Arbitrary surface terahertz imaging method based on six-degree-of-freedom robotic arm

Abstract

The present invention discloses a method for terahertz imaging of arbitrary curved surfaces based on a six-degree-of-freedom mechanical arm. The method uses the six-degree-of-freedom mechanical arm to sample terahertz signals on the surface of an arbitrary curved object, calculates the axis angle according to Euler angle posture information, and then uses the axis angle to project the current three-dimensional coordinates onto the initialization coordinate axis, calculates the three-dimensional coordinates of each terahertz point in the sampling point to construct a three-dimensional voxel, fills the three-dimensional voxel through a ray projection algorithm and a random forest interpolation method, and based on the constructed three-dimensional voxels, terahertz three-dimensional tomography can image and analyze arbitrary curved objects, thereby overcoming the defect that a two-dimensional scanning platform can only perform plane scanning imaging. At the same time, the ray projection algorithm and the random forest interpolation method are used to construct three-dimensional voxels, and a good three-dimensional tomography display effect is achieved.

Description

Technical field:

[0001] The invention belongs to the technical field of terahertz spectroscopy and imaging, and in particular relates to a terahertz imaging method of an arbitrary curved surface based on a mechanical arm. Background technology:

[0002] Terahertz waves are electromagnetic waves with a wavelength between infrared and microwaves. Because they are in the transition region from photonics to electronics, they have many unique properties, such as fingerprint spectroscopy, low energy, and special penetration. The interaction between terahertz waves and materials will produce a unique absorption spectrum - fingerprint spectrum. Polar molecules, such as water molecules, and non-polar molecules, such as carbon dioxide molecules, have very obvious differences in the absorption of terahertz waves, so terahertz absorption spectra are also of great value in testing molecular properties. Time-domain terahertz spectroscopy scanning technology uses high-precision delay lines to extend the sampling time of femtosecond pulses to tens of picoseconds, and reduces noise through hardware preprocessing. The time-domain signal generated by it is Fourier transformed to obtain the characteristic absorption spectrum of the substance being tested, and its spectral width can reach more than 5THz and the dynamic range can reach more than 70dB. This spectrum performance can meet the detection needs of most compounds, thus providing a large number of application scenarios for terahertz time-domain spectroscopy scanning.

[0003] At present, terahertz three-dimensional tomography technology has broad room for development, and improving the speed and resolution of three-dimensional tomography are urgent problems to be solved. The existing three-dimensional tomography algorithms, such as terahertz peak intensity imaging, terahertz time-of-flight imaging, and terahertz specific power spectrum imaging, can all be used for imaging display. Among them, terahertz three-dimensional tomography scanning based on a two-dimensional scanning platform is the most common. The three-dimensional tomography technology using a two-dimensional scanning platform to scan the terahertz signal of a planar object is also becoming mature, but it is difficult to achieve for curved objects. Patent 2022208512267 discloses a terahertz imaging method for cylindrical objects, which is only applicable to regular cylinders and is difficult to achieve for irregular curved objects. Based on this, the present invention proposes a terahertz imaging method for arbitrary curved surfaces based on a robotic arm. Summary of the invention:

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and seek to design a terahertz imaging method for arbitrary curved surfaces based on a six-degree-of-freedom robotic arm. By adopting a six-axis degree-of-freedom robotic arm, the problem that the existing terahertz three-dimensional tomography system based on a two-dimensional scanning platform is difficult to solve is solved was solved.

[0005] In order to achieve the above-mentioned object, the present invention relates to an arbitrary curved surface terahertz imaging method based on a six-degree-of-freedom robotic arm, comprising the following steps:

[0006] 1. Data Collection

[0007] (101) firstly, a six-degree-of-freedom robotic arm is used to perform three-dimensional stereo modeling on the collected curved surface object to obtain a curved surface model corresponding to the curved surface object, and a robotic arm mobile scanning path is created with a set step according to the established curved surface model, wherein each sampling position of the mobile scanning path includes position information and posture information, wherein the position information is the three-dimensional coordinate, and the posture information is the Euler angle, and the robotic arm mobile scanning path is the terahertz lens mobile scanning path;

[0008] (102) moving the terahertz lens in sequence according to the moving scanning path to each sampling position, stopping and collecting terahertz time domain signals, and recording the current sampling position and the corresponding terahertz time domain signal;

[0009] 2. Constructing 3D voxels

[0010] According to the following formula, construct the three-dimensional voxel x length *y length *z length ,

[0011] The size of the 3D voxel in the z direction is z length =max(Z)-min(Z)+M

[0012] Where Z = (z1, z2, ..., z n ), where z i , i=1, ..., n is the z coordinate of all robot movement scanning paths, max(Z) and min(Z) are the maximum and minimum values ​​in Z, respectively, and M is the length of the terahertz time domain signal or the length of the compressed terahertz time domain signal;

[0013] The size of the 3D voxel in the x direction is x length =max(X new )-min(X new )

[0014] X new ={x new1 , x new2 , ..., x newn}

[0015] x newi =cords*sinα+x i

[0016] X=(x1,x2,...,xn)

[0017] where x i , i = 1, ..., n is the x coordinate of all robot arm movement scanning paths, cords = (0, 1, 2, ..., M), max (X new ) and min(X new ) are Xnew The maximum and minimum values ​​in , α is the angle of the x-axis calculated according to the Euler angle;

[0018] The size of the y-direction of the 3D voxel is y length =max(Y new )-min(Y new )

[0019] Y new ={y new1 ,y new2 , ..., y newn}

[0020] y newi =cords*sinβ+y i

[0021] Y=(y1,y2,...,y)

[0022] where y i , i = 1, ..., n is the y coordinate of all robot arm movement scanning paths, cords = (0, 1, 2, ..., M), max (Y new ) and min(Y new ) are Y new The maximum and minimum values ​​in , β is the angle of the x-axis calculated according to the Euler angle;

[0023] 3. Filling 3D voxels

[0024] (301) Filling three-dimensional voxel data using a ray projection algorithm with the processed terahertz time domain signal;

[0025] (302) using a random forest model to fill in missing values ​​in the three-dimensional voxel data obtained in step (301);

[0026] 4. 3D tomography display

[0027] The data processed by the ray projection algorithm and the random forest interpolation method were displayed in three dimensions using three-dimensional tomography software.

[0028] Preferably, in step (101), multiple groups of terahertz time-domain signals at the same sampling point are averaged to obtain an average terahertz time-domain signal at each sampling point.

[0029] Preferably, in step (101), nearest neighbor compression is performed on the terahertz time domain signal or / and the average terahertz time domain signal as required.

[0030] Compared with the prior art, the present invention has the following beneficial effects: sampling of terahertz signals on the surface of any curved object is achieved through a six-degree-of-freedom robotic arm, three-dimensional voxels are filled through a ray projection algorithm and a random forest interpolation method, and based on the constructed three-dimensional voxels, terahertz three-dimensional tomography can image and analyze any curved object, overcoming the defect that a two-dimensional scanning platform can only perform planar scanning imaging, and at the same time, the ray projection algorithm and the random forest interpolation method are used to construct three-dimensional voxels, which has a good three-dimensional tomography display effect; the axis angle is calculated according to the Euler angle posture information, and the axis angle is used to project the current three-dimensional coordinates onto the initialization coordinate axis, and the three-dimensional coordinates of each terahertz point in the sampling point are calculated. Description of the drawings:

[0031] Figure 1 The present invention is a schematic diagram of the terahertz imaging structure of an arbitrary curved surface based on a six-degree-of-freedom robotic arm.

[0032] Figure 2 It is the movement trajectory of the robotic arm set in Example 1.

[0033] Figure 3 This is the three-dimensional tomography result of the data in Example 1. Specific implementation method:

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0035] Example

[0036] like Figure 1 As shown, the arbitrary curved surface terahertz imaging method based on a robotic arm involved in this embodiment adopts a terahertz three-dimensional tomography system for data acquisition.

[0037] like Figure 1As shown, the terahertz three-dimensional tomography system includes a femtosecond laser 1, a beam splitter 2, an optical oscillation delay line 3, a bias source 4, a terahertz transmitting antenna 5, a terahertz detection antenna 6, a terahertz lens 7, a signal generation and acquisition processing unit 8 (including an acquisition card), a host computer 9 and a six-degree-of-freedom mechanical arm (not shown in the figure), wherein the terahertz lens 7 includes a collimating lens and a focusing lens, and the terahertz lens 7 is fixed at the end of the six-degree-of-freedom mechanical arm, the femtosecond laser 1 is connected to the beam splitter 2, one end of the beam splitter 2 is connected to the terahertz transmitting antenna 5, and the other end of the beam splitter 2 is connected to the terahertz detection antenna 6 through the optical oscillation delay line 3, the bias source 4 is connected to the terahertz transmitting antenna 5, and the signal generation and acquisition processing unit 8 (including an acquisition card) is respectively connected to the optical oscillation delay line 3, the bias source 4, the terahertz detection antenna 6 and the host computer 9.

[0038] The femtosecond pulse laser generated by the femtosecond laser 1 is split by the beam splitter 2, one beam is the pump light and the other is the detection light. Under the action of the bias source 4, the pump light is directly focused on the terahertz transmitting antenna 5 to generate a terahertz pulse. The terahertz pulse passes through the collimating lens and the focusing lens in sequence and then irradiates the sampling point. The reflected terahertz pulse carrying the sampling point information passes through the focusing lens and the collimating lens in sequence and is focused on the terahertz detection antenna 6, where it meets the detection light delayed by the optical oscillation delay line 3, and outputs a photocurrent signal, which is collected by the signal generation and collection processing unit 8.

[0039] The six-degree-of-freedom robot arm involved in this embodiment is a FANUC robot, model LR Mate 200iD.

[0040] The arbitrary curved surface terahertz imaging method based on a robotic arm involved in this embodiment includes the following steps:

[0041] 1. Data Collection

[0042] (101) First, a six-degree-of-freedom robot is used to perform three-dimensional stereo modeling on the collected curved surface object to obtain a curved surface model corresponding to the curved surface object. According to the established curved surface model, a robot movement scanning path is created with a set step. Each sampling position of the mobile scanning path includes position information and posture information. The position information is the three-dimensional coordinate (x, y, z), and the posture information is the Euler angle. Each three-dimensional coordinate point A (x, y, z) corresponds to a collection point A' (x', y', z') on the curved surface object to be collected, and the distance between A (x, y, z) and A' (x', y', z') is a focal length of the terahertz lens. Since the terahertz lens 7 is fixed at the end of the robot, the robot movement scanning path is the terahertz lens 7 movement scanning path. Therefore, each time the terahertz lens 7 moves a sampling position, the corresponding sampling point is located at its focus.

[0043] (102) The terahertz lens 7 is moved sequentially along the moving scanning path to reach each sampling position, stops and performs terahertz time domain signal acquisition, and records the current sampling position and the corresponding terahertz time domain signal.

[0044] Specifically, each time the robotic arm reaches a sampling position, it sends a collection pulse to the acquisition card. After receiving the collection pulse, the acquisition card performs signal collection. The six-degree-of-freedom robotic arm will stay at the position, and the acquisition card will collect one or more groups of time domain signals and record the current position coordinates.

[0045] Preferably, when multiple groups of terahertz time domain signals are collected at the same sampling point, the multiple groups of terahertz time domain signals at the same sampling point are averaged, and each sampling point obtains an average terahertz time domain signal to improve the signal-to-noise ratio;

[0046] Preferably, in general, the terahertz time domain signal is too long and affects the imaging effect. Therefore, the terahertz time domain signal or / and the average terahertz time domain signal are subjected to nearest neighbor compression as needed.

[0047] 2. Constructing 3D voxels

[0048] According to the following formula, construct the three-dimensional voxel x length *y length *z length ,

[0049] The size of the 3D voxel in the z direction is z length =max(Z)-min(Z)+M

[0050] Where Z = (z1, z2, ..., z n ), where z i , i=1, ..., n is the z coordinate of all robot movement scanning paths, max(Z) and min(Z) are the maximum and minimum values ​​in Z, respectively, and M is the length of the terahertz time domain signal or the length of the compressed terahertz time domain signal;

[0051] The size of the 3D voxel in the x direction is x length =max(X new )-min(X new )

[0052] X new ={x new1 , x new2 , ..., x newn}

[0053] x newi =cords*sinα+x i

[0054] X=(x1,x2,...,x n)

[0055] where x i , i = 1, ..., n is the x coordinate of all robot arm movement scanning paths, cords = (0, 1, 2, ..., M), max (X new ) and min(X new ) are X new The maximum and minimum values ​​in , α is the angle of the x-axis calculated according to the Euler angle;

[0056] The size of the y-direction of the 3D voxel is y length =max(Y new )-min(Y new )

[0057] Y new ={y new1 ,y new2 , ..., y newn}

[0058] y newi =cords*sinβ+y i

[0059] Y=(y1,y2,...,y)

[0060] where y i , i = 1, ..., n is the y coordinate of all robot arm movement scanning paths, cords = (0, 1, 2, ..., M), max (Y new ) and min(Y new ) are Y new The maximum and minimum values ​​in , β is the angle of the x-axis calculated according to the Euler angle.

[0061] 3. Filling 3D voxels

[0062] (301) The processed terahertz time domain signal is used to fill three-dimensional voxel data using a ray projection algorithm.

[0063] (302) A random forest model is used to fill in the missing values ​​in the three-dimensional voxel data obtained in step (301).

[0064] 4. 3D tomography display

[0065] The data processed by the ray projection algorithm and the random forest interpolation method were displayed in three dimensions using three-dimensional tomography software.

[0066] Example 1

[0067] 1. Data collection:

[0068] First, a three-dimensional model is built on the curved surface object to be collected to obtain the corresponding surface model of the curved surface object, and a mechanical moving scanning path is created for the established surface model. First, it is determined to perform mobile scanning with a step of 1 mm, and the Euler angle transformation is performed according to the surface model of the collected object. The Euler angle transformation is mainly to make the terahertz lens perpendicular to the surface to be collected, because only in this way can the terahertz time domain signal pulse be better received. If the terahertz lens is not set perpendicular to the surface, the terahertz pulse emitted by the terahertz lens will be reflected to other places, and the reflected pulse cannot be better received by the terahertz lens.

[0069] The robot uses a right-hand coordinate system, and all three-dimensional coordinates are generated in the initialized coordinate system (original three-dimensional coordinate axis). After the robot establishes the trajectory model, a data file based on six degrees of freedom will be generated. The data file contains the three-dimensional coordinates and three Euler angles of the robot. The three-dimensional coordinates are the position information of the robot in the world coordinate system, and the Euler angles are the posture information of the robot. Since the terahertz lens is fixed at the far end of the robot, the position information of the three-dimensional coordinates and the posture information of the three Euler angles are the position information and posture information of the terahertz lens. During the acquisition process of the robot, when the robot reaches the specified position, it will send an acquisition pulse to the acquisition card. After receiving the acquisition pulse, the acquisition card will perform signal acquisition. The robot will stay at the position for 0.5s to 1s. The acquisition card will continuously collect multiple time domain signals and record the current position coordinates. When the robot reaches another position, the acquisition card will record the coordinate information of the next position and perform signal acquisition at the next position, which is convenient for later data processing.

[0070] Data points are classified according to the position information of the acquisition card, and multiple groups of signals collected at the same position are averaged, which improves the signal-to-noise ratio of the signal and is beneficial to the later three-dimensional tomography. This embodiment uses a robotic arm with a step of 1mm, and collects 1625 points with a surface area of ​​30mm*50mm. The time length of the terahertz time domain signal is 120ps, and the time sampling interval is 0.1ps, so the length of a terahertz time domain signal is 1200 points.

[0071] 2. Constructing 3D voxels

[0072] According to the surface sample design, 30mm*50mm is collected at a step of 1mm, and a total of 1625 points are collected. Since the number of scanning points of the robot arm in each row is not fixed at 30, it may be greater or less than 30, so the total number of points collected is 1625 points. Figure 2 It is the set movement trajectory of the robot arm.

[0073] First, I need to create a three-dimensional voxel containing the entire surface object to be collected. First, z is set to the maximum value minus the minimum value in the three-dimensional position information z of all terahertz lenses, and then add the length of a terahertz time domain signal. Define Z as the set of all z coordinates of the robot arm's moving scanning path. Since the length of the terahertz time domain signal is 1200 points, the z direction is too long, and the nearest neighbor of the terahertz time domain signal is compressed 10 times to 120 lengths. In this way, after data processing after collection, we process the terahertz time domain signal collected by the acquisition card into 1625 terahertz time domain signals, and the length of each signal is 120 points after compression. Z=(z1,z2,...,z 1625 ), where z i , i=1,...,1625 are the z coordinates of all robot arm movement scanning paths.

[0074] So the size of the final 3D voxel in the z direction is z lenth =max(Z)-min(Z)+120.

[0075] For the calculation of the length of the three-dimensional voxel in the x and y directions, first, according to the x and y coordinates of the three-dimensional coordinates of the 1625 points, the x and y coordinates are subtracted from their respective minimum values ​​to perform a translation in the xy plane, so that the minimum values ​​of the x and y coordinates are both 0. Then, the Euler angles are converted into axis angles, and the calculated axis angles are α, β, and γ with the x, y, and z axes, respectively.

[0076] Take the calculation in the x direction as an example X=(x1,x2,...,x n ), where x i , i = 1, ..., n are the x coordinates of all robot movement scanning paths. Define cords = (0, 1, 2, ..., 119), so the projection of the first coordinate x on the x-axis is x new1 =cords*sinα+x1, because the terahertz time domain signal of a point is 120 points, so cords is defined as 0 to 119, we define the origin as (0, 0, 0), so the x-axis projection coordinate of the first coordinate point is x new1 , x new1 is a set of x coordinates with a length of 120, so the projection of all x coordinates on the x-axis is X new ={x new1 , x new2 , ..., x new1625}, so the size of the x-direction of the 3D voxel is x length =max(X new )-min(X new ). Similarly, we can get the size of the y direction of the three-dimensional voxel as y length =max(Y new)-min(Y new ). new The projection of all y coordinates on the y axis is Y new ={y new1 ,y new2 , ..., y new1625},y new1 =cords*sinβ+y1. Each of the 120 points in the terahertz time domain signal of 1625 points calculates their x-coordinates and y-coordinates, and then takes the maximum value minus the minimum value in the x-coordinate and rounds up as the length of x. Similarly, takes the maximum value minus the minimum value in the y-coordinate and rounds up as the length of y. In this way, the length, width and height of the three-dimensional voxel are constructed. The size of the constructed three-dimensional voxel is x length *y length *z length .

[0077] 3. Filling 3D voxels

[0078] (301) The processed 1625 terahertz time-domain signals are used to fill the voxel data using a ray projection algorithm.

[0079] The ray casting method is a direct volume rendering algorithm based on image sequences. From each pixel of the image, a ray is emitted along a fixed direction (usually the line of sight), and the ray passes through the entire image sequence. In this process, the image sequence is sampled to obtain the terahertz time domain signal intensity. At the same time, the terahertz time domain intensity signal is accumulated and averaged according to the light absorption model until the ray passes through the entire image sequence, and finally a rendered image is obtained. The ray casting algorithm emits a set of rays from the viewpoint and traverses the entire volume data set. The data on the three-dimensional data set through which the ray passes can be sampled evenly at intervals, which is similar to human real vision and is suitable for perspective projection. It emits rays from the viewpoint to each pixel of the viewing plane, passes through the viewing plane to the volume data set, samples the intersecting volume data, and mixes the sampling point values ​​from front to back, and uses the mixed color as the final value of the pixel, thereby realizing three-dimensional reconstruction. After using the ray casting algorithm to fill the data of the three-dimensional voxels, there will still be some data processing missing value states. In order to better perform missing value interpolation processing for three-dimensional tomography, the machine learning algorithm random forest is used here to process missing values.

[0080] The random forest interpolation method traverses all features and starts to fill in the least missing values ​​(because the least accurate information is needed to fill in the least missing features). When filling in a feature, first replace the missing values ​​of other features with 0. After each regression prediction, the predicted value is placed in the original feature matrix, and then the next feature is filled in. Each time the filling is completed, the feature with missing values ​​will be reduced by one, so after each cycle, fewer and fewer features need to be filled with 0. When it comes to the last feature (this feature should be the one with the most missing values ​​among all the features), there are no other features that need to be filled with 0, and a large amount of valid information has been filled in for other features using regression, which can be used to fill in the most missing features. After traversing all the features, the data is complete and there are no more missing values.

[0081] 4. 3D tomography display:

[0082] The data after the ray projection algorithm and random forest interpolation method are displayed in three dimensions using 3D tomography software, which can clearly show the distance and effect of the upper and lower surfaces, with high imaging quality. At the same time, for any curved surface, we can collect data through robotic arm modeling, and after algorithm processing, we can perform 3D tomography display to view internal defects and effects.

[0083] The three-dimensional coordinates of the terahertz lens at the current moment can be obtained through the three-dimensional coordinate information, and the posture information of the current terahertz lens can be obtained through the three Euler angle information.

[0084] By converting the Euler angle into the axis angle, we can get the direction vector of the terahertz lens at the current moment, that is, the angle between the terahertz lens and the original three-dimensional coordinate axis. The terahertz time domain signal length and the three-dimensional coordinate value of each point (sampling point) can be converted through the direction vector. Then, according to the three-dimensional coordinate value of each point in the terahertz time domain signal, the terahertz signal intensity value of each point is subjected to the light projection algorithm to fill the three-dimensional voxel data.

[0085] For all the collected terahertz time-domain signals, we cannot fill in the values ​​of the three-dimensional voxels we created. For the missing values, we use the random forest interpolation method to fill in the missing values, and perform three-dimensional tomography on the filled data. The effect after algorithm processing can perform three-dimensional tomography of the upper and lower surfaces very well.

[0086] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An arbitrary curved surface terahertz imaging method based on a six-degree-of-freedom robotic arm, characterized in that: The following steps are involved: (1) Data collection: (101) firstly, a six-degree-of-freedom robotic arm is used to perform three-dimensional stereo modeling on the collected curved surface object to obtain a curved surface model corresponding to the curved surface object, and a robotic arm mobile scanning path is created with a set step according to the established curved surface model, wherein each sampling position of the mobile scanning path includes position information and posture information, wherein the position information is the three-dimensional coordinate, and the posture information is the Euler angle, and the robotic arm mobile scanning path is the terahertz lens mobile scanning path; (102) moving the terahertz lens in sequence according to the moving scanning path to each sampling position, stopping and collecting terahertz time domain signals, and recording the current sampling position and the corresponding terahertz time domain signal; (2) Constructing 3D voxels: According to the following formula, construct the three-dimensional voxel x length *y length *z length , The size of the 3D voxel in the z direction is z length =max(Z)-min(Z)+M Where Z = (z1, z2, ..., z n ), where z i , i=1, ..., n is the z coordinate of all robot movement scanning paths, max(Z) and min(Z) are the maximum and minimum values ​​in Z, respectively, and M is the length of the terahertz time domain signal or the length of the compressed terahertz time domain signal; The size of the 3D voxel in the x direction is x length =max(X new )-min(X new ) X new ={x new1 ,x new2 ,...,x newn } x newi =cords*sinα+x i X=(x1,x2,...,x n ) where x i , i = 1, ..., n is the x coordinate of all robot arm movement scanning paths, cords = (0, 1, 2, ..., M), max (X new ) and min(X new ) are X new The maximum and minimum values ​​in , α is the angle of the x-axis calculated according to the Euler angle; The size of the y-direction of the 3D voxel is y length =max(Y new )-min(Y new ) AND new ={and new1 ,and new2 ,...,and newn } and newi =cords*sinβ+y i Y=(y1,y2,...,y) where y i , i = 1, ..., n is the y coordinate of all robot arm movement scanning paths, cords = (0, 1, 2, ..., M), max (Y new ) and min(Y new ) are Y new The maximum and minimum values ​​in , β is the angle of the x-axis calculated according to the Euler angle; (3) Filling 3D voxels: (301) using a ray projection algorithm to fill three-dimensional voxel data with the processed terahertz time domain signal; (302) using a random forest model to fill in missing values ​​in the three-dimensional voxel data obtained in step (301); (4) Three-dimensional tomography display: The data processed by the ray projection algorithm and the random forest interpolation method were displayed in three dimensions using three-dimensional tomography software.

2. The arbitrary curved surface terahertz imaging method based on a six-degree-of-freedom robotic arm according to claim 1, characterized in that: In step (101), multiple groups of terahertz time domain signals at the same sampling point are averaged to obtain an average terahertz time domain signal at each sampling point.

3. The arbitrary curved surface terahertz imaging method based on a six-degree-of-freedom robotic arm according to claim 1, characterized in that: In step (101), nearest neighbor compression is performed on the terahertz time domain signal or / and the average terahertz time domain signal as required.

Images