A heliostat mirror surface detection method and detection system

Abstract

The present invention discloses a heliostat mirror surface detection method and a detection system, belonging to the technical field of heliostat mirror surface detection, and capable of solving the problems of low detection accuracy and low detection efficiency of the existing detection methods. The method includes: S1, collecting multiple fringe images of different frequencies reflected by the heliostat; S2, extracting the phase data information of each fringe image; S3, performing three-dimensional reconstruction on the reflecting surface of the heliostat according to the phase data information of the multiple fringe images to obtain a fitted surface of the heliostat, and determining the surface type detection result of the heliostat according to the fitted surface. The present invention is used for detecting the surface type of the reflecting surface of the heliostat.

Description

Technical Field

[0001] The present invention relates to a heliostat mirror surface detection method and detection system, belonging to the technical field of heliostat mirror surface detection. Background Art

[0002] A heliostat is a key component in tower solar power generation technology, and the accuracy of its curvature directly affects the concentration efficiency. There are the following two existing methods for detecting the reflecting surface of a heliostat: One is to use the laser beam deflection scanning method to detect the heliostat, but the detection accuracy of this method is relatively low. The other is to detect the heliostat by the secondary reflection mirror surface quality detection method based on image recognition, but the detection efficiency of this method is relatively low. Summary of the Invention

[0003] The present invention provides a heliostat mirror surface detection method and detection system, which can solve the problems of relatively low detection accuracy and relatively low detection efficiency of the existing detection methods.

[0004] On the one hand, the present invention provides a heliostat mirror surface detection method, and the method includes:

[0005] S1. Collect multiple fringe images of different frequencies reflected by the heliostat;

[0006] S2. Extract the phase data information of each fringe image;

[0007] S3. Perform three-dimensional reconstruction on the reflecting surface of the heliostat according to the phase data information of multiple fringe images to obtain a fitting surface of the heliostat, and determine the mirror surface detection result of the heliostat according to the fitting surface.

[0008] Optionally, the extraction of the phase data information of each fringe image in S2 is specifically:

[0009] Use a deep neural network to perform decoupling processing on multiple fringe images respectively to obtain the phase data information of each fringe image.

[0010] Optionally, the loss function of the deep neural network is a quaternary composite loss function.

[0011] Optionally, the quaternary composite loss function is composed of a phase angle loss, a difference frequency consistency loss, an adversarial reflection suppression loss, and a structural similarity loss.

[0012] Optionally, the three-dimensional reconstruction of the reflecting surface of the heliostat according to the phase data information of multiple fringe images in S3 to obtain a fitting surface of the heliostat specifically includes:

[0013] Generate a depth map according to the phase data information of multiple fringe images;

[0014] Using the depth map to perform three-dimensional reconstruction on the reflecting surface of the heliostat to obtain the fitted surface of the heliostat.

[0015] Optionally, the using the depth map to perform three-dimensional reconstruction on the reflecting surface of the heliostat to obtain the fitted surface of the heliostat specifically includes:

[0016] Converting the depth map into three-dimensional point cloud data by using a camera imaging model;

[0017] Performing surface reconstruction on the three-dimensional point cloud data by using a volume fusion algorithm to obtain the fitted surface of the heliostat.

[0018] On the other hand, the present invention provides a heliostat surface type detection system based on any one of the above-mentioned heliostat surface type detection methods, and the system includes:

[0019] An acquisition unit for acquiring multiple fringe images of different frequencies reflected by the heliostat;

[0020] An analysis unit for extracting phase data information of each fringe image;

[0021] A reconstruction unit for performing three-dimensional reconstruction on the reflecting surface of the heliostat according to the phase data information of multiple fringe images to obtain the fitted surface of the heliostat, and determining the surface type detection result of the heliostat according to the fitted surface.

[0022] Optionally, the acquisition unit includes:

[0023] A transmission module for transmitting the heliostat to the detection position;

[0024] A canvas arranged opposite to the detection position, on which there are multiple fringe patterns with different phases;

[0025] Two reel modules respectively connected to both ends of the canvas for driving the canvas to move so as to sequentially display the fringe patterns on the canvas;

[0026] An acquisition module arranged on one side of the canvas for acquiring multiple fringe images reflected by the heliostat.

[0027] Optionally, the acquisition unit further includes:

[0028] A support module arranged below the canvas for supporting the canvas so that the fringe patterns on the canvas are in a flattened state.

[0029] Optionally, the acquisition module includes:

[0030] An image acquisition device arranged on one side of the canvas for acquiring multiple fringe images reflected by the heliostat;

[0031] Adjusting structure, connected to the image collector, for adjusting the height of the image collector so that the image collector is within the focal length of the heliostat.

[0032] The beneficial effects that the present invention can produce include:

[0033] The heliostat surface shape detection method provided by the present invention extracts the phase data information of each fringe image in multiple fringe images of different frequencies, and performs three-dimensional reconstruction of the reflecting surface based on multiple phase data information. Through the frequency difference, the true shape information of the reflecting surface can be accurately extracted, thereby effectively eliminating the influence of parasitic reflection and surface shape, and reducing the error caused by single-frequency fringes.

[0034] The heliostat surface shape detection method provided by the present invention, when using a deep neural network to extract the phase data information of each fringe image, adopts a quaternion composite loss function that couples geometric constraints and data-driven constraints. This loss function can avoid the problem of the failure of the mutation penalty at the boundary of the traditional loss function, suppress the abnormal phase shift caused by parasitic reflection, and effectively suppress the stepped artifacts in phase unwrapping, improving the surface smoothness.

[0035] The heliostat surface shape detection system provided by the present invention controls the canvas to display the fringe pattern through a reel module, uses an acquisition module to acquire multiple fringe images reflected by the heliostat, and then uses the phase shift method to detect the surface shape of the mirror surface. This method improves the detection accuracy and speed of the heliostat surface shape detection, and greatly reduces the detection cost. Description of the Drawings

[0036] Figure 1 It is a flowchart of the heliostat surface shape detection method provided by an embodiment of the present invention;

[0037] Figure 2 It is a three-dimensional structure schematic diagram of the heliostat surface shape detection system provided by an embodiment of the present invention;

[0038] Figure 3 It is a planar structure schematic diagram of the heliostat surface shape detection system provided by an embodiment of the present invention.

[0039] Reference Signs:

[0040] 1. First reel; 2. Second reel; 3. First driving motor; 4. Second driving motor; 5. Telescopic rod; 6. Support platform; 7. Camera; 8. Heliostat; 9. Transmission module; 10. Moving crossbar; 11. Canvas; 12. Smoothing structure; 13. Computer; 14. Vertical rod. Detailed Embodiments

[0041] The present invention will be described in detail below with reference to the embodiments, but the present invention is not limited to these embodiments.

[0042] An embodiment of the present invention provides a method for detecting the mirror surface shape of a heliostat, as Figures 1 to 3 shown, the method includes:

[0043] S1. Collect multiple fringe images with different frequencies reflected by the heliostat 8.

[0044] In practical applications, a collecting device can be used to collect multiple fringe images with different frequencies reflected by the heliostat 8. The embodiment of the present invention does not limit the specific structure of the collecting device. The collecting device can be an existing fringe image collecting structure or a collecting unit provided by the detection system side of the present invention. The specific structure of the collecting unit will be described in detail when describing the detection system of the present invention.

[0045] In the present invention, different fringe images have different frequencies.

[0046] S2. Extract the phase data information of each fringe image.

[0047] Specifically: use a deep neural network to decouple each of the multiple fringe images to obtain the phase data information of each fringe image.

[0048] The loss function of the deep neural network is a quaternion composite loss function that couples geometric constraints and data-driven constraints. The specific loss architecture is:

[0049] Assume that the network input is a group of fringe images containing frequencies, and the output is the phase prediction value corresponding to each frequency, is the true phase value, then the loss function is defined as a weighted combination of the following four parts:

[0050] The first part: ;

[0051] is the phase angle loss. For the phase prediction error of each frequency, a period-sensitive angular residual calculation is used. This loss forces the network to accurately predict the main phase value. Using cosine distance instead of mean square error avoids the problem of sudden penalty failure at the boundary of the traditional loss. In each training batch, calculate the phase angle loss of all samples at all frequencies, and the loss value updates the network weights through backpropagation, preferentially minimizing the local phase error.

[0052] The second part:

[0053] ;

[0054] is the difference frequency consistency loss. In the above formula, For the true thickness information of the calibration plate, is modulo operation to ensure that the phase difference is always within the period of; is the combination number of taking 2 frequencies from different frequencies. As the denominator, it is to normalize the calculation results of the phase differences of all frequencies, so that the loss value can be measured on a suitable scale, avoiding the loss value being too large or too small due to the influence of the number of frequencies, and thus more reasonably reflecting the overall difference between the predicted phase differences and the theoretical values of different frequencies. Use the theoretical relationship between the phase differences of multiple frequencies to establish constraints. Assume that any two frequencies actually satisfy ( is the proportionality coefficient between the i-th frequency and the j-th frequency), then the predicted phase difference should be consistent with the theoretical difference. This loss function ensures that the predicted phase differences of different frequencies are consistent with their theoretical values, avoids local phase mutations caused by parasitic reflections, thereby improving the global phase consistency, and calculates the loss for all possible frequency combinations after each forward propagation to ensure that the multi-frequency phases output by the network conform to the actual phases.

[0055] Part Three: ;

[0056] is the adversarial reflection suppression loss. In the above formula, is that the generator (main network) receives the input fringe image I and outputs the predicted phase. D is the discriminator network, with the phase map as the input and the output being [ 0,1 ] the probability value, indicating whether the input comes from the data distribution of true reflection-free interference. The expectation can be approximated by the average value of the batch samples in the calculation. Construct an auxiliary discriminator to distinguish the parasitic reflection artifacts in the predicted phase image, and the generator needs to deceive the discriminator. This loss encourages the network to generate results consistent with the true reflection-free interference phase distribution, especially improving the phase continuity in high-reflection regions.

[0057] Part Four: ;

[0058] is the structural similarity loss. In the above formula, is the phase gradient (calculated by the Sobel operator, retaining the gradients in both the x and y directions). is the mean value of the phase gradients in the current batch, is the variance of the phase gradients, is the covariance between the predicted phase gradient and the true gradient, is the stability constant. This loss function introduces the structural similarity constraint of the phase gradient to maintain the local structural features of the predicted phase map. This loss effectively suppresses the stepped artifacts in phase unwrapping and improves the surface smoothness.

[0059] The final composite loss function is composed of a linear combination of the above four terms:

[0060] ;

[0061] In the above formula, the coefficient is obtained through model training. For example, after training, the values of each coefficient are: .

[0062] This four - element composite loss function avoids the problem of the mutation penalty failure at the boundary of the traditional loss function, suppresses the abnormal phase shift caused by parasitic reflection, and effectively suppresses the stepped artifacts in phase unwrapping, improving the surface smoothness.

[0063] After obtaining multiple fringe images, a deep neural network is used for fringe decoupling to recover the original phase data information from the fringe images containing multiple frequencies and angles. To improve the generalization ability of the network, each fringe image will undergo data augmentation processing such as rotation, cropping, and illumination change before decoupling. Fringe images usually have phase misalignment due to factors such as reflection and distortion. The goal of the network is to extract clear phase data information from them. The network structure contains multiple convolutional layers, which focus on extracting the spatial features in the fringe images and retain more detailed information by introducing skip connections to enhance the processing ability of the fringe images. Finally, the decoupled phase image is recovered through deconvolution operations. In the design of the loss function, the present invention designs a four - element composite loss function to measure the error between the output phase and the true phase. The trained deep neural network can effectively decouple the phase data information from complex fringe images and provide accurate phase data for subsequent 3D reconstruction.

[0064] S3. Perform 3D reconstruction on the reflecting surface of the heliostat 8 according to the phase data information of multiple fringe images to obtain the fitted surface of the heliostat 8, and determine the surface type detection result of the heliostat 8 according to the fitted surface.

[0065] Among them, performing 3D reconstruction on the reflecting surface of the heliostat 8 according to the phase data information of multiple fringe images to obtain the fitted surface of the heliostat 8 specifically includes:

[0066] (1) Generate a depth map according to the phase data information of multiple fringe images.

[0067] The mapping model for converting phase data information into depth values in space by triangulation is as follows:

[0068] ;

[0069] where, is the phase data information; is the baseline distance from the camera to the canvas; is the phase period wavelength; is the camera focal length; is the calibration correction term.

[0070] Substitute the absolute phase value of each pixel in the phase data information into the mapping model, calculate the corresponding depth value, and generate an initial depth map. Then, complete the missing area through a post-processing interpolation algorithm (such as the nearest neighbor interpolation method) and output the optimized depth map.

[0071] (2) Use the depth map to perform three-dimensional reconstruction on the reflecting surface of the heliostat 8 to obtain the fitted surface of the heliostat 8.

[0072] Specifically: First, use the camera imaging model to convert the depth map into three-dimensional point cloud data; then, use the volume fusion algorithm to perform surface reconstruction on the three-dimensional point cloud data to obtain the fitted surface of the heliostat 8.

[0073] Since fringe images of different frequencies can provide different depth information, which helps to accurately reconstruct the three-dimensional surface. First, use the camera imaging model to convert the depth value of each pixel in the depth map into three-dimensional space coordinates to obtain three-dimensional point cloud data; subsequently, use machine learning models such as convolutional neural network (CNN) or generative adversarial network (GAN) to perform noise repair on the three-dimensional point cloud data to obtain optimized three-dimensional point cloud data. Finally, by using the volume fusion algorithm (Poisson surface reconstruction), the optimized three-dimensional point cloud data is converted into a complete three-dimensional surface. Performing smoothing and hole repair processing on the final three-dimensional reconstruction result can generate a fine fitted surface.

[0074] Another embodiment of the present invention provides a heliostat surface type detection system based on any of the above heliostat surface type detection methods. The system includes:

[0075] An acquisition unit for acquiring multiple fringe images of different frequencies reflected by the heliostat 8.

[0076] An analysis unit for extracting the phase data information of each fringe image.

[0077] A reconstruction unit for performing three-dimensional reconstruction on the reflecting surface of the heliostat 8 according to the phase data information of multiple fringe images to obtain the fitted surface of the heliostat 8, and determining the surface type detection result of the heliostat 8 according to the fitted surface.

[0078] Reference Figure 2 and Figure 3 As shown, the acquisition unit includes:

[0079] A transfer module 9 for transferring the heliostat 8 to the detection position.

[0080] The transfer module 9 transfers the produced heliostat 8 to the detection position. In the embodiments of the present invention, the specific structure of the transfer module 9 is not limited as long as it can transfer the heliostat 8 to the detection position. In practical applications, the transfer module 9 can adopt structures such as a conveyor belt.

[0081] A canvas 11 is arranged opposite to the detection position, and has a plurality of fringe patterns with different phases thereon.

[0082] The canvas 11 contains a series of fringe patterns with different phases, and the interval between two adjacent fringe patterns is generally 40 cm to 60 cm. In practical applications, the interval between adjacent fringe patterns can be set to 50 cm.

[0083] Two reel modules are respectively connected to both ends of the canvas 11 for driving the canvas 11 to move so as to sequentially display the fringe patterns on the canvas 11.

[0084] Specifically, the reel module includes:

[0085] A reel, with the end of the canvas 11 wound around the reel;

[0086] A driving motor for driving the reel to rotate so as to drive the canvas 11 to be wound around the reel.

[0087] In the embodiments of the present invention, the reel in the reel module connected to the right end of the canvas 11 is denoted as the first reel 1, and the corresponding driving motor is denoted as the first driving motor 3; the reel in the reel module connected to the left end of the canvas 11 is denoted as the second reel 2, and the corresponding driving motor is denoted as the second driving motor 4.

[0088] Reference Figure 2 As shown, the first reel 1 stretches a set of fringe canvases 11 in the second reel 2 and displays them on the support table 6. The second reel 2 contracts the fringe canvases 11 in the first reel 1 and recovers them into the second reel 2. The first driving motor 3 provides power for the first reel 1. The second driving motor 4 provides power for the second reel 2.

[0089] An acquisition module is arranged on one side of the canvas 11 for acquiring a plurality of fringe images reflected by the heliostat 8.

[0090] Specifically, the acquisition module includes:

[0091] An image collector is provided on one side of the canvas 11 for collecting multiple fringe images reflected by the heliostat 8. In practical applications, the image collector can be a camera 7.

[0092] The heliostat 8 is the object to be detected. The camera 7 is used to collect the fringe images reflected by the heliostat 8.

[0093] An adjusting structure is connected to the image collector for adjusting the height of the image collector so that the image collector is within the focal length of the heliostat 8.

[0094] Among them, the adjusting structure includes:

[0095] A bracket including two vertical rods 14. Guide grooves are provided on the opposite surfaces of the two vertical rods 14 along their lengths.

[0096] A moving cross bar 10 has its two ends respectively arranged in the two guide grooves and can move along the guide grooves. The image collector is connected to the moving cross bar 10.

[0097] The moving cross bar 10 can move in the vertical direction, enabling it to adjust the position of the camera 7.

[0098] Furthermore, the acquisition unit further includes:

[0099] A support module is provided below the canvas 11 for supporting the canvas 11 so that the fringe pattern on the canvas 11 is in a flattened state.

[0100] In the embodiment of the present invention, the support module includes:

[0101] A support table 6 is provided below the canvas 11 for supporting the canvas 11;

[0102] A plurality of telescopic rods 5 are provided below the support table 6 for adjusting the height of the support table 6 so that the fringe pattern on the canvas 11 is in a flattened state.

[0103] The support table 6 supports the fringe canvas 11 and provides a placement platform for the fringe canvas 11.

[0104] The telescopic rods 5 provide support for the support table 6 and can be telescoped to adjust the height of the support table 6.

[0105] Preferably, the acquisition unit further includes:

[0106] A smoothing structure 12 is installed at the edge where the support table 6 contacts the canvas 11.

[0107] The smooth structure 12 enables the canvas 11 to run smoothly between the two reels and reduces the wear on the canvas 11. The specific structure of the smooth structure 12 in the embodiments of the present invention is not limited. In practical applications, the smooth structure 12 can be a rounded bar or other structures.

[0108] The working process of the heliostat mirror surface detection system will be described in detail below.

[0109] First, place the acquisition unit behind the heliostat production line. After the heliostat 8 is manufactured, the transfer module 9 brings the heliostat 8 to the detection position. Fix the position of the heliostat 8 and adjust the camera 7 up and down through the moving crossbar 10 so that it is within the focal length of the heliostat 8, and start calibrating the camera 7 to ensure the accuracy of the detection. The present invention uses an intelligent calibration method based on deep learning, which is a new method for calibrating the camera 7 using deep learning technology. At the beginning stage of this intelligent calibration method, a large amount of image data needs to be collected and preprocessed. After preprocessing, the collected data is trained using a convolutional neural network. Through continuous iteration and optimization, the model can more accurately estimate the internal parameters (such as focal length, principal point position) and external parameters (such as rotation and translation matrices) of the camera 7, as well as the distortion coefficients. The trained model can be applied to the calibration of the camera 7 in this heliostat mirror surface detection device. By obtaining the images collected by the camera 7, the model can quickly and accurately estimate the internal and external parameters of the camera 7, realizing the calibration of the camera 7.

[0110] After the calibration of the camera 7, the heliostat mirror surface detection part starts. The first driving motor 3 drives the first reel 1 to drive the canvas 11 inside the second reel 2 to roll. There is a set of 16 stripe patterns printed in sequence on the canvas 11. The spacing between each stripe pattern is 50 cm. The 50 cm spacing is set to prevent the influence between adjacent stripe patterns. In order to minimize the deformation degree of the canvas 11 as much as possible, the telescopic rod 5 is used to support the support platform 6 so that the canvas 11 is in a state of closely fitting the support platform 6. At the same time, under the action of the smooth structure 12, the canvas 11 can run smoothly between the two reels and reduce the wear on the canvas 11. Each time the canvas 11 displays a stripe pattern, the camera 7 is used to take an image of the stripe pattern reflected by the heliostat 8 and save the image into the computer 13. Repeat 16 times to complete the acquisition of a set of stripe patterns. After a set of stripe patterns is acquired, the second driving motor 4 drives the second reel 2 to drive the canvas 11 inside the first reel 1 to roll for the next acquisition.

[0111] After obtaining a set of fringe images containing the information of the reflective surface shape of the heliostat 8, the internal code of the computer 13 is used to analyze this set of fringe images. Then, the reflective surface shape of the heliostat 8 is obtained by code fitting. If the fitted reflective surface shape is within the qualified deviation range from the set standard surface shape, it is considered that this heliostat 8 is qualified and it will be shipped out. If the fitted reflective surface shape is quite different from the set standard surface shape, it will be sent back for calibration. After calibration, it is put into the surface shape detection system again for detection, and this process is repeated multiple times until the heliostat 8 is qualified.

[0112] The functions implemented by the internal code in the computer 13 are the same as those implemented by the analysis unit and the reconstruction unit, mainly solving the problems of phase recovery and 3D reconstruction in fringe images. Specifically, first, a deep neural network is used to decouple the fringe images to extract accurate phase data information. Next, combined with machine learning optimization techniques, the decoupled phase data information is converted into a depth map, and the depth information in the depth map is optimized and reconstructed to generate a high-precision 3D surface model. Thus, the fitted surface of the heliostat 8 is obtained. During the whole process, the combination of the neural network and the multi-frequency method not only ensures high-precision phase recovery but also effectively eliminates complex reflection interference, finally realizing high-precision 3D reconstruction based on machine learning.

[0113] After the detection of the present invention starts, it does not require multiple steps and can complete the detection of the reflective surface shape of the heliostat 8 in a short time.

[0114] The present invention uses a scroll control canvas 11 to replace the display screen to play the fringe pattern, greatly reducing the cost.

[0115] The present invention uses the method of phase shift to detect the mirror surface, improving the detection accuracy and speed of the heliostat 8.

[0116] The present invention uses an independently innovatively designed detection architecture, which can interact quickly with the heliostat production line, shortening the overall operation time of the entire production line and the detection device.

[0117] The present invention decouples the complex phase data information in the fringe images through a deep neural network to ensure that accurate phase data can be recovered from noise and parasitic reflections.

[0118] The present invention converts the phase information into an accurate depth map, further improves the accuracy of 3D reconstruction through multi-frequency optimization, and enhances the fineness and stability of 3D reconstruction by using machine learning algorithms.

[0119] As described above, these are only several embodiments of the present application and do not impose any form of limitation on the present application. Although the present application is disclosed above with preferred embodiments, it is not intended to limit the present application. Any person skilled in the relevant art can make some changes or modifications within the scope of the technical solution of the present application by using the disclosed technical content, which are equivalent to equivalent implementation cases and all fall within the scope of the technical solution.

Claims

1. A heliostat mirror surface detection method, characterized in that The method includes: S1. Collect multiple fringe images of different frequencies reflected by a heliostat; S2. Use a deep neural network to decouple each of the multiple fringe images respectively to obtain the phase data information of each fringe image; S3. Perform three-dimensional reconstruction on the reflecting surface of the heliostat according to the phase data information of the multiple fringe images to obtain the fitted surface of the heliostat, and determine the surface type detection result of the heliostat according to the fitted surface; Among them, the loss function of the deep neural network is a four - element composite loss function, and the four - element composite loss function consists of a phase - angle loss , a difference - frequency consistency loss , an adversarial reflection suppression loss , and a structural similarity loss ; Assume that the network input is a set of fringe images containing frequencies , and the output is the predicted phase value corresponding to each frequency , which is the true phase value; Then ; ； In the above formula, is the true thickness information of the calibration plate, is the modulo operation to ensure that the phase difference is always within the period of; is the number of combinations of taking 2 frequencies from different frequencies; is the th frequency and the th frequency ratio coefficient; ； In the above formula, is the generator that receives the input fringe image and outputs the predicted phase; is the discriminator network, with the phase map as the input and the probability value of being as the output, indicating whether the input comes from the data distribution of real non-reflection interference; the expectation is approximately the average value of the batch samples; ； In the above formula, is the phase gradient; is the mean value of the phase gradient in the current batch, is the variance of the phase gradient, is the covariance between the predicted phase gradient and the true gradient, is the stability constant; Composite loss function ; In the above formula, the coefficient is obtained through model training.

2. The method according to claim 1, wherein The three-dimensional reconstruction of the reflecting surface of the heliostat according to the phase data information of the multiple fringe images in S3 to obtain the fitted surface of the heliostat specifically includes: Generate a depth map according to the phase data information of the multiple fringe images; Perform three-dimensional reconstruction on the reflecting surface of the heliostat using the depth map to obtain the fitted surface of the heliostat.

3. The method according to claim 2, wherein The performing three-dimensional reconstruction on the reflecting surface of the heliostat using the depth map to obtain the fitted surface of the heliostat specifically includes: Convert the depth map into three-dimensional point cloud data using a camera imaging model; Perform surface reconstruction on the three-dimensional point cloud data using a volume fusion algorithm to obtain the fitted surface of the heliostat.

4. A heliostat surface type detection system based on the heliostat surface type detection method according to any one of claims 1 to 3, characterized in that, The system includes: A collection unit for collecting multiple fringe images of different frequencies reflected by a heliostat; An analysis unit for extracting the phase data information of each fringe image; A reconstruction unit for performing three-dimensional reconstruction on the reflecting surface of the heliostat according to the phase data information of the multiple fringe images to obtain the fitted surface of the heliostat, and determining the surface type detection result of the heliostat according to the fitted surface.

5. The system according to claim 4, wherein The collection unit includes: A transmission module for transmitting the heliostat to a detection position; A canvas arranged opposite to the detection position, on which there are multiple fringe patterns with different phases; Two reel modules respectively connected to both ends of the canvas for driving the canvas to move to sequentially display the fringe patterns on the canvas; A collection module arranged on one side of the canvas for collecting multiple fringe images reflected by the heliostat.

6. The system according to claim 5, wherein The collection unit further includes: A support module arranged below the canvas for supporting the canvas to make the fringe patterns on the canvas in a flattened state.

7. The system according to claim 5, wherein The collection module includes: An image collector arranged on one side of the canvas for collecting multiple fringe images reflected by the heliostat; An adjustment structure connected to the image collector for adjusting the height of the image collector to make the image collector located within the focal length of the heliostat.

Images