Lightweight convolutional network-based DIC displacement field dynamic correction method and system

Abstract

The invention relates to the technical field of digital image correlation DIC measurement, in particular to a DIC displacement field dynamic correction method and system based on a lightweight convolutional network, and the method comprises the steps: 1, collecting a structure surface image sequence, and carrying out the preprocessing; 2, calculating an initial displacement field of the preprocessed image sequence through a traditional DIC algorithm, extracting key feature information, and taking the key feature information as input data for dynamic correction; 3, constructing a lightweight convolutional neural network model, transmitting the extracted input data to the lightweight convolutional neural network model for recognition, designing a mixed loss function for training, and learning a mapping relation between an initial displacement field and a real displacement field; and 4, correcting the initial displacement field through the trained lightweight convolutional neural network model, and directly outputting corrected high-precision displacement field data. By using the function fitting capability of the neural network, high-precision displacement field correction is realized under limited computing resources, and the measurement precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital image correlation (DIC) measurement, and in particular to a DIC displacement field dynamic correction method and system based on a lightweight convolutional network. Background Art

[0002] Digital Image Correlation (DIC), a non-contact, full-field measurement technique, has been widely used in fields such as structural health monitoring and material mechanical properties testing. However, in practical applications, the displacement field data obtained by DIC measurements often suffers from errors and insufficient precision due to various factors, such as noise interference, illumination variations, and complex structural surface deformations. This is particularly true in complex environments and monitoring scenarios with high real-time requirements. Improving the accuracy and reliability of displacement field measurements is an urgent issue.

[0003] Currently, traditional DIC algorithms suffer from low computational efficiency and limited accuracy when dealing with complex deformations and noise interference. While some deep learning-based methods have improved measurement accuracy to a certain extent, they typically require extensive computing resources and time for model training and inference, making it difficult to meet real-time requirements.

[0004] Lightweight Convolutional Neural Networks (CNNs) optimize network structure and computational processes. Through techniques such as depthwise separable convolution and network pruning, they significantly reduce network parameters, lower computational complexity, and improve data processing efficiency while maintaining model performance. Unlike traditional correction methods based on physical models, lightweight CNNs do not require manual design of complex feature extraction and correction algorithms. Instead, they automatically learn optimal feature representations and correction strategies through extensive training data. By implementing end-to-end training and prediction, from input image to output displacement field, the entire process requires no human intervention, avoiding the error accumulation that often occurs in traditional methods due to improper parameter settings or unreasonable model assumptions.

[0005] In the field of structural health monitoring, DIC measurement technology is being used more and more widely. In order to improve the accuracy and efficiency of measurement, various optimizations and improvements have been made to the traditional DIC method.

[0006] Patent CN201510100472.9 discloses an adaptive smoothing method for displacement fields related to digital images. Based on penalized least squares regression and generalized cross validation (GCV), it automatically optimizes the displacement field smoothing parameters and combines discrete cosine transform (DCT) to suppress noise and improve the accuracy of strain field calculations. However, this method involves multiple iterations and smoothing processing, has high computational complexity, and has poor real-time performance.

[0007] Patent CN202111160148.8 discloses a DIC-based method for measuring the fatigue crack propagation morphology of steel structures. It uses the zero-mean normalized cross-correlation algorithm (ZNCC) to distinguish between coarse and fine crack morphologies, and combines topological structure displacement field analysis to achieve dynamic monitoring. However, it relies on traditional DIC algorithms (such as sub-pixel interpolation) and does not consider the impact of interference such as noise and illumination changes on the accuracy of the displacement field.

[0008] Patent CN202410077820.4 proposes a speckle-free three-dimensional DIC monitoring method and system for concrete rust expansion damage. It extracts the natural texture features of concrete through the SIFT operator to replace artificial speckle. It is based on affine mapping and iterative optimization algorithm, but the processing speed is limited and the correction results cannot be output in real time.

[0009] Patent CN201910659769.7 discloses a strain field calculation method that combines SPM and DIC technologies. This method uses discrete cosine transform (DCT) to adaptively smooth the displacement field and eliminate random errors. However, it does not address adaptive correction mechanisms in dynamic environments (such as temperature fluctuations and vibrations). This method is susceptible to data drift, which can lead to decreased accuracy during long-term monitoring. However, there are relatively few methods and systems for improving accuracy based on artificial intelligence.

[0010] Patent CN202310512199.5 proposes a chip warpage prediction method and system based on DIC measurement and machine learning. This method uses displacement field data from the chip surface as input and trains a machine learning model to predict chip warpage. For scenarios requiring both real-time performance and high precision, existing technologies have limitations, lacking a method and system that can simultaneously meet the requirements of computational efficiency, strong image feature learning capabilities, and rapid training and inference. Summary of the Invention

[0011] In response to the problems existing in the prior art, the purpose of the present invention is to provide a method and system for dynamic real-time correction of DIC displacement fields based on lightweight convolutional neural networks, which can significantly improve measurement accuracy while meeting real-time requirements and adapt to the needs of structural health monitoring in complex environments. At the same time, it also provides an efficient measurement method for the field of structural health monitoring, further promoting technological progress and development in this field.

[0012] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a dynamic correction method for DIC displacement field based on a lightweight convolutional network, comprising the following steps: Step 1: Collect an image sequence of the structure surface and preprocess the collected image sequence; Step 2: Calculate the initial displacement field of the preprocessed image sequence using the traditional DIC algorithm and extract key feature information as input data for dynamic correction; Step 3: Construct a lightweight convolutional neural network model. The lightweight convolutional neural network model uses the initial displacement field containing noise interference and the corresponding true displacement field as the training data set. The extracted input data is transmitted to the lightweight convolutional neural network model for recognition. The model is trained by designing a hybrid loss function to learn the mapping relationship between the initial displacement field and the true displacement field. Step 4: Correct the initial displacement field through the trained lightweight convolutional neural network model and directly output the corrected high-precision displacement field data.

[0013] In the above-mentioned DIC displacement field dynamic correction method based on a lightweight convolutional network, step 1 includes: Step 1-1: Based on the actual monitoring requirements and the structural characteristics and deformation characteristics, select a high-speed camera with appropriate focal length and resolution, determine the optimal installation position and angle, and use a fixture to ensure the stability of the camera to clearly capture the deformation of the structural surface during the applied load process; Step 1-2: Based on the rate of structural deformation and monitoring requirements, continuously acquire images at the set frame rate. The resolution of image acquisition meets the following requirements: number of pixels / unit length ≥ ,in d is the minimum characteristic size of the structure surface; Step 1-3: Use the weighted average method to grayscale the image and convert the color image into a grayscale image. The calculation formula is: Y =0.299 R +0.587 G +0.114 B ,in, R 、 G 、 B are the pixel values of the red, green, and blue channels of the image, respectively. Y is the pixel value after grayscale; Steps 1-4: Select an appropriate noise filtering algorithm based on the noise type and intensity in the image, process the grayscale image, remove random noise interference in the image, and improve image quality; Steps 1-5: Normalize the image. According to the model's input requirements and data distribution characteristics, the image pixel values are unified to the range of [0, 1] to enhance the model's adaptability and generalization ability to different image data. The calculation formula is: ,in, I is the pixel value of the original image, and are the minimum and maximum pixel values of the image, is the normalized pixel value; Step 1-6: Evaluate the quality of the pre-processed image by calculating the mean and standard deviation of the image.

[0014] In the above-mentioned lightweight convolutional network-based dynamic correction method for DIC displacement fields, in steps 1-4, the noise filtering algorithm uses non-local mean filtering to filter the image by calculating the similarity weights between pixels, thereby preserving the image texture and details, including: Step a: Divide the image into several pixel patches, each of which is centered on a pixel and contains surrounding neighborhood pixels. Step b: Define pixel blocks, for each pixel in the image x , define a x Centered at n × n Pixel blocks I patch ( x ); Step c: Calculate the similarity weight for two pixels in the image x and y , calculate the similarity weight of the pixel blocks they are in , the calculation formula of similarity weight is: ,in, represents the Euclidean distance between two pixel blocks, h is a smoothing parameter used to control the decay rate of the weight; Step d: Normalization, calculation of the center pixel x The normalization constant : ,in, is the image domain; Step e: Update pixel values and update the center pixel according to the similarity weight and normalization constant x Value: , repeat the above steps to get the final filtered image .

[0015] In the above-mentioned DIC displacement field dynamic correction method based on a lightweight convolutional network, step 2 includes: Step 2-1: Compare the grayscale correlation of pixels between adjacent images, calculate the initial displacement field of each point on the surface of the structure, and use the corner detection algorithm to detect the corner features in the image. The calculation formula for corner detection is: ,in, M is the autocorrelation matrix of the image, k is an empirical constant; Step 2-2: Use the edge detection algorithm to extract the edge information of the image, and combine it with the texture feature analysis method to comprehensively extract the information that can reflect the key deformation characteristics of the structure surface as input data.

[0016] In the above-mentioned DIC displacement field dynamic correction method based on a lightweight convolutional network, step 2-2 includes: Step f: Use Gaussian filter to smooth and denoise the image; Step g: Calculate the gradient magnitude and direction, and use the Sobel operator to calculate the gradient components of each pixel in the horizontal and vertical directions: , ,in, is the gradient component of the image in the horizontal direction, is the gradient component of the image in the vertical direction, is the input image after Gaussian filtering; Step h: Calculate the gradient magnitude and direction based on the gradient components: , ,in, is the gradient magnitude, is the gradient direction; Step i: Non-maximum suppression: compare the gradient amplitude of each pixel with the gradient amplitudes of two neighboring pixels in the same direction. If the gradient amplitude of a pixel is not the maximum, suppress the pixel and set its gradient amplitude to 0. Step j: Use the double threshold method to detect edges and obtain a complete edge contour.

[0017] The above-mentioned DIC displacement field dynamic correction method based on lightweight convolutional network, the dual threshold method edge detection includes: Set a high threshold T high and low threshold T low , the gradient amplitude is greater than T high The pixel points are marked as strong edges E strong , indicating that the pixel point belongs to an obvious edge, and the gradient amplitude is between T low and T high The pixels between are marked as weak edges E weak , indicating that the pixel may be part of the edge, but it is not certain, and the gradient amplitude is less than T lowThe pixel point is marked as non-edge, indicating that the pixel point does not belong to the edge; From the strong edge E strong Start by searching for pixels in its neighborhood along the gradient direction or its reverse direction. If there is a weak edge in the neighborhood E weak , assuming that the weak edge is connected to the strong edge, it will be marked as a strong edge, thereby realizing the connection of the edge. The above process is repeated until all possible connected weak edges are processed and finally a complete edge contour is obtained.

[0018] In the above-mentioned DIC displacement field dynamic correction method based on a lightweight convolutional network, in step 2-2, the texture feature analysis uses the gray-level co-occurrence matrix (GLCM) method to calculate the contrast, correlation, energy, and entropy features of the image, and the calculated features are fused to form a feature vector, including: Step k: Convert the image into a grayscale image and construct a grayscale co-occurrence matrix based on the image parameters, including the grayscale level N , spatial relationships, including determining distances between pairs of pixels d and angles ; Step 1: Initialize the co-occurrence matrix and create a N × N Matrix P , initialize all elements to 0; Step m: Fill the co-occurrence matrix and traverse every pixel in the image , get its grayscale value ; According to the set distance d and angles , find the corresponding neighbor pixels , get its grayscale value ; Increase the count of elements in the co-occurrence matrix ; Step n: Normalize the co-occurrence matrix and divide each element in the co-occurrence matrix by the total number of pixel pairs to obtain the probability matrix P ; Step o: Calculate texture features from the normalized co-occurrence matrix, including: Contrast reflects the uniformity of the grayscale distribution of the image: , Correlation reflects the linear relationship of gray values in the image: , , , , , Energy value that reflects the uniformity of the grayscale distribution of the image: , Entropy reflects the complexity of the image grayscale distribution: ,in, Represents the row index gray level in the gray level co-occurrence matrix i The mean of the grayscale values, Represents the column index gray level in the gray level co-occurrence matrix j The mean of the grayscale values, Represents the row index gray level in the gray level co-occurrence matrix i The standard deviation of the grayscale values, Represents the column index gray level in the gray level co-occurrence matrix j The standard deviation of the grayscale values.

[0019] In the above-mentioned DIC displacement field dynamic correction method based on lightweight convolutional network, step 3 includes: Step 3-1: Decompose the standard convolution into deep convolution and gradual convolution to build a lightweight convolutional neural network model. During the model training process, design a hybrid loss function and set the mean square error loss according to the characteristics of the training data and the performance requirements of the model. , gradient loss and combat losses The weights between them balance the contributions of each part, and the calculation formula is: ,in, , , are weight coefficients that control the contribution of mean square error loss, gradient loss, and adversarial loss respectively; Step 3-2: Use the learning rate adjustment strategy and optimization algorithm to perform multiple rounds of iterative training until the model converges. The trained model can learn the mapping relationship between the initial displacement field and the actual displacement field.

[0020] In the above-mentioned DIC displacement field dynamic correction method based on a lightweight convolutional network, step 4 includes: Step 4-1: Input the initial displacement field data: The initial displacement field data calculated by the traditional DIC algorithm As input to a lightweight convolutional neural network model; Step 4-2: Input the initial displacement field data into the lightweight convolutional neural network for forward propagation calculation; Step 4-3: Convolution layer calculation, for each convolution kernelk , the calculation formula is: ,in, H and W is the height and width of the convolution kernel, is the initial displacement field data at position The value of Is the convolution kernel at position The weight value of Step 4-4: Activation function calculation. After the convolution layer, the activation function is applied to introduce nonlinearity, allowing the neural network to learn and fit more complex functional relationships. The calculation formula is: ; Step 4-5: Get the correction amount , after the network's forward propagation, the network outputs the correction With the initial displacement field data The dimensions are the same, Each element in represents the displacement correction value of the corresponding position; Steps 4-6: Calculate the corrected displacement field : .

[0021] A DIC displacement field dynamic correction system based on a lightweight convolutional network, comprising: Image acquisition and processing module: It consists of a high-speed camera, image processing software and its control circuit, and is used to acquire image sequences of the surface of the structure and pre-process the acquired image sequences; Initial displacement field calculation and feature extraction module: The initial displacement field is calculated from the preprocessed image sequence using the traditional DIC algorithm, and the key feature information is extracted using the feature extraction algorithm library as input data for dynamic correction; Neural network correction module: It contains a trained lightweight convolutional neural network model, receives input data from the initial displacement field calculation and feature extraction module, uses the initial displacement field containing noise interference and the corresponding real displacement field as the training data set, transmits the extracted input data to the lightweight convolutional neural network model for recognition, and trains by designing a hybrid loss function to learn the mapping relationship between the initial displacement field and the real displacement field. The initial displacement field is corrected by the trained lightweight convolutional neural network model, and the corrected high-precision displacement field data is directly output.

[0022] The beneficial effects of the present invention's method and system for dynamic correction of DIC displacement fields based on a lightweight convolutional network are as follows: the lightweight convolutional neural network is used to dynamically correct the displacement field obtained by DIC measurement, which can fully utilize the powerful function fitting ability of the neural network, achieve high-precision displacement field correction under limited computing resources, and significantly improve measurement accuracy.

[0023] By designing a hybrid loss function, the absolute error of the displacement field, gradient information, and adversarial training are comprehensively considered, so that the model can better learn the complex mapping relationship between the initial displacement field and the true displacement field during training, thereby improving the reliability and stability of the correction results.

[0024] From the perspective of the overall process, the method of the present invention can perform dynamic real-time correction of the DIC displacement field while meeting the accuracy requirements. It is suitable for scenarios such as structural health monitoring and material testing in complex environments that require high-precision real-time displacement field measurement. It has broad application prospects and practical engineering value. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Flowchart of a method for dynamic real-time correction of DIC displacement field based on a lightweight convolutional network according to an embodiment of the present invention; Figure 2 Schematic diagram of the structure of a dynamic real-time correction system for DIC displacement field based on a lightweight convolutional neural network according to an embodiment of the present invention; Figure 3 This is a diagram of the training data set construction process under the experimental conditions in Example 2 of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is described below in conjunction with specific implementation methods and drawings.

[0027] Example 1 A dynamic correction method for DIC displacement field based on a lightweight convolutional network includes the following steps.

[0028] Step 1: Collect an image sequence of the structure surface and preprocess the collected image sequence.

[0029] A high-speed camera is used to capture a sequence of images of the structure's surface. The high-speed camera can capture surface images of the structure in different states at a high frame rate, providing a data basis for subsequent displacement field calculations.

[0030] The preprocessing process includes grayscale processing of the image to reduce the data dimension and improve processing efficiency; noise filtering processing to remove random noise interference in the image and improve image quality; and image normalization processing to unify the image pixel values to a certain range and enhance the generalization ability and stability of the model.

[0031] Step 2: Calculate the initial displacement field of the preprocessed image sequence using the traditional DIC algorithm, and extract key feature information as input data for dynamic correction.

[0032] When calculating the initial displacement field, traditional DIC algorithms determine the displacement of each point on the surface by comparing the grayscale changes of pixels between adjacent images. To better capture the key deformation features of the surface, a combination of image-based corner detection, edge detection, and texture feature analysis is employed to extract key feature information. This key feature information reflects the typical deformation characteristics of the surface, providing valuable input for subsequent neural network corrections.

[0033] Step 3: Construct a lightweight convolutional neural network model. The lightweight convolutional neural network model uses the initial displacement field containing noise interference and the corresponding real displacement field as the training data set. The extracted input data is transmitted to the lightweight convolutional neural network model for recognition. The model is trained by designing a hybrid loss function to learn the mapping relationship between the initial displacement field and the real displacement field.

[0034] During training, a hybrid loss function is designed to achieve high-precision displacement field correction with limited computing resources. This hybrid loss function includes a mean square error loss function, which measures the difference between the predicted and true displacement fields, ensuring the model is as close to the true value as possible; a gradient loss function, which constrains the gradient of the displacement field to maintain its smoothness and continuity; and an adversarial loss function, which introduces an adversarial training mechanism to improve the model's generalization ability and robustness to noise.

[0035] Step 4: Correct the initial displacement field through the trained lightweight convolutional neural network model and directly output the corrected high-precision displacement field data.

[0036] Through end-to-end training, the lightweight convolutional neural network model can learn the mapping relationship between the initial displacement field and the true displacement field, thereby achieving accurate correction of the initial displacement field.

[0037] Example 2 As shown in Figure 1, a dynamic correction method for DIC displacement field based on a lightweight convolutional network includes the following steps.

[0038] Step 1: Use a high-speed camera to capture a sequence of images of the structure's surface. Based on the actual monitoring requirements and the structural characteristics and deformation patterns, select a high-speed camera with an appropriate focal length and resolution, and determine the optimal mounting position and angle to ensure clear capture of surface deformation during load application. Use a tripod or other mounting device to ensure camera stability and avoid vibration during acquisition.

[0039] Image acquisition is carried out continuously at a certain frame rate. The frame rate should be determined according to the rate of structural deformation and monitoring requirements to ensure that the deformation process of the structure can be fully captured. The resolution should meet the number of pixels / unit length ≥ ( d is the minimum feature size on the surface of the structure), ensuring that the image can clearly distinguish the details of the structure.

[0040] Step 2: Preprocess the collected image sequence. First, grayscale the image to convert the color image into a grayscale image to reduce the amount of data and computational complexity. The grayscale processing uses the weighted average method, and the calculation formula is: Y =0.299 R +0.587 G +0.114 B， in, R 、 G 、 B are the pixel values of the red, green, and blue channels of the image, respectively. Y is the pixel value after grayscale conversion.

[0041] Then, according to the type and intensity of noise in the image, select the appropriate noise filtering algorithm, such as Gaussian filtering, median filtering, bilateral filtering, wavelet transform, non-local mean filtering, etc., to process the grayscale image, remove random noise interference in the image, and improve image quality.

[0042] Optionally, the weight distribution of the Gaussian filter is in the shape of a Gaussian function, where the weight of pixels farther from the center pixel is smaller, so that pixels near the edge are not smoothed too much, thereby better preserving the edge details of the image.

[0043] Optionally, median filtering is a nonlinear filtering algorithm that replaces the value of each pixel with the median of the pixel values in its neighborhood, effectively removing salt and pepper noise while preserving image edges and detail information.

[0044] Alternatively, bilateral filtering takes into account both the spatial distance and grayscale similarity of pixels and is suitable for images that require fine processing.

[0045] Alternatively, wavelet transform can remove noise while retaining more local features. It decomposes the image into different scales and frequencies for processing, and is more applicable to images with complex textures and rich details.

[0046] Optionally, non-local mean filtering performs filtering on an image by calculating similarity weights between pixels, which can better preserve the texture and details of the image.

[0047] The specific steps include: The image is divided into several pixel blocks (patches), each of which is centered on a pixel and contains surrounding neighborhood pixels.

[0048] Define pixel blocks, for each pixel in the image x , define a x Centered at n × n Pixel blocks I patch ( x ).

[0049] Calculate the similarity weight for two pixels in the image x and y , calculate the similarity weight of the pixel blocks they are in ω ( x , y ).

[0050] The calculation formula for similarity weight is: ,in, represents the Euclidean distance between two pixel blocks, h is a smoothing parameter that controls the decay rate of the weight.

[0051] Normalization processing, calculation of the center pixel x The normalization constant : ,in, is the image domain.

[0052] Update pixel values and update the center pixel according to the similarity weight and normalization constant x Value: .

[0053] Repeat the above steps to get the final filtered image .

[0054] After filtering, the image is normalized. According to the model’s input requirements and data distribution characteristics, the image pixel values are unified to the range of [0, 1] to enhance the model’s adaptability and generalization ability to different image data. The calculation formula is: ,in, I is the pixel value of the original image, and are the minimum and maximum pixel values of the image, is the normalized pixel value.

[0055] Finally, the quality of the preprocessed image can be evaluated by calculating the mean and standard deviation of the image to ensure that it meets the requirements of subsequent processing.

[0056] Step 3: Use a traditional DIC algorithm to calculate the initial displacement field of the preprocessed image sequence and extract key feature information. Traditional DIC algorithms calculate the initial displacement field for each point on the structure's surface by comparing the grayscale correlation of pixels between adjacent images. Based on the deformation characteristics of the structure and monitoring requirements, appropriate DIC algorithm parameters, such as the search area size and matching template size, are selected to improve the accuracy of the initial displacement field calculation.

[0057] At the same time, a corner detection algorithm (such as Harris corner detection) is used to detect corner features in the image, and an edge detection algorithm (such as Canny edge detection) is used to extract image edge information. Combined with texture feature analysis methods (such as local binary patterns), information that can reflect the key deformation characteristics of the structural surface is comprehensively extracted as input data.

[0058] The calculation formula of Harris corner detection is: ,in, M is the autocorrelation matrix of the image, k It is an empirical constant, usually between 0.04 and 0.06.

[0059] The Canny edge detection algorithm includes the following steps.

[0060] Image smoothing and denoising: Use Gaussian filter to smooth and denoise the image.

[0061] Calculate the gradient magnitude and direction, and use the Sobel operator to calculate the gradient components of each pixel in the horizontal and vertical directions: , , in, is the gradient component of the image in the horizontal direction, is the gradient component of the image in the vertical direction, is the input image after Gaussian filtering.

[0062] Calculate the gradient magnitude and direction based on the gradient components: , , in, is the gradient magnitude, is the gradient direction.

[0063] Non-maximum suppression compares the gradient magnitude of each pixel with the gradient magnitudes of two neighboring pixels in the same direction. If the gradient magnitude of a pixel is not the maximum, the pixel is suppressed and its gradient magnitude is set to 0.

[0064] Double threshold method for edge detection.

[0065] Set the high threshold ( T high ) and low threshold ( T low ).

[0066] The gradient magnitude is greater than T high The pixel points are marked as strong edges ( E strong ), indicating that the pixel belongs to a clear edge.

[0067] Set the gradient amplitude between T low and T high The pixels between are marked as weak edges ( E weak ), indicating that the pixel may be part of the edge, but it is not certain.

[0068] The gradient amplitude is less than T low The pixel point is marked as non-edge, indicating that the pixel point does not belong to the edge.

[0069] From the strong edge E strong Start by searching for pixels in its neighborhood along the gradient direction or its reverse direction. If there is a weak edge in the neighborhood E weak , assuming that the weak edge is connected to the strong edge, it is marked as a strong edge, thus achieving edge connection. Repeat the above process until all possible connected weak edges are processed and finally a complete edge contour is obtained.

[0070] Texture feature analysis uses the gray-level co-occurrence matrix (GLCM) method to calculate image features such as contrast, correlation, energy, and entropy, and fuses the calculated features to form a feature vector for subsequent analysis and processing.

[0071] The specific steps are as follows: If the input image is a color image, it must first be converted to a grayscale image.

[0072] To construct the gray-level co-occurrence matrix (GLCM), we first need to determine the following parameters.

[0073] Grayscale level: The grayscale level is usually set to 256 (for 8-bit images), or the grayscale level can be compressed as needed (such as 16 levels, 32 levels, etc.).

[0074] Spatial Relationships: Determining the distance between pairs of pixels d and angles Common distances d is 1, angle Usually 0°, 45°, 90°, and 135°.

[0075] The steps for constructing GLCM are as follows.

[0076] Initialize the co-occurrence matrix and create a N × N Matrix P ,in N is the number of gray levels, and all elements are initialized to 0.

[0077] Fill the co-occurrence matrix and traverse every pixel in the image , get its grayscale value ; According to the set distance d and angles , find the corresponding neighbor pixels , get its grayscale value ; Increase the count of elements in the co-occurrence matrix .

[0078] Normalize the co-occurrence matrix and divide each element in the co-occurrence matrix by the total number of pixel pairs to obtain the probability matrix P .

[0079] Compute texture features from the normalized co-occurrence matrix P The following texture features are calculated in .

[0080] Contrast reflects the uniformity of the grayscale distribution of the image: , Correlation reflects the linear relationship between the grayscale values in the image: , , , , , The energy value also reflects the uniformity of the image grayscale distribution: , Entropy reflects the complexity of the grayscale distribution of the image: ,in, Represents the row index (gray level) in the gray level co-occurrence matrix i ), Represents the column index (gray level) in the gray level co-occurrence matrix j ), Represents the row index (gray level) in the gray level co-occurrence matrix i ) of the grayscale values, Represents the column index (gray level) in the gray level co-occurrence matrix j ) is the standard deviation of the grayscale values.

[0081] Based on the above steps, according to the specific structure and deformation conditions, the parameters of the feature extraction algorithm are adjusted, such as the threshold in the corner detection algorithm and the filter size in the edge detection algorithm. The extracted feature information is filtered and fused to remove redundant and irrelevant information to ensure that the feature data input into the neural network has high quality and representativeness.

[0082] Step 4: The extracted input data is transmitted to a lightweight convolutional neural network model for recognition. During the training phase, the lightweight convolutional neural network model uses an initial displacement field containing noise interference and the corresponding real displacement field as the training dataset. For example, a DIC measurement of a structure with known deformation can be performed under laboratory conditions using high-precision measurement equipment to obtain an image sequence containing real displacement field data. Based on the characteristics and intensity of noise in actual application scenarios, an appropriate noise type (such as Gaussian noise, salt and pepper noise, etc.) and intensity parameters are selected. Noise is added to the initial displacement field data to simulate measurement conditions in actual complex environments, forming a training dataset.

[0083] The lightweight convolutional neural network model uses depthwise separable convolution, decomposing the standard convolution into depthwise convolution and gradual convolution, significantly reducing the amount of computation and the number of model parameters. During model training, a hybrid loss function is designed. Based on the characteristics of the training data and the performance requirements of the model, the weights between the mean square error loss, gradient loss, and adversarial loss are reasonably set to balance the contribution of each part and achieve the best training effect. The calculation formula is: ,in, , , are weight coefficients that control the contribution of mean square error loss, gradient loss, and adversarial loss respectively.

[0084] The mean square error loss function is used to calculate the difference between the predicted displacement field and the true displacement field: ,in, and are the predicted displacement field and the real displacement field, respectively. i elements, N is the total number of elements.

[0085] The gradient loss function is used to constrain the gradient change of the displacement field: ,in, and are the gradients of the predicted displacement field and the true displacement field, respectively.

[0086] The adversarial loss function introduces an adversarial training mechanism between the discriminant network and the generative network. It calculates the adversarial loss value based on the discriminant network's discrimination results on the real data and the generated data, as well as the corresponding real labels. The calculation formula is: ,in, N is the data sample size, is the true label (real data is 1, generated data is 0), Is the discriminant network for real data The judgment result of The network generates the initial displacement field data containing noise The generated result, Is the discriminant network to generate data The judgment result of .

[0087] Using appropriate learning rate adjustment strategies and optimization algorithms, multiple rounds of iterative training are performed until the model converges. Ultimately, the trained model can better learn the mapping relationship between the initial displacement field and the true displacement field, correct the initial displacement field, and after end-to-end training, can directly output the corrected high-precision displacement field data.

[0088] Correction steps include: Input initial displacement field data: The initial displacement field data calculated by the traditional DIC algorithm As the input of the lightweight convolutional neural network model, the initial displacement field data is a two-dimensional matrix that represents the initial displacement value of each point on the surface of the structure.

[0089] Network forward propagation: The initial displacement field data is input into the lightweight convolutional neural network for forward propagation calculation.

[0090] Convolution layer calculation: For each convolution kernel k , the calculation formula is: ,in, H and W is the height and width of the convolution kernel, is the initial displacement field data at position The value of Is the convolution kernel at position The weight value of .

[0091] Activation function calculation: After the convolution layer, the activation function is applied to introduce nonlinearity, enabling the neural network to learn and fit more complex functional relationships. The calculation formula is: .

[0092] Get correction amount : After forward propagation through the network, the network outputs the correction amount . It is also a two-dimensional matrix, which is consistent with the initial displacement field data The dimensions are the same. Each element in represents the displacement correction value of the corresponding position.

[0093] Calculate the corrected displacement field : .

[0094] The process of constructing the training dataset under laboratory conditions mentioned in step 4 is as follows Figure 3 As shown, the specific steps include.

[0095] Prepare a structural specimen with known deformation: Select a structural specimen with a known deformation mode, for example, a material specimen that will undergo uniform tension or bending under a specific load.

[0096] Perform DIC measurements to obtain the true displacement field: Mount the specimen on a high-precision measuring device and use a high-speed camera to capture a sequence of images of the structure's surface. Ensure that the camera parameters (such as frame rate and resolution) are set appropriately to capture a clear view of the deformation process. Synchronously start the loading device and the high-speed camera to record image data of the structure during the deformation process. Use the high-precision measuring device to obtain true displacement field data on the structure's surface. This data will serve as the true labels in the training dataset.

[0097] Simulate the actual initial displacement field: Add noise to the actual displacement field data to simulate the measurement conditions in an actual complex environment. You can add Gaussian noise, salt and pepper noise, etc. The noise intensity should be set according to the degree of interference that may be encountered in the actual application scenario.

[0098] Data pairing and organization: Pair the noise-added data with the real displacement field data to ensure that each pair of data is aligned in timestamp and spatial position for accurate correspondence in model training; organize the paired data into training, validation, and test sets.

[0099] Forming a training dataset: Store the paired and organized data in a database or file system to build a complete training dataset for training the lightweight convolutional neural network model.

[0100] Example 3 Based on Examples 1 and 2, as shown in Figure 2, this embodiment provides a real-time DIC displacement field correction system based on a lightweight convolutional network. The system includes an image acquisition and processing module, an initial displacement field calculation and feature extraction module, and a neural network correction module. The image acquisition and processing module includes an image acquisition module and a preprocessing module.

[0101] Image acquisition module: It consists of a high-speed camera and its control circuit, responsible for acquiring image sequences of the structure surface.

[0102] Preprocessing module: runs on computer image processing software or dedicated image processing hardware, and performs preprocessing operations such as grayscale conversion, noise filtering and normalization on the collected image sequence.

[0103] Initial displacement field calculation and feature extraction module: Integrates the traditional DIC algorithm program and feature extraction algorithm library to calculate the initial displacement field and extract key feature information.

[0104] Neural Network Correction Module: This module contains a trained lightweight convolutional neural network model, receives input data from the initial displacement field calculation and feature extraction modules, and rapidly outputs corrected, high-precision displacement field data through model inference. The entire system works collaboratively to achieve dynamic, real-time correction of the DIC displacement field.

[0105] The lightweight convolutional neural network model uses the initial displacement field data containing noise and the corresponding real displacement field data as training sets and is trained based on the designed hybrid loss function. During the model training process, the weights of each part of the hybrid loss function are reasonably set according to the characteristics of the training data and the performance requirements of the model, and an appropriate learning rate adjustment strategy and optimization algorithm are adopted to perform multiple rounds of iterative training until the model converges. The trained model can construct a mapping relationship function between the initial displacement field and the real displacement field, thereby correcting the new initial displacement field data to make it closer to the real displacement field. In practical applications, the performance of the model is evaluated in different scenarios, such as calculating the error indicators (such as mean square error, mean absolute error, etc.) between the corrected displacement field data and the real displacement field data to evaluate the degree of improvement in the model's accuracy; at the same time, the processing speed and delay of the system for real-time data are measured to evaluate the real-time performance of the system, so as to more comprehensively demonstrate the advantages and application value of the present invention.

[0106] By combining a lightweight convolutional neural network with traditional DIC technology, an innovative dynamic real-time correction method and system are proposed. This method effectively addresses the issue of insufficient DIC displacement field measurement accuracy in complex environments while also meeting real-time requirements. This provides a more accurate and reliable displacement field measurement method for fields such as structural health monitoring and material testing, demonstrating significant technical advantages and application value. It significantly improves measurement accuracy while meeting real-time requirements, adapting to the needs of structural health monitoring in complex environments.

[0107] The above embodiments are intended only to illustrate the structural concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the present invention and implement it accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent changes or modifications made based on the essence of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dynamic correction method for DIC displacement field based on lightweight convolutional network, characterized by: The following steps are involved: Step 1: Collect an image sequence of the structure surface and preprocess the collected image sequence; Step 2: Calculate the initial displacement field of the preprocessed image sequence using the traditional DIC algorithm and extract key feature information as input data for dynamic correction; Step 3: Construct a lightweight convolutional neural network model. The lightweight convolutional neural network model uses the initial displacement field containing noise interference and the corresponding true displacement field as the training data set. The extracted input data is transmitted to the lightweight convolutional neural network model for recognition. The model is trained by designing a hybrid loss function to learn the mapping relationship between the initial displacement field and the true displacement field. Step 4: Correct the initial displacement field through the trained lightweight convolutional neural network model and directly output the corrected high-precision displacement field data.

2. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 1 is characterized in that: The step 1 comprises: Step 1-1: Based on the actual monitoring requirements and the structural characteristics and deformation characteristics, select a high-speed camera with appropriate focal length and resolution, determine the optimal installation position and angle, and use a fixture to ensure the stability of the camera to clearly capture the deformation of the structural surface during the applied load process; Step 1-2: Based on the rate of structural deformation and monitoring requirements, continuously acquire images at the set frame rate. The resolution of image acquisition meets the following requirements: number of pixels / unit length ≥ ,in d is the minimum characteristic size of the structure surface; Step 1-3: Use the weighted average method to grayscale the image and convert the color image into a grayscale image. The calculation formula is: Y =0.299 R +0.587 G +0.114 B ,in, R 、 G 、 B are the pixel values of the red, green, and blue channels of the image, respectively. Y is the pixel value after grayscale; Steps 1-4: Select an appropriate noise filtering algorithm based on the noise type and intensity in the image, process the grayscale image, remove random noise interference in the image, and improve image quality; Steps 1-5: Normalize the image. According to the model's input requirements and data distribution characteristics, the image pixel values are unified to the range of [0, 1] to enhance the model's adaptability and generalization ability to different image data. The calculation formula is: , where I is the pixel value of the original image, and are the minimum and maximum pixel values of the image, is the normalized pixel value; Step 1-6: Evaluate the quality of the pre-processed image by calculating the mean and standard deviation of the image.

3. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 1, characterized in that: In steps 1-4, the noise filtering algorithm uses non-local mean filtering to filter the image by calculating the similarity weights between pixels, thereby preserving the texture and details of the image, including: Step a: Divide the image into several pixel patches, each of which is centered on a pixel and contains surrounding neighborhood pixels. Step b: Define pixel blocks, for each pixel in the image x , define a x Centered at n × n Pixel blocks I patch ( x ); Step c: Calculate the similarity weight. For two pixels x and y in the image, calculate the similarity weight of the pixel block they are in. , the calculation formula of similarity weight is: ,in, represents the Euclidean distance between two pixel blocks, and h is a smoothing parameter used to control the decay speed of the weight; Step d: Normalization, calculation of the center pixel x The normalization constant : ,in, is the image domain; Step e: Update pixel values and update the center pixel according to the similarity weight and normalization constant x Value: , repeat the above steps to get the final filtered image .

4. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 1, characterized in that: The step 2 includes: Step 2-1: Compare the grayscale correlation of pixels between adjacent images, calculate the initial displacement field of each point on the surface of the structure, and use the corner detection algorithm to detect the corner features in the image. The calculation formula for corner detection is: ,in, M is the autocorrelation matrix of the image, k is an empirical constant; Step 2-2: Use the edge detection algorithm to extract the edge information of the image, and combine it with the texture feature analysis method to comprehensively extract the information that can reflect the key deformation characteristics of the structure surface as input data.

5. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 4 is characterized in that: The step 2-2 includes: Step f: Use Gaussian filter to smooth and denoise the image; Step g: Calculate the gradient magnitude and direction, and use the Sobel operator to calculate the gradient components of each pixel in the horizontal and vertical directions: ， ,in, is the gradient component of the image in the horizontal direction, is the gradient component of the image in the vertical direction, is the input image after Gaussian filtering; Step h: Calculate the gradient magnitude and direction based on the gradient components: ， ,in, is the gradient magnitude, is the gradient direction; Step i: Non-maximum suppression: compare the gradient amplitude of each pixel with the gradient amplitudes of two neighboring pixels in the same direction. If the gradient amplitude of a pixel is not the maximum, suppress the pixel and set its gradient amplitude to 0. Step j: Use the double threshold method to detect edges and obtain a complete edge contour.

6. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 5, characterized in that: The dual threshold method for edge detection includes: Set a high threshold T high and low threshold T low , the gradient amplitude is greater than T high The pixel points are marked as strong edges E strong , indicating that the pixel point belongs to an obvious edge, and the gradient amplitude is between T low and T high The pixels between are marked as weak edges E weak , indicating that the pixel may be part of the edge, but it is not certain, and the gradient amplitude is less than T low The pixel point is marked as non-edge, indicating that the pixel point does not belong to the edge; From the strong edge E strong Start by searching for pixels in its neighborhood along the gradient direction or its reverse direction. If there is a weak edge in the neighborhood E weak , assuming that the weak edge is connected to the strong edge, it will be marked as a strong edge, thereby realizing the connection of the edge. The above process is repeated until all possible connected weak edges are processed and finally a complete edge contour is obtained.

7. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 4, characterized in that: In step 2-2, the texture feature analysis uses the gray level co-occurrence matrix (GLCM) method to calculate the contrast, correlation, energy, and entropy features of the image, and fuses the calculated features to form a feature vector, including: Step k: Convert the image into a grayscale image and construct a grayscale co-occurrence matrix based on the image parameters, including the grayscale level N , spatial relationships, including determining distances between pairs of pixels d and angles ; Step 1: Initialize the co-occurrence matrix and create a N × N Matrix P , initialize all elements to 0; Step m: Fill the co-occurrence matrix and traverse every pixel in the image , get its grayscale value ; According to the set distance d and angles , find the corresponding neighbor pixels , get its grayscale value ; Increase the count of elements in the co-occurrence matrix ; Step n: Normalize the co-occurrence matrix and divide each element in the co-occurrence matrix by the total number of pixel pairs to obtain the probability matrix P ; Step o: Calculate texture features from the normalized co-occurrence matrix, including: Contrast reflects the uniformity of the grayscale distribution of the image: ， Correlation reflects the linear relationship of gray values in the image: ， ， ， ， ， Energy value that reflects the uniformity of the grayscale distribution of the image: ， Entropy reflects the complexity of the grayscale distribution of the image: ,in, Represents the row index gray level in the gray level co-occurrence matrix i The mean of the grayscale values, Represents the column index gray level in the gray level co-occurrence matrix j The mean of the grayscale values, Represents the row index gray level in the gray level co-occurrence matrix i The standard deviation of the grayscale values, Represents the column index gray level in the gray level co-occurrence matrix j The standard deviation of the grayscale values.

8. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 1, characterized in that: The step 3 includes: Step 3-1: Decompose the standard convolution into deep convolution and gradual convolution to build a lightweight convolutional neural network model. During the model training process, design a hybrid loss function and set the mean square error loss according to the characteristics of the training data and the performance requirements of the model. , gradient loss and combat losses The weights between them balance the contributions of each part, and the calculation formula is: ,in, , , are weight coefficients that control the contribution of mean square error loss, gradient loss, and adversarial loss respectively; Step 3-2: Use the learning rate adjustment strategy and optimization algorithm to perform multiple rounds of iterative training until the model converges. The trained model can learn the mapping relationship between the initial displacement field and the actual displacement field.

9. The DIC displacement field dynamic correction method based on a lightweight convolutional network according to claim 1, characterized in that: The step 4 comprises: Step 4-1: Input the initial displacement field data: The initial displacement field data calculated by the traditional DIC algorithm As input to a lightweight convolutional neural network model; Step 4-2: Input the initial displacement field data into the lightweight convolutional neural network for forward propagation calculation; Step 4-3: Convolution layer calculation, for each convolution kernel k , the calculation formula is: ,in, H and W is the height and width of the convolution kernel, is the initial displacement field data at position The value of Is the convolution kernel at position The weight value of Step 4-4: Activation function calculation. After the convolution layer, the activation function is applied to introduce nonlinearity, allowing the neural network to learn and fit more complex functional relationships. The calculation formula is: ; Step 4-5: Get the correction amount , after the network's forward propagation, the network outputs the correction With the initial displacement field data The dimensions are the same, Each element in represents the displacement correction value of the corresponding position; Steps 4-6: Calculate the corrected displacement field : .

10. A DIC displacement field dynamic correction system based on a lightweight convolutional network, characterized in that: include: Image acquisition and processing module: It consists of a high-speed camera, image processing software and its control circuit, and is used to acquire image sequences of the surface of the structure and pre-process the acquired image sequences; Initial displacement field calculation and feature extraction module: The initial displacement field is calculated from the preprocessed image sequence using the traditional DIC algorithm, and the key feature information is extracted using the feature extraction algorithm library as input data for dynamic correction; Neural network correction module: It contains a trained lightweight convolutional neural network model, receives input data from the initial displacement field calculation and feature extraction module, uses the initial displacement field containing noise interference and the corresponding real displacement field as the training data set, transmits the extracted input data to the lightweight convolutional neural network model for recognition, and trains by designing a hybrid loss function to learn the mapping relationship between the initial displacement field and the real displacement field. The initial displacement field is corrected by the trained lightweight convolutional neural network model, and the corrected high-precision displacement field data is directly output.

Images