Intelligent detection method and system for full-field deformation of lunar soil bricks

Abstract

The invention belongs to the technical field of lunar surface material mechanical testing, and particularly discloses an intelligent detection method and system for lunar soil brick full-field deformation, and the method comprises the steps: training a detection network constructed based on Diffusion Transformers through a detection data set, and obtaining a lunar soil brick full-field deformation detection model; the detection data set comprises a lunar soil brick image sample and a corresponding displacement cloud picture label, and when the lunar soil brick image sample is obtained, determining a distortion coefficient and correcting camera distortion based on a mapping relation model and according to an environment temperature; in the lunar soil brick mechanical loading experiment process, determining a camera distortion coefficient corresponding to the environment temperature, and obtaining a lunar soil brick image through a camera; the lunar soil brick image is input into the soil brick full-field deformation detection model to obtain a displacement cloud picture, and lunar soil brick full-field deformation detection is achieved. The method effectively deals with the limitations of extreme temperature difference of the lunar surface and incapability of making speckles, and has high precision and high adaptability.

Description

Technical Field

[0001] The present invention belongs to the technical field of lunar material mechanics testing, and more specifically, relates to an intelligent detection method and system for the full-field deformation of lunar soil bricks. Background Art

[0002] With the continuous deepening of human lunar exploration, lunar surface construction and resource utilization have gradually become important research topics. In the field of in-situ resource utilization (ISRU) on the lunar surface, lunar soil, as a potential building material, the research on its mechanical properties and deformation behavior in the lunar environment has become a key link in lunar engineering projects. As one of the lunar building materials, the mechanical property characterization of lunar soil bricks is of crucial significance for the construction of lunar bases, equipment design, and mission execution. However, factors such as extreme temperature changes, vacuum environment, and radiation on the lunar surface make traditional deformation detection techniques face many challenges under lunar conditions.

[0003] Currently, the deformation detection of lunar soil bricks mainly relies on two types of methods: contact measurement and non-contact measurement. However, these techniques each have certain limitations. Contact measurement methods are easily interfered by the external environment and cannot effectively cope with extreme temperature changes and vacuum conditions on the lunar surface; non-contact measurement methods, such as digital image correlation (DIC) technology, often have difficulty obtaining accurate deformation information due to the lack of obvious speckle features on the surface of lunar soil bricks. In addition, the temperature difference between day and night on the lunar surface changes violently, which not only affects the imaging quality of camera equipment but also may cause changes in the camera distortion coefficient, thereby affecting the applicability of traditional DIC technology in the lunar environment.

[0004] Therefore, there is an urgent need for a new method that can adapt to the extreme lunar environment and efficiently and accurately perform full-field deformation detection on the surface of lunar soil bricks without speckles. Summary of the Invention

[0005] In view of the above deficiencies or improvement requirements of the prior art, the present invention provides an intelligent detection method and system for the full-field deformation of lunar soil bricks, aiming to improve the adaptability and accuracy of full-field deformation detection of lunar soil bricks in the complex lunar environment.

[0006] To achieve the above object, according to one aspect of the present invention, an intelligent detection method for the full-field deformation of lunar soil bricks is proposed, including the following steps:

[0007] Model training stage:

[0008] Train a detection network based on Diffusion Transformers through a detection data set, and use the trained detection network as a full-field deformation detection model for lunar soil bricks;

[0009] The detection dataset includes lunar soil brick image samples and their corresponding displacement nephogram labels; among them, the lunar soil brick image samples are obtained by a camera, and in this process, based on the pre-constructed mapping relationship model between the environmental temperature and the camera distortion coefficient, the camera distortion coefficient is determined according to the environmental temperature, so as to correct the camera distortion.

[0010] Model application stage:

[0011] During the mechanical loading experiment of the lunar soil brick, the environmental temperature is monitored in real time, the camera distortion coefficient corresponding to the environmental temperature is determined based on the mapping relationship model, and the lunar soil brick image is obtained by the camera; the lunar soil brick image is input into the full-field deformation detection model of the soil brick to obtain a displacement nephogram, realizing the full-field deformation detection of the lunar soil brick.

[0012] As a further preference, the acquisition method of the displacement nephogram label in the detection dataset is: preprocessing the lunar soil brick image sample to obtain a target image sample, and then obtaining the corresponding displacement nephogram label through DIC analysis.

[0013] As a further preference, the preprocessing of the lunar soil brick image sample includes:

[0014] The lunar soil brick image sample is divided into sub-image blocks of the same size, and the sub-image blocks are denoised; then the histogram equalization is performed on the denoised sub-image blocks, and then the sub-image blocks are re-stitched to obtain the target image sample.

[0015] As a further preference, the denoising process of the sub-image blocks includes:

[0016] Calculate the variance of the pixel gray values of each sub-image block respectively, and then obtain the average variance of all sub-image blocks in the lunar soil brick image sample; determine the filtering intensity according to the average variance, and use this filtering intensity to filter all sub-image blocks.

[0017] As a further preference, the method for filtering the sub-image blocks is mean filtering, Gaussian filtering or median filtering.

[0018] As a further preference, the construction method of the mapping relationship model between the environmental temperature and the camera distortion coefficient includes:

[0019] Under fixed temperature conditions, calibrate the internal parameter matrix and external parameter matrix of the camera;

[0020] During the temperature change process, the camera takes multiple images and records the images and the corresponding environmental temperatures;

[0021] Using the Harris corner detection method, feature points are extracted from the image. Based on the world coordinates and pixel coordinates of the image feature points, as well as the intrinsic matrix and extrinsic matrix, the distortion coefficients are calculated; thus, the camera distortion parameters corresponding to different environmental temperatures are obtained, and then the mapping relationship model between the environmental temperature and the camera distortion coefficients is fitted.

[0022] As a further preference, the detection network based on Diffusion Transformers includes an encoder, a diffusion module, and a decoder. Among them, the encoder is used to extract the features of the image samples, the diffusion module is used to introduce a noise diffusion process to the extracted features, and the decoder is used to decode the diffused features.

[0023] As a further preference, the detection dataset includes a real dataset and a simulated dataset. In the real dataset, real lunar soil brick image samples and their corresponding displacement cloud map labels are obtained. In the simulated dataset, simulated lunar soil brick image samples and their corresponding displacement cloud map labels are obtained;

[0024] The detection network is pre-trained with the simulated dataset; after the pre-training is completed, the detection network is fine-tuned with the real dataset. During this process, the parameters of the encoder in the detection network are fixed, and only the parameters of the decoder in the detection network are updated; the detection network obtained by fine-tuning is used as the lunar soil brick full-field deformation detection model.

[0025] As a further preference, when the detection network is pre-trained with the simulated dataset and the decoder is fine-tuned with the real dataset, the mean square error loss function is used.

[0026] According to another aspect of the present invention, an intelligent detection system for the full-field deformation of lunar soil bricks is provided, including a processor, and the processor is used to execute the above intelligent detection method for the full-field deformation of lunar soil bricks.

[0027] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following technical advantages are mainly possessed:

[0028] 1. The present invention pre-constructs the mapping relationship model between the environmental temperature and the camera distortion coefficients, overcomes the influence of lunar surface temperature changes on the imaging system, provides accurate data for the training of the detection network, and thus forms a non-contact and efficient lunar soil brick deformation detection scheme, effectively coping with the extreme temperature difference on the lunar surface and the limitation that speckles cannot be made during the existing in-situ full-field deformation detection on the lunar surface. It has the advantages of high adaptability and high precision in the complex lunar surface environment, and can meet the actual needs of lunar resource development, lunar surface construction, and long-term missions.

[0029] 2. Ensure the detection accuracy in extreme environments based on lunar surface temperature compensation. Aiming at the influence of the extreme lunar surface temperature (-173°C to 127°C) on the imaging performance of the camera, the present invention monitors the lunar surface environment temperature in real time, combines temperature compensation, dynamically adjusts the camera distortion coefficient, and establishes a mapping relationship model between temperature and imaging distortion. It can effectively eliminate the image distortion caused by temperature changes, ensure high-precision image acquisition in extreme temperature environments, and provide stable and reliable input data for subsequent deformation detection of the surface of non-speckled lunar soil bricks.

[0030] 3. Image texture enhancement and adaptive processing under non-speckle conditions. Aiming at the characteristics of the natural non-speckle or low-speckle texture on the surface of real lunar soil bricks, the present invention enhances the texture and grayscale of the experimental image samples of lunar soil bricks, significantly improving the identifiability of the surface texture and the image quality. This method effectively avoids the difficult and unstable generation of artificial speckles due to factors such as extreme temperature differences and low light in the lunar surface environment, and is easily affected by external factors such as temperature, dust, and light, thus affecting the image quality and the accuracy of deformation detection, overcoming the dependence of traditional DIC technology on artificial speckles. The present invention can significantly improve the identifiability of image features by enhancing texture and grayscale under the complex and non-speckle surface conditions of lunar soil bricks, ensure high-quality deformation detection data can still be obtained in the real lunar surface environment, meet the actual needs of lunar surface deformation detection, and provide reliable technical support for lunar in-situ construction tasks.

[0031] 4. High-precision full-field deformation detection based on the Diffusion Transformers algorithm. The present invention uses the improved Diffusion Transformers architecture to intelligently detect the full-field deformation of lunar soil bricks. Through the encoder, multi-scale feature extraction is performed on the input non-speckle image samples, combined with the diffusion modeling process and the self-attention mechanism, to realize the modeling and feature enhancement of global context information. The high-resolution displacement cloud map is restored through the decoder, and finally high-precision full-field deformation detection results are generated. Compared with traditional detection methods, the present invention can still accurately predict the displacement and deformation information during the loading process under the conditions of sparse and irregular natural textures on the surface of lunar soil bricks, showing extremely high adaptability and detection accuracy. Description of the Drawings

[0032] Figure 1 It is a flow chart of the intelligent detection method for the full-field deformation of lunar soil bricks in the embodiment of the present invention;

[0033] Figure 2 It is a schematic diagram of the intelligent detection method for the full-field deformation of lunar soil bricks in the embodiment of the present invention;

[0034] Figure 3 It is a schematic diagram of the training process of the full-field deformation detection model of lunar soil bricks in the embodiment of the present invention. Specific Embodiments

[0035] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to 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 used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0036] An intelligent detection method for the full-field deformation of lunar soil bricks provided by an embodiment of the present invention, as Figure 1 and Figure 2 shown, includes the following steps:

[0037] S1: Construct a mapping relationship model between the environmental temperature and the camera distortion coefficient.

[0038] Use the temperature sensor of the lunar rover to monitor the lunar surface environmental temperature in real time, and calibrate the lunar rover camera in combination with the camera calibration method based on active vision. Through multiple calibrations by adjusting the temperature, a mapping relationship model between the lunar surface environmental temperature and the camera distortion coefficient is established using the temperature compensation algorithm. First, the internal parameters and external parameters are initially calibrated. Using the feature points extracted from multiple images and the lunar rover movement path data, the internal parameters and external parameters are jointly calibrated first to ensure the accuracy of the basic geometric parameters of the camera in the projection model; then the distortion parameters are calibrated separately. On the basis of knowing the internal parameters and external parameters, the distortion parameter κ is calibrated separately through the distortion feature points extracted from multiple images, and the change law of the distortion parameters at various temperatures is calculated.

[0039] Specifically, it includes:

[0040] S11: Under the condition of a fixed temperature, use the temperature sensor equipped on the lunar rover to monitor the environmental temperature in real time, and ensure that the temperature is stable near T0 (the fluctuation range does not exceed ±1 ° °C). In this stage, control the lunar rover to perform rotational motion, and synchronously record the camera shooting state, rotation angle and translation displacement data. Furthermore, calibrate the internal parameter matrix K and the external parameter matrix [R|t] of the camera. At this time, it is assumed that the distortion parameter is a fixed value or temporarily ignored.

[0041] S12: Establish a camera projection model. Combining the internal parameter matrix K, the external parameter matrix [R|t] and the distortion parameter κ of the camera, establish the projection relationship between the image coordinates and the world coordinates:

[0042]

[0043] where: (X, Y, Z) are world coordinates, (x, y) are pixel coordinates, K is the internal parameter matrix, f x , f yis the focal length (in pixels), c x , c y is the principal point coordinate, [R|t] is the external parameter matrix, R(θ) is the rotation matrix, θ represents the rotation angle of the lunar rover; t is the translation vector, representing the vector from the origin of the world coordinate system to the origin of the camera coordinate system; κ is the distortion parameter.

[0044] S13: Enter the dynamic temperature calibration stage. Actively adjust the experimental environment through the heating / cooling module of the lunar rover or utilize the natural day-night temperature difference on the lunar surface to make the environmental temperature T change in a gradient of -150 ° °C to 100 ° °C, with a step size of 20 ° °C. During the temperature change process, control the lunar rover to rotate continuously. The left and right cameras take pictures synchronously during the rotation of the lunar rover. Every time it rotates by Δθ = 10 ° , the left and right cameras independently take a picture each, forming a pair of left and right images at the same moment. Each pair of images is synchronously recorded with the corresponding real-time temperature T, rotation angle θ, and translation displacement t data.

[0045] Furthermore, through the Harris corner detection method, extract feature points from the images and accurately calculate their pixel coordinates (x, y) in the image coordinate system to form a feature point data set. Through the motion control system of the lunar rover, record the rotation angle and displacement corresponding to the shooting moment of each image to form the path data of the two-dimensional translation motion of the camera, and pair the motion data with the image feature points as the input samples for binocular vision calibration.

[0046] S14: Based on the camera projection model in S12 and the multi-group calibration experiment sample data obtained in S13, calculate the distortion parameter κ at different temperatures T and establish a mapping model between temperature and distortion parameter. The mapping relationship model is expressed as:

[0047] κ(T) = {k1(T), k2(T)}

[0048] where κ(T) is the mapping relationship between temperature T and radial distortion k1 and tangential distortion k2 to adapt to the camera distortion correction under different temperature conditions; calibrate the radial distortion k1 and tangential distortion k2 separately through the distortion feature points at different temperatures.

[0049] S2: Based on the mapping relationship model, obtain the lunar soil brick image samples and their corresponding displacement cloud map labels, and construct a detection data set.

[0050] Specifically, the camera captures multiple images of lunar soil bricks at different temperatures. During this process, based on the mapping relationship model, the camera distortion coefficient is determined according to the environmental temperature, thereby correcting the camera distortion in real time. In addition, when shooting, it is preferable to clamp the left and right cameras of the lunar rover at 60°, and keep them at the same horizontal plane as the bottom surface of the speckle-free lunar soil brick experiment, fix the shooting position, angle, and perspective, and collect the images of the lunar soil bricks. Then, preprocess the images of the lunar soil bricks, and obtain the displacement cloud map labels corresponding to each image of the lunar soil brick through DIC analysis.

[0051] Furthermore, since it is difficult to obtain real lunar soil brick image samples on the moon and it is difficult to obtain a large amount of data, the detection dataset actually includes a real dataset and a simulated dataset. The real dataset contains real lunar soil brick image samples obtained on the moon and their corresponding displacement cloud map labels, and the simulated dataset contains simulated lunar soil brick image samples obtained by simulating the lunar environment on Earth in large quantities and their corresponding displacement cloud map labels.

[0052] Furthermore, preprocess the images of the lunar soil bricks, including: cutting the image samples into a unified size, dividing them into sub-image blocks, calculating the average variance of the sub-image blocks to evaluate the noise intensity of the images, and then performing denoising processing. Subsequently, perform histogram equalization on the sub-image blocks to enhance the surface texture of the lunar soil bricks and obtain the target image samples. Pair the target image samples with their corresponding true displacement cloud map labels obtained through DIC analysis. Specifically, it includes the following steps:

[0053] S21: Image cutting. Cut the original image sample dataset into sub-image blocks of a unified size according to preset requirements. The size of the original image is W×H, and the size of each sub-image block is w×h. The preset requirements are to specify the image size and the size of the sub-image blocks.

[0054] S22: Noise intensity analysis. Analyze the noise intensity of each sub-image block and calculate the variance of each sub-image block Its calculation formula is:

[0055]

[0056] where I(i, j) is the pixel gray value at position (i, j) in the sub-image block, and μ k is the gray average value of this sub-image block, and the calculation formula is:

[0057]

[0058] S23: Average variance calculation. Calculate the sum of the average variances of all sub-image blocks. The calculation formula is:

[0059]

[0060] where N is the total number of sub-image blocks, is the variance of the k-th sub-image block.

[0061] S24: Filtering intensity selection. Select the filtering intensity F according to the average variance value The relationship between the filtering intensity and the noise intensity is matched through the noise characteristic curve function to obtain:

[0062]

[0063] where is the filtering intensity function based on the noise intensity.

[0064] S25: Denoising processing. Use the selected filtering intensity F to perform denoising processing on each sub-image block. The filtering method used is mean filtering, Gaussian filtering or median filtering, and the most suitable filtering method is automatically selected according to the noise characteristics.

[0065] S26: Histogram equalization. Perform histogram equalization processing on each denoised sub-image block to enhance the surface texture of the lunar soil bricks; specifically including:

[0066] Calculate the gray histogram H(x) of each sub-image block, where x represents the gray value, and the calculation formula is:

[0067]

[0068] where δ is the delta function. When the pixel value I(i, j) is equal to x, δ(I(i, j)-x) = 1, otherwise it is 0;

[0069] Perform histogram equalization according to the cumulative distribution function of each sub-image block to obtain the new gray value I ′ (i, j), and the calculation formula is:

[0070] I ′ (i, j) = T(I(i, j))

[0071] where T(x) is the mapping function of histogram equalization, and its calculation formula is:

[0072]

[0073] where round means rounding, and H(t) represents the frequency of the gray value t.

[0074] S27: Image re-stitching. Re-stitch all processed sub-image blocks to obtain the enhanced target image, and finally form the target image sample with enhanced texture and gray scale.

[0075] Through the above image preprocessing steps, the noise in the original image can be effectively removed, the natural texture features on the surface of the speckle-free lunar soil bricks can be enhanced, and high-quality image data can be provided, providing reliable input support for the full-field deformation detection of lunar soil bricks.

[0076] S3: Train the detection network constructed based on Diffusion Transformers (DiT) using the detection dataset, and use the trained detection network as the full-field deformation detection model for lunar soil bricks.

[0077] Specifically, the detection network based on Diffusion Transformers includes:

[0078] An encoder module for extracting features from the input image sample I input to generate a feature representation F through multiple layers of convolution and Transformer blocks encoded ;

[0079] A denoising diffusion probability model module for introducing a noise diffusion process into the feature representation, specifically:

[0080]

[0081] where p(x t ∣x t-1 ) represents the relationship between the current state x t and the previous state x t-1 in the diffusion process, is the noise attenuation parameter, and I is the identity matrix;

[0082] A decoder module for recovering the high-resolution displacement cloud map prediction result from the diffused feature representation. Combining the self-attention mechanism of the Transformer module, it finds the correlation and dependence between different positions in the input features, integrates the global information into the feature representation of each position; through the method of multi-head attention, it analyzes the interaction between features from multiple perspectives, captures the global context information, and recovers the high-resolution displacement cloud map. The calculation formula of the self-attention mechanism is:

[0083] MultiHead(Q, K, V) = Concat(head1,..., head h )W O ,

[0084] head i = Attention(QW i Q , KW i K , VWi V )

[0085] Among them, W i Q , W i K , W i V are weight matrices of different heads respectively, and h represents the number of attention heads.

[0086] Furthermore, during training, first perform pre-training on the full-field deformation cloud map generation architecture of lunar soil bricks based on Diffusion Transformers through a simulated dataset to obtain a pre-trained detection model. Subsequently, freeze the backbone network, fine-tune the decoder part of the model, and adaptively fine-tune the model using a real dataset to obtain the target full-field deformation detection model of lunar soil bricks. As Figure 3 shown, it specifically includes:

[0087] S31: The simulated dataset contains a large number of image samples I input during the loading process of simulated lunar soil bricks and the displacement cloud map label y obtained through DIC analysis. Pre-train the detection network based on Diffusion Transformers through the simulated dataset, including:

[0088] Feature encoding, extract features from the input image sample I input through the encoder to obtain the feature representation:

[0089] F encoded = DiT Encoder(I input )

[0090] where F encoded is the feature map output by the encoder;

[0091] Add Gaussian noise to the denoising diffusion probability model and input the feature map F encoded into the denoising diffusion probability model;

[0092] Global context modeling, capture global context information through the self-attention mechanism of the Transformer module, and calculate the attention map:

[0093]

[0094] where Q, K, and V represent Query, Key, and Value respectively, and d k is the dimension of the key;

[0095] Use the DiT decoder to obtain from the feature representation F encodedPrediction results of the displacement nephogram restored in the middle:

[0096]

[0097] Among them, is the generated predicted displacement nephogram.

[0098] During the pre-training process, the mean squared error (MSE) loss function is used to measure the difference between the prediction result and the true displacement nephogram label. The loss function is defined as:

[0099]

[0100] Among them, is the displacement nephogram predicted by the model, y i is the true displacement nephogram, and N is the number of samples.

[0101] S32: Freeze the pre-trained part of the DiT backbone network, only fine-tune the decoder part of the model, and use the real dataset to fine-tune the model to obtain the target full-field deformation detection model of lunar soil bricks. Specifically, in the adaptive fine-tuning stage, by fixing the parameters of the encoder module and only updating the parameters of the decoder module, the loss function is minimized to optimize the difference between the generated displacement nephogram and the true displacement nephogram. The loss function is the same as that in S31 pre-training.

[0102] S4: The full-field deformation detection model of lunar soil bricks can predict the displacement and deformation of lunar soil bricks under different loading conditions from the input lunar soil brick images, and can be applied to the non-speckle three-dimensional DIC intelligent detection task of lunar soil bricks to provide accurate displacement prediction.

[0103] Specifically, during the lunar surface mechanical loading experiment of lunar soil bricks, the lunar surface environmental temperature is monitored in real time, the camera distortion coefficient corresponding to the environmental temperature is determined based on the mapping relationship model, and the lunar soil brick image is obtained through the camera; the lunar soil brick image is preprocessed, and the preprocessing method refers to step S2, which will not be elaborated here; then the full-field deformation detection model of lunar soil bricks is input to obtain a high-precision full-field deformation nephogram of lunar soil bricks, which accurately reflects the displacement and deformation information during the loading process of lunar soil bricks.

[0104] Those skilled in the art can easily understand that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent detection method for full-field deformation of lunar soil bricks, characterized in that: The steps include: Model training phase: The detection network built based on Diffusion Transformers is trained through the detection data set, and the trained detection network is used as the full-field deformation detection model of lunar soil bricks. The detection data set includes lunar soil brick image samples and their corresponding displacement cloud map labels; wherein, the lunar soil brick image samples are obtained through a camera, and in this process, based on a pre-constructed mapping relationship model between the ambient temperature and the camera distortion coefficient, the camera distortion coefficient is determined according to the ambient temperature, thereby correcting the camera distortion; Model application phase: During the mechanical loading experiment of lunar soil bricks, the ambient temperature is monitored in real time, the camera distortion coefficient corresponding to the ambient temperature is determined based on the mapping relationship model, and the lunar soil brick image is acquired through the camera; the lunar soil brick image is input into the soil brick full-field deformation detection model to obtain the displacement cloud map, thereby realizing the full-field deformation detection of the lunar soil brick.

2. The intelligent detection method for full-field deformation of lunar soil bricks according to claim 1 is characterized in that: The method for obtaining the displacement cloud map labels in the detection dataset is as follows: preprocess the lunar soil brick image samples to obtain the target image samples, and then obtain the corresponding displacement cloud map labels through DIC analysis.

3. The intelligent detection method for full-field deformation of lunar soil bricks as claimed in claim 2 is characterized in that: Preprocess the lunar soil brick image samples, including: The lunar soil brick image samples are divided into sub-image blocks of uniform size, and the sub-image blocks are denoised; then the denoised sub-image blocks are histogram equalized, and then the sub-image blocks are reassembled to obtain the target image samples.

4. The intelligent detection method for full-field deformation of lunar soil bricks as claimed in claim 3 is characterized in that: The sub-image block is subjected to denoising, including: The variance of the grayscale value of each sub-image block is calculated respectively, and then the average variance of all sub-image blocks in the lunar soil brick image sample is obtained; the filtering strength is determined according to the average variance, and the filtering strength is used to filter all sub-image blocks.

5. The intelligent detection method for full-field deformation of lunar soil bricks as claimed in claim 4 is characterized in that: The method for filtering the sub-image block is mean filtering, Gaussian filtering or median filtering.

6. The intelligent detection method for full-field deformation of lunar soil bricks according to claim 1 is characterized in that: The method for constructing a mapping relationship model between ambient temperature and camera distortion coefficient includes: Under fixed temperature conditions, calibrate the camera's intrinsic and extrinsic matrix; During the temperature change process, the camera captures multiple images and records the images and the corresponding ambient temperature; The feature points are extracted from the image through the Harris corner detection method. The distortion coefficients are calculated according to the world coordinates and pixel coordinates of the image feature points, as well as the intrinsic parameter matrix and the extrinsic parameter matrix. Thus, the camera distortion parameters corresponding to different ambient temperatures are obtained, and then the mapping relationship model between the ambient temperature and the camera distortion coefficient is fitted.

7. The intelligent detection method for full-field deformation of lunar soil bricks according to any one of claims 1 to 6, characterized in that: The detection network built based on Diffusion Transformers includes an encoder, a diffusion module and a decoder, wherein the encoder is used to extract the features of image samples, the diffusion module is used to introduce a noise diffusion process to the extracted features, and the decoder is used to decode the diffused features.

8. The intelligent detection method for full-field deformation of lunar soil bricks as claimed in claim 7, characterized in that: The detection data set includes a real data set and a simulated data set. The real data set is used to obtain real lunar soil brick image samples and their corresponding displacement cloud map labels, and the simulated data set is used to obtain simulated lunar soil brick image samples and their corresponding displacement cloud map labels. Pre-train the detection network using simulated datasets; After pre-training is completed, the detection network is fine-tuned using real data sets. During this process, the encoder parameters in the detection network are fixed and only the decoder parameters in the detection network are updated. The fine-tuned detection network is used as the full-field deformation detection model for lunar soil bricks.

9. The intelligent detection method for full-field deformation of lunar soil bricks as claimed in claim 8, characterized in that: The mean squared error loss function is used when pre-training the detection network with simulated datasets and when fine-tuning the decoder with real datasets.

10. An intelligent detection system for full-field deformation of lunar soil bricks, characterized in that: It includes a processor, which is used to execute the intelligent detection method of the full-field deformation of the lunar soil brick as described in any one of claims 1-9.

Images