the initial triangle into a triangular mesh; S33. Repeat step S32 until all points in the point set have been processed; The point set is the set of all points in the segmented 3D point cloud data.

7. The method for measuring the volume of a cement conveyor using deep learning point cloud segmentation according to claim 5, characterized in that, The Delaunay triangulation method specifically involves dividing the cement pile into two dimensions to obtain triangular sections.

8. The method for measuring the volume of a cement conveyor based on deep learning point cloud segmentation according to claim 7, characterized in that, The volume calculation in step S3 is specifically as follows: Each vertex of the segmented triangle corresponds to a projection point in the point cloud model of the irregular cement surface. There is a one-to-one correspondence between each segmented triangle and a triangular facet of the irregular cement surface. The projected triangle and the triangular facet constitute a composite volume. Each composite volume consists of two tetrahedrons and one triangular prism. The volume of each composite volume is equal to the sum of the volumes of these three spatial volumes, calculated using the following formula: V i =V ABC-A′B′C′ +V B-A′B′C +V A-A′BC′ , Among them, V i Let V be the volume of the i-th composite body. ABC-A′B′C′ V represents the volume of a triangular prism. B-A′B′C and V A-A′BC′ Let A, B, and C represent the volumes of two tetrahedrons, where A, B, and C are points in the bisecting triangle, and A', B', and C' are the corresponding projection points. Adding the volumes of all the components together gives the total volume of cement in the conveyor.

9. A cement conveyor volume measurement system based on deep learning point cloud segmentation, the system being used to perform the steps of the cement conveyor volume measurement method based on deep learning point cloud segmentation as described in any one of claims 1 to 8.

Images