Antique defect identification method and system based on multispectral imaging analysis

Abstract

The invention discloses an antique defect identification method and system based on multispectral imaging analysis, and relates to the technical field of defect detection, and the method comprises the steps: carrying out the adjacent overlapping imaging of M divided regions of a target antique through a multi-component lens, and obtaining M region multispectral images which comprise overlapping regions; carrying out splicing path iteration loss analysis on the overlapping domain, determining a splicing path, completing multispectral image splicing, and obtaining a multispectral image of the target antique; and performing cross-scene defect identification on the image by using a domain shift anti-interference identifier to obtain an antique defect identification result. The technical problem that in the prior art, when a large-area antique image is spliced, multi-spectral imaging cannot effectively cope with illumination changes, shooting angle deviation and image splicing errors, and consequently the recognition precision is low is solved, and the purposes that through precise image splicing and cross-scene domain shifting anti-interference recognition, the recognition precision is high, and the recognition accuracy is high are achieved. And the accuracy and robustness of antique defect identification are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to an antique defect recognition method and system based on multi-spectral imaging analysis. Background Art

[0002] With the advancement of cultural relics protection technology, traditional manual inspection and visible light imaging methods can no longer meet the needs of accurate identification of tiny defects on the surface of antiques. Multispectral imaging technology can provide rich image information and reveal defects that are invisible to the naked eye, but it is still a challenge in image stitching and lighting changes in complex environments, noise interference, etc. Existing stitching technology is prone to errors in different environments, affecting image quality and recognition accuracy. Summary of the invention

[0003] The present application provides a method and system for identifying antique defects using multispectral imaging analysis, which is used to solve the technical problem in the prior art that multispectral imaging cannot effectively cope with illumination changes, shooting angle deviations and image stitching errors when stitching large-area antique images, resulting in low recognition accuracy.

[0004] In a first aspect of the present application, a method for identifying antique defects by multispectral imaging analysis is provided, the method comprising: performing neighbor overlapping imaging on M divided areas of a target antique through a multi-component lens to obtain M regional multispectral images, wherein M is a positive integer, and the M regional multispectral images include M regional multispectral image overlapping domains; performing iterative loss analysis on the M regional multispectral image overlapping domains based on the regional positions of the M divided areas to determine M stitching paths; stitching the M regional multispectral images based on the M stitching paths to obtain a target antique multispectral image; and performing cross-scene defect recognition on the target antique multispectral image using a domain shift anti-interference identifier to obtain a target antique defect recognition result.

[0005] According to a second aspect of the present application, a system for identifying antique defects by multispectral imaging analysis is provided, the system comprising: a nearest neighbor overlapping imaging module, the nearest neighbor overlapping imaging module being used to perform nearest neighbor overlapping imaging on M divided areas of a target antique through a multi-component lens, to obtain M regional multispectral images, wherein M is a positive integer, and the M regional multispectral images include M regional multispectral image overlapping domains; a stitching path iteration module, the stitching path iteration module being used to perform stitching path iteration loss analysis on the M regional multispectral image overlapping domains based on the regional positions of the M divided areas, to determine M stitching paths; an image stitching module, the image stitching module being used to stitch the M regional multispectral images based on the M stitching paths, to obtain a target antique multispectral image; and a cross-scene defect recognition module, the cross-scene defect recognition module being used to perform cross-scene defect recognition on the target antique multispectral image using a domain shift anti-interference identifier, to obtain a target antique defect recognition result.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The present application provides a method and system for identifying antique defects by multispectral imaging analysis, which relates to the field of defect detection technology. M areas of the antique are imaged in an overlapping manner through a multi-component lens to obtain a multispectral image and an overlapping domain. The overlapping domain is analyzed for a stitching path according to the area position to determine the best stitching path, complete the image stitching, and use a domain shift anti-interference identifier to perform cross-scene defect recognition to obtain an accurate defect recognition result. This solves the technical problem in the prior art that multispectral imaging cannot effectively cope with illumination changes, shooting angle deviations and image stitching errors when stitching large-area antique images, resulting in low recognition accuracy. This achieves the technical effect of improving the accuracy and robustness of antique defect recognition through precise image stitching and cross-scene domain shift anti-interference recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A schematic flow chart of a method for identifying antique defects by multispectral imaging analysis provided in an embodiment of the present application;

[0010] Figure 2 A schematic diagram of the structure of an antique defect identification system using multispectral imaging analysis provided in an embodiment of the present application.

[0011] Explanation of the reference numerals: neighbor overlapping imaging module 11, stitching path iteration module 12, image stitching module 13, cross-scene defect recognition module 14. DETAILED DESCRIPTION

[0012] The present application provides a method and system for identifying antique defects using multispectral imaging analysis, which is used to solve the technical problem in the prior art that multispectral imaging cannot effectively cope with illumination changes, shooting angle deviations and image stitching errors when stitching large-area antique images, resulting in low recognition accuracy.

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0014] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices.

[0015] Embodiment 1, as Figure 1 As shown, the present application provides a method for identifying antique defects by multispectral imaging analysis, the method comprising:

[0016] P10: Perform neighbor overlapping imaging on the M divided areas of the target antique through a multi-component lens to obtain M regional multispectral images, where M is a positive integer and the M regional multispectral images include M regional multispectral image overlapping domains.

[0017] Specifically, first of all, for the target antiques, this application specifically refers to calligraphy and painting antiques, and uses multi-component lens imaging technology to perform high-resolution multispectral imaging on its surface. Calligraphy and painting antiques usually have large areas and complex patterns and colors, so it is impossible to obtain complete multispectral data through a single imaging. In order to overcome this limitation, the present invention divides the imaging area of ​​the target antique into M sub-areas, and the selection of these areas is based on the actual size and detail requirements of the calligraphy and painting, where M is a positive integer. This division method makes the area of ​​each sub-area small enough to fully obtain its multispectral information in a single imaging process.

[0018] For each divided area, multi-component lenses are used for neighbor overlapping imaging. Split-lens imaging is a technology that uses multiple lenses to shoot the same target from different angles or positions, which can effectively expand the imaging range. In this step, there is a certain overlap between the images taken by each split lens. The existence of this overlapping area is to ensure that the images can be accurately aligned during subsequent stitching to avoid gaps or faults in the imaging process. The imaging data of each area forms a multispectral image, which captures the spectral information of the area in different bands, rather than just the red, green, and blue images visible to the naked eye in traditional imaging technology. The existence of overlapping areas not only provides the necessary spatial reference for image stitching, but can also be used to correct possible geometric distortion and spectral differences during the stitching process.

[0019] Multispectral imaging technology is one of the core technologies of the present invention. It obtains the characteristics of the surface of an object by capturing spectral information in different wavelength ranges. Compared with traditional RGB imaging, multispectral imaging can provide richer spectral information, thereby more accurately reflecting the pigment composition and structural characteristics of antiques such as calligraphy and painting. Multispectral images contain spectral data of multiple bands, each of which corresponds to a specific spectral range, such as the visible light band (400-700nm) and the near-infrared band (700-1000nm). By analyzing the spectral data of these different bands, accurate classification and identification of calligraphy and painting pigments can be achieved.

[0020] After completing the multispectral imaging of each divided area, M regional multispectral images are obtained, which not only contain the spectral information of their respective areas, but are also related to each other through overlapping areas. The existence of overlapping domains provides a basis for subsequent image stitching and data fusion, making it possible to reconstruct a complete multispectral image of antiques from multiple local images.

[0021] Through the use of multi-component lenses, the M divided areas of calligraphy and painting antiques are imaged in overlapping neighboring areas. This not only solves the problem that a single imaging cannot cover a large area of ​​antiques, but also uses multispectral imaging technology to obtain rich spectral information, providing a solid data foundation for subsequent image stitching and pigment classification.

[0022] P20: Based on the regional positions of the M divided regions, an iterative loss analysis of the stitching path is performed on the overlapping domains of the M regional multispectral images to determine M stitching paths.

[0023] Furthermore, step P20 of the embodiment of the present application also includes:

[0024] P21: Extract the first regional multispectral image overlapping domain from the M regional multispectral image overlapping domains, and match the second regional multispectral image overlapping domain overlapping with the first regional multispectral image overlapping domain in combination with the regional positions of the M divided regions; P22: Perform initial feature point recognition on the first regional multispectral image overlapping domain and the second regional multispectral image overlapping domain, perform iterative loss analysis on the stitching path based on the initial feature points, and determine the first stitching path; P23: Perform iterative loss analysis on the stitching path of the M regional multispectral image overlapping domains respectively to determine the M stitching paths.

[0025] It should be understood that based on the regional positions of the M divided areas, the overlapping domain of each area is subjected to iterative loss analysis of the stitching path to ensure high precision and seamless connection during the stitching process. In this step, by analyzing the positional relationship of the M areas, the overlapping part between each area and the adjacent area is clarified. The overlapping area is a key part of the stitching and can provide a basis for alignment for subsequent image stitching. When analyzing these overlapping areas, the iterative loss analysis method can be used to gradually optimize the stitching path, reduce the spectral shift and image distortion caused by the stitching error, and finally determine the M stitching paths.

[0026] First, an overlapping area is randomly extracted from the M regional multispectral images as a reference, namely the first regional multispectral image overlapping domain. When selecting this area, its position and importance in the overall image are usually considered. Subsequently, based on the regional position information of the M divided areas, the second regional multispectral image overlapping domain overlapping with the first regional multispectral image overlapping domain is found. In other words, the first regional multispectral image overlapping domain and the second regional multispectral image overlapping domain are the overlapping areas of the two spectral images at the same coordinate position, wherein the regional position information refers to the coordinate position of each divided area in the overall image of the antique.

[0027] After determining the overlapping domain of the multispectral images of the first and second regions, it is necessary to identify the initial feature points of the two overlapping domains. Initial feature points refer to pixels with significant features in the overlapping region, such as corner points, edge points, etc. These feature points can be used as reference points for image alignment. By identifying the initial feature points, the relative position relationship between the two overlapping domains can be preliminarily determined. In other words, this step processes the overlapping region through feature point recognition technology, usually using algorithms such as SIFT (Scale Invariant Feature Transform) to identify stable feature points in the image. These feature points are significantly representative in spatial and spectral dimensions and can provide a stable reference for image stitching.

[0028] After identifying the initial feature points, an iterative loss analysis of the stitching path is performed based on these feature points. Iterative loss analysis is an optimization algorithm that iteratively calculates the loss values ​​under different stitching paths and selects the path with the smallest loss value as the optimal stitching path. The loss value usually includes geometric distortion error, spectral difference error, etc. These errors reflect the quality of the stitched image. Through iterative optimization, the first stitching path, that is, the optimal stitching path for the multispectral images of the first and second regions, can be determined. As this process progresses, the stitching path will become more and more accurate, reducing the impact caused by spectral offset or uneven illumination.

[0029] Ultimately, all the overlapping domains of the multispectral images of the M regions will undergo a similar iterative loss analysis of the stitching path. This process is carried out step by step, each time an overlapping domain is selected as a reference, another overlapping domain that overlaps with it is found, and then the initial feature point recognition and stitching path iterative loss analysis are performed. In this way, the stitching paths between all adjacent images are gradually determined, and finally a complete M stitching paths are formed. Through this process, seamless stitching of images can be achieved, so that the final antique image can achieve high accuracy in both spatial and spectral dimensions. In this way, the final stitched image can not only be seamlessly connected visually, but also provide more accurate image data support for subsequent antique defect identification, ensuring the accuracy of the identification results.

[0030] Furthermore, step P22 of the embodiment of the present application also includes:

[0031] P22-1: Use SIFT feature extraction method to identify initial feature points in the overlapping domain of the first-region multispectral image and the overlapping domain of the second-region multispectral image to obtain initial feature points; P22-2: Based on the position of the initial feature points in the overlapping domain of the first-region multispectral image, construct neighborhoods in the overlapping domain of the first-region multispectral image and the overlapping domain of the second-region multispectral image to obtain the neighborhood of initial feature points; P22-3: Perform iterative loss analysis of the stitching path in the neighborhood of the initial feature points to obtain the first stitching path. The neighborhood of the initial feature points includes 8 pixel points in the overlapping domain of the first-region multispectral image and the overlapping domain of the second-region multispectral image that are adjacent to the initial feature points.

[0032] Optionally, the SIFT feature extraction method is used to perform initial feature point recognition, neighborhood construction, and iterative loss analysis of the stitching path on the overlapping domain of the multispectral images of the first area and the second area to obtain an accurate stitching path.

[0033] First, the scale-invariant feature transform (SIFT) algorithm is used to identify the initial feature points of the overlapping domain of the multispectral image in the first region and the overlapping domain of the multispectral image in the second region. The SIFT algorithm is a feature extraction method widely used in image matching and recognition neighborhoods. Its core advantage lies in its good invariance to image scale, rotation, and illumination changes. In this application, the SIFT algorithm identifies stable feature points (such as corners and edges) in the image as the key basis for alignment in the stitching process. These feature points have high recognition under different viewing angles and illumination conditions. Through the SIFT feature extraction method, representative initial feature points are extracted from two adjacent regions respectively. These points have significant features in the spatial and spectral dimensions of the image.

[0034] Next, based on the position of the initial feature point in the overlapping domain of the multispectral image of the first region, the neighborhood of the initial feature point is constructed in the overlapping domain of the multispectral image of the first region and the second region respectively. The construction of the neighborhood is to perform accurate analysis and optimization in a smaller range. Specifically, the neighborhood of the initial feature point includes 8 pixel points adjacent to the initial feature point in the overlapping domain of the first region and the second region, forming a 3×3 pixel block centered on the initial feature point. This neighborhood construction method can fully consider the local information around the initial feature point and provide a more accurate reference for the subsequent splicing path analysis.

[0035] After obtaining the neighborhood of the initial feature points, an iterative loss analysis of the stitching path is performed. The core of this process is to gradually optimize the stitching path by analyzing the errors between images in the neighborhood of the initial feature points. The iterative loss analysis compares the performance of different stitching paths in the overlapping domain to find the path that minimizes the error to ensure the accuracy of image stitching. Exemplarily, multiple initial feature points are traversed and stitched, and the optimal stitching path is determined by calculating the loss values ​​under different stitching paths. The loss value usually includes geometric distortion error, spectral difference error, etc., which reflects the quality of the stitched image. The optimal stitching path can be determined by calculating the comprehensive error value and iterative optimization in the neighborhood of the initial feature points. Through this process, the first stitching path is obtained, which can align the images of the first area and the second area most accurately in spectrum and space, ensure seamless connection of the stitched images, and avoid stitching errors caused by spectral offset or uneven illumination.

[0036] The entire process ensures high precision and consistency of image stitching through the extraction of initial feature points, precise construction of the neighborhood, and iterative optimization, providing a reliable image basis for subsequent antique defect identification.

[0037] Furthermore, step P22-3 of the embodiment of the present application also includes:

[0038] P22-31: randomly generate a stitching path in the neighborhood of the initial feature point to obtain a first initial stitching path; P22-32: divide the neighborhood of the initial feature point using the first stitching path to obtain a first initial feature point neighborhood sub-neighborhood and a second initial feature point sub-neighborhood; P22-33: use a gradient difference recognition formula to identify the gradient difference of the initial feature point, the first initial feature point sub-neighborhood and the second initial feature point sub-neighborhood to obtain a first initial stitching path gradient difference; P22-34: randomly generate a second initial stitching path from the neighborhood of the initial feature point again; P22-35: determine whether the gradient difference of the second initial splicing path is greater than or equal to the gradient difference of the first initial splicing path, and if so, use the second initial splicing path as the first stage splicing path; P22-36: perform multiple iterations on the first stage splicing path in the neighborhood of the initial feature point until the preset iteration stop condition is met, stop the iteration, and use the stage splicing path corresponding to the minimum value of the stage splicing path gradient difference during the iteration as the first splicing path. The preset iteration stop condition is that the number of iterations is greater than or equal to the preset number of iterations or the difference between the stage splicing path gradient differences of two adjacent iterations is less than or equal to the preset difference.

[0039] Wherein, the gradient difference identification formula is: ;in, is the gradient difference of the first initial splicing path, is the pixel value of the i-th pixel in the sub-neighborhood of the first initial feature point, is the pixel value of the jth pixel in the sub-neighborhood of the second initial feature point, is the pixel value of the initial feature point, n+m=16, where n and m are the number of pixels in the first area and the second area respectively.

[0040] In a possible embodiment of the present application, the process of iterative loss analysis of the splicing path in the neighborhood of the initial feature point is further elaborated in detail to ensure that the best splicing path is obtained.

[0041] First, a preliminary hypothesis of the stitching path is generated from the neighborhood of the initial feature point, and the first initial stitching path is obtained by randomly generating stitching paths. This path is a stitching path randomly selected from the overlapping domain of the image starting from the initial feature point. In order to perform this operation, a range needs to be defined first: a neighborhood area is set based on the position of the initial feature point. Then, a path is randomly selected in this area as the starting path. This path may not necessarily be the optimal path, but it provides a starting point for subsequent path optimization.

[0042] Next, the initial feature point neighborhood is further divided using the first initial stitching path to obtain two sub-neighborhoods: the first initial feature point sub-neighborhood and the second initial feature point sub-neighborhood. Specifically, the first initial stitching path divides the neighborhood into two regions, one of which contains pixels in the overlapping domain of the multispectral image of the first region, and the other contains pixels in the overlapping domain of the multispectral image of the second region. For each neighborhood region, 16 pixels around the initial feature point are first selected, including 8 pixels adjacent to the initial feature point in the overlapping domain of the first region, and 8 pixels adjacent to the initial feature point in the overlapping domain of the second region. This division step can effectively divide the pixels of the stitching area into two sub-regions, ensuring that the matching relationship between each region can be accurately processed in the subsequent stitching path analysis.

[0043] After obtaining the initial feature point, the first initial feature point sub-neighborhood, and the second initial feature point sub-neighborhood, the path optimization is performed using the above gradient difference identification formula. This formula measures the spectral difference of the splicing path by calculating the brightness difference between the initial feature point and the adjacent pixel points. Through this process, the first initial splicing path gradient difference can be obtained and used to evaluate the quality of the current path.

[0044] After the gradient difference of the first initial stitching path is identified, the second initial stitching path is randomly generated from the initial feature point neighborhood again. The generation process is similar to the first initial stitching path, but the selection method and overlapping area selection when the path is generated may be different. At this time, the gradient difference of the second initial stitching path also needs to be identified to calculate the gradient difference of the second initial stitching path.

[0045] After obtaining the gradient difference of the second initial splicing path, the gradient difference of the second initial splicing path is compared with the gradient difference of the first initial splicing path. If the gradient difference of the second path is less than or equal to the gradient difference of the first path, the second path is selected as the first stage splicing path. At this time, it is considered that the second path is more superior in splicing quality, so it is used as the splicing path of the current stage.

[0046] Finally, after the first-stage splicing path is determined, multiple iterations are performed until the preset stop condition is met, such as the number of iterations reaches the preset maximum number of iterations or the difference between the gradient differences of the splicing paths of two adjacent iterations is less than or equal to the preset threshold. During each iteration, the splicing path will be fine-tuned according to the results of the previous iteration, thereby gradually reducing the gradient difference and improving the splicing accuracy.

[0047] Through this process, the optimal stitching path is obtained, that is, the first stitching path obtained under the condition of minimizing the gradient difference. This optimization process ensures the high precision of image stitching, making the final image more perfectly connected in spatial and spectral dimensions, greatly improving the accuracy of the stitching results, and providing high-quality image data support for subsequent defect identification and analysis.

[0048] P30: stitching the M regional multispectral images based on the M stitching paths to obtain a target antique multispectral image.

[0049] Specifically, based on the M stitching paths obtained above, the multispectral images of the M regions are stitched together to obtain a complete multispectral image of the target antique.

[0050] First, according to the determined M stitching paths, the multispectral images of each pair of adjacent areas are stitched one by one according to the optimized stitching path. The stitching process first aligns the overlapping domains of each area according to the spatial position relationship of the stitching path to ensure that each pixel point can be accurately connected between its corresponding areas.

[0051] In the specific implementation, we first start from the first stitching path and stitch the images of the first area and the second area. Then, based on the next stitching path, stitch the third area with the first two areas in turn until all the images of the M areas are successfully stitched together. Each time the stitching is performed, the spectral information of the overlapping areas will be combined to ensure the consistency and continuity of the spectral information during the entire stitching process.

[0052] In this way, a complete multispectral image of the target antique is finally obtained, which combines the high-resolution, multi-band spectral information of all M regions. This multispectral image can not only present rich details on the surface of the antique, but also effectively reveal tiny defects and aging marks that are invisible to the naked eye, providing accurate image data support for subsequent defect detection and analysis.

[0053] The entire stitching process depends on the accuracy and optimization of each stitching path. Therefore, the fine adjustment and optimization of the stitching path in the early stage play a decisive role in the final result. Through this high-precision stitching technology, high-quality multispectral image data can be obtained in complex environments, especially when facing large-area target antiques.

[0054] P40: Use a domain shift anti-interference identifier to perform cross-scene defect recognition on the target antique multispectral image to obtain a target antique defect recognition result.

[0055] Furthermore, step P40 of the embodiment of the present application also includes:

[0056] P41: Acquire multiple standard antique defect recognition results and corresponding multiple standard antique spectral images; P42: Collect the current imaging environment according to the preset imaging environment indicator set to obtain the current imaging environment indicator set, and interfere with the multiple standard antique spectral images based on the current imaging environment indicator set to obtain multiple interfered antique spectral images; P43: Construct the domain shift anti-interference identifier based on the multiple interfered antique spectral images and multiple standard antique defect recognition results. Among them, the preset imaging environment indicator set includes illumination uniformity, illumination intensity, shooting angle, field of view and resolution.

[0057] It should be understood that the domain shift anti-interference identifier is used to perform cross-scene defect recognition on the target antique multispectral image. Since the spectral information and recognition results of the standard image may change under different imaging environments, traditional defect recognition methods are often unable to adapt to these changes, so a domain shift anti-interference identifier that can work stably under variable environments is needed.

[0058] First, obtain multiple standard antique defect recognition results and corresponding standard antique spectral images. These images are collected under specific imaging conditions and have been accurately annotated to reflect the true situation of antique surface defects. These standard data will serve as the basis for training the recognizer to ensure that the recognizer can accurately identify defects under ideal conditions.

[0059] Next, the current imaging environment is collected according to the preset imaging environment indicator set to obtain the current imaging environment indicator set. The preset imaging environment indicator set includes key parameters such as illumination uniformity, illumination intensity, shooting angle, field of view and resolution, which will affect the imaging quality and the recognizability of defects in the image. Based on the current imaging environment indicator set, multiple standard antique spectral images are interfered with to simulate the image changes that may occur in the current imaging environment, thereby obtaining multiple interfered antique spectral images. This step is equivalent to obtaining samples that meet the current imaging environment for training and testing the identifier.

[0060] In this way, spectral images under different environmental conditions are simulated, and then a domain shift anti-interference recognizer is constructed based on these interfering antique spectral images and standard antique defect recognition results. During this training process, the recognizer adapts to defect recognition tasks under different imaging environments by learning the differences between standard images and interfering images, as well as their corresponding defect recognition results. In this way, the recognizer can maintain high recognition accuracy and robustness in the face of changes in the actual imaging environment.

[0061] Through the above steps, the embodiment of the present application can accurately identify the defects of antiques in a variable imaging environment, providing important technical support for the protection and restoration of antiques. This method not only improves the accuracy of defect identification, but also enhances the adaptability and generalization ability of the recognition system.

[0062] In summary, the embodiments of the present application have at least the following technical effects:

[0063] This application uses a multi-component lens to overlap M areas of the antique to obtain a multispectral image and overlapping domain. Then, the overlapping domain is spliced ​​according to the regional position, the optimal splicing path is determined, the image is spliced, and a complete multispectral image of the antique is formed. Finally, a domain shift anti-interference identifier is used to perform cross-scene defect recognition to obtain accurate defect recognition results.

[0064] The technical effect of improving the accuracy and robustness of antique defect identification has been achieved through precise image stitching and cross-scene domain shift anti-interference recognition.

[0065] Embodiment 2, based on the same inventive concept as the method for identifying antique defects by multispectral imaging analysis in the above embodiment, Figure 2 As shown, the present application provides an antique defect identification system based on multispectral imaging analysis, and the system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0066] The neighbor overlapping imaging module 11 is used to perform neighbor overlapping imaging on M divided areas of the target antique through a multi-component lens to obtain M regional multispectral images, wherein M is a positive integer, and the M regional multispectral images include M regional multispectral image overlapping domains.

[0067] The stitching path iteration module 12 is used to perform stitching path iteration loss analysis on the overlapping domains of the M regional multispectral images based on the regional positions of the M divided regions, and determine M stitching paths.

[0068] The image stitching module 13 is used to stitch the M regional multispectral images based on the M stitching paths to obtain a target antique multispectral image.

[0069] The cross-scene defect recognition module 14 is used to perform cross-scene defect recognition on the target antique multispectral image by using a domain shift anti-interference identifier to obtain a target antique defect recognition result.

[0070] Furthermore, the splicing path iteration module 12 is also used to perform the following steps:

[0071] A first regional multispectral image overlapping domain is extracted from the M regional multispectral image overlapping domains, and a second regional multispectral image overlapping domain overlapping with the first regional multispectral image overlapping domain is matched in combination with the regional positions of the M divided regions; initial feature points are identified for the first regional multispectral image overlapping domain and the second regional multispectral image overlapping domain, and a stitching path iterative loss analysis is performed based on the initial feature points to determine a first stitching path; stitching path iterative loss analysis is performed on the M regional multispectral image overlapping domains respectively to determine the M stitching paths.

[0072] Furthermore, the splicing path iteration module 12 is also used to perform the following steps:

[0073] The SIFT feature extraction method is used to identify initial feature points in the overlapping domain of the multispectral images of the first region and the overlapping domain of the multispectral images of the second region to obtain initial feature points; based on the position of the initial feature points in the overlapping domain of the multispectral images of the first region, a neighborhood is constructed in the overlapping domain of the multispectral images of the first region and the overlapping domain of the multispectral images of the second region to obtain an initial feature point neighborhood; and a stitching path iterative loss analysis is performed in the neighborhood of the initial feature points to obtain a first stitching path.

[0074] Furthermore, the splicing path iteration module 12 is further configured to perform the following steps:

[0075] A stitching path is randomly generated in the neighborhood of the initial feature point to obtain a first initial stitching path; the initial feature point neighborhood is divided by using the first stitching path to obtain a first initial feature point neighborhood sub-neighborhood and a second initial feature point sub-neighborhood; a gradient difference recognition formula is used to identify the gradient difference of the initial feature point, the first initial feature point sub-neighborhood and the second initial feature point sub-neighborhood to obtain a first initial stitching path gradient difference; a second initial stitching path is randomly generated from the neighborhood of the initial feature point again, and a gradient difference recognition is performed on the second initial stitching path to obtain a second initial stitching path gradient difference; it is determined whether the gradient difference of the second initial stitching path is greater than or equal to the gradient difference of the first initial stitching path, and if so, the second initial stitching path is used as a first stage stitching path; the first stage stitching path is iterated multiple times in the neighborhood of the initial feature point until a preset iteration stop condition is met, the iteration is stopped, and the stage stitching path corresponding to the minimum value of the stage stitching path gradient difference during the iteration process is used as the first stitching path.

[0076] Furthermore, the splicing path iteration module 12 is also used to perform the following steps:

[0077] The gradient difference identification formula is: ;in, is the gradient difference of the first initial splicing path, is the pixel value of the i-th pixel in the sub-neighborhood of the first initial feature point, is the pixel value of the jth pixel in the sub-neighborhood of the second initial feature point, is the pixel value of the initial feature point, n+m=16.

[0078] Furthermore, the splicing path iteration module 12 is also used to perform the following steps:

[0079] The neighborhood of the initial feature point includes eight pixel points adjacent to the initial feature point in the overlapping domain of the first-region multispectral image and the overlapping domain of the second-region multispectral image.

[0080] Furthermore, the splicing path iteration module 12 is also used to perform the following steps:

[0081] The preset iteration stop condition is that the number of iterations is greater than or equal to the preset number of iterations or the difference between the gradient differences of the stage splicing paths of two adjacent iterations is less than or equal to the preset difference.

[0082] Furthermore, the cross-scenario defect recognition module 14 is further configured to perform the following steps:

[0083] Acquire multiple standard antique defect recognition results and corresponding multiple standard antique spectral images; collect the current imaging environment according to a preset imaging environment indicator set to obtain a current imaging environment indicator set, and interfere with the multiple standard antique spectral images based on the current imaging environment indicator set to obtain multiple interfered antique spectral images; construct the domain shift anti-interference identifier based on the multiple interfered antique spectral images and multiple standard antique defect recognition results.

[0084] Furthermore, the cross-scenario defect recognition module 14 is further configured to perform the following steps:

[0085] The preset imaging environment indicator set includes illumination uniformity, illumination intensity, shooting angle, field of view, and resolution.

[0086] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0087] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0088] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for identifying antique defects by multispectral imaging analysis, characterized in that: The method comprises: By using a multi-component lens, the M divided regions of the target antique are respectively subjected to neighbor overlapping imaging to obtain M regional multispectral images, wherein M is a positive integer, and the M regional multispectral images include M regional multispectral image overlapping domains; Based on the regional positions of the M divided regions, performing a stitching path iterative loss analysis on the overlapping domains of the M regional multispectral images to determine M stitching paths; Based on the M stitching paths, the M regional multispectral images are stitched together to obtain a target antique multispectral image; A domain shift anti-interference identifier is used to perform cross-scene defect recognition on the target antique multispectral image to obtain a target antique defect recognition result.

2. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 1, characterized in that: Based on the regional positions of the M divided regions, performing a stitching path iterative loss analysis on the overlapping domains of the M regional multispectral images to determine the M stitching paths includes: Extracting a first regional multispectral image overlapping domain from the M regional multispectral image overlapping domains, and matching a second regional multispectral image overlapping domain overlapping with the first regional multispectral image overlapping domain in combination with the regional positions of the M divided regions; Performing initial feature point recognition on the overlapping domain of the first-region multispectral image and the overlapping domain of the second-region multispectral image, and performing iterative loss analysis of the stitching path based on the initial feature points to determine a first stitching path; Performing stitching path iterative loss analysis on the overlapping domains of the M regional multispectral images respectively to determine the M stitching paths.

3. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 2, characterized in that: Performing initial feature point recognition on the overlapping domain of the multispectral image of the first region and the overlapping domain of the multispectral image of the second region, performing iterative loss analysis of the stitching path based on the initial feature points, and determining a first stitching path, including: Using SIFT feature extraction method to identify initial feature points of the overlapping domain of the multispectral image of the first region and the overlapping domain of the multispectral image of the second region, to obtain initial feature points; Based on the position of the initial feature point in the first-region multispectral image overlapping domain, a neighborhood is constructed in the first-region multispectral image overlapping domain and the second-region multispectral image overlapping domain to obtain an initial feature point neighborhood; Performing iterative loss analysis on the splicing path in the neighborhood of the initial feature point to obtain a first splicing path.

4. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 3, characterized in that: Performing iterative loss analysis of the splicing path in the neighborhood of the initial feature point to obtain a first splicing path includes: Randomly generate a splicing path in the neighborhood of the initial feature point to obtain a first initial splicing path; Dividing the initial feature point neighborhood by using the first splicing path to obtain a first initial feature point sub-neighborhood and a second initial feature point sub-neighborhood; Using a gradient difference recognition formula to perform gradient difference recognition on the initial feature point, the first initial feature point sub-neighborhood, and the second initial feature point sub-neighborhood to obtain a first initial splicing path gradient difference; randomly generating a second initial stitching path from the neighborhood of the initial feature point again, performing gradient difference identification on the second initial stitching path, and obtaining a second initial stitching path gradient difference; Determine whether the gradient difference of the second initial splicing path is greater than or equal to the gradient difference of the first initial splicing path, and if so, use the second initial splicing path as the first-stage splicing path; The first stage stitching path is iterated multiple times in the neighborhood of the initial feature point until a preset iteration stop condition is met, the iteration is stopped, and the stage stitching path corresponding to the minimum value of the stage stitching path gradient difference during the iteration process is used as the first stitching path.

5. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 4, characterized in that: The gradient difference identification formula is: ； in, is the first initial splicing path gradient difference, is the pixel value of the i-th pixel in the sub-neighborhood of the first initial feature point, is the pixel value of the jth pixel in the sub-neighborhood of the second initial feature point, is the pixel value of the initial feature point, n+m=16.

6. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 3, characterized in that: The neighborhood of the initial feature point includes eight pixel points adjacent to the initial feature point in the overlapping domain of the first-region multispectral image and the overlapping domain of the second-region multispectral image.

7. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 4, characterized in that: The preset iteration stop condition is that the number of iterations is greater than or equal to the preset number of iterations or the difference between the gradient differences of the stage splicing paths of two adjacent iterations is less than or equal to the preset difference.

8. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 1, characterized in that: The domain shift anti-interference identifier is used to perform cross-scene defect recognition on the target antique multispectral image to obtain a target antique defect recognition result, including: Obtaining multiple standard antique defect recognition results and corresponding multiple standard antique spectral images; The current imaging environment is collected according to a preset imaging environment indicator set to obtain a current imaging environment indicator set, and the plurality of standard antique spectral images are interfered with based on the current imaging environment indicator set to obtain a plurality of interfered antique spectral images; The domain shift anti-interference identifier is constructed based on the multiple interfering antique spectral images and multiple standard antique defect recognition results.

9. The method for identifying antique defects by multispectral imaging analysis as claimed in claim 8, characterized in that: The preset imaging environment indicator set includes illumination uniformity, illumination intensity, shooting angle, field of view, and resolution.

10. An antique defect recognition system based on multispectral imaging analysis, characterized in that: The system comprises: A neighbor overlapping imaging module, wherein the neighbor overlapping imaging module is used to perform neighbor overlapping imaging on M divided areas of the target antique through a multi-component lens, to obtain M regional multispectral images, wherein M is a positive integer, and the M regional multispectral images include M regional multispectral image overlapping domains; A stitching path iteration module, the stitching path iteration module is used to perform stitching path iteration loss analysis on the overlapping domains of the M regional multispectral images based on the regional positions of the M divided regions, and determine M stitching paths; An image stitching module, the image stitching module is used to stitch the M regional multispectral images based on the M stitching paths to obtain a target antique multispectral image; A cross-scene defect recognition module is used to use a domain shift anti-interference identifier to perform cross-scene defect recognition on the target antique multispectral image to obtain a target antique defect recognition result.

Images