Moving bridge type multi-sensor 3D scanning system

Abstract

The invention discloses a mobile bridge type multi-sensor 3D scanning system, and belongs to the technical field of 3D scanning, and the system comprises a measurement platform which is used for placing a measured object; the three-coordinate bridge type measurement structure realizes omnibearing measurement of a measured object; the dynamic coding transmitting module is used for dynamically adjusting the signal transmitting direction or frequency; the multi-mode sensing module is used for receiving the light reflected back from the measured object and performing full-dimensional data capture on the data of the measured object through data received by different types of sensors; and the data fusion processor is internally provided with a hypergraph neural network model to realize feature fusion of multi-source heterogeneous data received by different types of sensors. According to the invention, high-precision, omnibearing and intelligent measurement of the measured object is realized.

Description

Technical Field

[0001] This application relates to a mobile bridge - type multi - sensor 3D scanning system, belonging to the technical field of 3D scanning. Background Art

[0002] With the continuous improvement of the requirements for the accuracy and comprehensiveness of obtaining the three - dimensional shape and material information of objects in many fields such as industrial manufacturing, quality inspection, and digital cultural relics protection, 3D scanning technology has received extensive attention and application. In the prior art, most traditional 3D scanning systems adopt a single - sensor or simple multi - sensor combination method. The measurement structure lacks flexibility, making it difficult to achieve all - around accurate measurement of the object to be measured in complex spatial postures. The data collected by multiple sensors are often processed independently, making it difficult to fully explore the potential correlations between data of different types of sensors. As a result, the capture of the shape and material characteristics of the object to be measured is not comprehensive and in - depth enough, and it is impossible to quickly and accurately give risk decision results, making it difficult to meet the requirements of high - precision, all - around, and intelligent measurement of objects in practical applications. Summary of the Invention

[0003] According to one aspect of this application, a mobile bridge - type multi - sensor 3D scanning system is provided, which realizes high - precision, all - around, and intelligent measurement of the object to be measured.

[0004] The mobile bridge - type multi - sensor 3D scanning system includes: A measurement platform for placing the object to be measured; A three - coordinate bridge - type measurement structure. The three - coordinate bridge - type measurement structure is installed on the measurement platform for adjusting the measurement orientation, and a detection head is installed on the three - coordinate bridge - type measurement structure. The detection head integrates a dynamic coding emission module and a multi - modal sensing module to achieve all - around measurement of the object to be measured; A dynamic coding emission module integrated in the detection head, configured to dynamically adjust the signal emission direction or frequency; the polarization coding unit of the dynamic coding emission module generates circularly polarized light containing topological phase; A multi - modal sensing module integrated in the detection head, including a vision layer and a tactile layer. The vision layer captures a dynamic image sequence of the surface of the object to be measured through a multi - camera array, and the tactile layer generates a non - contact force distribution matrix through a pressure sensor array and receives the light reflected from the object to be measured according to the non - contact force distribution matrix, capturing all - dimensional data of the object to be measured through the data received by different types of sensors; A data fusion processor with a built-in hypergraph neural network model to achieve feature fusion of multi-source heterogeneous data received by different types of sensors. The hypergraph neural network model constructs multi-view views of the morphology and material of the object under test based on the data of the object under test. Among them, nodes represent the features of different types of sensors, and hyperedges represent cross-modal associations, forming a multi-view hypergraph. At the same time, identify the key feature subset related to the morphology and material features of the object under test, and use the key feature subset to reconstruct the multi-view hypergraph structure, and then update the hypergraph neural network. According to the preset risk assessment threshold for the morphology and material features of the object under test, use the updated hypergraph neural network to determine the risk decision result of the object under test.

[0005] Further, the three-coordinate bridge measuring structure includes: An X-axis moving mechanism is arranged on the upper end surface of the measuring platform. The object under test is located above the X-axis moving mechanism, and the X-axis moving mechanism guides the object under test to move along the X-axis direction of the measuring platform. A bridge is horizontally erected above the measuring platform and fixed on both sides of the measuring platform. A Z-axis moving mechanism is arranged on the bridge, and the Z-axis moving mechanism moves horizontally along the bridge to realize the Y-axis movement of the detection head. The detection head is installed on the Z-axis moving mechanism to realize the Z-axis movement of the detection head. The X-axis moving mechanism is located below the bridge.

[0006] Further, the dynamic coding emission module includes: A wavelength adaptive unit that automatically selects the illumination wavelength according to the surface material of the object under test. A polarization coding unit that modulates the topological phase distribution of circularly polarized light. A safety protection unit that monitors the optical power in real time and triggers an emergency shutdown. Among them, the frequency range of the signal is dynamically limited by the coding sequence, and invalid coding sequences are automatically filtered.

[0007] Further, the multi-modal sensing module is realized through a sensor array. The sensor array includes a vision layer and a tactile layer. The vision layer includes: A central camera defines a first search area with the image center as the origin, and its radius expands adaptively with the target movement speed. Multiple annular cameras are circumferentially arranged around the central camera, and the fan-shaped imaging areas of each camera cover the blind areas of the first search area. Among them, the vision layer estimates the position of a high-speed moving target through an edge prediction algorithm and aligns the data with the tactile layer in space and time. The tactile layer includes: A capacitive sensing array, deployed in the central area, maps the surface topography of an object through the change in capacitance; An ultrasonic sensing unit, deployed in the peripheral area, improves the resolution through echo superposition; Among them, the data of the capacitive sensing array and the ultrasonic sensing unit are fused by a hypergraph neural network.

[0008] Furthermore, the data fusion processor includes: A data calibration module that establishes the mapping relationship between the sensor feature vector and the hypergraph node; A dynamic reconstruction module that filters the key feature subset and reconstructs the hypergraph structure based on gradient risk assessment; A risk assessment module that outputs the risk probability through a fully connected layer and a Softmax function; Among them, the hyperedge set of the hypergraph neural network is adaptively adjusted by k-means clustering.

[0009] Furthermore, the hypergraph neural network performs the following operations: Construct a node feature matrix and a hyperedge association matrix, where the nodes represent different sensor feature dimensions; Iteratively update the node features through the hypergraph convolutional layer to fuse local and global associations; Adopt a continuous learning framework to dynamically optimize the hyperedge weights and delete redundant nodes based on feature similarity.

[0010] Furthermore, it also includes cross-control logic: The dynamic coding and transmitting module adjusts the polarization coding parameters in real time according to the mechanical distribution data of the tactile layer; The three-coordinate bridge measurement structure dynamically plans the obstacle avoidance scanning path based on the output of the risk assessment module; The safety protection unit synchronously triggers the self-checking protocol of the data fusion processor when detecting abnormal optical power.

[0011] The beneficial effects that this application can produce include: The mobile bridge-type multi-sensor 3D scanning system provided by this application has a three-coordinate bridge measurement structure, enabling the system to have a flexible measurement orientation adjustment ability. Combined with a detection head integrating a dynamic coding emission module and a multi-modal sensing module, it can dynamically optimize signal emission and achieve synchronous acquisition of multi-modal data, completing all-round and high-precision measurement of the object to be measured. The hypergraph neural network model built into the data fusion processor can deeply explore the cross-modal associations of multi-source heterogeneous data, accurately fuse the features of different sensors, construct multi-view views, and effectively capture rich information about the shape and material of the object to be measured. By identifying the key feature subset to reconstruct the hypergraph structure and update the model, the data processing efficiency and accuracy are greatly improved. Finally, based on the risk assessment threshold, a risk decision result is quickly output, providing an efficient, intelligent, and accurate three-dimensional measurement solution for scenarios such as industrial inspection. Description of the Drawings

[0012] Figure 1 It is a schematic diagram of the mobile bridge-type multi-sensor 3D scanning system in an embodiment of this application; Figure 2 It is a schematic diagram of the three-coordinate bridge measurement structure in an embodiment of this application; List of components and reference numerals: 1 - measurement platform; 2 - detection head; 3 - X-axis movement mechanism; 4 - bridge; 5 - Z-axis movement mechanism. Detailed Embodiment

[0013] The following describes this application in detail with reference to the embodiments, but this application is not limited to these embodiments.

[0014] Refer to Figure 1-2 , the mobile bridge-type multi-sensor 3D scanning system, characterized by including: A measurement platform 1 for placing the object to be measured; A three-coordinate bridge measurement structure, which is installed on the measurement platform 1 for adjusting the measurement orientation, and a detection head 2 is installed on the three-coordinate bridge measurement structure. The detection head 2 integrates a dynamic coding emission module and a multi-modal sensing module to achieve all-round measurement of the object to be measured; A dynamic coding emission module, integrated in the detection head, configured to dynamically adjust the signal emission direction or frequency; the polarization coding unit of the dynamic coding emission module generates circularly polarized light containing topological phase; A multi-modal sensing module, integrated in the detection head, including a vision layer and a tactile layer. The vision layer captures a dynamic image sequence of the surface of the object to be measured through a multi-camera array, and the tactile layer generates a non-contact force distribution matrix through a pressure sensor array and receives the light reflected from the object to be measured according to the non-contact force distribution matrix, capturing full-dimensional data of the object to be measured through the data received by different types of sensors; A data fusion processor, built with a hypergraph neural network model, realizes the feature fusion of multi-source heterogeneous data received by different types of sensors. The hypergraph neural network model constructs multi-view views of the morphology and material of the object under test based on the data of the object under test. Among them, nodes represent the features of different types of sensors, and hyperedges represent cross-modal associations, forming a multi-view hypergraph. At the same time, identify the key feature subset related to the morphology and material features of the object under test, and use the key feature subset to reconstruct the multi-view hypergraph structure, thereby updating the hypergraph neural network. According to the preset risk assessment threshold for the morphology and material features of the object under test, use the updated hypergraph neural network to determine the risk decision result of the object under test.

[0015] Specifically, the measurement platform 1 is the basic support structure of the entire scanning system. Its main function is to place the object under test and provide a stable placement position for subsequent measurement work, ensuring that the object under test remains relatively stationary or moves in a predetermined manner during the scanning process, so as to accurately obtain its data. The three-coordinate bridge measurement structure is installed on the measurement platform 1 and serves as the core mechanical structure of the measurement system. Its key role is to adjust the measurement orientation. By precisely moving and positioning in three coordinate directions (usually the X, Y, and Z axes), the position and posture of the detection head 2 can be flexibly changed, thereby realizing the measurement of different parts and angles of the object under test and ensuring full coverage of the surface of the object under test. The detection head 2 is installed on the three-coordinate bridge measurement structure and is a key component for data acquisition. The detection head 2 integrates a dynamic coding emission module and a multi-modal sensing module, and these two modules work together to complete the measurement task of the object under test.

[0016] Among them, the dynamic coding emission module dynamically adjusts the signal emission direction or frequency. During the 3D scanning process, according to the shape, material, and measurement requirements of the object under test, the dynamic coding emission module can change the characteristics of the signal in real time. For example, for an object with a complex shape, it may be necessary to adjust the emission direction to better illuminate each surface of the object; for objects with different materials, it may be necessary to change the emission frequency to obtain a more accurate reflected signal, thereby providing a better signal source for subsequent data acquisition. The multi-modal sensing module is responsible for receiving the light reflected from the object under test. These lights carry various information on the surface of the object under test, such as shape, texture, color, etc. Through the data received by different types of sensors, the multi-modal sensing module can achieve the full-dimensional capture of the data of the object under test. Different types of sensors may have different working principles and measurement characteristics. For example, a laser sensor can accurately measure the distance and shape of an object, and an optical sensor can obtain the color and texture information of an object, etc. The data collected by these different types of sensors complement each other to jointly construct the complete information of the object under test.

[0017] The data fusion processor incorporates a hypergraph neural network model, which is the data processing core of the entire system. The main function of this model is to achieve feature fusion of multi-source heterogeneous data received by different types of sensors. During the 3D scanning process, due to the use of multiple sensors, the collected data has different formats, structures, and semantic information, belonging to multi-source heterogeneous data. The hypergraph neural network model can effectively process these complex data and organically fuse them. Based on the data of the object to be measured, the hypergraph neural network model constructs multi-view views of the morphology and material of the object to be measured. In this hypergraph structure, nodes represent different types of sensor features, and hyperedges represent cross-modal associations. This representation method can intuitively display the mutual relationship between different sensor features and describe the object to be measured from multiple perspectives. The model will identify key feature subsets related to the morphology and material features of the object to be measured. These key feature subsets are crucial for accurately describing the characteristics of the object to be measured. Using the identified key feature subsets, the multi-view hypergraph structure is reconstructed, and then the hypergraph neural network is updated. In this way, the model can continuously optimize itself and improve its ability to extract and analyze the features of the object to be measured. According to the preset risk assessment threshold for the morphology and material features of the object to be measured, the updated hypergraph neural network is used to determine the risk decision result of the object to be measured. For example, in industrial inspection, according to factors such as the degree of morphological defects and material anomalies of the object to be measured, combined with the preset risk assessment threshold, it can be judged whether the object to be measured meets the quality standards, whether there are potential risks, and corresponding decision results are given, such as qualified, unqualified, or further inspection required, etc.

[0018] The three-coordinate bridge measurement structure includes: The X-axis movement mechanism 3 is arranged on the upper end surface of the measurement platform 1, and the object to be measured is located above the X-axis movement mechanism 3. The X-axis movement mechanism 3 guides the object to be measured to move along the X-axis direction of the measurement platform 1; The bridge 4 is horizontally erected above the measurement platform 1 and fixed on both sides of the measurement platform 1. A Z-axis movement mechanism 5 is arranged on the bridge 4. The Z-axis movement mechanism 5 moves horizontally along the bridge 4, so as to realize the Y-axis movement of the detection head 2. The detection head 2 is installed on the Z-axis movement mechanism 5 to realize the Z-axis movement of the detection head 2; The X-axis movement mechanism 3 is located below the bridge 4.

[0019] Specifically, the X-axis motion mechanism 3 is arranged on the upper end surface of the measurement platform 1, below the bridge 4, which not only ensures the compactness of the overall structure but also facilitates the coordinated movement with the measurement platform 1 and the bridge 4. The object to be measured is placed above the X-axis motion mechanism 3. When the X-axis motion mechanism 3 is activated, it can directly drive the object to be measured to move in the X-axis direction on the measurement platform 1. This design enables the object to be measured to achieve precise positioning and movement in the X-axis direction, providing a basis for subsequent measurements at different positions. By guiding the object to be measured to move along the X-axis direction, it is possible to perform scanning measurements on different parts of the object to be measured in the X-axis direction, thereby obtaining the complete morphology and characteristic information of the object to be measured in the X-axis direction. The bridge 4 is horizontally installed above the measurement platform 1 and fixed on both sides of the measurement platform 1. This fixing method ensures the stability and reliability of the bridge 4, enabling it to bear the weight of the Z-axis motion mechanism 5 and the detection head 2 as well as various forces generated during the movement process. As an important supporting component of the entire measurement structure, the bridge 4 not only provides an installation basis for the Z-axis motion mechanism 5 but also, through its horizontally installed structure, enables the Z-axis motion mechanism 5 to move horizontally on the bridge 4, thereby indirectly realizing the movement of the detection head 2 in the Y-axis direction. The Z-axis motion mechanism 5 is arranged on the bridge 4. Through cooperation with the bridge 4, the Z-axis motion mechanism 5 can move horizontally along the bridge 4. The detection head 2 is installed on the Z-axis motion mechanism 5. The Z-axis motion mechanism 5 can not only move horizontally on the bridge 4 itself (realizing the Y-axis movement of the detection head 2) but also drive the detection head 2 to move up and down in the Z-axis direction. This dual-axis motion ability enables the detection head 2 to perform flexible positioning and movement in three-dimensional space. Through the horizontal movement of the Z-axis motion mechanism 5 on the bridge 4, the positioning and scanning of the detection head 2 in the Y-axis direction are realized; through the up and down movement of the Z-axis motion mechanism 5 itself, the positioning and scanning of the detection head 2 in the Z-axis direction are realized. Combining the movement of the X-axis motion mechanism 3 driving the object to be measured in the X-axis direction, the three work together, enabling the detection head 2 to perform omnidirectional and multi-angle scanning measurements on the object to be measured, thereby obtaining complete and accurate three-dimensional data of the object to be measured.

[0020] During the measurement process, the X-axis motion mechanism 3, the Z-axis motion mechanism 5 on the bridge 4, and the detection head 2 cooperate with each other to form a complete three-coordinate measurement system. The X-axis motion mechanism 3 is responsible for the movement of the object to be measured in the X-axis direction. The Z-axis motion mechanism 5 drives the detection head 2 to move in the Y-axis and Z-axis directions. Through the coordinated movement in the three coordinate directions, the detection head 2 can accurately reach each measurement point of the object to be measured, realizing omnidirectional and high-precision measurement of the object to be measured.

[0021] It should be noted that the motion drives of the X, Y, and Z axes in the three-coordinate bridge measurement structure can use existing drive structures as long as they can achieve the functions.

[0022] The dynamic coding transmission module includes: A wavelength adaptive unit that automatically selects the illumination wavelength according to the surface material of the object to be measured; A polarization coding unit that modulates the topological phase distribution of circularly polarized light; A safety protection unit that monitors the optical power in real time and triggers an emergency shutdown; Among them, the frequency range of the signal is dynamically defined by the coding sequence, and invalid coding sequences are automatically filtered.

[0023] The modulation of the topological phase includes: ; In the formula, l represents the topological charge number, l ∈ {1, 2, 3, 4}; f represents the equivalent lens focal length; represents the wavelength; r represents the radial distance; represents the azimuth angle; A safety protection unit that monitors the light power in real time and triggers an emergency shutdown when it exceeds the threshold; Among them, the frequency range of the transmitted signal is (A, B), both A and B are positive integers, and A is the lower limit of the range and B is the upper limit of the range. If the frequency corresponding to the coding sequence exceeds the range limit, the coding sequence corresponding to the transmitted signal exceeding the range limit is set as an invalid coding, and the coding sequence corresponding to the transmitted signal within the range limit is set as a valid coding; When the coding period of the valid coding is 0, it represents a single emission. When it is non-0, the coding period should be greater than the total sum of all frequency interval times corresponding to the coding sequence, and the difference between the coding period and the total sum of all frequency interval times is greater than F.

[0024] Specifically, the wavelength adaptive unit has the ability to automatically select an appropriate illumination wavelength according to the object to be measured. During the 3D scanning process, objects to be measured with different materials and characteristics have different reflection, absorption and other characteristics for light of different wavelengths. For example, some materials have a higher reflectivity for light of a specific wavelength, while some other materials may have absorption characteristics for certain wavelengths. The wavelength adaptive unit analyzes relevant information of the object to be measured (such as color, material type, etc.) through built-in sensors or algorithms, and automatically selects the most suitable illumination wavelength. This adaptive selection method can significantly improve the interaction effect between the signal and the object to be measured, enhance the intensity and quality of the reflected signal, and thus provide a more reliable basis for subsequent measurement and data processing. By optimizing the illumination wavelength, the adaptability and measurement accuracy of the scanning system to objects to be measured of various materials can be effectively improved, measurement errors caused by wavelength mismatch can be reduced, and accurate and clear data of the object to be measured can be obtained in different measurement scenarios. The polarization encoding unit is responsible for generating circularly polarized light containing topological phase. Topological phase is a special phase characteristic that can carry additional information. By introducing topological phase into circularly polarized light, unique coding characteristics can be given to the transmitted signal. In the mentioned topological phase modulation formula, the topological charge number l is a key parameter, which belongs to the set {1, 2, 3, 4}. Different topological charge numbers correspond to different topological phase distributions, which provides multiple possibilities for signal coding. The equivalent lens focal length f also affects the modulation effect of topological phase. By adjusting the values of l and f, the characteristics of topological phase in circularly polarized light can be precisely controlled, and the coding of the transmitted signal can be realized. This coding method increases the complexity and information content of the signal, helps to improve the anti-interference ability and measurement accuracy of the system in a complex environment. Using topological phase for coding provides a novel and effective signal coding means for the 3D scanning system, which can enhance the uniqueness and identifiability of the signal. Especially in complex measurement environments such as noise interference or multipath reflection, it helps to accurately extract information of the object to be measured and improve the reliability and measurement accuracy of the system. The main responsibility of the safety protection unit is to monitor the light power in real time. During the operation of the dynamic coding emission module, the stability and safety of the light power are crucial. If the light power is too high, it may not only cause damage to the object to be measured (such as thermal or chemical effects on objects made of certain sensitive materials), but also damage the optical components of the measurement system itself, affecting the normal operation and service life of the system. The safety protection unit monitors the light power in real time. Once it detects that the power exceeds the preset threshold, it will immediately trigger an emergency shutdown mechanism to cut off the light emission, thus protecting the safety of the object to be measured and the measurement system. The safety protection unit provides reliable safety protection for the entire scanning system, ensuring that the system can operate within a safe range under various measurement conditions, avoiding equipment damage or measurement accidents caused by abnormal light power, and improving the stability and reliability of the system.

[0025] Among them, the frequency range of the transmitted signal is limited between (A, B), where both A and B are positive integers. A represents the lower limit of the range, and B represents the upper limit of the range. This setting of the frequency range is to ensure that the transmitted signal works within a specific frequency band to meet the measurement requirements and electromagnetic compatibility requirements of the system. For example, it avoids frequency interference with other wireless devices or electronic systems, and at the same time ensures the stability and accuracy of the signal during transmission and processing. When the frequency corresponding to the coding sequence exceeds the range limit, the coding sequence corresponding to the transmitted signal exceeding the range limit is set as an invalid code; while the coding sequence corresponding to the transmitted signal within the range limit is set as a valid code. This determination rule clarifies the definition of the validity of the transmitted signal coding, ensuring that only the coding sequences that meet the frequency range requirements can be recognized and processed by the system, and avoiding interference of invalid signals on the measurement results. For valid codes, there are specific requirements for setting the coding period. When the coding period is 0, it means a single transmission, that is, the signal is transmitted once and not repeated. When the coding period is non-zero, the coding period should be greater than the sum of all frequency interval times corresponding to the coding sequence, and the difference between the coding period and the sum of all frequency interval times is greater than F. This rule setting is to ensure that the coding signal has a reasonable time interval and rhythm during transmission, avoid signal overlap or confusion caused by improper arrangement of frequency interval times, ensure that each coding signal can be accurately recognized and processed, and thus ensure the accuracy and reliability of the measurement data.

[0026] The multi-modal sensing module is implemented through a sensor array; The sensor array includes a vision layer and a tactile layer; The vision layer includes: A central camera that defines a first search area with the image center as the origin, and its radius expands adaptively with the target movement speed; Multiple annular cameras are circumferentially arranged around the central camera, and the fan-shaped imaging areas of each camera cover the blind areas of the first search area; Among them, the vision layer estimates the position of a high-speed moving target through an edge prediction algorithm and aligns the data with the tactile layer in space and time; The tactile layer includes: A capacitive sensing array deployed in the central area that maps the surface topography of an object through the change in capacitance; An ultrasonic sensing unit deployed in the peripheral area that improves the resolution through echo superposition; Among them, the data of the capacitive sensing array and the ultrasonic sensing unit are fused through a hypergraph neural network.

[0027] Specifically, the multi-modal sensing module realizes multi-dimensional information acquisition by means of a sensor array. The sensor array adopts a hierarchical structure, including a vision layer and a tactile layer. This hierarchical design integrates sensors of different sensing types, can give full play to the advantages of various sensors, comprehensively obtain information of the object to be measured from different angles, and provide a rich data source for subsequent data fusion and analysis.

[0028] Among them, the vision layer is composed of a central camera and a plurality of annular cameras arranged circumferentially around the central industrial camera. As the core of the vision system, the central camera usually has high resolution and imaging quality, and can obtain clear and accurate image information of the central area of the object to be measured. The plurality of annular cameras are arranged around the central camera. This layout can expand the visual coverage range and achieve omnidirectional and dead-angle-free image capture of the object to be measured. Through the collaborative work of the central camera and the annular cameras, the vision layer can capture a dynamic image sequence of the surface of the object to be measured. During the 3D scanning process, the object to be measured may be stationary or in motion. The dynamic image sequence can record the changes in the surface features of the object to be measured at different moments, including information such as shape, texture, and color. These image data provide important visual basis for subsequent three-dimensional reconstruction, feature extraction, and data analysis, and help to more accurately restore the true morphology of the object to be measured.

[0029] The tactile layer adopts a pressure sensor array, which is composed of a plurality of pressure sensors arranged according to a certain rule. The pressure sensor can sense the magnitude and distribution of the external pressure. By integrating a plurality of pressure sensors together to form an array, the pressure information of a large area can be collected. The pressure sensor array is used to generate a non-contact force distribution matrix. In the 3D scanning scenario, although it is called the "tactile layer", in fact, information is obtained in a non-contact manner (such as based on principles such as optics and electromagnetics to sense the interaction force or pressure distribution with the object). The non-contact force distribution matrix can reflect the force distribution on the surface of the object to be measured at different positions, such as the local deformation of the object surface and the influence of surface roughness on the force. By analyzing the non-contact force distribution matrix, information on the surface mechanical properties of the object to be measured, such as hardness and elasticity, can be obtained. These information complement the image information obtained by the vision layer and provide strong support for comprehensively understanding the characteristics of the object to be measured.

[0030] It should be noted that the visual layer and the tactile layer obtain information about the object under test from different perceptual dimensions. Visual information intuitively shows the appearance characteristics of the object, while tactile information reflects the mechanical properties of the object. The combination of the two can overcome the limitations of a single sensor in information acquisition and provide a more comprehensive and accurate description of the object under test. The fusion of multi-modal information can be mutually verified and supplemented, reducing measurement deviations caused by errors in a single sensor. For example, when visual information is affected by factors such as lighting and occlusion, tactile information can provide auxiliary judgment; conversely, there may be a certain degree of ambiguity in the acquisition of tactile information, and visual information can provide more accurate spatial positioning and feature recognition. The design of the multi-modal sensing module enables the 3D scanning system to adapt to a wider range of application scenarios. In the field of industrial inspection, it can not only accurately measure the size and shape of an object, but also evaluate the surface quality and mechanical properties of the object; in the medical field, it can be used for three-dimensional modeling and mechanical property analysis of human tissues or organs; in robot grasping and operation tasks, it can help the robot better perceive the characteristics of the object and environmental information, achieving more intelligent and precise operations.

[0031] The central camera takes a picture and forms an image, marked as P1, and the annular cameras take pictures and form images, marked as P2, P3, …, P n , where n is the number of annular cameras, taking positive integers; Set a first search area for the marked image P1. Taking the image center as the origin, define a circular area with a radius of R1, where R1 = 0.4 × image width. When the target movement speed of the object under test > V1, R1 automatically expands to R2 = R1 × 1.5, and the edge prediction algorithm is enabled to estimate the target position. The first search area covers the object under test; For the marked images P2, P3, …, P n Set a second search area. Based on the camera installation angle θi, define the angular range [θ i -Δθ, θ i +Δθ] of each sector area in the second search area, where Δθ = 30°. The sector area is the imaging area of each annular camera. Among them, the second search area covers the blind area of the first search area.

[0032] Specifically, the central camera takes a picture and forms an image marked as P1, and the annular cameras take pictures and form images marked as P2, P3, …, P n(n is the number of circular cameras and is a positive integer). This marking method facilitates the differentiation and management of images obtained by different cameras, providing clear identification for subsequent image processing and target search operations. Taking the image center as the origin, a circular area with a radius of R1 is defined as the first search area, where R1 = 0.4 × image width. In most cases, the main features or targets of the object to be measured may be concentrated near the center area of the image. Setting the initial search area as a circular area centered at the image center with a radius of 0.4 times the image width can effectively narrow the search range, improve the search efficiency, and reduce unnecessary computational complexity. When the target movement speed of the object to be measured > V1, R1 automatically expands to R2 = R1 × 1.5, and the edge prediction algorithm is enabled to estimate the target position. This is because when the target moves at a relatively high speed, in the next frame of the image, the target may have moved out of the initially set smaller search area. By expanding the search area radius by 1.5 times, the probability of finding a fast-moving target is increased. At the same time, enabling the edge prediction algorithm can estimate the possible position of the target in the current frame of the image based on information such as the position, speed, and movement direction of the target in the previous frame of the image, further improving the search accuracy and efficiency, and ensuring effective tracking and capture of the target even when the target moves quickly. The first search area needs to cover the object to be measured, which is the basic premise for ensuring accurate acquisition of information about the object to be measured. If the search area cannot cover the object to be measured, it may lead to target loss or incomplete information acquisition, affecting subsequent measurement and analysis results.

[0033] Based on the camera installation angle θ i , the angular range of each sector area in the second search area is defined as [θ i −Δθ, θ i +Δθ], where Δθ = 30°. Since the circular cameras are arranged circumferentially around the central camera, each circular camera has its specific installation angle θ i . Centered on this installation angle, a sector area is defined by expanding 30° to the left and right as the search area. This setting method fully considers the imaging perspective and layout characteristics of the circular cameras, can reasonably divide the search area, and ensures that each circular camera is responsible for target search within its specific viewing angle range. The second search area needs to cover the blind area of the first search area. Since the first search area of the central camera is a circular area, there may be some positions at the edge or corner of the image that cannot be effectively covered, and these positions form the blind area of the first search area. By reasonably arranging the circular cameras and setting their search areas as sectors, the sector search areas of multiple circular cameras cooperate with each other to fill the blind area of the central camera's circular search area, achieving all-round and non-blind-spot search and monitoring of the object to be measured, and ensuring complete information acquisition of the object to be measured.

[0034] It should be noted that by setting reasonable search areas for different cameras, dynamically adjusting the search areas according to the target motion state, and combining with the edge prediction algorithm, the search efficiency and accuracy for the target of the object to be measured can be significantly improved. In a complex 3D scanning scenario, quickly and accurately locating and tracking the object to be measured is the key to ensuring the measurement quality, and this search area setting method helps to achieve this goal. Considering factors such as the rapid movement of the target and the blind areas of the search area, by dynamically expanding the search area and complementary setting of the search areas of multiple cameras, the adaptability and robustness of the system under different working conditions are enhanced. Even when the target motion state changes or there are some areas that are difficult to search, the system can still reliably obtain the information of the object to be measured, ensuring the stability and reliability of the 3D scanning system. At the same time, accurately searching and obtaining the information of the object to be measured is the basis for multi-modal data fusion. Through reasonable search area setting, it is ensured that the image data obtained by different cameras in the vision layer can completely and accurately reflect the characteristics of the object to be measured, providing high-quality visual information for subsequent fusion analysis with the data of the tactile layer, which helps to achieve a more comprehensive and in-depth understanding and measurement of the object to be measured.

[0035] The pressure sensing array is arranged in a grid interleaved manner, with a capacitive array deployed in the central area and ultrasonic units deployed in the peripheral area; The capacitive array establishes a mapping relationship between the capacitance change amount and the object surface topography: ; Wherein, is the capacitance change amount, reflecting the degree of electric field distortion caused by the approach of the object; is the vacuum permittivity, characterizing the propagation ability of the electric field in vacuum; is the relative permittivity, describing the dielectric properties of the material relative to vacuum, is the distance distribution function from the object surface to the sensor, which changes with the position (x, y); is the integral area element, representing the differential area covered by the sensor array; The ultrasonic unit is used to improve the resolution: ; Wherein, is the imaging intensity, representing the superposition result of the echo signals at the position (x, y); is the number of sensors, the total number of ultrasonic units participating in imaging. is the echo signal of the i-th sensor, is the sound speed, enhancing the imaging accuracy through superposition; t is the echo time, representing the time delay from the emission to the reception of the sound wave; (x i , y i) The coordinate position of the i-th sensor defines its spatial position in the array.

[0036] Specifically, the pressure sensing array adopts a grid staggered layout, which can increase the density and distribution uniformity of sensors. Compared with the regular grid layout, the staggered layout can reduce the mutual interference between sensors, improve the perception accuracy of pressure distribution, enable the sensor array to capture more subtly the pressure changes applied to the object surface, and provide richer data for subsequent pressure distribution analysis and 3D reconstruction. A capacitance array is deployed in the central area of the array. The capacitance sensor is very sensitive to the electric field changes caused by the approach of an object. When an object approaches the sensor surface, it will change the electric field distribution around the sensor, resulting in a change in the capacitance value. The central area is usually the most concentrated and crucial part where the sensor array senses pressure. Deploying a capacitance array can make full use of its sensitive perception ability to electric field changes to accurately capture the topography information of the object surface in this area. Ultrasonic units are deployed in the peripheral area. The ultrasonic sensor obtains information by emitting and receiving ultrasonic waves and has good penetration and spatial resolution. On the basis of obtaining preliminary topography information through the capacitance array in the central area, the ultrasonic units in the peripheral area can further expand the measurement range and supplement the measurement of some areas that are difficult to accurately sense by the capacitance array, and at the same time improve the resolution of the overall system by using the characteristics of ultrasonic waves. Among them, in the formula, ΔC is the capacitance change amount, which directly reflects the degree of electric field distortion caused by the approach of an object. When the object surface approaches the sensor, it will change the electric field distribution between the sensor plates, causing the capacitance value to change. The magnitude of ΔC is closely related to the degree of this electric field distortion. is the vacuum permittivity, which is a physical constant representing the propagation ability of the electric field in a vacuum and is one of the basic parameters for calculating the capacitance change amount. is the relative permittivity, which describes the dielectric properties of a material relative to vacuum. Different materials have different relative permittivities, which affect the distribution of the electric field between the object and the sensor, and thus affect the change in capacitance. d(x,y) is the distance distribution function from the object surface to the sensor, which varies with the position (x,y). It reflects the distance differences between different positions on the object surface and the sensor, and is one of the key factors affecting the change in capacitance. dA is the integral area element, representing the differential area covered by the sensor array. By performing an integral operation over the entire area covered by the sensor array, the total change in capacitance caused by the change in the object surface topography can be calculated comprehensively. By establishing this mapping relationship, the capacitance change data measured by the capacitance array can be converted into the topography information of the object surface. In practical applications, by measuring the capacitance changes at different positions and combining the known parameters and formulas, the height or topography characteristics of the object surface at each position can be deduced inversely, providing important data basis for 3D reconstruction. At the same time, in the formula, I(x,y) is the imaging intensity, representing the superposition result of the echo signals at the position (x,y). N is the number of sensors, that is, the total number of ultrasonic units participating in imaging. A larger number of sensors means that more echo signal information can be obtained. s i is the echo signal of the i-th sensor. Each sensor emits ultrasonic waves and receives the echo signals reflected from the object surface. These echo signals contain information such as the position and reflection characteristics of the object surface. v is the speed of sound, which is a relatively stable value in a given medium. By measuring the echo time t (the time delay from the emission to the reception of the sound wave) and combining it with the speed of sound, the distance between the object surface and the sensor can be calculated. (x i , y i ) is the coordinate position of the i-th sensor, which defines its spatial position in the array. By superimposing and spatially correcting the echo signals of multiple sensors, the position of the object surface in space can be accurately determined. The ultrasonic units emit ultrasonic waves and receive echo signals. There are time delay differences in the echo signals received by sensors at different positions, and this difference reflects the distances between different positions on the object surface and the sensors. By superimposing the echo signals of multiple sensors, the intensity of the imaging signal can be enhanced and the noise interference can be reduced, thus improving the accuracy and resolution of imaging. At the same time, using the spatial position information of the sensors, the coordinates of the object surface in space can be accurately determined, further refining the perception of the object surface topography and achieving more accurate 3D imaging of the object surface.

[0037] It should be noted that the capacitive array and the ultrasonic unit complement each other in function. The capacitive array is sensitive to the topographical changes of the object surface at close range and can quickly capture the minute undulations on the surface; while the ultrasonic unit has good penetrability and spatial resolution, can measure the surface information of objects at relatively long distances, and make up for the deficiencies of the capacitive array in measurement range and depth perception. The combination of the two can provide more comprehensive and accurate object surface topography information. By establishing the mapping relationship between the capacitance change amount and the object surface topography and using the mechanism of the ultrasonic unit to improve the resolution, this pressure sensing array can more accurately obtain the three-dimensional information of the object surface. In the fields of 3D scanning, industrial inspection, medical imaging, etc., high-precision surface topography measurement is of great significance for product quality control, disease diagnosis, and scientific research. This design of the sensing array combining the capacitive array and the ultrasonic unit enables it to adapt to more complex measurement scenarios. For example, when measuring objects with diverse surface materials, irregular shapes, or obstructions, the capacitive array and the ultrasonic unit can respectively play their advantages and jointly complete the accurate measurement of the object surface, improving the adaptability and reliability of the system.

[0038] The data fusion processor includes: A data calibration module for calibrating the sensors and establishing the mapping relationship between the sensor feature vectors and the hypergraph structure; Set the sensor node sets for different sensors, and the sensor node set V = {v1,..., v n} corresponds to n types of sensor features; Among them, if the system contains n types of sensor features, then: Node set: ; In the formula, represents the feature vector of the i-th type of sensor; n represents the total number of sensor types (n ≥ 1); d i represents the feature dimension of the i-th type of sensor; represents d-dimensional real number space; Hyperedge set: ; In the formula, represents the j-th hyperedge; represents the set of positive integers; j ≥ 1; A data preprocessing module for achieving the spatio-temporal alignment and benchmark unification of multi-sensor data; S1. Unify the data space through timestamp synchronization and coordinate transformation matrix: ; In the formula, represents the alignment time; Denote finding a time transformation \(T\) such that the difference between the measurements of two sensors is minimized; \(\sum\) is the summation symbol, indicating the accumulation of the differences over all time points; and respectively represent the measurements of sensor \(i\) and sensor \(j\) at time \(t\); represents the result after applying the time transformation \(T\) to the measurement of sensor \(j\) at time \(t\); represents the square of the two - norm, used to quantify the difference between the measurement of sensor \(i\) and the measurement of sensor \(j\) after the time transformation \(T\); S2. Extract visual and tactile features; S3. Output a set of standardized feature vectors; Hyper - graph neural network, used to construct a multi - perspective hyper - graph to achieve cross - modal feature fusion; Dynamic reconstruction module, used to identify a key feature subset related to the morphology and material characteristics of the measured object and reconstruct the multi - perspective hyper - graph structure; Risk assessment module, used to generate a final risk decision.

[0039] Specifically, the core task of the data calibration module is to calibrate the multi - modal sensors and establish the mapping relationship between the sensor feature vectors and the hyper - graph structure, which is the basis for subsequent efficient data fusion and analysis. Through this mapping, the raw data of different types of sensors can be transformed into a form that can be processed and utilized in the hyper - graph structure, facilitating the exploration of the internal relationships between the data. Set the sensor node set \(V = \{v_1,\cdots,v n \}\) for different sensors, and each node \(v i corresponds to a type of sensor feature. In the node set, the feature vector of the \(i\) - th type of sensor is represented as , where \(n\) is the total number of sensor types (\(n\geq1\)), \(d i represents the feature dimension of the \(i\) - th type of sensor, represents d - dimensional real - number space. Clearly describes the feature information carried by each sensor node, providing a data basis for subsequent construction of the hyper - graph structure. By mapping different sensors to nodes in the hyper - graph and taking the sensor features as the attributes of the nodes, the relationships between different sensors and the information they provide can be intuitively represented. This representation method based on nodes and feature vectors helps to uniformly process and analyze multi - sensor data in the hyper - graph structure. Among them, the hyper - edge set is used to describe the complex association relationships between sensor nodes. Each hyper - edge \(e j(j ≥ 1) can connect multiple nodes. This connection method is different from the limitation in traditional graphs where an edge can only connect two nodes, and it can more flexibly represent the interaction and dependency relationships between multi-sensor data. For example, in some application scenarios, the data of multiple sensors may be simultaneously affected by a certain specific feature of the object under test. Through hyperedges, these related sensor nodes can be connected to form an associated subgraph, facilitating subsequent feature fusion and data analysis.

[0040] The data preprocessing module is to achieve the spatio-temporal alignment and benchmark unification of multi-sensor data, providing high-quality data for subsequent feature extraction and fusion. Due to differences in the working principles, sampling frequencies, and installation positions of different sensors, the data they acquire may be inconsistent in time and space. Therefore, preprocessing operations are required. By finding a time transformation T to minimize the difference between the measurement values of two sensors. During the processing, and respectively represent the measurement values of sensor i and sensor j at time t, represents the result after applying the time transformation T to the measurement value of sensor j at time t, represents the square of the L2 norm, which is used to quantify the difference between the two measurement values. By accumulating and minimizing the differences at all time points, the optimal time transformation T can be found to achieve the temporal alignment of the two sensors. In addition to temporal alignment, it is also necessary to unify the data spaces of different sensors. This is usually achieved through a coordinate transformation matrix to convert the coordinate systems of different sensors into a unified coordinate system, making the data they acquire comparable in space. Spatio-temporal alignment and benchmark unification are key steps in multi-sensor data fusion, which can avoid analysis errors caused by data inconsistency. After completing spatio-temporal alignment and benchmark unification, feature extraction is performed on the data of the visual layer (central camera and annular camera) and the tactile layer (pressure sensing array). Visual features may include information such as the shape, texture, and color of the object, and tactile features may include mechanical property information such as pressure distribution and surface hardness of the object. Among them, the extraction of features can be achieved by using existing technologies for feature stitching and fusion. By extracting these features, the original sensor data can be transformed into more representative and analyzable feature vectors. The extracted feature vectors are standardized to have the same scale and distribution characteristics. The standardized set of feature vectors can be used as the input for subsequent modules such as hypergraph neural networks for feature fusion and analysis.

[0041] Hypergraph neural networks are used to construct multi-view hypergraphs and achieve cross-modal feature fusion. In a multi-modal sensor system, sensor data of different modalities provide different perspective information about the object under measurement. Hypergraph neural networks integrate feature information of different modalities by constructing a hypergraph structure, and utilize the powerful learning ability of neural networks to explore the potential correlations and complex patterns among data, thus achieving effective fusion of cross-modal features. Hypergraph neural networks perform information propagation and feature update on the constructed hypergraph structure. Each node interacts with other nodes through the hyperedges it connects, and hyperedges can be regarded as channels for information transmission between nodes. During the information propagation process, the neural network updates and fuses the features of nodes according to the features of nodes and the weights of hyperedges. Through multiple rounds of information propagation and feature update, hypergraph neural networks can learn the internal connections between different-modal features, generate more discriminative fusion features, and provide strong support for subsequent tasks such as risk assessment.

[0042] The main task of the dynamic reconstruction module is to identify the key feature subset related to the morphology and material characteristics of the object under test, and reconstruct the multi-view hypergraph structure based on these key features. In practical applications, different features of the object under test may have different importance for different analysis tasks. Through the dynamic reconstruction module, according to the specific application requirements and the characteristics of the object under test, the most critical feature subset for the current task can be screened out, and the hypergraph structure can be adjusted to make it more focused on these key features, improving the efficiency and accuracy of data analysis. This module can adopt some feature selection algorithms, such as statistical-based methods, machine learning-based methods, etc., to evaluate the importance of each feature for morphology and material characteristics. According to the evaluation results, select the features with higher importance to form the key feature subset. Then, adjust the node and hyperedge connection relationships in the hypergraph according to the key feature subset to reconstruct the multi-view hypergraph structure. For example, if it is found that certain tactile features are very critical for material recognition, and certain visual features are more important for morphology measurement, then the connection between the corresponding nodes of these key features can be strengthened to construct a hypergraph structure that better meets the requirements of the current task. Risk assessment module Function overview: The risk assessment module generates the final risk decision based on the data processed and analyzed by the previous modules. In many application scenarios, such as industrial inspection, medical diagnosis, etc., it is necessary to conduct a risk assessment on the state of the object under test and make corresponding decisions according to the assessment results. The risk assessment module comprehensively utilizes the information after multi-sensor fusion, combines the preset risk assessment models and rules, quantitatively evaluates the risk level of the object under test, and outputs the final risk decision result. Evaluation basis: This module may consider the morphology features of the object under test (such as whether there are defects, deformations, etc.), material features (such as hardness, elasticity, etc.), and other relevant information, and calculate the risk index according to the correlation between these features and the risk. For example, in industrial product quality inspection, if serious defects (morphology features) are detected on the product surface and the material hardness does not meet the requirements (material features), then the risk assessment module may determine that the product has a high quality risk and output corresponding risk decisions, such as rejecting the product, suggesting further inspection, etc.

[0043] The implementation of the hypergraph neural network includes: S4. Hypergraph construction, each sensor feature vector is used as an independent node, and hyperedges are established based on physical constraints; Node feature matrix: X ∈ ℝ {n×d} ; Hyperedge incidence matrix: H ∈ {0, 1} {n×m} , H[i, j] = 1 indicates that node v i belongs to hyperedge e j ; In the formula, d is the unified feature dimension; n represents the total number of nodes; m represents the total number of hyperedges; Discover implicit associations using k-means clustering and adaptively adjust the hyperedge dimension; S5. Hypergraph Convolution: ; Among them, represents the node feature matrix of the -th layer; H represents the hyperedge association matrix; represents the node degree diagonal matrix; represents the hyperedge degree diagonal matrix; W represents the hyperedge weight matrix; represents the trainable parameter matrix of the -th layer; represents the non-linear activation function; S6. Multi-level Fusion: Stack multiple hypergraph convolution layers to gradually fuse local and global features.

[0044] Specifically, the Hypergraph Neural Network (HGNN) is a graph neural network that can handle complex high-order relationships and is particularly suitable for feature fusion of multi-modal sensor data. Among them, each sensor feature vector is used as an independent node, and the node feature matrix is represented as X ∈ ℝ {n×d} , where n is the total number of nodes and d is the unified feature dimension. This definition method enables each sensor node to carry its unique feature information, providing a basis for subsequent hypergraph construction and feature fusion. The hyperedge association matrix is used to describe the relationship between nodes and hyperedges. In this way, hyperedges can connect multiple nodes, thereby representing the complex association relationships between multi-sensor data. Use the k-means (K-means) clustering algorithm to cluster the node features and discover potential implicit associations. The clustering results can be used to construct hyperedges, so that each hyperedge corresponds to a clustering cluster, thereby connecting nodes with similar features. According to the clustering results and actual needs, dynamically adjust the dimension of the hyperedges. For example, if the number of nodes in a certain clustering cluster is large or the feature differences are large, the dimension of this hyperedge can be appropriately increased to more finely describe the relationship between nodes. Hypergraph convolution realizes the propagation and update of node features through the hyperedge association matrix H, the node degree diagonal matrix , the hyperedge degree diagonal matrix and the weight matrix W. In the formula, represents the node feature matrix of the l-th layer. H is the hyperedge association matrix, which is used to map node features to the hyperedge space. W is the hyperedge weight matrix, which is used to adjust the importance of different hyperedges. Maps the features in the hyperedge space back to the node space. is the trainable parameter matrix of the l-th layer, which is used to learn the transformation of node features. It is a non - linear activation function used to introduce non - linearity and enhance the expressive power of the model. The node - degree diagonal matrix Describes the number of hyper - edges connected to each node and is used to normalize node features. The hyper - edge - degree diagonal matrix Describes the number of nodes connected to each hyper - edge and is used to normalize hyper - edge features. In practical applications, H is usually normalized to balance the influence of nodes and hyper - edges. By stacking multiple hyper - graph convolutional layers, local and global features are gradually fused. Each layer of convolutional operation fuses the node features of the current layer with the features of adjacent nodes to capture more complex feature patterns. During the feature fusion process, in the shallow network, hyper - graph convolution mainly focuses on the local feature interaction between nodes and directly - connected hyper - edges. In the deep network, as the number of network layers increases, node features are propagated and fused multiple times through hyper - edges, thus capturing global feature patterns. Multi - level fusion can effectively combine feature information at different levels, retaining both local details and capturing global structural information, thereby improving the performance of the model in complex tasks.

[0045] Therefore, through constructing a hyper - graph structure, implementing hyper - graph convolution operations, and multi - level feature fusion, hyper - graph neural networks can effectively handle complex correlation relationships in multi - modal sensor data. Their unique hyper - edge representation method and adaptive adjustment mechanism enable the model to flexibly capture the interactions and dependencies between different sensor data.

[0046] The implementation of the dynamic reconstruction module includes: S7. Key feature recognition: Based on the gradient - risk - assessment feature region, output the node importance score S i ∈[0,1]; Specifically, through gradient - risk - assessment technology, identify the feature regions that have an important impact on risk assessment and generate an importance score Si ∈[0,1] for each node. By calculating the gradient of the model output with respect to the input features, evaluate the influence degree of each feature on the final risk - assessment result. The magnitude of the gradient can reflect the importance of the feature. The larger the gradient, the more significant the influence of the feature on the output result. Based on the gradient information, calculate the importance score Si for each node (i.e., each sensor feature vector). The importance score is usually obtained by normalizing the gradient value, ranging from 0 to 1. 1 indicates that the node is crucial for risk assessment, and 0 indicates that the node has little impact.

[0047] S8. Hyper - graph reconstruction: Delete the nodes or hyper - edges with S i < τ, where τ is a preset threshold; Dynamically adjust the hyper - edge coverage range based on feature similarity; Adopt a continuous - learning framework to incrementally update hyper - graph parameters.

[0048] Specifically, according to the key feature recognition results, the hypergraph structure is adjusted by deleting unimportant nodes or hyperedges and dynamically adjusting the coverage range of hyperedges to improve the efficiency and accuracy of the model. During the operation, an importance threshold τ is set. For nodes with importance scores Si < τ or their related hyperedges, they are deleted from the hypergraph. This step can effectively reduce the computational amount and eliminate the impact of noisy data on the model. For the remaining nodes and hyperedges, the connection range of hyperedges is dynamically adjusted according to their feature similarity. Feature similarity can be measured by calculating the cosine similarity or Euclidean distance between node feature vectors. Nodes with high similarity can increase the hyperedges they connect to better capture the correlation between features. The Continual Learning framework is adopted to incrementally update the hypergraph parameters. Continual Learning allows the model to retain the knowledge learned previously while continuously receiving new data, avoiding catastrophic forgetting. Through incremental updates, the model can adapt to environmental changes and the arrival of new data, maintaining the stability and accuracy of its performance.

[0049] The implementation of the risk assessment module includes: S9. Input the updated hypergraph node features X'; S10. A fully connected layer and Softmax to output the risk probability p ∈ [0, 1]; S11. If p > θ, it is determined as a high risk, where θ is determined by the ROC curve.

[0050] Specifically, the risk assessment module receives the updated hypergraph node features X' from the dynamic reconstruction module. These features have undergone multi-level feature fusion and dynamic reconstruction, containing the morphology, material, and other relevant feature information of the object under test, and can comprehensively reflect the state of the object under test. Before being input into the risk assessment module, the hypergraph node features X' usually undergo standardization or normalization processing to ensure the consistency of feature scales and avoid evaluation biases caused by feature scale differences. Through the fully connected layer and the Softmax function, the updated hypergraph node features are mapped to the risk probability space to output the probability p that the object under test belongs to the high-risk category. Among them, the fully connected layer converts the input feature X' into a high-dimensional representation to capture the complex interaction relationships between features. The number of neurons in the fully connected layer can be adjusted according to the specific task, and usually, the optimal number is determined through experiments. The Softmax function is applied in the output layer to convert the output of the fully connected layer into a probability distribution.

[0051] The formula of the Softmax function is: ; where is the value of the th neuron output by the fully connected layer; is the corresponding risk probability; e is the natural constant; j is the index variable traversing all categories; through the Softmax function, the output probability satisfies 0 ≤ ≤ 1 and , representing the probabilities that the object under test belongs to different risk categories.

[0052] According to the output risk probability and in combination with the preset threshold θ, it is determined whether the object under test belongs to the high-risk category. The threshold θ is determined by the ROC curve (Receiver Operating Characteristic Curve). The ROC curve helps select the optimal threshold by plotting the true positive rate (TPR) and false positive rate (FPR) at different thresholds, so that the model achieves the best balance between sensitivity and specificity. If the risk probability p > θ, it is determined that the object under test is in the high-risk category; otherwise, it is determined to be in the low-risk category. This determination result can be directly used in the subsequent decision support system to trigger corresponding alarms or intervention measures.

[0053] This application also includes cross-control logic: The dynamic coding and transmitting module adjusts the polarization coding parameters in real time according to the mechanical distribution data of the tactile layer; The three-coordinate bridge measuring structure dynamically plans the obstacle avoidance scanning path based on the output of the risk assessment module; The safety protection unit synchronously triggers the self-checking protocol of the data fusion processor when detecting abnormal optical power.

[0054] Specifically, this application forms a complete perception - decision - execution chain through tactile data, coding adjustment, measurement path optimization, and safety monitoring.

[0055] The above are only several embodiments of this application and do not impose any form of limitation on this application. Although this application is disclosed as above with preferred embodiments, it is not intended to limit this application. Any person skilled in the art, without departing from the scope of the technical solution of this application, making some changes or modifications using the disclosed technical content is equivalent to equivalent implementation cases and all belong to the scope of the technical solution.

Claims

1. Mobile bridge-type multi-sensor 3D scanning system, characterized in that, Including: A measurement platform (1) for placing the object to be measured; A three - coordinate bridge - type measurement structure, which is installed on the measurement platform (1) for adjusting the measurement orientation, and a detection head (2) is installed on the three - coordinate bridge - type measurement structure. The detection head (2) integrates a dynamic coding emission module and a multi - modal sensing module to achieve all - round measurement of the object to be measured; A dynamic coding emission module, integrated in the detection head, configured to dynamically adjust the signal emission direction or frequency; the polarization coding unit of the dynamic coding emission module generates circularly polarized light containing topological phase; A multi - modal sensing module, integrated in the detection head, including a vision layer and a tactile layer. The vision layer captures a dynamic image sequence of the surface of the object to be measured through a multi - camera array, and the tactile layer generates a non - contact force distribution matrix through a pressure sensor array and receives the light reflected from the object to be measured according to the non - contact force distribution matrix. The data received by different types of sensors are used to capture all - dimensional data of the object to be measured; A data fusion processor, built - in with a hypergraph neural network model, to realize the feature fusion of multi - source heterogeneous data received by different types of sensors. The hypergraph neural network model constructs multi - perspective views of the morphology and material of the object to be measured based on the data of the object to be measured; among them, nodes represent the features of different types of sensors, and hyper - edges represent cross - modal associations, forming a multi - perspective hypergraph; at the same time, identify the key feature subset related to the morphology and material features of the object to be measured, and use the key feature subset to reconstruct the multi - perspective hypergraph structure, thereby updating the hypergraph neural network; according to the preset risk assessment threshold of the morphology and material features of the object to be measured, use the updated hypergraph neural network to determine the risk decision result of the object to be measured.

2. The mobile bridge-type multi-sensor 3D scanning system according to claim 1, wherein The three - coordinate bridge - type measurement structure includes: An X - axis movement mechanism (3), arranged on the upper end surface of the measurement platform (1), with the object to be measured located above the X - axis movement mechanism (3). The X - axis movement mechanism (3) guides the object to be measured to move along the X - axis direction of the measurement platform (1); A bridge (4), which is horizontally erected above the measurement platform (1) and fixed on both sides of the measurement platform (1). A Z - axis movement mechanism (5) is arranged on the bridge (4), and the Z - axis movement mechanism (5) moves horizontally along the bridge (4), thereby realizing the Y - axis movement of the detection head (2). The detection head (2) is installed on the Z - axis movement mechanism (5) to realize the Z - axis movement of the detection head (2); The X - axis movement mechanism (3) is located below the bridge (4).

3. The mobile bridge type multi-sensor 3D scanning system according to claim 1, characterized in that, The dynamic coding emission module includes: A wavelength adaptive unit that automatically selects the illumination wavelength according to the surface material of the object to be measured; A polarization coding unit that modulates the topological phase distribution of circularly polarized light; A safety protection unit that monitors the optical power in real - time and triggers an emergency shutdown; Among them, the frequency range of the signal is dynamically limited by the coding sequence, and invalid coding sequences are automatically filtered.

4. The mobile bridge type multi-sensor 3D scanning system according to claim 1, wherein The multi - modal sensing module is realized through a sensor array; The sensor array includes a vision layer and a tactile layer; The vision layer includes: A central camera defines a first search area with the image center as the origin, and its radius expands adaptively with the target movement speed; A plurality of annular cameras are circumferentially arranged around the central camera, and the fan-shaped imaging areas of each camera cover the blind areas of the first search area; Among them, the vision layer estimates the position of the high-speed moving target through an edge prediction algorithm and aligns the data with the tactile layer in space and time; The tactile layer includes: A capacitive sensing array is deployed in the central area to map the surface topography of the object through the change in capacitance; An ultrasonic sensing unit is deployed in the peripheral area to improve the resolution through echo superposition; Among them, the data of the capacitive sensing array and the ultrasonic sensing unit are fused by a hypergraph neural network.

5. The mobile bridge-type multi-sensor 3D scanning system according to claim 1, characterized in that The data fusion processor includes: A data calibration module that establishes the mapping relationship between the sensor feature vector and the hypergraph node; A dynamic reconstruction module that filters the key feature subset based on gradient risk assessment and reconstructs the hypergraph structure; A risk assessment module that outputs the risk probability through a fully connected layer and a Softmax function; Among them, the hyperedge set of the hypergraph neural network is adaptively adjusted by k-means clustering.

6. The mobile bridge-type multi-sensor 3D scanning system according to claim 1, characterized in that The hypergraph neural network performs the following operations: Construct a node feature matrix and a hyperedge association matrix, where the nodes represent different sensor feature dimensions; Iteratively update the node features through the hypergraph convolutional layer to fuse local and global associations; Adopt a continuous learning framework to dynamically optimize the hyperedge weights and delete redundant nodes based on feature similarity.

7. The mobile bridge type multi-sensor 3D scanning system according to claim 3, characterized in that, It also includes a cross-control logic: The dynamic coding and transmitting module adjusts the polarization coding parameters in real time according to the mechanical distribution data of the tactile layer; The three-coordinate bridge measurement structure dynamically plans an obstacle avoidance scanning path based on the output of the risk assessment module; The safety protection unit synchronously triggers the self-checking protocol of the data fusion processor when detecting abnormal optical power.

Images