A method for digital radiographic inspection and automatic evaluation of a ship weld

Abstract

This invention discloses a digital radiographic inspection and automatic evaluation method for ship welds, relating to the fields of non-destructive testing and image processing. The method includes: acquiring digital radiographic images of complex ship weld structures; evaluating the imaging quality of the digital radiographic images; detecting, identifying, classifying, and measuring weld defects in digital radiographic images that pass the quality evaluation; classifying and evaluating the identified defects according to preset classification society weld radiographic inspection evaluation standards; and generating a structured inspection report that meets the specifications. This invention automates the entire weld inspection process by acquiring digital radiographic images of ship welds and performing quality evaluation, defect identification, and rule-based classification evaluation. Through standard rule-driven and multi-scale image analysis, it improves the accuracy of defect identification and the consistency of evaluation, reduces manual intervention, and thus improves inspection efficiency and result standardization.

Description

Technical Field

[0001] This application relates to the field of nondestructive testing and image processing technology, and in particular to a digital radiographic inspection and automatic evaluation method for ship welds. Background Technology

[0002] In shipbuilding, weld quality directly affects the safety and service reliability of the hull structure. Due to the widespread presence of complex structures such as curved hull plates, T-section corner joints, and pipe joints, weld morphologies are complex and space is limited, posing limitations to traditional non-destructive testing methods in image acquisition and defect identification. Especially in thick plate welding or multi-layered structure welds, problems such as insufficient imaging contrast, structural overlap interference, and difficulty in effectively identifying minute defects often occur, thus affecting the accuracy and consistency of the inspection results.

[0003] While existing digital radiographic inspection technology has been gradually applied to the inspection of ship welds, in practice it still relies heavily on manual experience for image quality judgment and defect assessment, resulting in low processing efficiency, high subjectivity, and inconsistent standard implementation. Furthermore, for complex image data acquired through multiple exposures and perspectives, there is a lack of effective unified preprocessing, fusion, and automatic assessment mechanisms, making it difficult to meet the shipbuilding industry's demand for highly consistent and reliable inspection results. Therefore, it is necessary to propose a digital radiographic inspection method for ship welds that can achieve image quality optimization, automatic defect identification, and standardized assessment. Summary of the Invention

[0004] This application provides a digital radiographic inspection and automatic evaluation method for ship welds. It employs a combination of multi-scale image processing, spatial registration and fusion, and defect rule-based evaluation techniques to adaptively preprocess and enhance digital radiographic images. By combining connected component analysis and rule base matching, it achieves automatic defect identification and hierarchical evaluation, thereby improving the accuracy and consistency of weld defect detection and realizing the technical effect of automated and standardized evaluation of digital radiographic inspection of welds in complex ship structures.

[0005] This application provides a digital radiographic inspection and automatic evaluation method for ship welds, comprising: acquiring digital radiographic images of welds in complex ship structures, including hull curved plates, T-shaped corner joints, and pipe joints; acquiring digital radiographic images through digital radiographic imaging (DR), computed radiographic imaging (CR), or a combination thereof, and adjusting the imaging layout and imaging parameters according to the geometric characteristics of the complex ship structure; evaluating the imaging quality of the digital radiographic images, wherein the evaluation indicators include at least spatial resolution sensitivity set based on weld defect detection requirements and grayscale dynamic range coverage integrity for variations in hull plate thickness; detecting, identifying, classifying, and measuring weld defects in digital radiographic images that pass the quality evaluation; classifying and evaluating the identified defects according to preset classification society weld radiographic inspection evaluation standards, and generating a structured inspection report that meets the specifications. Attached Figure Description

[0006] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0007] Figure 1 This is a flowchart illustrating a digital radiographic inspection and automatic evaluation method for ship welds provided in an embodiment of this application. Detailed Implementation

[0008] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0009] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. 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 explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0011] This application provides a digital radiographic inspection and automatic evaluation method for ship welds, such as... Figure 1 As shown, the method includes: S1: Acquire digital radiographic images of welds in complex ship structures, including hull curved plates, T-shaped corner joints, and pipe joints. The digital radiographic images are acquired through digital radiography (DR), computed radiography (CR), or a combination thereof, and the imaging layout and imaging parameters are adjusted according to the geometric characteristics of the complex ship structure.

[0012] Furthermore, step S1 also includes: S11. For T-section fillet welds and welded joints between bulb flats and plates, double-wall imaging or oblique radiography is used, and the radiography angle is calculated based on the 3D model data of the weld. S12. For curved plate welds, a flexible detector array or CR imaging plate is used with a bending attachment method to ensure the imaging surface fits the curved surface of the workpiece. S13. In space-constrained areas, a crawling robot or magnetic adsorption X-ray device is used to perform long-distance or lateral radiography, and image data is transmitted wirelessly. S14. When a single exposure cannot cover the dynamic range of weld thickness, multi-energy DR imaging or multi-exposure parameter CR imaging is used to obtain images of different areas. S15. For pipe joints or large complex structures, a mobile X-ray device is controlled to image at multiple preset angles, and 3D reconstruction processing is performed based on the multi-view images to obtain 3D volumetric data of the weld area.

[0013] Specifically, the first step is to acquire digital radiographic images of welds in complex ship structures. These complex structures include, but are not limited to, typical complex welded structures such as hull curved plate structures, T-section corner joints, and pipe joints. Based on the geometric shape and spatial distribution characteristics of different structures and welds, digital radiographic imaging (DR), computed radiographic imaging (CR), or a combination of both are used for image acquisition. During the acquisition process, the X-ray source position, detector position, exposure time, and energy parameters are adaptively adjusted according to weld thickness variations, obstruction, and structural curvature to obtain weld radiographic image data with good contrast and structural integrity. For T-section corner welds and welded joints formed by bulb flats and plates, due to significant geometric obstruction and structural layering characteristics, double-wall radiographic imaging or oblique radiographic imaging is used for image acquisition. In practice, a three-dimensional geometric model of the weld is established, and the spatial relationship between the X-ray source and detector is calculated to determine the optimal radiographic angle, allowing the X-rays to penetrate as much as possible into the critical areas of the weld, thereby reducing the impact of structural overlap on defect imaging and improving defect visibility. For curved plate weld structures, conventional planar detectors struggle to guarantee precise fit due to the continuous curvature variations on their surfaces. In such cases, flexible detector arrays can be employed, allowing the detectors to deform with the curved surface. Alternatively, computed radiography (CR) imaging plates can be used, achieving close adhesion to the curved surface through mechanical or magnetic adsorption. This ensures the imaging plane aligns as closely as possible with the workpiece surface, minimizing the impact of geometric distortion on image quality. In confined spaces such as ship cabins or narrow structures where direct access by personnel or equipment is difficult, crawling robots or magnetically adsorbed mobile X-ray inspection equipment can be used for image acquisition. This equipment can move along the ship's structural surface, carrying the X-ray source and detector for long-distance or lateral imaging. Simultaneously, the acquired image data is transmitted in real-time to the control terminal via a wireless communication module, enabling remote inspection in complex environments. For thick plate weld structures, the significant differences in X-ray absorption across different thickness regions often make it difficult to obtain clear images of both the weld toe and weld root areas in a single exposure. Therefore, multi-energy digital X-ray imaging can be used to acquire images with different penetration capabilities by adjusting the X-ray energy, or multi-exposure parameter computed tomography can be used to acquire images of the same weld seam multiple times under different exposure conditions, which can then be used for subsequent image fusion processing. For pipe joints and large, complex three-dimensional welded structures, due to their complex spatial morphology and immobility, multi-view digital imaging can be used for inspection. Specifically, by controlling a crawling robot or robotic arm to carry an X-ray source and detector, the weld seam area is imaged point-by-point or segmentally from multiple preset angles to obtain multiple sets of two-dimensional X-ray image data. Subsequently, based on the geometric relationships between the multi-view images, tomographic reconstruction or photoelectric tomography reconstruction methods are used to reconstruct the three-dimensional volume data of the weld seam area, thereby achieving precise spatial location and quantitative analysis of defects.

[0014] S2: Evaluate the imaging quality of the digital X-ray image. The evaluation indicators include at least the spatial resolution sensitivity set based on the weld defect detection requirements and the grayscale dynamic range coverage integrity for changes in the thickness of the ship plate.

[0015] Furthermore, step S2 also includes: S21. Before quality evaluation, adaptive preprocessing is performed on the acquired digital radiographic images. S22. The image quality indicator, serial numbers, and positioning marks in the images are inspected and their integrity assessed to determine if they meet the requirements of ship inspection regulations. S23. The visibility of the image quality indicator wires in the images is evaluated, and their signal-to-noise ratio is calculated to determine if the image resolution meets ship inspection standards. S24. The visibility of the smallest identifiable structure in the images is evaluated to determine if the detection sensitivity meets the inspection requirements for workpieces of corresponding thickness. S25. Artifacts and scattering interference in the images are detected and evaluated, and their impact is quantified. S26. The dynamic range of the image is analyzed based on its grayscale distribution to determine whether it covers the thickness range of the weld and base material, ensuring sufficient contrast for defects.

[0016] Furthermore, step S21 also includes: S21-1. Acquire digital X-ray inspection images that meet the quality assessment requirements, including single-image images and multiple images under different exposure conditions, and organize the images into an image sequence according to exposure conditions, camera angle, and acquisition order. ,in For exposure condition index, For camera view indexing, For the collection order index, S21-2: Extract noise statistical features from each image in the image sequence, including the mean pixel grayscale value. and pixel grayscale variance And automatically generate corresponding denoising algorithm parameters based on the noise statistical characteristics. , used for adaptive denoising, where the local block-level noise variance is defined as: ;in In pixels A local image patch centered on the image. This represents the average grayscale value of the block. The number of pixels within the block. S21-3. Based on the denoising algorithm parameters, perform denoising operations on each image in the image sequence, including processing images under different exposure conditions, and record the denoising parameters and processing results for each image; S21-4. Perform feature point detection and matching on denoised images from different camera perspectives, establish constraint conditions in conjunction with weld geometric reference information, output the spatially aligned image set, and record the correspondence between the registration matrix and the original image; S21-5. Stitch the registered images according to spatial coordinates and acquisition order, and perform grayscale fusion on the overlapping areas. The grayscale fusion formula is as follows: ;in, ， For local gray-level gradient weights, As the thickness resolution weighting coefficient, record the coordinates of the stitching region and the fusion parameters to generate a stitched image set; S21-6, perform a high dynamic range fusion algorithm on the image sequences under different exposure conditions in the stitched image set to generate HDR images. The pixel grayscale value calculation formula is: ;in, The first after splicing Secondary exposure image; S21-7, Establish an index table for preprocessed images and original images, record the correspondence between the denoising, registration, stitching and HDR fusion parameters of each image and the original image, and use it to associate the original image and the preprocessed image in the defect assessment process.

[0017] Furthermore, step S21-3 also includes: S21-3-1. Perform imaging pattern recognition on each image in the image sequence to generate corresponding imaging pattern identification parameters; S21-3-2. Based on the imaging pattern identification parameters, perform multi-scale decomposition processing on the image to extract noise feature parameters at different scales; S21-3-3. Based on the noise feature parameters, perform initial reconstruction processing on the image to generate an initial image; S21-3-4. Based on the imaging pattern identification parameters and noise feature parameters, select the corresponding denoising processing path from a preset denoising strategy set and generate denoising algorithm parameters; S21-3-5. Based on... The denoising algorithm parameters and the selected denoising processing path are used to perform adaptive denoising processing on the initial image to obtain a denoised result image. During the denoising process, the denoising intensity is dynamically adjusted based on the local gray-level distribution characteristics of the image. S21-3-6: Local abnormal noise suppression processing is performed on the denoised result image to obtain an optimized denoised image, so as to remove isolated noise points and maintain image structure information. S21-3-7: The denoising algorithm parameters, denoising processing path and processing result image of each image are recorded, and an index correspondence relationship between the denoised result image and the original image is established.

[0018] Specifically, after acquiring digital radiographic images of ship welds, the acquired digital radiographic images undergo an imaging quality evaluation. This evaluation primarily determines whether the current image meets the requirements for subsequent defect detection and assessment. The evaluation includes at least indicators such as image spatial resolution, detection sensitivity, artifact interference level, and grayscale dynamic range integrity. By comprehensively evaluating these indicators, images with severe distortion, insufficient resolution, or incomplete grayscale coverage can be filtered out, thereby improving the accuracy and stability of subsequent defect identification and assessment results. Before performing the imaging quality evaluation, adaptive preprocessing is first applied to the digital radiographic images. Specifically, denoising, grayscale normalization, and image enhancement are performed on single images or multiple images acquired under different exposure conditions. In one embodiment, the grayscale distribution characteristics of each region in the image are first statistically analyzed, and the mean and variance of the grayscale in local regions are calculated to characterize the image noise level. Subsequently, based on the noise intensity of different regions, the denoising parameters are adaptively adjusted, with stronger smoothing processing applied to high-noise regions and more edge detail information preserved in low-noise regions, thus suppressing noise while maintaining the edge features of weld defects. For images acquired using multi-exposure conditions, preprocessing can be performed on images of different exposures, and registration and fusion processing can be performed based on the spatial positional relationship of the images to improve the overall dynamic range of the thick plate weld area. For images from different viewpoints, spatial correspondence can be established based on feature point matching relationships, and image stitching can be performed to form a complete weld area image. After image preprocessing, the image quality indicator, serial number, and positioning marks in the image are detected and their integrity is assessed. Specifically, the marked areas in the image are identified through image segmentation or edge detection, and the positional parameters of the image quality indicator outline, character area, and positioning mark area are extracted. Then, it is determined whether the image quality indicator is fully displayed, whether the characters are identifiable, and whether the positioning marks are located within the preset detection area, thereby confirming whether the current image meets the requirements of the ship weld radiographic inspection procedure. For images with missing marks, marked obstructions, or abnormal positioning, they can be judged as not meeting the quality evaluation requirements, and corresponding abnormal information is output. The spatial resolution of the image is evaluated by assessing the visibility of the image quality indicator wire and the signal-to-noise ratio of the corresponding area. Specifically, after identifying the image quality indicator area, grayscale analysis is performed on the corresponding area of ​​the metal wire to calculate the grayscale difference between the metal wire area and the background area, and the corresponding signal-to-noise ratio (SNR) parameter is further calculated. When the edge of the metal wire has a clear and continuous grayscale change, and the SNR is higher than a preset threshold, the current image resolution is determined to meet the requirements for ship weld inspection; otherwise, the image resolution is determined to be insufficient. The detection sensitivity is evaluated by analyzing the imaging of the smallest identifiable structure in the image. Specifically, the thinnest metal wire or the smallest hole structure in the image quality indicator is identified, and the grayscale change characteristics, edge gradient characteristics, and contour continuity characteristics of its corresponding area are calculated.When the minimum structure can be stably identified and the corresponding grayscale change amplitude reaches a preset threshold, it is determined that the current image sensitivity meets the detection requirements of the workpiece with the corresponding thickness. For thick plate welds, the sensitivity evaluation threshold can be automatically adjusted according to the base material thickness to adapt to the differences in X-ray attenuation under different thickness conditions. In one embodiment, artifacts and scattering interference in the image are detected and evaluated. Specifically, local texture analysis and grayscale anomaly detection are performed on the image to identify abnormal areas such as edge erosion areas, scattering haze areas, stripe interference areas, and scratch areas. Subsequently, the area ratio, grayscale shift degree, and edge continuity change degree of abnormal areas are statistically analyzed, and the influence of artifacts on image quality is quantified based on the statistical results. When the influence of artifacts exceeds a preset threshold, it can be determined that there is significant interference in the current image, and a prompt is made to re-perform X-ray detection. The dynamic range of the image is analyzed based on the image grayscale distribution. Specifically, the grayscale histograms of the weld area and the base material area are statistically analyzed to determine the distribution of image grayscale values ​​in the overall grayscale range, and the grayscale coverage range and grayscale change continuity are calculated. When the image grayscale range can simultaneously cover both the weld fusion area and the base material area, and maintain continuous grayscale variation, the image dynamic range is deemed to meet the detection requirements. For images acquired using multi-exposure methods, high dynamic range fusion can be performed on images from different exposures to expand the overall grayscale coverage, thereby improving the defect contrast display capability of thick plate weld areas.

[0019] Furthermore, to meet the imaging quality assessment requirements of different ship weld structures, digital radiographic images are first acquired. These digital radiographic images include images acquired under single-exposure conditions, as well as multiple images acquired under different exposure energies, exposure times, or gain parameters. For thick plate welds, multi-layer welds, or weld areas with significant thickness variations, the absorption capacity of different areas for X-rays varies, making it difficult to simultaneously assess the imaging quality of both the weld root and weld toe areas under a single exposure. Therefore, multiple image acquisitions can be performed on the same weld area under different exposure conditions. Specifically, image data with different grayscale response ranges can be obtained by adjusting the X-ray tube voltage, X-ray tube current, exposure time, detector gain parameters, or imaging energy level. Simultaneously, for large weld areas or complex spatial structures, image acquisition can be performed from different perspectives to reduce structural occlusion and overlap interference. The acquired images are organized and managed according to exposure conditions, camera perspective, and acquisition order to form an image sequence. ,in For exposure condition index, For camera view indexing, For the collection order index, The image contains pixel coordinates. Noise statistical features are extracted from each image in the image sequence, and corresponding denoising algorithm parameters are automatically generated based on these features for subsequent adaptive denoising processing. Because digital X-ray images are easily affected by detector thermal noise, scattering noise, quantum noise, and interference from scattering by complex structures during acquisition, image noise levels under different exposure conditions often vary significantly. Therefore, it is necessary to dynamically adjust the denoising intensity based on the grayscale distribution of local image regions. A sliding window statistical analysis is performed on local regions of the image. For any pixel coordinates, including the pixel grayscale mean... and pixel grayscale variance And automatically generate corresponding denoising algorithm parameters based on the noise statistical characteristics. , used for adaptive denoising, where the local block-level noise variance is defined as: ;in In pixels A local image patch centered on the image. This represents the average grayscale value of the block. The number of pixels within the block. The pixel coordinate index is used in the local image block; after obtaining the corresponding denoising algorithm parameters, adaptive denoising processing is performed on each digitized ray image in the image sequence. Since the noise characteristics differ under different exposure conditions, different illumination angles, and different structural regions, denoising operations are performed separately for different images. Based on the denoising algorithm parameters generated in step S21-2, the corresponding filtering method is selected to process the image. The filtering method includes one or more combinations of mean filtering, Gaussian filtering, median filtering, bilateral filtering, non-local mean filtering, or diffusion filtering based on partial differential equations. For regions with relatively smooth grayscale changes and strong noise, the filter window size and smoothing intensity are increased to reduce random noise; for regions with obvious edge gradient changes, the smoothing intensity is reduced and the edge preservation weight is increased to avoid over-smoothing of weld edges and minor defect features. Independent denoising processing can be performed on images acquired under different exposure conditions. For example, for low-exposure images, since quantum noise is more pronounced, stronger smoothing processing can be used; for high-exposure images, the focus is on preserving edge details in high grayscale regions. During the denoising process, information such as denoising parameters, filtering methods, filtering intensity, and processing time for each image is recorded, and a correspondence between the processed image and the original image is established for subsequent image quality traceability and defect assessment. After the above processing, a denoised image with low noise level and intact defect edge features can be obtained, providing a stable data foundation for subsequent multi-view image registration and defect identification. Spatial registration processing is performed on denoised images from different camera perspectives to establish spatial correspondence between images from different perspectives. Since large ship structure welds usually require image acquisition from multiple perspectives, the images obtained from different perspectives differ in spatial position, scale, and rotation direction. Therefore, it is necessary to perform unified spatial alignment on images from different perspectives. Specifically, feature point detection processing is first performed on each denoised image. The feature points include gray-level abrupt change points, edge intersection points, texture feature points, or weld structure feature points, etc. Subsequently, matching analysis is performed on the feature points in different images to establish the correspondence between images from different perspectives. In one implementation, geometric constraints can be established based on the weld centerline position, weld boundary contour, structural corner position, or preset positioning marks to reduce mismatches. Subsequently, spatial transformation parameters between images are calculated based on the correspondence of the matched feature points. These spatial transformation parameters include translation, rotation, scaling, and perspective transformation parameters, and a corresponding image registration matrix is ​​further generated. The registration matrix is ​​used to spatially transform images from different perspectives, aligning the same weld regions in different images within a unified coordinate system.After spatial alignment is completed, the spatially registered image set is output, and the registration matrix, reference image number, and spatial transformation parameters corresponding to each image are recorded to establish the mapping relationship between the registration result and the original image. The registered images are stitched together according to spatial coordinates and acquisition order, and grayscale fusion is performed on the overlapping areas. The grayscale fusion formula is as follows: ;in, , For local gray-level gradient weights, The thickness resolution weighting coefficients are used to record the coordinates of the stitching region and the fusion parameters to generate a stitched image set; wherein, the local gray-level gradient weights... By registering the image at the pixel level The gradient of grayscale space changes within the neighborhood is calculated, and the specific calculation method is as follows: First, based on the pixel point... Define a local window at the center Calculate the gray-level gradients in the horizontal and vertical directions within the window respectively: ; This allows us to obtain the gradient magnitude of the pixel, which is then used as the local grayscale gradient weight. ; wherein, the This is used to characterize the intensity of local structural changes at a pixel location. When weld boundaries, defect edges, or abrupt structural changes are present, the gradient amplitude increases significantly, thus assigning a higher structural weight during the stitching and fusion process. Indicates the first Amplitude registration image at pixel coordinates The grayscale value at that location; , representing the pixel spatial coordinates in the image. Image indexes representing different imaging perspectives or exposure conditions; Represented by pixels The local computational window region centered on; This represents the gradient of grayscale changes in the image along the horizontal direction. This represents the gradient of grayscale changes in the vertical direction of the image; Represents pixels The local gray-level gradient weight is used to characterize the intensity of local structural changes. The thickness resolution weight coefficient... The equivalent penetration thickness is obtained through calculation based on the geometric relationship of X-ray imaging. The specific calculation method is as follows: the pixel points are determined according to the spatial geometric relationship between the X-ray source, the workpiece, and the detector. Corresponding equivalent penetration path length of the ray , represented as: ;in, This represents the actual thickness of the workpiece at that location. The angle between the incident direction of the ray and the normal direction of the workpiece is given. Furthermore, the equivalent penetration thickness is normalized to obtain the thickness resolution weighting coefficient: ;in, This represents the maximum equivalent penetration thickness across the entire map, used for normalization. The thickness attenuation at a pixel location is used to characterize the degree of attenuation. When the thickness is greater in a thick plate area or a region with stronger ray attenuation, its weight is increased accordingly to enhance its contribution to the splicing and fusion process. Indicates the first The image at the pixel point The thickness resolution weighting coefficient at the location; where... This indicates the equivalent ray penetration thickness corresponding to that pixel. Indicates the workpiece at the pixel point The actual material thickness at the corresponding location; This indicates the angle between the incident direction of the ray and the normal to the workpiece surface. Indicates angle The cosine value is used to correct the equivalent thickness under oblique projection conditions; Represents the maximum equivalent penetration thickness within the entire image region, used for... Perform normalization processing; Image indexes representing different imaging conditions (viewpoint or exposure conditions); Represents the spatial coordinates of image pixels. Wherein, the coefficients... and These represent the contribution ratios of the local gray-level gradient weight term and the thickness resolution weight term in the image fusion process, respectively. α and β The gradient response intensity and thickness distribution characteristics of the images to be fused are adaptively calculated and satisfy the following conditions. ;in, α The calculation method is as follows: , β The calculation method is as follows ： in, , Indicates the first i The local gray-level gradient weights at pixel (x,y) of an image. Indicates the first i The image at the pixel point (x,y) Thickness resolution weighting coefficient at the location, This represents the average gradient response value of the entire image. This represents the average thickness response value of the entire image, where M represents the total number of pixels involved in the statistical calculation. Using the above method, when the local structural edge information in the image is relatively rich, αAutomatically enlarged to increase the contribution of structural edge regions in the stitching and fusion process, especially when the proportion of thickness variation areas in the image is high. Automatically increased to enhance the fusion weight in areas of thickness attenuation, thereby achieving adaptive allocation of weight parameters during the stitching and fusion process. A high dynamic range fusion algorithm is executed on image sequences under different exposure conditions within the stitched image set to generate HDR images. The formula for calculating the grayscale value per pixel is: ;in, The first after splicing The image is then exposed multiple times. After image denoising, spatial registration, image stitching, and high dynamic range fusion processing are completed, an index correspondence is established between the preprocessed image and the original image. This is used to achieve data traceability and defect assessment throughout the entire image processing process. Specifically, a unique image identifier number is generated for each original digitized X-ray image, and an image index table is established based on the image identifier number. The image index table is used to record the correspondence between the original image and the processing results at each stage. The information recorded in the index table includes: original image number, image acquisition time, exposure condition parameters, camera viewpoint parameters, image acquisition sequence number, denoising processing parameters, image registration matrix parameters, image stitching region coordinates, grayscale fusion weight parameters, HDR fusion parameters, and the corresponding preprocessed result image number. For preprocessed images generated after multiple processing steps, their source original image number and the corresponding processing sequence can be recorded. For example, the original exposure image numbers corresponding to a certain HDR fusion image, the spatial registration relationship of each image, and the fusion parameters can be recorded. In subsequent defect detection and quality assessment, when a target defect is detected, the corresponding original image and its processing information can be quickly located through the index table, thereby achieving a bidirectional mapping between the defect area and the preprocessed image. For example, when the automatic assessment result is abnormal, the corresponding original image can be traced back through the index table to view the original grayscale distribution of the defect in the unenhanced state; or the corresponding position of the defect in images from different viewpoints can be relocated according to the registration matrix to assist in manual review.

[0020] Furthermore, imaging pattern recognition is first performed on each digitized X-ray image in the image sequence to determine the imaging type and noise feature category corresponding to the current image, and to generate corresponding imaging pattern identification parameters. Since the noise sources and image characteristics differ for different digitized X-ray imaging methods—for example, digitized X-ray (DR) images typically have a high dynamic range, while computed radiography (CR) images are more prone to shot noise and scanning stripes—different denoising strategies are needed for different imaging modes. Specifically, the current image is classified into modes by reading image acquisition parameters, exposure parameters, detector type parameters, and image grayscale statistical features. The imaging pattern identification parameters include at least: imaging mode type; exposure level type; image noise level; dynamic range level; image contrast level; and structural complexity level. These parameters are then used as the basis for subsequent denoising strategy selection and parameter generation. After determining the imaging mode, multi-scale decomposition processing is performed on the image to extract noise feature parameters at different scales. Specifically, wavelet transform, non-subsampled contourlet transform, pyramid decomposition, or frequency domain decomposition can be used to perform multi-scale decomposition processing on the image, decomposing the original image into multiple sub-band images with different frequency ranges and directional features. The image is divided into two sub-bands: a low-frequency sub-band primarily reflects the overall grayscale variation of the weld area, and a high-frequency sub-band primarily reflects edge details and noise information. Subsequently, statistical analysis is performed on the grayscale distribution characteristics, local variance characteristics, and texture variation characteristics of the sub-bands at different scales to extract the corresponding noise feature parameters. For example, in areas with significant random noise, the local grayscale fluctuations in the high-frequency sub-band are large; while in areas with true defect edges, there are usually continuous directional texture variations. After obtaining the noise feature parameters, initial reconstruction processing is performed on the image to generate an initial image. Specifically, based on the noise distribution in the sub-bands at different scales, coefficient adjustment processing is performed on some high-noise sub-bands. For example, high-frequency coefficients that are clearly random noise are attenuated, while structural responses with strong directional consistency at the weld edge are preserved. Then, the adjusted sub-bands at each scale are reconstructed using inverse transformation to generate the initial image. Compared with the original image, the random noise level in the initial image is initially suppressed, while the weld contour and defect edge information are preserved, thus providing a more stable data foundation for subsequent adaptive denoising processing. Based on the imaging mode identifier parameters and noise characteristic parameters, a corresponding denoising processing path is selected from a preset denoising strategy set, and corresponding denoising algorithm parameters are generated. The preset denoising strategy set includes various types of denoising processing strategies, such as: edge-preserving denoising strategies; high-frequency noise suppression denoising strategies; stripe artifact suppression denoising strategies; scattering haze suppression denoising strategies; and multi-exposure fusion-assisted denoising strategies. For example, for low-exposure, high-noise images, a high-frequency noise suppression processing path can be preferentially selected; for images with obvious edge defects, an edge-preserving denoising processing path is preferred.Subsequently, based on the noise level, local texture complexity, and exposure parameters of the current image, corresponding denoising algorithm parameters are automatically generated. These parameters include: filter window size; smoothing intensity coefficient; edge preservation weight; high-frequency suppression threshold; number of iterations; and texture protection parameters. After obtaining the denoising algorithm parameters and the corresponding denoising processing path, adaptive denoising processing is performed on the initial image. Specifically, the denoising intensity is dynamically adjusted according to the gray-level distribution characteristics of local image regions. For regions with gentle gray-level changes and large local variance, the smoothing intensity is increased to suppress random noise; for regions with significant gray-level gradient changes, the smoothing intensity is decreased, and the edge protection weight is increased to avoid over-smoothing of weld edges and minor defect features. In one embodiment, bilateral filtering, non-local mean filtering, anisotropic diffusion filtering, or multi-scale joint filtering can be used to perform adaptive denoising processing. After processing, a denoised result image is generated. The denoised result image effectively reduces random noise and scattering interference while maintaining the integrity of the weld structure edges. Further local abnormal noise suppression processing is performed on the denoised result image to eliminate isolated abnormal noise points and improve the continuity of the image structure. Specifically, isolated bright spots, isolated dark spots, and abnormal stripe regions in the image are identified through local connected component analysis, morphological filtering, or local statistical anomaly detection. For abnormal regions that are small in area, lack continuous structural features, and exhibit inconsistent grayscale changes with the surrounding area, they can be identified as local abnormal noise regions, and local replacement, grayscale interpolation, or neighborhood smoothing are performed. While removing abnormal noise points, the continuous structural information of weld boundaries and defect edge regions is preserved, thus avoiding the accidental deletion of real, minor defects. Finally, an optimized denoised image is generated. The denoising algorithm parameters, denoising processing path, and processing result image corresponding to each image are recorded, and an index correspondence between the denoised result image and the original image is established. Specifically, a unique identifier is generated for each original image and its corresponding processing result, and a denoising processing record table is established. The record table includes at least: original image number; imaging mode identifier parameters; denoising processing path number; denoising algorithm parameters; filter type; filter intensity; processing time; and output image number. Simultaneously, a mapping relationship between the processing result image and the original image is established to enable image traceability in subsequent defect detection and quality assessment processes. For example, when an abnormal judgment result is found during the subsequent defect assessment process, the corresponding original image can be quickly located through the index relationship, and its denoising process and parameter settings can be viewed, thereby improving the traceability and reliability of the defect detection results.

[0021] S3: Detect, identify, classify, and measure the dimensions of weld defects in digital radiographic images that have passed quality evaluation.

[0022] Furthermore, step S3 also includes: S31. Perform multi-scale decomposition processing on the digital X-ray image to extract image response information at different scales and directions, and fuse the response information at each scale to generate a multi-scale response image. The multi-scale response image is used to characterize the morphology, size, and directional variation features of weld defects. S32. Enhance the multi-scale response image, including local gradient enhancement of the high-frequency response region to highlight the edge features of microcracks and needle-like pores; contrast enhancement of the low-frequency response region to increase the grayscale difference between the defect region and the background region; and fuse the enhanced response image to generate a defect response enhancement image. S33. Perform threshold segmentation processing on the defect response enhancement image to obtain... S33: Obtain a binary image of the defect candidate region; perform connected component labeling on the binary image, extract the spatial location, contour boundary, and area parameters of each defect candidate region, and generate a defect region set; S34: For the defect region set, calculate geometric parameters for each defect region, including region area, aspect ratio, boundary complexity, and grayscale distribution characteristics; combine welding process parameters and base material thickness information to determine the defect type, and generate corresponding defect type identifiers and determination criteria; S35: Perform consistency verification processing on the defect type determination results, including merging duplicate detected defects, merging spatially overlapping defects, and removing abnormal isolated defects, to generate a final defect detection result set.

[0023] Furthermore, the enhancement processing of the multi-scale response image in step S32 also includes: The multi-scale response image is decomposed into sub-band coefficients at different scales and directions. The high-frequency components of the sub-band coefficients are adaptively enhanced to enhance defect edges and texture features and suppress background interference. The low-frequency components of the sub-band coefficients are contrast-enhanced to improve the overall grayscale contrast of the image. The enhanced sub-band coefficients are reconstructed to obtain the enhanced digital ray image.

[0024] Furthermore, in step S32, the multi-scale response image is enhanced. Another implementation also includes: Anisotropic diffusion processing is applied to the multi-scale response image to preserve defect edge features while suppressing noise; morphological filtering is applied to the anisotropic diffusion-processed image based on multi-directional structuring elements to enhance the features of strip-shaped and point-shaped defects; and a defect enhancement response image is generated based on the filtering results to improve the response intensity of small defect regions.

[0025] Furthermore, the defect type determination described in step S34 includes: Connectivity analysis is performed on the binary image obtained from defect detection to obtain multiple independent defect regions. These defect regions are then clustered based on spatial proximity to obtain defect cluster regions. For each defect cluster region, feature parameters are calculated, including the number of defects, coverage area, distribution density, spatial distribution characteristics, and main distribution direction. Based on these feature parameters, a type determination is made for the defect cluster regions by matching them with preset discrimination rules. The corresponding defect type and its feature parameters are output according to the discrimination results and used for subsequent evaluation processing.

[0026] Specifically, the process involves detecting, identifying, classifying, and measuring the dimensions of weld defects in digitized radiographic images that have passed quality evaluation, thereby achieving automated analysis and structured output of internal weld defects. The digitized radiographic images undergo multi-scale decomposition to obtain detailed and overall structural information of the weld structure at different scale levels. In one embodiment, pyramid decomposition, wavelet decomposition, or directional multi-scale transformation can be used to decompose the original image into multiple scale components and multiple directional components, thereby obtaining image response information at different spatial resolutions. Subsequently, the response information at each scale and direction is fused, using weighted superposition or adaptive fusion based on energy distribution, to generate a multi-scale response image. This multi-scale response image characterizes the morphological changes, size features, and directional changes of weld defects at different scales, providing fundamental data support for subsequent defect enhancement and identification. After obtaining the multi-scale response image, defect feature enhancement processing is performed to improve the distinguishability between the defect area and the background area. In one implementation, the method includes: performing local gradient enhancement processing on high-frequency response regions in a multi-scale response image to enhance the response intensity of high-gradient regions, making the edge features of small defects such as microcracks and needle-like pores more prominent; performing contrast enhancement processing on low-frequency response regions to increase the grayscale difference between the weld defect region and the base material background region by adjusting the grayscale dynamic range, thereby improving overall distinguishability; further, fusing the response images after high-frequency and low-frequency enhancement to obtain a defect response enhancement image, so that defect features at different scales are uniformly expressed in the same image space. After obtaining the defect response enhancement image, threshold segmentation processing is performed on it to initially separate the defect region from the background region. In one implementation, a binary image of defect candidate regions is generated using an adaptive thresholding method or a segmentation method based on local statistical characteristics. Subsequently, connected component labeling processing is performed on the binary image to identify each independent defect candidate region, and the spatial location information, contour boundary information, and area information of each region are extracted. Finally, a set of defect regions is generated, providing an input basis for subsequent defect type discrimination. After obtaining the enhanced image of the defect response, threshold segmentation is performed to initially separate the defect region from the background region. In one implementation, a binary image of the defect candidate region is generated using an adaptive thresholding method or a segmentation method based on local statistical characteristics. Subsequently, connected component labeling is performed on the binary image to identify each independent defect candidate region, and the spatial location information, contour boundary information, and area information of each region are extracted. Finally, a set of defect regions is generated, providing the input basis for subsequent defect type discrimination. After completing the defect type discrimination, the defect detection results undergo consistency verification to improve the stability and reliability of the results.In one implementation, the method includes: merging defects that overlap in spatial location or are highly adjacent to each other to avoid duplicate counting; merging defects with spatial intersection or partial overlap characteristics to form a unified defect entity; eliminating isolated defects that are too small in area and do not meet the preset judgment conditions to reduce noise interference; and finally generating a set of verified defect detection results as input data for subsequent grading and evaluation.

[0027] Furthermore, when enhancing the multi-scale response image, a joint enhancement process based on multi-scale decomposition and reconstruction is adopted to further improve the discriminability and structural representation capability of the weld's minute defects. Specifically, the multi-scale response image is first subjected to multi-scale decomposition processing. In one embodiment, wavelet transform, non-subsampled contour wave transform, or directional pyramid decomposition methods can be used to decompose the input multi-scale response image into multiple scale levels and multiple directional sub-band coefficients. The sub-band coefficients are used to characterize the local structural information of the image at different spatial frequencies and in different directions, where high-frequency sub-band coefficients mainly contain defect edges, fine cracks, and local texture change information, while low-frequency sub-band coefficients mainly reflect the overall weld structure and background grayscale distribution information. Subsequently, targeted enhancement processing is performed on each of the decomposed sub-band coefficients. For the high-frequency component sub-band coefficients, an adaptive enhancement strategy is used for processing. In one implementation, the enhancement intensity is dynamically adjusted based on the local statistical characteristics of the high-frequency coefficients, including local mean, local variance, and gradient intensity. This enhances the coefficient response in high-gradient regions (such as crack edges and pore boundaries) while suppressing noise amplification in low signal-to-noise ratio regions, thereby effectively enhancing the edges and texture information of minute defects. For the low-frequency component sub-band coefficients, contrast enhancement processing is employed. In one implementation, a method based on grayscale histogram adjustment or local contrast stretching can be used to dynamically expand the range of low-frequency information, enhancing the grayscale difference between the weld base material region and the defect region, thus improving the structural discernibility of the overall image. After enhancing the high-frequency and low-frequency components separately, the sub-band coefficients at each scale and in each direction are reconstructed. In one implementation, the enhanced sub-band coefficients are resynthesized into a spatial domain image through an inverse transformation process, generating an enhanced digital ray image. This enhanced image maintains the continuity of the original structural information while improving the contrast and edge clarity of minute defect regions, making subsequent defect detection and segmentation more accurate and stable.

[0028] Furthermore, in one embodiment, when enhancing the multi-scale response image in step S32, a processing method combining spatial domain structure preservation and morphological enhancement is adopted to further improve the discriminability of minute defects in the weld image. Specifically, anisotropic diffusion processing is first performed on the multi-scale response image. In one embodiment, the anisotropic diffusion processing constrains the diffusion process based on image gradient information, ensuring that diffusion occurs fully in low-gradient regions to achieve noise smoothing, while reducing the diffusion intensity in high-gradient regions (i.e., defect edge regions) to avoid blurring of edge information. Through this processing method, random noise and background interference in the image are effectively suppressed while the structural edge features of weld defects are well preserved. After completing the anisotropic diffusion processing, morphological filtering processing is further performed on the processed image. In one embodiment, the morphological filtering processing is designed based on multi-directional structural elements, which include linear structural elements of different lengths, directions, and scales, used to match possible strip cracks, linear incomplete fusion defects, and approximately circular porosity defects in the weld image. By performing a combination of morphological opening and closing operations on the image, small noise points can be removed to a certain extent, while enhancing the defect response consistent with the morphology of structural elements, thus improving the contrast of strip-shaped and point-shaped defects. After morphological filtering, a defect enhancement response image is generated based on the differences between the images before and after filtering. In one implementation, the enhanced local response region can be extracted using difference operations or high-hat transform, thereby highlighting the small defect regions in the original image that have weak responses but structural continuity, making the grayscale response intensity of the defect region significantly higher than that of the background region. The final generated defect enhancement response image, while suppressing noise interference, enhances the detectability of small cracks, porosity, and other local defects in the weld, providing a clearer input image basis for subsequent threshold segmentation and defect recognition.

[0029] Furthermore, defect type discrimination involves comprehensively analyzing the spatial distribution morphology and structural characteristics of defects based on the extraction of candidate defect regions and calculation of geometric parameters, thereby achieving the classification and quantitative characterization of weld defects. Specifically, firstly, connected component analysis is performed on the binary image obtained from defect detection. In one embodiment, pixel-level connectivity scanning of the binary image divides the spatially interconnected pixel set into independent connected components, each corresponding to an independent candidate defect region, thus obtaining multiple spatially separated initial defect regions. Subsequently, the defect regions are clustered based on spatial proximity relationships to form defect clustering regions. In one embodiment, the geometric center or centroid of each connected component is used as the spatial representation point. Based on the Euclidean distance or relative positional relationship between each defect region, when the spatial distance between two or more defect regions is less than a preset threshold, they are determined to have spatial correlation and merged into the same cluster, thus obtaining defect clustering regions, which are used to characterize spatially interconnected defect sets or those exhibiting a certain distribution pattern. After obtaining the defect clustering regions, feature parameters are calculated for each clustering region. In one embodiment, the feature parameters include, but are not limited to, defect quantity, coverage area, distribution density, spatial distribution characteristics, and main distribution direction. Specifically: the defect quantity represents the total number of independent defects within the clustered region; the coverage area characterizes the spatial circumference or minimum envelope area of ​​the clustered region; the distribution density describes the concentration of defects per unit area; the spatial distribution characteristics reflect the uniformity or clustering of defects within the region; and the main distribution direction, obtained through statistical analysis of the defect center points, characterizes the overall extension direction of the defects, for example, by obtaining the principal axis direction using principal component analysis. After obtaining the above feature parameters, they are matched with preset defect type discrimination rules. In one embodiment, the preset discrimination rules are established based on the typical morphological characteristics of ship weld defects to distinguish different types of defects, such as dense defects, chain defects, or locally clustered defects. By comparing the feature parameters with rule threshold conditions, the defect type to which the defect clustered region belongs is determined. Finally, based on the discrimination results, the corresponding defect type identifier and its corresponding feature parameters are output and used for subsequent weld quality assessment processing. By adopting the above method, the defect type identification is not only based on the local features of a single defect, but also combined with the spatial distribution pattern of defects, thereby improving the accuracy and stability of defect classification under complex weld conditions.

[0030] S4: Based on the preset classification society's weld radiographic inspection evaluation standards, classify and evaluate the identified defects, and generate a structured inspection report that meets the requirements of the specifications.

[0031] Furthermore, step S4 of this application also includes: S41. Load the classification society's weld radiographic inspection evaluation standard into the system, and convert the judgment clauses in the evaluation standard into an executable rule set; S42. Match the defect type and corresponding parameter information with the rule set to determine the applicable evaluation clauses; S43. Based on the evaluation clauses, group the defects and calculate the cumulative parameters to obtain the defect evaluation results; S44. Associate the defect evaluation results with the weld identification information and generate an electronic inspection report that meets the specifications.

[0032] Specifically, based on preset classification society weld radiographic inspection evaluation standards, the defect results identified in step S3 are automatically graded and evaluated, and a structured inspection report is generated to achieve standardized judgment and traceable record output of weld quality. Classification society weld radiographic inspection evaluation standards are loaded into the system; these classification societies include, but are not limited to, CCS, DNV, ABS, or LR. These evaluation standards are typically given in text or specification clause form, including allowable defect size, defect quantity limits, defect spacing requirements, and defect grouping rules. To enable automated computer processing, the judgment clauses in the evaluation standards are structurally transformed into an executable rule set. In one embodiment, the rule set is expressed in a condition-conclusion form, where the condition part includes defect type, size parameters, and spatial distribution characteristics, and the conclusion part corresponds to the defect grade judgment result or the pass / fail judgment result, thereby mapping the standard clauses to machine-executable logic. After completing the rule set construction, rule matching processing is performed based on the defect type and corresponding geometric parameter information output in step S3. In one implementation, the system performs a rule-by-rule matching and filtering based on information such as defect type identifier, defect size parameters, defect location, and distribution characteristics to determine the assessment clause that best matches the current defect. This matching process automatically associates different types of defects with different assessment standards, thus avoiding errors and inconsistencies caused by manual table lookups. After determining the applicable assessment clause, the defects are grouped and cumulative parameters are calculated. In one implementation, multiple defects that are spatially related or meet standard grouping conditions are divided into the same assessment group to comply with the requirements for defect combination assessment in classification society standards. Simultaneously, according to the assessment clause, cumulative parameters are calculated for defects within the same group, including total defect length, total number of defects, maximum single defect size, and defect spacing. By comparing the above cumulative parameters with rule thresholds, the corresponding defect assessment result is obtained, such as pass, exceedance, or downgrade. After completing the defect assessment, the assessment result is associated with weld identification information. In one embodiment, the weld identification information includes weld number, component segmentation information, inspection batch information, and inspection time information, used to establish a unique correspondence between the inspection results and specific engineering objects. Subsequently, a structured electronic inspection report is generated based on the associated evaluation results. In another embodiment, the electronic inspection report includes defect type, defect parameters, evaluation level, applicable standard clauses, and evaluation basis explanations, and is output in a structured format according to a preset format to meet the data integration requirements of the ship quality management system. Through the above methods, the entire process of weld radiographic inspection results, from standard analysis and rule matching to graded evaluation and report generation, is automated, improving the standardization, consistency, and traceability of inspection evaluation.

[0033] This application embodiment constructs a fully automated processing system for digital radiographic inspection of ship welds, sequentially realizing a collaborative processing mechanism from image acquisition, quality assessment, defect identification to standardized evaluation and report generation. This achieves automatic detection, automatic classification, and automatic grading of weld defects without requiring manual frame-by-frame interpretation and evaluation, and outputs structured inspection reports that meet classification society requirements. This improves inspection efficiency and evaluation consistency, reduces human subjective error, and enhances the standardization and automation level of ship weld quality control. In the specific implementation process, this application first acquires digital radiographic images of complex ship weld structures. These images are acquired based on digital radiographic imaging (DR), computed radiographic imaging (CR), or a combination thereof. Furthermore, considering the geometric characteristics of different structural forms such as hull curved plates, T-shaped corner joints, and pipe joints, the imaging layout and exposure parameters are adaptively adjusted to ensure that effective imaging data meeting inspection requirements can be obtained for different structural weld areas. Subsequently, the acquired digital radiographic images undergo automated image quality evaluation. This evaluation incorporates spatial resolution and sensitivity indicators specific to weld defect detection requirements. Simultaneously, it analyzes the grayscale dynamic range coverage integrity in conjunction with the characteristics of plate thickness variations, thereby determining whether the image meets the basic conditions for subsequent defect identification and assessment. Based on this, images meeting quality standards proceed to the defect detection and identification stage. During defect identification, multi-scale feature extraction and response enhancement are performed on the image to obtain feature information reflecting changes in weld defect morphology, size, and direction. Defect candidate regions are extracted through threshold segmentation and connected component analysis. Then, combined with geometric parameters and welding process information, defect types are identified and dimensions are measured, achieving automated defect identification and structured representation. Finally, based on the defect identification results, according to the pre-set classification society weld radiographic inspection assessment standards, defect types and their parameter information are matched and grouped with standard rules to achieve automatic defect classification and assessment. A structured inspection report containing defect information, assessment criteria, and mapping relationships with standard clauses is generated, thus achieving automated and standardized processing of the entire ship weld radiographic inspection process. In summary, the embodiments of this application not only improve the automation level and inspection efficiency of the radiographic inspection process of ship welds, but also significantly enhance the stability and consistency of defect identification results. They achieve fully automated processing from image acquisition, quality evaluation, defect detection to standardized grading assessment and report generation, thereby effectively reducing subjective errors caused by manual interpretation, improving the traceability and standard compliance of inspection results, and meeting the engineering application requirements of ship weld quality control.

[0034] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A digital radiographic inspection and automatic evaluation method for ship welds, characterized in that, The method includes: S1. Obtain digital radiographic images of welds in complex ship structures, including hull curved plates, T-shaped corner joints, and pipe joints. The digital radiographic images are obtained through digital radiographic imaging (DR), computed radiographic imaging (CR), or a combination thereof, and the imaging layout and imaging parameters are adjusted according to the geometric characteristics of the complex ship structure. S2. Evaluate the imaging quality of the digital X-ray image. The evaluation indicators include at least the spatial resolution sensitivity set based on the weld defect detection requirements and the grayscale dynamic range coverage integrity for changes in the thickness of the ship plate. S3. Detect, identify, determine the type of, and measure the dimensions of weld defects in digital radiographic images that have passed quality evaluation; S4. Based on the preset classification society's weld radiographic inspection evaluation standards, classify and evaluate the identified defects, and generate a structured inspection report that meets the requirements of the specifications.

2. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 1, characterized in that, Obtain digital radiographic images of welds in complex ship structures, including hull curved plates, T-shaped corner joints, and pipe joints, including: S11. For T-section fillet welds and welded joints between bulb flats and plates, double-wall imaging or oblique radiography is used, and the radiography angle is calculated based on the three-dimensional model data of the weld. S12. For curved plate welds, a flexible detector array or CR imaging plate is used to be bent and attached to make the imaging surface fit the curved surface of the workpiece. S13. In areas with limited space, use crawling robots or magnetic adsorption X-ray equipment to perform long-distance or lateral radiography and transmit image data wirelessly. S14. When a single exposure cannot cover the dynamic range of weld thickness, use multi-energy DR imaging or multi-exposure parameter CR imaging to obtain images of different areas. S15. For pipe nodes or large complex structures, control the mobile X-ray device to perform imaging at multiple preset angles, and perform three-dimensional reconstruction processing based on multi-view images to obtain three-dimensional volume data of the weld area.

3. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 1, characterized in that, The imaging quality of the digitized X-ray image is evaluated, including: S21. Before conducting quality assessment, adaptive preprocessing is performed on the acquired digital X-ray images; S22. Inspect and assess the integrity of the image quality indicator, serial number, and positioning mark in the image to determine whether they meet the requirements of the ship inspection regulations. S23. Evaluate the visibility of the image quality meter wire in the image and calculate its signal-to-noise ratio to determine whether the image resolution meets the requirements of the ship specifications. S24. Evaluate the visibility of the smallest identifiable structure in the image to determine whether the detection sensitivity meets the detection requirements for workpieces of corresponding thickness. S25. Detect and evaluate artifacts and scattering interference in the image, and quantify their impact. S26. Analyze the dynamic range of the image based on its grayscale distribution to determine whether it covers the thickness range of the weld and the base material, so as to ensure that the defect has sufficient contrast.

4. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 3, characterized in that, Before quality assessment, the acquired digital X-ray images undergo adaptive preprocessing, including: S21-1. Acquire digital X-ray inspection images that meet the quality assessment requirements, including single-image images and multiple images under different exposure conditions, and organize the images into an image sequence according to exposure conditions, camera angle, and acquisition order. ,in For exposure condition index, For camera view indexing, For the collection order index, These are the pixel coordinates in the image; S21-2. Extract noise statistical features for each image in the image sequence, including the mean pixel grayscale value. and pixel grayscale variance And automatically generate corresponding denoising algorithm parameters based on the noise statistical characteristics. , used for adaptive denoising, where the local block-level noise variance is defined as: ； in In pixels A local image patch centered on the image. This represents the average grayscale value of the block. The number of pixels within the block. For pixel coordinate indices within a local image patch; S21-3. Based on the denoising algorithm parameters, perform denoising operation on each image in the image sequence, including processing images under different exposure conditions, and record the denoising parameters and processing results of each image. S21-4. Perform feature point detection and matching on denoised images from different camera perspectives, establish constraints by combining weld geometric reference information, output spatially aligned image set, and record the correspondence between the registration matrix and the original image. S21-5. The registered images are stitched together according to their spatial coordinates and acquisition order. Gray-level fusion is performed on the overlapping areas. The gray-level fusion formula is as follows: ； in, ， For local gray-level gradient weights, The thickness resolution weighting coefficients are used to record the coordinates of the stitching region and the fusion parameters, generating a stitched image set; S21-6. Perform a high dynamic range fusion algorithm on image sequences under different exposure conditions in the stitched image set to generate HDR images. The formula for calculating pixel grayscale values ​​is: ； in, The first after splicing Second exposure image; S21-7. Establish an index table for preprocessed images and original images, and record the correspondence between the denoising, registration, stitching and HDR fusion parameters of each image and the original image, so as to associate the original image and the preprocessed image in the defect assessment process.

5. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 4, characterized in that, Based on the denoising algorithm parameters, a denoising operation is performed on each image in the image sequence, including: S21-3-1. Perform imaging pattern recognition on each image in the image sequence and generate corresponding imaging pattern identification parameters; S21-3-2. Based on the imaging mode identification parameters, perform multi-scale decomposition processing on the image to extract noise feature parameters at different scales. S21-3-3. Based on the noise feature parameters, perform initial reconstruction processing on the image to generate an initial image; S21-3-4. Based on the imaging mode identification parameters and noise feature parameters, select the corresponding denoising processing path from the preset denoising strategy set and generate denoising algorithm parameters. S21-3-5. Based on the denoising algorithm parameters and the selected denoising processing path, adaptive denoising processing is performed on the initial image to obtain a denoised result image, wherein the denoising intensity is dynamically adjusted in combination with the local gray-level distribution features of the image during the denoising process. S21-3-6. Perform local abnormal noise suppression processing on the denoised result image to obtain an optimized denoised image, so as to remove isolated noise points and maintain image structure information. S21-3-7. Record the denoising algorithm parameters, denoising processing path and processing result image of each image, and establish the index correspondence between the denoised result image and the original image.

6. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 1, characterized in that, The process involves detecting and identifying weld defects, determining their types, and measuring their dimensions on digital radiographic images that have passed quality evaluation. S31. Perform multi-scale decomposition processing on the digital X-ray image, extract image response information at different scales and directions, and fuse the response information at each scale to generate a multi-scale response image. The multi-scale response image is used to characterize the shape, size and direction change features of weld defects. S32. Enhance the multi-scale response image, including local gradient enhancement of the high-frequency response region to highlight the edge features of microcracks and needle-like pores; contrast enhancement of the low-frequency response region to increase the grayscale difference between the defect region and the background region; and fusion processing of the enhanced response image to generate a defect response enhancement image. S33. Perform threshold segmentation on the defect response enhancement image to obtain a binary image of the defect candidate region; perform connected component labeling on the binary image, extract the spatial location, contour boundary and area parameters of each defect candidate region, and generate a defect region set; S34. For the set of defect regions, calculate the geometric parameters for each defect region, including the region area, aspect ratio, boundary complexity, and grayscale distribution characteristics; combine the welding process parameters and base material thickness information to identify the defect type and generate the corresponding defect type identifier and judgment criteria. S35. Perform consistency verification on the defect type discrimination results, including merging duplicate detected defects, merging spatially overlapping defects, and removing abnormal isolated defects, to generate the final defect detection result set.

7. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 6, characterized in that, Enhancement processing of the multi-scale response image includes: The multi-scale response image is subjected to multi-scale decomposition processing to obtain sub-band coefficients at different scales and directions; Adaptive enhancement processing is performed on the high-frequency components in the sub-band coefficients to enhance defect edges and texture features and suppress background interference; The low-frequency components in the sub-band coefficients are subjected to contrast enhancement processing to improve the overall grayscale contrast of the image. The coefficients of each sub-band after enhancement are reconstructed to obtain the enhanced digital ray image.

8. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 6, characterized in that, Enhancement processing of the multi-scale response image includes: Anisotropic diffusion processing is performed on the multi-scale response image to preserve defect edge features while suppressing noise; Morphological filtering is performed on images after anisotropic diffusion processing based on multi-directional structural elements to enhance the features of strip-shaped and point-shaped defects. Defect enhancement response images are generated based on the filtering results to improve the response intensity of small defect areas.

9. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 6, characterized in that, The determination of defect type includes: Connectivity analysis is performed on the binary image obtained from defect detection to obtain multiple independent defect regions. These defect regions are then clustered based on spatial proximity to obtain defect clustering regions. For the defect clustering region, its characteristic parameters are calculated, including the number of defects, coverage area, distribution density, spatial distribution characteristics, and main distribution direction; Based on the feature parameters and preset discrimination rules, the defect clustering region is classified. The corresponding defect type and its characteristic parameters are output based on the discrimination results and used for subsequent evaluation and processing.

10. The digital radiographic inspection and automatic evaluation method for ship welds as described in claim 1, characterized in that, Based on the pre-set classification society's weld radiographic inspection evaluation standards, the identified defects are graded and evaluated, and a structured inspection report conforming to the specifications is generated, including: S41. Load the classification society's weld radiographic inspection evaluation standard into the system, and convert the judgment clauses in the evaluation standard into an executable set of rules; S42. Based on the type of defect and the corresponding parameter information, match it with the rule set to determine the applicable evaluation clauses; S43. Based on the aforementioned evaluation criteria, the defects are grouped and cumulative parameters are calculated to obtain the defect evaluation results; S44. Link the defect assessment results with the weld identification information and generate an electronic inspection report that meets the specifications.

Images