Wearable building internal disease detection system and method

Abstract

The invention belongs to the technical field of engineering geophysical prospecting, and relates to a wearable building internal disease detection system and method, and the system comprises a backpack type base body which is internally provided with a containing cavity; the movable unit is mounted at the top of the base body; the sensing unit comprises a ground penetrating radar installed on the movable unit and used for detecting internal diseases of the structure, a depth camera arranged on the upper side face of the ground penetrating radar and used for detecting surface diseases of the structure and laser distance measuring sensors used for detecting the position and posture of the ground penetrating radar, and the laser distance measuring sensors are arranged on the four side faces of the ground penetrating radar and form a laser sensing array; the control unit is arranged in the accommodating cavity, and controls the movable unit to act to adjust the fitting position and posture of the ground penetrating radar according to a fitting path track generated by the position and posture of the ground penetrating radar; a disease diagnosis model for identifying disease types and positions according to internal disease information and surface disease information is arranged in the control unit; and the power module supplies power to the system. According to the invention, efficient and automatic data acquisition can be carried out in the building.

Description

Technical Field

[0001] The present invention belongs to the technical field of engineering geophysical exploration, and in particular, relates to a wearable building internal disease detection system and method. Background Art

[0002] In the construction field, buildings and civil infrastructure gradually age over time, which inevitably leads to the degradation of their structural performance, posing a potential threat to public safety. Therefore, it is crucial to regularly inspect and evaluate the structural condition of buildings and infrastructure. However, the internal structure of a building is usually a hidden project, which is difficult to detect and evaluate. Traditional detection methods are usually destructive detection, which has obvious limitations, such as reliance on manual operation, damage to the structure, and low detection efficiency.

[0003] With the advancement of non-destructive testing technology, the industry has gradually begun to adopt technologies such as ground penetrating radar (GPR) to detect damage to buildings. These non-destructive testing technologies can obtain internal information without damaging the structure, greatly reducing the impact of human factors on the test results. However, the current application of non-destructive testing equipment and technology is mainly concentrated in outdoor environments, such as bridge and road inspections, and these large infrastructure inspection tasks are usually completed through data collection platforms such as unmanned aerial vehicles (UAVs), ground unmanned vehicles (UGVs) and vehicles.

[0004] For the internal structure inspection of buildings, the existing equipment is large in size and heavy in weight, making it difficult to enter the narrow indoor space and realize comprehensive detection of the interior of the building, especially in terms of rapid detection and efficient data collection. Therefore, at present, there is a lack of a small and portable internal building disease detection equipment that can perform through-type detection inside the building to achieve more efficient and comprehensive detection data collection. Summary of the invention

[0005] In view of the above-mentioned problems existing in the prior art, the present invention provides a wearable building internal disease detection system and method. With a lightweight design, ordinary staff can conduct efficient and automatic data collection inside the building, thereby realizing professional and low-cost detection of massive buildings, and significantly improving the convenience and efficiency of building internal structure detection.

[0006] In order to achieve the above object, the first aspect of the present invention provides a wearable building internal disease detection system, comprising: The base body is set as a backpack structure, and a receiving cavity is set in the base body; The movable unit comprises a mechanical arm, a first rotating platform connected to a first end of the mechanical arm, and a second rotating platform connected to a second end of the mechanical arm, wherein the first rotating platform is mounted on the top of the base; The sensing unit includes a ground-penetrating radar for detecting internal defects of the structure, a depth camera for detecting surface defects of the structure, and a laser ranging sensor for detecting the position and posture of the ground-penetrating radar. The ground-penetrating radar is installed on the second rotating pan-tilt platform, the depth camera is installed on the upper side of the ground-penetrating radar, and the laser ranging sensor is installed on the four sides of the ground-penetrating radar to form a laser sensor array; The control unit is arranged in the accommodating cavity, generates a fitting path trajectory of the ground penetrating radar according to the position and posture of the ground penetrating radar, and controls the movement of the mechanical arm, the first rotating pan-tilt table and the second rotating pan-tilt table according to the fitting path trajectory, and adjusts the fitting position and fitting posture of the ground penetrating radar so that the ground penetrating radar fits the measured area of ​​the building; a disease diagnosis model is arranged in the control unit, and the disease diagnosis model identifies the type and position of the disease according to the internal disease information detected by the ground penetrating radar and the surface disease information detected by the depth camera to generate a disease image and a radar profile; The power module is arranged in the accommodating cavity and supplies power to the active unit, the sensing unit and the control unit.

[0007] In some embodiments, the base is made of nylon-polyester blended material, and the base includes a wearable vest and a box at the back of the vest; the wearable vest includes a back pad and an adjustable shoulder strap connected to the back pad, and the box is installed on the back pad; the accommodating cavity is arranged in the box, and a carrying handle is provided on the top of the box.

[0008] In some embodiments, the control unit comprises: A PLC controller, whose communication port is connected to the laser ranging sensor through a serial port conversion module and a serial port module, and is used to receive the position and posture of the ground penetrating radar detected by the laser ranging sensor; An embedded processing unit is connected to the PLC controller and is used to generate a fitting path trajectory of the ground penetrating radar according to the position and attitude of the ground penetrating radar; the disease diagnosis model is set in the embedded processing unit; The robotic arm controller is respectively connected to the embedded processing unit, the robotic arm, the first rotating gimbal and the second rotating gimbal, and is used to control the movements of the robotic arm, the first rotating gimbal and the second rotating gimbal according to the fitting path trajectory, and adjust the fitting position and fitting posture of the ground penetrating radar so that the ground penetrating radar fits the measured area of ​​the building.

[0009] In some embodiments, the disease diagnosis module is constructed by: Data acquisition steps: Obtain historical surface disease information and internal disease information as well as annotated pixel-level disease types; Dataset construction steps: Construct a training set based on historical surface disease information, internal disease information, and labeled pixel-level disease types; Model construction steps: Based on the deep neural network, the disease diagnosis model is constructed with surface disease information and internal disease information as input and disease images and radar profiles as output; Model training step: training the disease diagnosis model through the training set.

[0010] In some embodiments, the disease diagnosis model includes a backbone network, a neck network, and a head network connected in sequence; The backbone network adopts CSP structure for feature extraction. The backbone network is equipped with a global attention mechanism module and an SPPF module. The global attention mechanism module is used to reduce the dispersion of information in the global dimension and enhance the interaction between global features. The SPPF module is used to enhance the model reasoning ability. The neck network adopts FPN structure and PAN structure. The FPN structure increases the amount of feature information in the feature map by top-down sampling, and the PAN structure increases the amount of position information in the upper feature map by bottom-up sampling. The head network uses a decoupling head to separate the detection head and the classification head, and adopts an anchor-free method to localize objects.

[0011] In some embodiments, the backbone network is also provided with n+1 convolution modules for convolution calculation and n C2f modules for feature extraction, the output of the first convolution module is connected to the input of the second convolution module, the input of the first C2f module is connected to the output of the second convolution module, the output of the first C2f module is connected to the input of the third convolution module, the convolution modules and the C2f modules are arranged alternately in sequence, the output of the n+1th convolution module is connected to the input of the nth C2f module, and the output of the nth C2f module is connected to the global attention mechanism module and the SPPF module in sequence.

[0012] In some embodiments, the C2f module includes: A first convolutional block, configured to receive an input feature map and generate an intermediate feature map; The feature map splitting module splits the intermediate feature map into two feature maps; N residual blocks, the N residual blocks are connected in sequence, the first residual block receives the first part of the feature map output by the feature map splitting module, performs convolution, normalization and activation operations to generate the first feature map, the second residual block receives the first feature map, performs convolution, normalization and activation operations to generate the second feature map, and so on, the Nth residual block receives the N-1th feature map, performs convolution, normalization and activation operations to generate the Nth feature map; A feature fusion module performs feature fusion on the second part of the feature map output by the feature map splitting module and the feature map generated by each residual block to obtain a fused feature map; The second convolution block receives the fused feature map and generates the final feature map.

[0013] In some embodiments, the total loss function of the disease diagnosis module is expressed as:

[0014] in,

[0015]

[0016] In the formula, is the total loss function, is the classification loss function, is the weighting factor of the classification loss function, is the focal distribution loss function, is the weighting factor of the focal distribution loss function, is the bounding box regression loss function, is the weighting factor of the bounding box regression loss function, is the high-quality anchor box loss function, is the loss function of the common quality anchor box, , is the center coordinate of the prediction box, , is the center coordinate of the predicted box corresponding to the real box, is the width of the real frame, is the height of the actual frame.

[0017] In some embodiments, it also includes a communication module and a mobile terminal. The communication module is arranged in the accommodating cavity. The ground penetrating radar communicates with the embedded processing unit through the communication module to transmit the detected internal disease information to the embedded processing unit; the mobile terminal communicates with the embedded processing unit through the communication module to receive the disease information identified by the embedded processing unit and forwarded by the communication module for real-time display.

[0018] In some embodiments, it also includes a positioning module and a storage module arranged in the accommodating cavity. The positioning module is connected to the global positioning system and the control unit to obtain the location information of the wearable building internal disease detection system and send it to the control unit. The storage module is connected to the control unit to store the detection data and disease identification results.

[0019] A second aspect of the present invention provides a wearable building internal disease detection method, using the wearable building internal disease detection system described in the first aspect of the present invention, comprising the following steps: Preparation steps: Wear the wearable building internal disease detection system and turn on the power module; Surface disease detection steps: Use the control unit to adjust the robotic arm in real time, aim the depth camera at the building surface, and collect surface disease data; Internal disease detection steps: detecting the position and posture of the ground penetrating radar by means of a laser ranging sensor, the control unit generating a fitting path trajectory of the ground penetrating radar according to the position and posture of the ground penetrating radar, and controlling the movement of the mechanical arm, the first rotating pan-tilt table and the second rotating pan-tilt table according to the fitting path trajectory, adjusting the fitting position and fitting posture of the ground penetrating radar, so that the ground penetrating radar fits the measured area of ​​the building, and the ground penetrating radar collects internal disease data; Disease identification step: The control unit receives and identifies surface disease data and internal disease data, and generates a disease image and a radar profile.

[0020] Compared with the prior art, the advantages and positive effects of the present invention are: (1) The wearable building internal disease detection system provided by the present invention adopts a lightweight design and is easy to carry. It is suitable for the detection of surface diseases and internal diseases of indoor structures of buildings. The surface disease information is detected by a depth camera, and the internal disease information is detected by a ground penetrating radar. In addition, an active unit is designed, and the depth camera and the ground penetrating radar are arranged at the end of the active unit. When detecting diseases, the positions of the depth camera and the ground penetrating radar are adjusted by controlling the movement of the active unit, which greatly improves the detection range and the flexibility of operation. By identifying the type and location of the disease through the pre-constructed disease diagnosis model, it is possible to achieve accurate detection of surface diseases and internal diseases of the structure.

[0021] (2) The wearable building internal defect detection system provided by the present invention detects the position and posture of the ground penetrating radar through a laser ranging sensor, and generates a fitting path trajectory of the ground penetrating radar according to the position and posture of the ground penetrating radar. According to the fitting path trajectory, the fitting position and fitting posture of the ground penetrating radar are adjusted through the designed active unit, so that the ground penetrating radar fits the measured area of ​​the building and comes into close contact with the structure, thereby improving the quality of the detection image and further improving the accuracy of detection and the precision of defect identification.

[0022] (3) The wearable building internal defect detection system provided by the present invention can enable ordinary staff to conduct efficient and automated data collection on the interior of a building during defect detection, thereby achieving professional and low-cost detection of a large number of buildings, significantly improving the convenience and efficiency of building internal structure detection.

[0023] (4) The wearable building internal disease detection method provided by the present invention adopts a lightweight wearable building internal disease detection system, which is suitable for the detection of surface diseases and internal diseases of indoor structures of buildings. When performing disease detection, ordinary staff can conduct efficient and automatic data collection inside the building, thereby realizing professional and low-cost detection of massive buildings, and significantly improving the convenience and efficiency of building internal structure detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the three-dimensional structure of the wearable building internal disease detection system according to an embodiment of the present invention; Figure 2 is a structural block diagram of the base body according to an embodiment of the present invention; Figure 3 is a structural block diagram of an activity unit according to an embodiment of the present invention; Figure 4 This is a control principle block diagram of the wearable building internal disease detection system according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the principle of detecting internal defects of a structure by a ground penetrating radar according to an embodiment of the present invention; Figure 6 A schematic diagram of possible paths of electromagnetic waves when a ground penetrating radar detects internal defects of a structure according to an embodiment of the present invention; Figure 7 A flow chart of constructing a disease diagnosis model for an embodiment of the present invention; Figure 8 A schematic diagram of the network structure of the disease diagnosis model according to an embodiment of the present invention; Fig. 9 This is a schematic diagram of the structure of the global attention mechanism module according to an embodiment of the present invention; Fig.10 The present invention is a flowchart of a wearable building internal disease detection method according to an embodiment of the present invention.

[0025] In the figure, 1, substrate, 101, wearable vest, 1011, back pad, 1012, shoulder strap, 102, box, 1021, carrying handle, 2, activity unit, 201, mechanical arm, 202, first rotating gimbal, 203, second rotating gimbal, 3, sensing unit, 301, ground penetrating radar, 302, depth camera, 303, laser ranging sensor, 4, control unit, 401, disease diagnosis model, 4011, backbone network, 4012, neck network, 4013, head network, 402, PLC controller, 403, serial port conversion module, 404, serial port module, 405, embedded processing unit, 406, mechanical arm controller, 5, power module, 6, communication module, 7 mobile terminal, 8, positioning module, 9, transmitting antenna, 10, receiving antenna, 11, surface layer, 12, covering layer, 13, waterproof layer, 14, concrete structure. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments provided by the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0027] Obviously, the drawings described below are only some examples or embodiments of the present invention. For ordinary technicians in this field, the present invention can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed by the present invention, some changes in design, manufacturing or production based on the technical content disclosed by the present invention are just conventional technical means, and should not be understood as insufficient content disclosed by the present invention.

[0028] Reference to "embodiments" in the present invention means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present invention may be combined with other embodiments without conflict.

[0029] Unless otherwise defined, the technical terms or scientific terms involved in the present invention shall have the usual meanings understood by persons with ordinary skills in the technical field to which the present invention belongs. The words "one", "a", "a", "the" and the like involved in the present invention do not indicate a quantity limitation and may indicate the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in the present invention are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device including a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in the present invention are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in the present invention refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships. For example, "A and / or B" may indicate: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects in the front and back relation are in an "or" relationship. The terms "first", "second", "third" and the like involved in the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0030] For the internal structure inspection of buildings, the existing equipment is large in size and heavy in weight, so it is difficult to enter the narrow indoor space and realize the comprehensive inspection of the interior of the building. Based on this, the present invention provides a lightweight, wearable building internal disease detection system and method. The system adopts a lightweight design and can be carried on the back of the inspector, so that the inspector can move freely and enter relatively narrow spaces or spaces that are difficult to reach with existing traditional detection equipment. In terms of the detection of surface diseases of building structures, a depth camera is used to capture the image data of the surface diseases of the building structure and transmit it to the control unit. The disease diagnosis model set in the control unit is used to identify the type and location of the disease to generate a disease image; in terms of the detection of internal diseases of the building structure, a ground penetrating radar is used to contact the structure at close range to capture the radar image of the interior of the building structure and transmit it to the control unit. The disease diagnosis model set in the control unit is used to identify the type and location of the disease to generate a radar profile. The present invention provides a lightweight, wearable building internal disease detection system and method. When performing disease detection, ordinary staff can perform efficient and automated data collection on the interior of the building, thereby achieving professional and low-cost detection of massive buildings, and significantly improving the convenience and efficiency of building internal structure detection. The following is a detailed description of a wearable building internal disease detection system and method provided by the present invention in conjunction with the accompanying drawings.

[0031] See also Figures 1 to 4According to a first aspect of the present invention, there is provided a wearable building internal disease detection system, comprising: A base body 1 is provided as a backpack-type structure, and a receiving cavity (not shown in the figure) is provided in the base body; The movable unit 2 includes a mechanical arm 201, a first rotating platform 202 connected to a first end of the mechanical arm 201, and a second rotating platform 203 connected to a second end of the mechanical arm 201, wherein the first rotating platform 202 is installed on the top of the base 1; The sensing unit 3 includes a ground penetrating radar 301 for detecting internal defects of the structure, a depth camera 302 for detecting surface defects of the structure, and a laser ranging sensor 303 for detecting the position and posture of the ground penetrating radar 301. The ground penetrating radar 301 is installed on the second rotating platform 203, the depth camera 302 is installed on the upper side of the ground penetrating radar 301, and the laser ranging sensor 303 is installed on four sides of the ground penetrating radar 301 to form a laser sensor array; The control unit 4 is arranged in the accommodating cavity, generates a fitting path trajectory of the ground penetrating radar according to the position and posture of the ground penetrating radar, and controls the movement of the mechanical arm 201, the first rotating gimbal 202 and the second rotating gimbal 203 according to the fitting path trajectory, and adjusts the fitting position and fitting posture of the ground penetrating radar 301 so that the ground penetrating radar fits the measured area of ​​the building; the control unit 4 is provided with a disease diagnosis model 401, and the disease diagnosis model 401 identifies the disease type and position according to the internal disease information detected by the ground penetrating radar 301 and the surface disease information detected by the depth camera 302, and generates a disease image and a radar profile; The power module 5 is arranged in the accommodating cavity and supplies power to the active unit 2, the sensing unit 3 and the control unit 4.

[0032] When detecting structural surface defects, the surface defect collection device consists of a depth camera, a robotic arm, a first rotating gimbal and a second rotating gimbal. The inspector uses the robotic arm, the first rotating gimbal and the second rotating gimbal to adjust the direction and angle of the depth camera in real time to scan and take pictures. The images acquired by the depth camera are transmitted to the control unit in real time, and the control unit identifies the defects through the disease diagnosis model.

[0033] When detecting internal defects of a structure, the internal defect collection device is composed of a ground-penetrating radar, a mechanical arm, a first rotating pan-tilt platform, a second rotating pan-tilt platform, and a laser ranging sensor. The inspector detects the position and posture of the ground-penetrating radar through the laser ranging sensor. The control unit generates a fitting path trajectory of the ground-penetrating radar according to the position and posture of the ground-penetrating radar, and adjusts the fitting position and fitting posture of the ground-penetrating radar according to the fitting path trajectory, so that the ground-penetrating radar fits the measured area of ​​the building. The ground-penetrating radar then transmits the detected internal structural defect information to the control unit in real time, and the control unit identifies the defect through the defect diagnosis model.

[0034] Specifically, see Figure 5 When the ground penetrating radar detects internal defects of a building structure, it generates electromagnetic waves (such as Ricker waves) through the transmitting antenna 9 and transmits electromagnetic wave signals. When the electromagnetic waves emitted by the ground penetrating radar propagate inside the building structure, they will be reflected, diffracted, refracted, etc. when they encounter objects with uneven media and different dielectric constants. Figure 5 In the figure, the building structure includes a surface layer 11, a covering layer 12, a waterproof layer 13 and a concrete structure 14, and the surface layer 11, the covering layer 12, the waterproof layer 13 and the concrete structure 14 are connected in sequence from the outside to the inside. After the receiving antenna 10 receives the reflected electromagnetic wave, the detection radar processes the electromagnetic wave and extracts its features (such as time domain processing and frequency domain analysis), and then performs time-depth conversion, and uses an imaging algorithm (common methods include B-scan (cross-sectional view) and C-scan (cross-sectional depth view)) to generate a building internal defect map (two-dimensional or three-dimensional image) from the processed signal data. The building internal defect map is sent to the control unit, and the defect (i.e., defect) is identified through the defect diagnosis model.

[0035] It should be noted that ground penetrating radar mainly transmits electromagnetic waves, which propagate through different materials and record the reflected power and two-way travel time. According to the equation , the attenuation of electromagnetic waves EM is the free space permeability , conductivity , dielectric constant The angular frequency of the wave The attenuation of the electromagnetic wave amplitude is directly related to the frequency of the wave. The frequency-dependent attenuation mechanism states that the higher the frequency, the greater the attenuation rate and the higher the level of detail. In a specific embodiment, the ground penetrating radar uses a frequency of 1.5 GHz to achieve the best level of detail and penetration depth for evaluating the interior of a building.

[0036] The electromagnetic wave EM received by the receiving antenna consists of multiple reflections. In air coupling, the first part of the received wave is usually caused by the direct energy from the transmitter to the receiving antenna. The second reflection peak represents the reflected energy along the air-material interface. See Figure 6 For the receiving antenna, due to the small time difference, the air-material interface and the direct energy transmitted to the receiving antenna are combined into the direct wave A1, and the reflected energy of the signal and the air-material interface is the air-material interface reflected wave A2. The other partial reflected wave A3 represents the energy reflected from other interfaces including steel bars. Figure 6In the figure, TWTT (Two-Way Travel Time) represents the two-way propagation time, which is the time required for the radar signal to be transmitted and reflected back to the receiver, that is, the time taken for the signal to be transmitted from the radar transmitter and reflected back to the receiver through the underground medium. t1 represents the time of the first reflection interface, and t2 represents the time of the second reflection interface. When detecting internal defects of components, the ground penetrating radar is in direct contact with the surface of the structure, which can provide better image quality and improve the accuracy of internal defect detection.

[0037] Specifically, in some embodiments, the ground penetrating radar uses GSSI (Geophysical Survey Systems, Inc.), and GSSI's structural scanner SIR30 is used for non-destructive detection and evaluation of the interior of the building. The ground penetrating radar has high sensitivity and accurate measurement capabilities to effectively detect and analyze the damage inside the building. The SIR30 system is used to move on the surface of the building to be inspected to collect radar data of the structure. At the same time, to ensure coverage of the entire area, the wireless (Bluetooth) is connected to the control unit to adjust the position and direction of the ground penetrating radar as needed.

[0038] Specifically, in some embodiments, the depth camera uses an Intel Realsense D455 depth camera, which has high resolution and flexibility and is suitable for use with a ground penetrating radar. In addition, its dual-camera design and Intel Realsense vision technology can provide accurate depth information and high-resolution images. The depth camera has a small size and weight, is easy to install above the ground penetrating radar, and can be connected to a control unit via USB for data transmission and image processing.

[0039] Continue to see Figure 3 In some embodiments, the base 1 is made of nylon-polyester blended material, and the base 1 includes a wearable vest 101 and a box 102 at the back of the vest; the wearable vest 101 includes a back pad 1011 and an adjustable shoulder strap 1012 connected to the back pad 1011, and the box 102 is installed on the back pad; the accommodating cavity is arranged in the box 102, and a carrying handle 1021 is arranged on the top of the box 102. The base is made of nylon-polyester blended material. Since nylon-polyester blended material has good wear resistance and waterproofness and is relatively light, the base is light, wear-resistant, and waterproof, which can reduce the burden of the user and increase the service life of the base. The shoulder strap is an adjustable shoulder strap, so that the user can adjust the length and tightness of the shoulder strap according to personal body shape and preference to ensure wearing comfort.

[0040] It should be noted that the accommodating cavity provided in the box is used to accommodate the control unit, the power module and the peripheral electronic cables. In order to facilitate the connection between the switch of the power module and the electronic cable and other external components, holes are opened on the box to place the power switch of the power module and the electronic cable ports connecting other components.

[0041] It should be noted that the structure of the shoulder strap needs to take into account the dispersion of weight, relief of shoulder pressure and provision of good support. Specifically, in some embodiments, the back pad adopts a thickened back pad to increase the comfort of the back and provide good support. The shoulder strap adopts a wide and soft design to increase the comfort of the shoulder, disperse the pressure and provide good support. In addition, the design of the thickened back pad and shoulder strap also provides a better cushioning effect.

[0042] Continue to see Figure 4 In some embodiments, the control unit 4 includes: A PLC controller 402, whose communication port is connected to the laser ranging sensor 303 through a serial port conversion module 403 and a serial port module 404, and is used to receive the position and posture of the ground penetrating radar detected by the laser ranging sensor 303; The embedded processing unit 405 is connected to the PLC controller 402 and is used to generate a fitting path trajectory of the ground penetrating radar 301 according to the position and posture of the ground penetrating radar; the disease diagnosis model 401 is set in the embedded processing unit; The robotic arm controller 406 is respectively connected to the embedded processing unit 405, the robotic arm 201, the first rotating gimbal 202 and the second rotating gimbal 203, and is used to control the movement of the robotic arm 201, the first rotating gimbal 202 and the second rotating gimbal 203 according to the fitting path trajectory, and adjust the fitting position and fitting posture of the ground penetrating radar 301 so that the ground penetrating radar 301 fits the measured area of ​​the building.

[0043] The PLC controller and embedded processing unit work together to obtain the position and attitude information of the ground penetrating radar through the laser ranging sensor. The embedded processing unit generates the fitting path trajectory of the ground penetrating radar based on these data, and adjusts the position of the detection radar in real time in combination with the disease diagnosis model. At the same time, the robot arm controller coordinates the actions of the robot arm and the rotating pan-tilt head to ensure that the detection radar is always in the best fitting state, improving the detection accuracy and efficiency.

[0044] Specifically, in some embodiments, the embedded processing unit uses NVIDIA Jetson Xavier NX. It has a six-core NVIDIA Carmel ARM® v8.2 64-bit CPU, 384 NVIDIA CUDA® cores, and 48 Tensor cores, providing powerful computing performance. At the same time, it also has 8GB of LPDDR4x memory and 16GB of eMMC storage, which can meet the needs of data processing and storage. NVIDIA Jetson Xavier NX also provides a wealth of interfaces, including multiple USB ports, Gigabit Ethernet ports, HDMI interfaces, etc., which are convenient for connecting various sensors and peripheral devices. Enable the device to process ground penetrating radar data in real time and store detection results. The internal solid-state drive selects Samsung 850 EVO 250GB SATA III 2.5 inches, which has a storage capacity of 250GB and is suitable for lightweight design requirements. It uses a SATA III interface to provide fast data transfer speeds and stable performance. This solid-state drive has excellent durability and reliability, and is suitable for upgrading or replacing laptops, ultrabooks or other lightweight devices. Its high-speed read and write performance and reliable quality can meet the needs of daily design work, and its low energy consumption helps to extend the battery life of the device.

[0045] Select Siemens Simatic S7-1200 CPU 1212C as the PLC controller, and connect the laser distance sensor to the communication port of S7-1200 PLC through a serial port conversion module (e.g. RS-232) and a serial port module (e.g. CP2102) to receive the input data of the laser distance sensor.

[0046] In some embodiments, see Figure 7 , the construction method of the disease diagnosis module is: S1. Data acquisition step: Obtain historical surface disease information and internal disease information as well as annotated pixel-level disease types.

[0047] S2. Dataset construction steps: Construct a training set based on historical surface disease information, internal disease information and labeled pixel-level disease types.

[0048] S3. Model construction steps: Based on the deep neural network, the disease diagnosis model is constructed with surface disease information and internal disease information as input and disease images and radar profiles as output.

[0049] S4, model training step, training the disease diagnosis model through the training set.

[0050] With deep neural network as the core, combined with surface disease information and internal disease information, an efficient disease diagnosis model is generated by establishing a mapping relationship between input and output. The model is trained through big data. During the model training process, historical data and pixel-level annotation data are used to build a high-quality training set to ensure the model's accurate recognition ability on different disease types.

[0051] In some embodiments, see Figure 8 The disease diagnosis model includes a backbone network 4011, a neck network 4012 and a head network 4013 which are connected in sequence.

[0052] The backbone network adopts the CSP structure for feature extraction. The backbone network is equipped with a global attention mechanism module GAM and an SPPF module. The global attention mechanism module GAM is used to reduce the diffusion of information in the global dimension and enhance the interaction between global features. The SPPF module is used to enhance the model reasoning ability.

[0053] In some embodiments, see Figure 8 The backbone network is also equipped with n+1 convolution modules for convolution calculation and n C2f modules for feature extraction. The output of the first convolution module is connected to the input of the second convolution module, the input of the first C2f module is connected to the output of the second convolution module, and the output of the first C2f module is connected to the input of the third convolution module. The convolution modules and C2f modules are arranged alternately in sequence. The output of the n+1 convolution module is connected to the input of the nC2f module, and the output of the nC2f module is connected to the global attention mechanism module GAM and SPPF module in sequence. Each convolution module and C2f module works together through a clear connection relationship, and each module completes its specific function to ensure that the features are optimized layer by layer. The global attention mechanism module GAM and the SPPF module are responsible for feature weighting and multi-scale feature fusion, respectively, to improve the expression ability and robustness of the model.

[0054] Specifically, in some embodiments, continue to see Figure 7 , the C2f module includes: A first convolutional block, configured to receive an input feature map and generate an intermediate feature map; The feature map splitting module splits the intermediate feature map into two feature maps; N residual blocks, the N residual blocks are connected in sequence, the first residual block receives the first part of the feature map output by the feature map splitting module, performs convolution, normalization and activation operations to generate the first feature map, the second residual block receives the first feature map, performs convolution, normalization and activation operations to generate the second feature map, and so on, the Nth residual block receives the N-1th feature map, performs convolution, normalization and activation operations to generate the Nth feature map; A feature fusion module performs feature fusion on the second part of the feature map output by the feature map splitting module and the feature map generated by each residual block to obtain a fused feature map; The second convolution block receives the fused feature map and generates the final feature map.

[0055] The C2f module adopts a split-first-fuse processing mode to achieve multi-level feature enhancement through the separation and recursive processing of feature maps. The introduction of residual blocks effectively alleviates the gradient vanishing problem in deep network training while maintaining efficient feature representation capabilities. Finally, through the feature fusion module and the second convolution block, information integration and high-dimensional feature representation are achieved.

[0056] Specifically, the convolution block consists of a convolution layer, a batch normalization layer, and an activation function, which enables the disease diagnosis model to obtain more information about the gradient flow, while reducing the computational workload of the disease diagnosis model and maintaining lightweight computation.

[0057] The global attention mechanism module GAM retains the sequential channel-spatial attention mechanism based on the attention mechanism module CBAM, and further redesigns the submodules. Fig. 9 In the channel attention submodule, 3D permutation is used to preserve the three-dimensional information, and then a two-layer MLP (multi-layer perceptron) is used to amplify the cross-dimensional channel spatial dependency. The spatial attention submodule focuses on the spatial information by utilizing two convolutional layers for spatial information fusion.

[0058] Specifically, the global attention mechanism module GAM pays attention to the input features Operation, where C represents the number of channels, H represents the height, and W represents the width. Intermediate state features and output features Can be used separately , To define, where represents the channel attention map, and represents the spatial attention map, The symbol represents element-by-element multiplication. In the channel attention submodule, the channel attention map Applied to input features , thus producing a channel-wise attention representation. This attention map is then combined with the input features Multiply elements together to generate intermediate state features In the spatial attention submodule, the spatial attention map Applied to intermediate state features , capturing the spatial dependencies within the intermediate states. Spatial Attention Map Intermediate state characteristics Multiply the elements to produce the final output features By combining channel and spatial attention mechanisms, the global attention mechanism module (GAM) aims to capture important features in all three dimensions of the input feature map.

[0059] Specifically, the neck network adopts the FPN structure and the PAN structure. The FPN structure increases the amount of feature information in the feature map by top-down sampling, and the PAN structure increases the amount of position information in the upper feature map by bottom-up sampling. Through top-down upsampling, the FPN structure is used to increase the amount of feature information in the lower feature map. Through bottom-up downsampling, the PAN structure is used to increase the amount of position information in the upper feature map. Finally, the two feature outputs are combined to ensure accurate prediction of images of different sizes.

[0060] Specifically, the head network uses a decoupling head to separate the detection head and the classification head, and adopts an anchor-free method to locate the object. Specifically, the object is located by referencing the center of the object and estimating the distance from the center to the bounding box, thereby improving the accuracy and effectiveness of object detection in the image.

[0061] In some embodiments, the total loss function of the disease diagnosis module is expressed as:

[0062] in,

[0063]

[0064] In the formula, is the total loss function, is the classification loss function, is the weighting factor of the classification loss function, is the focal distribution loss function, is the weighting factor of the focal distribution loss function, is the bounding box regression loss function, is the weighting factor of the bounding box regression loss function, is the high-quality anchor box loss function, is the loss function of the common quality anchor box, , is the center coordinate of the prediction box, , is the center coordinate of the predicted box corresponding to the real box, is the width of the real frame, is the height of the actual frame.

[0065] Classification loss function It is expressed as: , where represents the number of categories in the classification task, is the weight assigned to each class to balance the influence of different classes. Represents the probability value after applying the sigmoid activation function, Represents an index variable used to traverse all categories. It represents each of the categories and is used to traverse all categories when calculating the loss. The classification loss function calculation formula calculates a weighted version of the binary cross entropy loss by adjusting the weight of each category and considering the probability distribution of all categories.

[0066] Focal Distribution Loss Function It is expressed as: , where , Indicates that it comes from a continuous label The values ​​on the left and right sides near satisfy the equation , , Through the SoftMax layer (Right now ) to achieve, , Input scores for the SoftMax layer, representing labels ,Label By combining the loss terms of the classification loss function, focal distribution loss function, and bounding box regression loss function with their respective weight factors, the disease diagnosis model optimizes the overall performance by effectively balancing the contributions of the classification and regression components.

[0067] Specifically, in some embodiments, a Panasonic NCR18650B battery is selected as the energy source of the power module to ensure that the device can operate stably during use and maintain sufficient working time. At the same time, a power management module is equipped, which includes short circuit, undervoltage, overvoltage, reverse voltage and overtemperature protection functions to prevent voltage failures from damaging system components.

[0068] In some embodiments, see Figure 4The wearable building internal disease detection system also includes a communication module 6 and a mobile terminal 7. The communication module 6 is arranged in the accommodating cavity. The ground penetrating radar 301 communicates with the embedded processing unit 405 through the communication module 6 to transmit the detected internal disease information to the embedded processing unit 405; the mobile terminal 7 communicates with the embedded processing unit 405 through the communication module 6 to receive the disease information forwarded by the communication module 6 and identified by the embedded processing unit 405 and display it in real time. The mobile terminal realizes real-time display of detected ground diseases or abnormal conditions through wireless communication, and interacts with users to improve the operational convenience and practicality of the system. A real-time and efficient detection and display path is constructed through the communication module, the embedded processing unit and the mobile terminal. Users can view the results in real time and intuitively through the mobile terminal, get instant feedback and quickly judge the severity of the problem, make decisions quickly, and take corresponding repair or investigation measures.

[0069] Specifically, in some embodiments, the SIMCom SIM8200EA-M2 5G module is selected as the communication module, which supports 5G NR and LTE Cat 20 networks and can provide download speeds of up to 2Gbps and upload speeds of 200Mbps. At the same time, SIMCom has designed a low-power mode to help extend the battery life of the device. For the wearable building internal disease detection system, it can ensure that the device can run for a long time without frequent charging. And the small physical size helps to save space in the design and is suitable for integration into lightweight wearable devices.

[0070] Specifically, the communication module is equipped with a WiFi module, and the communication module communicates with the mobile terminal through the WiFi module. Real-time data is communicated with the mobile terminal through the WiFi module, so that the user can monitor the system status in real time, receive real-time data, and respond in time.

[0071] Specifically, in some embodiments, the WiFi module uses Intel Wi-Fi 6 AX200, which can provide higher speed, lower latency, and better network capacity management. It also uses advanced low-power technology to reduce energy consumption and extend battery life while ensuring performance.

[0072] In some embodiments, see Figure 4 The wearable building internal disease detection system also includes a positioning module 8 and a storage module (not shown in the figure) arranged in the accommodating cavity. The positioning module 8 is connected to the global positioning system and the control unit 4 to obtain the location information of the wearable building internal disease detection system and send it to the control unit 4. The storage module is connected to the control unit 4 to store the detection data and disease identification results. The introduction of the positioning module and the storage module enhances the adaptability and data management capabilities of the detection system.

[0073] The wearable building internal disease detection system provided by the embodiment of the present invention adopts a lightweight design, with an overall weight of about 8-10 kilograms, is easy to carry, and is suitable for the detection of surface diseases and internal diseases of indoor structures of buildings. Surface disease information is detected by a depth camera, and internal disease information is detected by a ground-penetrating radar. An active unit is designed, and the depth camera and ground-penetrating radar are arranged at the end of the active unit. When detecting diseases, the positions of the depth camera and ground-penetrating radar are adjusted by controlling the movement of the active unit, which greatly improves the detection range and flexibility of operation. By identifying the type and location of the disease through a pre-constructed disease diagnosis model, accurate detection of surface diseases and internal diseases of the structure can be achieved.

[0074] See also Fig.10 The second embodiment of the present invention provides a wearable building internal disease detection method, using the wearable building internal disease detection system described in the first aspect of the present invention, comprising the following steps: S1. Preparation steps: Wear the wearable building internal disease detection system and turn on the power module.

[0075] S2, surface disease detection step: adjust the robotic arm in real time through the control unit, aim the depth camera at the building surface, and collect surface disease data.

[0076] S3, internal disease detection step: detect the position and posture of the ground penetrating radar through the laser ranging sensor, the control unit generates a fitting path trajectory of the ground penetrating radar according to the position and posture of the ground penetrating radar, and controls the movement of the mechanical arm, the first rotating pan-tilt table and the second rotating pan-tilt table according to the fitting path trajectory, adjusts the fitting position and fitting posture of the ground penetrating radar, makes the ground penetrating radar fit the measured area of ​​the building, and collects internal disease data.

[0077] S4, disease identification step: the control unit receives and identifies the surface disease data and the internal disease data, and generates a disease image and a radar profile.

[0078] The above detection method combines surface data and internal data collection, uses depth cameras to detect surface defects, and uses ground-penetrating radar to perform non-destructive detection of internal defects in buildings. The control unit generates path planning and regulates the robotic arm and gimbal to complete automated fit detection. At the same time, the collected data is fused and identified through algorithms to generate intuitive defect images and radar profiles. Ordinary staff can perform efficient and automated data collection on the interior of buildings, thereby achieving professional and low-cost detection of massive buildings, significantly improving the convenience and efficiency of internal structural detection of buildings.

[0079] The above embodiments are used to explain the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A wearable building internal disease detection system, characterized in that: include: The base body is set as a backpack structure, and a receiving cavity is set in the base body; The movable unit comprises a mechanical arm, a first rotating platform connected to a first end of the mechanical arm, and a second rotating platform connected to a second end of the mechanical arm, wherein the first rotating platform is mounted on the top of the base; The sensing unit includes a ground-penetrating radar for detecting internal defects of the structure, a depth camera for detecting surface defects of the structure, and a laser ranging sensor for detecting the position and posture of the ground-penetrating radar. The ground-penetrating radar is installed on the second rotating pan-tilt platform, the depth camera is installed on the upper side of the ground-penetrating radar, and the laser ranging sensor is installed on the four sides of the ground-penetrating radar to form a laser sensor array; The control unit is arranged in the accommodating cavity, generates a fitting path trajectory of the ground penetrating radar according to the position and posture of the ground penetrating radar, and controls the movement of the mechanical arm, the first rotating pan-tilt table and the second rotating pan-tilt table according to the fitting path trajectory, and adjusts the fitting position and fitting posture of the ground penetrating radar so that the ground penetrating radar fits the measured area of ​​the building; a disease diagnosis model is arranged in the control unit, and the disease diagnosis model identifies the type and position of the disease according to the internal disease information detected by the ground penetrating radar and the surface disease information detected by the depth camera to generate a disease image and a radar profile; The power module is arranged in the accommodating cavity and supplies power to the active unit, the sensing unit and the control unit.

2. The wearable building internal disease detection system as claimed in claim 1, characterized in that: The base is made of nylon-polyester blended material, and includes a wearable vest and a box at the rear of the vest; the wearable vest includes a back pad and an adjustable shoulder strap connected to the back pad, and the box is installed on the back pad; the accommodating cavity is arranged in the box, and a carrying handle is arranged on the top of the box.

3. The wearable building internal disease detection system as claimed in claim 1, characterized in that: The control unit comprises: A PLC controller, whose communication port is connected to the laser ranging sensor through a serial port conversion module and a serial port module, and is used to receive the position and posture of the ground penetrating radar detected by the laser ranging sensor; An embedded processing unit is connected to the PLC controller and is used to generate a fitting path trajectory of the ground penetrating radar according to the position and attitude of the ground penetrating radar; the disease diagnosis model is set in the embedded processing unit; The robotic arm controller is respectively connected to the embedded processing unit, the robotic arm, the first rotating gimbal and the second rotating gimbal, and is used to control the movements of the robotic arm, the first rotating gimbal and the second rotating gimbal according to the fitting path trajectory, and adjust the fitting position and fitting posture of the ground penetrating radar so that the ground penetrating radar fits the measured area of ​​the building.

4. The wearable building internal disease detection system as claimed in claim 1, characterized in that: The construction method of the disease diagnosis module is as follows: Data acquisition steps: Obtain historical surface disease information and internal disease information as well as annotated pixel-level disease types; Dataset construction steps: Construct a training set based on historical surface disease information, internal disease information, and labeled pixel-level disease types; Model construction steps: Based on the deep neural network, the disease diagnosis model is constructed with surface disease information and internal disease information as input and disease images and radar profiles as output; Model training step: training the disease diagnosis model through the training set.

5. The wearable building internal disease detection system as claimed in claim 1, characterized in that: The disease diagnosis model includes a backbone network, a neck network, and a head network connected in sequence; The backbone network adopts CSP structure for feature extraction. The backbone network is equipped with a global attention mechanism module and an SPPF module. The global attention mechanism module is used to reduce the dispersion of information in the global dimension and enhance the interaction between global features. The SPPF module is used to enhance the model reasoning ability. The neck network adopts FPN structure and PAN structure. The FPN structure increases the amount of feature information in the feature map by top-down sampling, and the PAN structure increases the amount of position information in the upper feature map by bottom-up sampling. The head network uses a decoupling head to separate the detection head and the classification head, and adopts an anchor-free method to localize objects.

6. The wearable building internal disease detection system as claimed in claim 5, characterized in that: The backbone network is also provided with n+1 convolution modules for convolution calculation and n C2f modules for feature extraction. The output of the first convolution module is connected to the input of the second convolution module, the input of the first C2f module is connected to the output of the second convolution module, the output of the first C2f module is connected to the input of the third convolution module, the convolution modules and the C2f modules are arranged alternately in sequence, the output of the n+1th convolution module is connected to the input of the nth C2f module, and the output of the nth C2f module is connected to the global attention mechanism module and the SPPF module in sequence; the C2f module includes: A first convolutional block, configured to receive an input feature map and generate an intermediate feature map; The feature map splitting module splits the intermediate feature map into two feature maps; N residual blocks, the N residual blocks are connected in sequence, the first residual block receives the first part of the feature map output by the feature map splitting module, performs convolution, normalization and activation operations to generate the first feature map, the second residual block receives the first feature map, performs convolution, normalization and activation operations to generate the second feature map, and so on, the Nth residual block receives the N-1th feature map, performs convolution, normalization and activation operations to generate the Nth feature map; A feature fusion module performs feature fusion on the second part of the feature map output by the feature map splitting module and the feature map generated by each residual block to obtain a fused feature map; The second convolution block receives the fused feature map and generates the final feature map.

7. The wearable building internal disease detection system as claimed in claim 1, characterized in that: The total loss function of the disease diagnosis module is expressed as: in, In the formula, is the total loss function, is the classification loss function, is the weighting factor of the classification loss function, is the focal distribution loss function, is the weighting factor of the focal distribution loss function, is the bounding box regression loss function, is the weighting factor of the bounding box regression loss function, is the high-quality anchor box loss function, is the loss function of the common quality anchor box, , is the center coordinate of the prediction box, , is the center coordinate of the predicted box corresponding to the real box, is the width of the real frame, is the height of the actual frame.

8. The wearable building internal disease detection system as claimed in claim 1, characterized in that: It also includes a communication module and a mobile terminal. The communication module is arranged in the accommodating cavity. The ground penetrating radar communicates with the embedded processing unit through the communication module to transmit the detected internal disease information to the embedded processing unit; the mobile terminal communicates with the embedded processing unit through the communication module to receive the disease information identified by the embedded processing unit and forwarded by the communication module and display it in real time.

9. The wearable building internal disease detection system as claimed in claim 1, characterized in that: It also includes a positioning module and a storage module arranged in the accommodating cavity. The positioning module is connected to the global positioning system and the control unit to obtain the location information of the wearable building internal disease detection system and send it to the control unit. The storage module is connected to the control unit to store the detection data and disease identification results.

10. A wearable building internal disease detection method, using the wearable building internal disease detection system as claimed in any one of claims 1 to 9, characterized in that: The steps are: Preparation steps: Wear the wearable building internal disease detection system and turn on the power module; Surface disease detection steps: Use the control unit to adjust the robotic arm in real time, aim the depth camera at the building surface, and collect surface disease data; Internal disease detection steps: detecting the position and posture of the ground penetrating radar by means of a laser ranging sensor, the control unit generating a fitting path trajectory of the ground penetrating radar according to the position and posture of the ground penetrating radar, and controlling the movement of the mechanical arm, the first rotating pan-tilt table and the second rotating pan-tilt table according to the fitting path trajectory, adjusting the fitting position and fitting posture of the ground penetrating radar, so that the ground penetrating radar fits the measured area of ​​the building, and the ground penetrating radar collects internal disease data; Disease identification step: The control unit receives and identifies surface disease data and internal disease data, and generates a disease image and a radar profile.

Images