Non-contact deformation detection method based on image processing

Abstract

The present invention discloses a non-contact deformation detection method based on image processing, and the method includes image acquisition, image preprocessing, image segmentation, establishment of a deformation detection model and non-contact deformation detection. The present invention belongs to the field of deformation detection technology, and specifically refers to a non-contact deformation detection method based on image processing. The scheme adjusts the penalties for different categories by designing an adaptive loss function, and dynamically updates the adaptive parameters, thereby improving the performance of the segmentation model on different categories, making the model more flexible when processing diversified deformation areas, thereby improving the accuracy of subsequent deformation detection results; by decomposing the data matrix into a low-rank matrix and a sparse matrix, the normal deformation mode and the abnormal deformation area are effectively distinguished; when constructing the objective function, the historical consistency term is introduced, which helps to maintain the coherence of the deformation mode and ensure the detection stability, thereby making the deformation detection model flexible and ensuring that the deformation detection results have higher accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of deformation detection, and in particular to a non-contact deformation detection method based on image processing. Background Art

[0002] The non-contact deformation detection method is a technology that detects the deformation of the object without direct contact. Compared with the traditional contact measurement method, this method obtains the deformation detection results of the object through remote or indirect means. However, the general deformation detection method cannot accurately identify the deformation area, the category imbalance problem does not receive reasonable attention, and it cannot handle the diverse deformation areas, which leads to poor accuracy of the deformation detection results. The general deformation detection method has serious noise interference, which makes it impossible to extract the key features in the deformation image, and the poor model stability makes it unable to adapt to the dynamic monitoring scene in non-contact deformation detection. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a non-contact deformation detection method based on image processing. In view of the fact that general deformation detection methods cannot accurately identify deformation areas, the category imbalance problem does not receive reasonable attention, and the diversified deformation areas cannot be processed, which leads to the problem that the accuracy of deformation detection results is poor, this scheme adjusts the penalties for different categories by designing an adaptive loss function, dynamically updates the adaptive parameters, improves the performance of the segmentation model in different categories, and makes the model more flexible in processing diversified deformation areas; through repeated iterations and parameter updates, it has better generalization ability when segmenting deformed images, thereby improving the accuracy of subsequent deformation detection results; in view of the fact that general deformation detection methods have serious noise interference that makes it impossible to extract key features in deformed images, and the poor model stability makes it impossible to adapt to dynamic monitoring scenarios in non-contact deformation detection, this scheme effectively distinguishes normal deformation patterns from abnormal deformation areas by decomposing the data matrix into a low-rank matrix and a sparse matrix; when constructing the objective function, the historical consistency term is introduced, which helps to maintain the coherence of the deformation pattern and ensure the detection stability, thereby making the deformation detection model flexible and ensuring that the deformation detection results have higher accuracy.

[0004] The technical solution adopted by the present invention is as follows: The non-contact deformation detection method based on image processing provided by the present invention comprises the following steps:

[0005] Step S1: image acquisition;

[0006] Step S2: image preprocessing;

[0007] Step S3: image segmentation;

[0008] Step S4: establishing a deformation detection model;

[0009] Step S5: non-contact deformation detection.

[0010] Further, in step S1, the image acquisition is to acquire a historical machine tool deformation image set; the historical machine tool deformation image set includes state images before and after the target is deformed; the image is annotated, and the annotated labels are deformation area and deformation assessment result; the deformation area is used for image segmentation task; the deformation assessment result is used for deformation detection task; the deformation assessment result includes no deformation and deformation assessment level; the higher the deformation assessment level, the more serious the deformation.

[0011] Furthermore, in step S2, the image preprocessing is to rotate, scale, translate and add noise to the collected image; and normalize the pixels; obtain a pre-deformed image set; and divide it into a pre-test set and a pre-training set.

[0012] Furthermore, in step S3, the image segmentation specifically includes the following steps:

[0013] Step S31: Model architecture design: select the U-Net model as the basic model architecture, extract multi-scale features from the input pre-training set, and generate segmentation predictions to obtain predicted deformation areas;

[0014] Step S32: define an adaptive loss function; use the designed adaptive loss function to handle the class imbalance problem in the segmentation task and accurately locate the deformation area; the formula used is as follows:

[0015] ;

[0016] Where aeloss is the adaptive loss function; C is the total number of categories in the image segmentation task, and i1 is the category index; is the angle between the prediction vector of the i1th category and the actual label vector; k i1 is the moderating factor for each category;

[0017] Step S33: Model training; specifically including:

[0018] Step S331: forward propagation: input image X to generate prediction of deformation area through the model ; means as follows:

[0019] ;

[0020] In the formula, is the deformation region label predicted by the model; Y is the actual deformation region label; W and b are the model weight and bias respectively; is the model output;

[0021] Step S332: Calculate the loss; use the adaptive loss function to calculate the prediction error; if the loss converges or reaches the maximum number of iterations, the image segmentation model training is completed, and go to step S36; otherwise, go to step S34; the calculated loss is expressed as follows:

[0022] ;

[0023] In the formula, loss is the calculated loss value;

[0024] Step S34: parameter update; expressed as follows:

[0025] ;

[0026] Where W t1+1 and W t1 are the model parameters at the t1+1th iteration and the t1th iteration respectively; α1 is the learning rate; T1 is the maximum number of iterations; is a smoothing term;

[0027] Step S35: Update the adaptive parameters; evaluate the Dice similarity coefficient of each category on the pre-test set, and update the adaptive parameters based on this; and go to step S33; the updated adaptive parameters are expressed as follows:

[0028] ;

[0029] In the formula, is the adaptive parameter of the i1th category after update; is the parameter used to determine the upper limit; is the prediction accuracy of the i1th category at the t1th iteration; is the Dice similarity coefficient;

[0030] Step S36: model determination; a precision threshold is pre-set. If the prediction precision of the trained image segmentation model for the test set is higher than the precision threshold, the image segmentation model is established. The pre-deformed image set is segmented based on the established image segmentation model to obtain a deformed image set, which is divided into a test set and a training set. Otherwise, the parameters are adjusted to re-divide the pre-deformed image set for model training.

[0031] Furthermore, in step S4, the establishment of the deformation detection model specifically includes the following steps:

[0032] Step S41: image conversion: converting the images in the deformed image set into a matrix form to form a data matrix, in which each column represents an image;

[0033] Step S42: matrix decomposition; decomposing the data matrix into a low-rank matrix and a sparse matrix, the low-rank matrix represents the normal deformation mode, and the sparse matrix corresponds to the abnormal deformation area; the matrix decomposition is expressed as an optimization problem: ; Use the augmented Lagrange multiplier method to solve and separate the normal and abnormal deformation modes; where L is a low-rank matrix; S is a sparse matrix; rank(·) is the rank of the matrix; is a regularization parameter that controls the trade-off between low-rank terms and sparsity terms; is the L1 norm of the matrix;

[0034] Step S43: construct an objective function; evaluate the degree of match between each image and a predefined deformation pattern, where the deformation pattern is a template corresponding to each deformation evaluation level; construct an objective function for each image to evaluate the degree of match between the image and the deformation pattern; the formula used is as follows:

[0035] ;

[0036] In the formula, is the optimal deformation mode matrix corresponding to the nth image, that is, the most matching deformation mode; is the jth eigenvalue of the ith mode in the nth image; Image n is the image currently being deformed; is the eigenvalue of the jth position of the ith deformation mode; m is the total number of deformation modes, n1 is the total number of eigenvalues; α is the regularization parameter; H(·) is the historical consistency term, specifically the Hamming distance; and are the optimal deformation mode matrices for the t-th and t-1-th times respectively;

[0037] Step S44: model training; randomly generate N1 groups of initial deformation patterns and corresponding eigenvalues ​​based on the training set, use the objective function to evaluate the performance of the current deformation pattern, select the best group as the elite group, statistically analyze the characteristics of the deformation pattern based on the elite group, update the corresponding eigenvalues ​​of the deformation pattern, and generate the next round of deformation patterns; when the maximum number of iterations is reached or the objective function converges, the deformation detection model training is completed;

[0038] Step S45: model determination; a deformation accuracy threshold is pre-set. When the prediction accuracy of the deformation detection model for the test set is higher than the deformation accuracy threshold, the deformation detection model is established; otherwise, the parameters are adjusted and the data set is re-divided to train the deformation detection model.

[0039] Furthermore, in step S5, the non-contact deformation detection is based on the established deformation detection model, and the image to be detected is collected in real time, and is input into the deformation detection model after image segmentation, and the model output is used as the deformation detection result.

[0040] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0041] (1) In view of the fact that general deformation detection methods cannot accurately identify deformation areas, the class imbalance problem does not receive reasonable attention, and they cannot handle diverse deformation areas, which leads to poor accuracy of deformation detection results. This scheme designs an adaptive loss function to adjust the penalties for different categories and dynamically updates the adaptive parameters, thereby improving the performance of the segmentation model on different categories and making the model more flexible in handling diverse deformation areas. Through repeated iterations and parameter updates, it has better generalization ability when segmenting deformed images, thereby improving the accuracy of subsequent deformation detection results.

[0042] (2) In order to address the problems that general deformation detection methods have serious noise interference that makes it impossible to extract key features in deformation images, and poor model stability that makes it impossible to adapt to dynamic monitoring scenarios in non-contact deformation detection, this scheme effectively distinguishes normal deformation patterns from abnormal deformation areas by decomposing the data matrix into low-rank matrices and sparse matrices. When constructing the objective function, the historical consistency term is introduced to help maintain the consistency of the deformation pattern and ensure detection stability, thereby making the deformation detection model more flexible and ensuring that the deformation detection results have higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram of the process of the non-contact deformation detection method based on image processing provided by the present invention;

[0044] Figure 2 is a schematic diagram of the process of step S3;

[0045] Figure 3 It is a schematic diagram of the process of step S4.

[0046] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0048] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0049] Example 1, see Figure 1 The present invention provides a non-contact deformation detection method based on image processing, which comprises the following steps:

[0050] Step S1: image acquisition, collecting a historical machine tool deformation image set;

[0051] Step S2: image preprocessing;

[0052] Step S3: Image segmentation, by designing an adaptive loss function to adjust the penalties for different categories and dynamically updating the adaptive parameters; image segmentation is achieved by repeated iteration and parameter update;

[0053] Step S4: Establish a deformation detection model, decompose the data matrix into a low-rank matrix and a sparse matrix, introduce a historical consistency term to construct an objective function, and then establish the deformation detection model;

[0054] Step S5: non-contact deformation detection, implementing non-contact deformation detection of the image to be detected based on the established deformation detection model.

[0055] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the historical machine tool deformation image set includes state images before and after the target deformation; the image is annotated, and the annotated labels are deformation area and deformation assessment result; the deformation area is used for image segmentation task; the deformation assessment result is used for deformation detection task; the deformation assessment result includes no deformation and deformation assessment level; the higher the deformation assessment level, the more serious the deformation.

[0056] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, image preprocessing is to rotate, scale, translate and add noise to the collected image; and normalize the pixels; obtain a pre-deformed image set; and divide it into a pre-test set and a pre-training set.

[0057] Example 4, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S3, image segmentation specifically includes the following steps:

[0058] Step S31: Model architecture design: select the U-Net model as the basic model architecture, extract multi-scale features from the input pre-training set, and generate segmentation predictions to obtain predicted deformation areas;

[0059] Step S32: define an adaptive loss function; use the designed adaptive loss function to handle the class imbalance problem in the segmentation task and accurately locate the deformation area; the formula used is as follows:

[0060] ;

[0061] Where aeloss is the adaptive loss function; C is the total number of categories in the image segmentation task, and i1 is the category index; is the angle between the prediction vector of the i1th category and the actual label vector; k i1 is the moderating factor for each category;

[0062] Step S33: Model training; specifically including:

[0063] Step S331: forward propagation: input image X to generate prediction of deformation area through the model ; means as follows:

[0064] ;

[0065] In the formula, is the deformation region label predicted by the model; Y is the actual deformation region label; W and b are the model weight and bias respectively; is the model output;

[0066] Step S332: Calculate the loss; use the adaptive loss function to calculate the prediction error; if the loss converges or reaches the maximum number of iterations, the image segmentation model training is completed, and go to step S36; otherwise, go to step S34; the calculated loss is expressed as follows:

[0067] ;

[0068] In the formula, loss is the calculated loss value;

[0069] Step S34: parameter update; expressed as follows:

[0070] ;

[0071] Where W t1+1 and W t1 are the model parameters at the t1+1th iteration and the t1th iteration respectively; α1 is the learning rate; T1 is the maximum number of iterations; is a smoothing term;

[0072] Step S35: Update the adaptive parameters; evaluate the Dice similarity coefficient of each category on the pre-test set, and update the adaptive parameters based on this; and go to step S33; the updated adaptive parameters are expressed as follows:

[0073] ;

[0074] In the formula, is the adaptive parameter of the i1th category after update; is the parameter used to determine the upper limit; is the prediction accuracy of the i1th category at the t1th iteration; is the Dice similarity coefficient;

[0075] Step S36: model determination; a precision threshold is pre-set. If the prediction precision of the trained image segmentation model for the test set is higher than the precision threshold, the image segmentation model is established. The pre-deformed image set is segmented based on the established image segmentation model to obtain a deformed image set, which is divided into a test set and a training set. Otherwise, the parameters are adjusted to re-divide the pre-deformed image set for model training.

[0076] By performing the above operations, the general deformation detection method has the problems of being unable to accurately identify deformation areas, the category imbalance problem is not given reasonable attention, and it is unable to handle diverse deformation areas, which leads to poor accuracy of deformation detection results. This scheme designs an adaptive loss function to adjust the penalties for different categories, dynamically updates the adaptive parameters, improves the performance of the segmentation model on different categories, and makes the model more flexible in handling diverse deformation areas; through repeated iterations and parameter updates, it has better generalization ability when segmenting deformed images, thereby improving the accuracy of subsequent deformation detection results.

[0077] Example 5, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S4, establishing a deformation detection model specifically includes the following steps:

[0078] Step S41: image conversion: converting the images in the deformed image set into a matrix form to form a data matrix, in which each column represents an image;

[0079] Step S42: matrix decomposition; decomposing the data matrix into a low-rank matrix and a sparse matrix, the low-rank matrix represents the normal deformation mode, and the sparse matrix corresponds to the abnormal deformation area; the matrix decomposition is expressed as an optimization problem: ; Use the augmented Lagrange multiplier method to solve and separate the normal and abnormal deformation modes; where L is a low-rank matrix; S is a sparse matrix; rank(·) is the rank of the matrix; is a regularization parameter that controls the trade-off between low-rank terms and sparsity terms; is the L1 norm of the matrix;

[0080] Step S43: construct an objective function; evaluate the degree of match between each image and a predefined deformation pattern, where the deformation pattern is a template corresponding to each deformation evaluation level; construct an objective function for each image to evaluate the degree of match between the image and the deformation pattern; the formula used is as follows:

[0081] ;

[0082] In the formula, is the optimal deformation mode matrix corresponding to the nth image, that is, the most matching deformation mode; is the jth eigenvalue of the ith mode in the nth image; Image n is the image currently being deformed; is the eigenvalue of the jth position of the ith deformation mode; m is the total number of deformation modes, n1 is the total number of eigenvalues; α is the regularization parameter; H(·) is the historical consistency term, specifically the Hamming distance; and are the optimal deformation mode matrices for the t-th and t-1-th times respectively;

[0083] Step S44: model training; randomly generate N1 groups of initial deformation patterns and corresponding eigenvalues ​​based on the training set, use the objective function to evaluate the performance of the current deformation pattern, select the best group as the elite group, statistically analyze the characteristics of the deformation pattern based on the elite group, update the corresponding eigenvalues ​​of the deformation pattern, and generate the next round of deformation patterns; when the maximum number of iterations is reached or the objective function converges, the deformation detection model training is completed;

[0084] Step S45: model determination; a deformation accuracy threshold is pre-set. When the prediction accuracy of the deformation detection model for the test set is higher than the deformation accuracy threshold, the deformation detection model is established; otherwise, the parameters are adjusted and the data set is re-divided to train the deformation detection model.

[0085] By performing the above operations, the general deformation detection method has serious noise interference that makes it impossible to extract key features in the deformation image, and the poor model stability makes it unable to adapt to the dynamic monitoring scenarios in non-contact deformation detection. This scheme decomposes the data matrix into a low-rank matrix and a sparse matrix to effectively distinguish normal deformation patterns from abnormal deformation areas. When constructing the objective function, the historical consistency term is introduced to help maintain the consistency of the deformation pattern and ensure the detection stability, thereby making the deformation detection model more flexible and ensuring that the deformation detection results have higher accuracy.

[0086] Example 6, see Figure 1This embodiment is based on the above embodiment. In step S5, non-contact deformation detection is based on the established deformation detection model, and the image to be detected is collected in real time. After image segmentation, it is input into the deformation detection model, and the model output is used as the deformation detection result.

[0087] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0088] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0089] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A non-contact deformation detection method based on image processing, characterized in that: The method comprises the following steps: Step S1: image acquisition, collecting a historical machine tool deformation image set; Step S2: image preprocessing; Step S3: Image segmentation, by designing an adaptive loss function to adjust the penalties for different categories and dynamically updating the adaptive parameters; image segmentation is achieved by repeated iteration and parameter update; Step S4: Establish a deformation detection model, decompose the data matrix into a low-rank matrix and a sparse matrix, introduce a historical consistency term to construct an objective function, and then establish the deformation detection model; Step S5: non-contact deformation detection, realizing non-contact deformation detection of the image to be detected based on the established deformation detection model; Step S3 includes step S32: defining an adaptive loss function; the formula used is as follows: ； Where aeloss is the adaptive loss function; C is the total number of categories in the image segmentation task, and i1 is the category index; is the angle between the prediction vector of the i1th category and the actual label vector; k i1 is the moderating factor for each category; In step S4, the establishment of the deformation detection model specifically includes the following steps: Step S41: image conversion: converting the images in the deformed image set into a matrix form to form a data matrix, in which each column represents an image; Step S42: matrix decomposition; decomposing the data matrix into a low-rank matrix and a sparse matrix, the low-rank matrix represents the normal deformation mode, and the sparse matrix corresponds to the abnormal deformation area; the matrix decomposition is expressed as an optimization problem: ; Use the augmented Lagrange multiplier method to solve and separate the normal and abnormal deformation modes; where L is a low-rank matrix; S is a sparse matrix; rank(·) is the rank of the matrix; is a regularization parameter that controls the trade-off between low-rank terms and sparsity terms; is the L1 norm of the matrix; Step S43: construct an objective function; evaluate the degree of match between each image and a predefined deformation pattern, where the deformation pattern is a template corresponding to each deformation evaluation level; construct an objective function for each image to evaluate the degree of match between the image and the deformation pattern; the formula used is as follows: ； In the formula, is the optimal deformation mode matrix corresponding to the nth image, that is, the most matching deformation mode; is the jth eigenvalue of the ith mode in the nth image; Image n is the image currently being deformed; is the eigenvalue of the jth position of the ith deformation mode; m is the total number of deformation modes, n1 is the total number of eigenvalues; α is the regularization parameter; H(·) is the historical consistency term, specifically the Hamming distance; and are the optimal deformation mode matrices for the t-th and t-1-th times respectively; Step S44: model training; randomly generate N1 groups of initial deformation patterns and corresponding eigenvalues ​​based on the training set, use the objective function to evaluate the performance of the current deformation pattern, select the best group as the elite group, statistically analyze the characteristics of the deformation pattern based on the elite group, update the corresponding eigenvalues ​​of the deformation pattern, and generate the next round of deformation patterns; when the maximum number of iterations is reached or the objective function converges, the deformation detection model training is completed; Step S45: model determination; a deformation accuracy threshold is pre-set. When the prediction accuracy of the deformation detection model for the test set is higher than the deformation accuracy threshold, the deformation detection model is established; otherwise, the parameters are adjusted and the data set is re-divided to train the deformation detection model.

2. The non-contact deformation detection method based on image processing according to claim 1, characterized in that: In step S3, the image segmentation specifically includes the following steps: Step S31: Model architecture design: select the U-Net model as the basic model architecture, extract multi-scale features from the input pre-training set, and generate segmentation predictions to obtain predicted deformation areas; Step S32: defining an adaptive loss function; Step S33: Model training; specifically including: Step S331: forward propagation: input image X to generate prediction of deformation area through the model ; means as follows: ； In the formula, is the deformation region label predicted by the model; Y is the actual deformation region label; W and b are the model weight and bias respectively; is the model output; Step S332: Calculate the loss; use the adaptive loss function to calculate the prediction error; if the loss converges or reaches the maximum number of iterations, the image segmentation model training is completed, and go to step S36; otherwise, go to step S34; the calculated loss is expressed as follows: ； In the formula, loss is the calculated loss value; Step S34: parameter update; expressed as follows: ； Where W t1+1 and W t1 are the model parameters at the t1+1th iteration and the t1th iteration respectively; α1 is the learning rate; T1 is the maximum number of iterations; is a smoothing term; Step S35: updating adaptive parameters; Step S36: model determination; a precision threshold is pre-set. If the prediction precision of the trained image segmentation model for the test set is higher than the precision threshold, the image segmentation model is established. The pre-deformed image set is segmented based on the established image segmentation model to obtain a deformed image set, which is divided into a test set and a training set. Otherwise, the parameters are adjusted to re-divide the pre-deformed image set for model training.

3. The non-contact deformation detection method based on image processing according to claim 2, characterized in that: In step S35, the updating of the adaptive parameters is specifically to evaluate the Dice similarity coefficient of each category on the pre-test set, and update the adaptive parameters based on this; and then go to step S33; the updating of the adaptive parameters is expressed as follows: ； In the formula, is the adaptive parameter of the i1th category after update; is the parameter used to determine the upper limit; is the prediction accuracy of the i1th category at the t1th iteration; is the Dice similarity coefficient.

4. The non-contact deformation detection method based on image processing according to claim 1, characterized in that: In step S1, the historical machine tool deformation image set includes state images before and after the target deformation; Annotate the image, with the annotated labels being the deformation area and the deformation assessment result; Deformed regions are used for image segmentation tasks; The deformation assessment result is used for deformation detection tasks; the deformation assessment result includes undeformed and deformation assessment levels; The higher the deformation assessment level, the more severe the deformation.

5. The non-contact deformation detection method based on image processing according to claim 1, characterized in that: In step S5, the non-contact deformation detection is based on the established deformation detection model, and the image to be detected is collected in real time. After image segmentation, the image is input into the deformation detection model, and the model output is used as the deformation detection result.

6. The non-contact deformation detection method based on image processing according to claim 1, characterized in that: In step S2, the image preprocessing is to rotate, scale, translate and add noise to the collected image; and normalize the pixels; Obtain a pre-deformed image set; And divided into pre-test set and pre-training set.

Images