Titanium rod surface wear morphology detection method and system based on multispectral image fusion

Abstract

The present application provides a method and system for detecting the wear morphology of a titanium rod surface based on multispectral image fusion, which relates to the field of image processing technology. The method includes: connecting a spectrometer, clamping a target titanium rod, and performing multispectral image acquisition by controlling a rotating table; setting fusion dimensions, taking multi-scale enhancement and confidence perception as processing goals, developing an image fusion device and deploying it on a detection platform; based on synchronization timestamp constraints, importing the multispectral image into the detection platform, performing adaptive fusion processing according to the image fusion device to determine the fused image, performing wear positioning and background assimilation processing as the wear morphology detection result. This application solves the technical problem that due to the heterogeneity of multispectral images, direct fusion will lead to information loss or the introduction of noise, thereby affecting the accuracy of titanium rod surface wear morphology detection. By developing an image fusion device to fuse multispectral images, the accuracy of titanium rod surface wear morphology detection is improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and system for detecting the surface wear morphology of a titanium rod based on multispectral image fusion. Background Art

[0002] With the development of imaging technology, multispectral imaging technology has gradually become an emerging method in surface defect detection. Multispectral imaging technology can comprehensively capture the changes in different physical properties of the surface by combining image data from different spectral bands (such as visible light, infrared, ultraviolet, etc.). Although the introduction of multispectral imaging provides new possibilities for titanium rod surface detection, since multispectral images are imaged by light of different wavelengths, there is a large spectral heterogeneity, and images of each wavelength may have different characteristics, such as resolution, contrast, and noise level. When these images are directly fused, key features will be lost or noise will be introduced, affecting subsequent image analysis and processing, such as the detection of titanium rod surface wear morphology, resulting in insufficient detection accuracy and efficiency.

[0003] In summary, there is a technical problem in the existing technology that due to the heterogeneity of multispectral images, direct fusion will lead to information loss or introduction of noise, thereby affecting the accuracy of titanium rod surface wear morphology detection. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for detecting the surface wear morphology of titanium rods based on multispectral image fusion, so as to solve the technical problem in the prior art that due to the heterogeneity of multispectral images, direct fusion will lead to information loss or introduction of noise, thereby affecting the accuracy of titanium rod surface wear morphology detection.

[0005] In view of the above problems, the present application provides a titanium rod surface wear morphology detection method and system based on multispectral image fusion.

[0006] In the first aspect, the present application provides a titanium rod surface wear morphology detection method based on multispectral image fusion, and the titanium rod surface wear morphology detection method based on multispectral image fusion is implemented by a titanium rod surface wear morphology detection system based on multispectral image fusion, wherein the titanium rod surface wear morphology detection method based on multispectral image fusion includes: connecting a spectrometer, clamping the target titanium rod and performing multispectral image acquisition under multi-view constraints by controlling a rotating table; setting a fusion dimension, taking multi-scale enhancement and confidence perception as processing goals, developing an image fusion device and embedding it in a detection platform, wherein the fusion dimension includes at least a data dimension, a feature dimension and a decision dimension, introducing cross-modal attention into the feature dimension, and processing goals for multi-scale homogeneous enhancement and confidence weighting of common elements, and directional enhancement and confidence verification of complementary elements; based on a synchronous timestamp constraint, the multispectral image is imported into the detection platform, adaptive fusion processing is performed according to the image fusion device to determine the fused image, wear positioning and background assimilation processing based on corner window recognition are performed, and a pop-up window display is generated on the detection platform as the wear morphology detection result.

[0007] Optionally, a fusion mode is determined according to the fusion dimension; fusion layers are constructed for the fusion modes corresponding to the respective fusion dimensions, the fusion layers are cascaded and a lateral interaction channel is constructed to determine the image fuser.

[0008] Optionally, a fusion requirement is received, wherein the fusion requirement includes at least one fusion dimension; according to the fusion requirement, a fusion layer of the image fuser is activated, and processing based on common elements and complementary elements under the fusion dimension is performed in parallel to determine the fused image.

[0009] Optionally, in the processing of common elements and complementary elements under the fusion dimension, multi-scale homogeneity enhancement and confidence weighting of the common elements are performed in parallel, including: for any fusion dimension, identifying the common elements of the multispectral image and differential scale processing based on homogeneity, and determining a first enhancement result, wherein the differential scale is determined based on the difference between the homogeneity standard and the image standard; for the first enhancement result, determining N fusion groups through multi-image mapping; traversing the N fusion groups, performing confidence weighted fusion processing, and determining the common element fusion result.

[0010] Optionally, for each fusion group, the fusion weight of the features in each group is determined through confidence checking, and the fusion process is performed with the fusion weight as a constraint.

[0011] Optionally, in the parallel execution of the processing of common elements and complementary elements based on the fusion dimension, directional enhancement and confidence verification of the complementary elements are performed, including: introducing an element recognition baseline, wherein the element recognition baseline is a quality standard; based on the element recognition baseline being directional, the complementary elements are enhanced to determine a second enhancement result; for the second enhancement result, confidence verification is performed to determine a complementary element fusion result.

[0012] Optionally, according to the distribution of image space elements, the common element fusion result and the complementary element fusion result are spatially distributed and spliced ​​to determine the fused image.

[0013] Optionally, the fused image is identified, and a corner point window is determined by performing corner point identification, wherein the pixel trend is taken as the corner point, and the corner point window is a preset pixel scale; the fused image is marked with the corner point window, wherein the corner point window marks the non-smooth points on the surface of the titanium rod.

[0014] Optionally, the wear morphology area is framed according to the distribution of the identified corner point windows; for the fused image, background assimilation is performed on the non-wear morphology area to serve as the wear morphology distribution; the titanium rod defect database is connected, the wear morphology distribution is matched and marked accordingly, and determined as the wear morphology detection result, wherein the titanium rod defect database is built into the detection platform.

[0015] In the second aspect, the present application also provides a titanium rod surface wear morphology detection system based on multispectral image fusion, which is used to execute the titanium rod surface wear morphology detection method based on multispectral image fusion as described in the first aspect, wherein the titanium rod surface wear morphology detection system based on multispectral image fusion includes: an image acquisition module, which is used to connect the spectrometer, clamp the target titanium rod and perform multispectral image acquisition under multi-view constraints by controlling the turntable; an image fusion module, which is used to set the fusion dimension, develop an image fusion device with multi-scale enhancement and confidence perception as the processing goals, and embed it in the detection platform, wherein the fusion dimension includes at least data dimension, feature dimension and decision dimension, introduce cross-modal attention into the feature dimension, and the processing goal is multi-scale homogeneous enhancement and confidence weighting of common elements, and directional enhancement and confidence verification of complementary elements; a wear morphology detection module, which is used to import the multispectral image into the detection platform based on the synchronization timestamp constraint, perform adaptive fusion processing according to the image fusion device to determine the fused image, perform wear positioning and background assimilation processing based on corner window recognition, and generate a pop-up window display on the detection platform as the wear morphology detection result.

[0016] One or more technical solutions provided in this application have at least the following beneficial effects:

[0017] By connecting a spectrometer, clamping the target titanium rod, and controlling a rotating stage, multispectral image acquisition under multi-view constraints is performed. Fusion dimensions are set, with multi-scale enhancement and confidence perception as processing goals. An image fusion device is developed and embedded in the detection platform. The fusion dimensions include at least data, feature, and decision dimensions, introducing cross-modal attention into the feature dimension. The processing goals are multi-scale homogeneity enhancement and confidence weighting of common elements, and directional enhancement and confidence verification of complementary elements. Based on synchronized timestamp constraints, the multispectral image is imported into the detection platform. Adaptive fusion processing is performed on the image fusion device to determine the fused image. Wear localization and background assimilation based on corner window recognition are performed, and the wear topography detection results are displayed in a pop-up window on the detection platform. In other words, through multispectral image acquisition, the titanium rod surface is scanned and analyzed from multiple dimensions. An image fusion device is developed to set multiple dimensional branches. Based on demand, the corresponding dimensional branches are triggered to perform targeted processing, fusion, and spatial splicing of common and directional elements, thereby improving the accuracy and reliability of surface wear topography detection results.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, which can be implemented in accordance with the contents of the description, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically listed below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in this application or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and a person of ordinary skill in the art can obtain other drawings based on the provided drawings without creative work.

[0020] Figure 1 This is a flow chart of a titanium rod surface wear morphology detection method based on multispectral image fusion in this application;

[0021] Figure 2 This is a schematic diagram of the structure of the titanium rod surface wear morphology detection system based on multispectral image fusion in this application.

[0022] Description of the accompanying drawings: image acquisition module 11, image fusion module 12, wear morphology detection module 13. DETAILED DESCRIPTION

[0023] This application provides a method and system for detecting the wear and morphology of titanium rod surfaces based on multispectral image fusion, which solves the technical problem in the prior art that due to the heterogeneity of multispectral images, direct fusion will lead to information loss or the introduction of noise, thereby affecting the accuracy of titanium rod surface wear and morphology detection. Through multispectral image acquisition, the titanium rod surface is scanned and analyzed from multiple dimensions, and an image fusion device is developed to set multiple dimensional branches. Based on demand, the corresponding dimensional branches are triggered to perform targeted processing, fusion, and spatial splicing of common elements and directional elements, thereby improving the accuracy and reliability of surface wear and morphology detection results.

[0024] Below, the technical solutions in this application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this application, rather than all the embodiments of this application. It should be understood that this application is not limited to the example embodiments described herein. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should also be noted that, for the convenience of description, only the parts related to this application, rather than all of them, are shown in the accompanying drawings.

[0025] For example, see the attached Figure 1 The present application provides a titanium rod surface wear morphology detection method based on multispectral image fusion, wherein the titanium rod surface wear morphology detection method based on multispectral image fusion is performed by a titanium rod surface wear morphology detection system based on multispectral image fusion, and the titanium rod surface wear morphology detection method based on multispectral image fusion specifically includes the following steps:

[0026] S100: Connect the spectrometer, clamp the target titanium rod and perform multispectral image acquisition under multi-view constraints by controlling the rotation stage.

[0027] Specifically, the target titanium rod is fixed to the test platform by a clamping device to ensure that the titanium rod remains stable during the acquisition process to avoid image distortion due to external interference. The turntable is usually driven by a precision motor to rotate the target object (the titanium rod in this case) to different angles. The rotation angle is usually set to 10° every 5 seconds to obtain surface information from different angles. The more angles, the more viewing angle information can be provided. The spectrometer is installed in a suitable position and connected to the detection platform to ensure that the image information can be captured correctly. Different bands of light (such as visible light, infrared and ultraviolet) are selected, and images of multiple bands are obtained simultaneously through the spectrometer. Each band of image can provide different levels of surface information. For example, visible light can help capture the morphology of the surface, while infrared images can show the heat distribution generated by friction.

[0028] A spectrometer is a device that can measure the intensity of light or electromagnetic radiation at different wavelengths. It is widely used in the acquisition of multispectral images. It operates in different bands according to different application scenarios, such as ultraviolet light, visible light, infrared, etc. It is used to collect multispectral images of the target titanium rod surface and obtain reflection or emission information in different bands.

[0029] During the image acquisition process, the rotational motion is controlled to capture images of the titanium rod surface from multiple angles, overcoming wear or defects that cannot be observed from a single angle. At each viewing angle, the spectrometer captures images of the titanium rod surface in different wavelengths, simultaneously acquiring image data from various angles in multiple spectral bands.

[0030] For example, the wear condition of a target titanium rod surface is relatively complex. Not only is there slight wear on the surface, but there are also local cracks. A spectrometer equipped with infrared and visible light bands was used. The rotation angle of the turntable was set to collect data every 15°. Image data of 4 bands were used during the collection process. There are a total of 20 angles, and image data of 4 bands are collected at each angle, resulting in a total of at least 100 images. The infrared band image shows the temperature distribution, revealing that the heat generated by friction is unevenly distributed on the surface of the titanium rod; the visible light image shows the microscopic features of the surface wear area, revealing the depth and morphology of the surface wear; the ultraviolet band image shows the surface microcracks; the near-infrared band image shows the local temperature rise area caused by intensified friction.

[0031] Through multi-view, multi-band image acquisition, a comprehensive understanding of the wear morphology of the titanium rod surface is achieved, ensuring that no minor wear or cracks are missed. Multi-view acquisition avoids information loss due to the limitations of a single view, especially in complex or irregular wear areas, ensuring the reliability of the test results.

[0032] S200: Set the fusion dimension, take multi-scale enhancement and confidence perception as the processing goals, develop an image fusion device and embed it in the detection platform, wherein the fusion dimension at least includes the data dimension, feature dimension and decision dimension, introduces cross-modal attention into the feature dimension, and the processing goals are the homogeneous enhancement and confidence weighting of common elements at multiple scales, and the directional enhancement and confidence verification of complementary elements.

[0033] Furthermore, the present application S200 includes:

[0034] According to the fusion dimension, a fusion mode is determined; a fusion layer is constructed for each fusion mode corresponding to each fusion dimension, the fusion layer is cascaded and a lateral interaction channel is constructed to determine the image fuser.

[0035] Specifically, multiple fusion dimensions are set to ensure that information at different levels can be effectively fused. Multiple fusion dimensions are different data levels or angles considered when fusing multiple spectral images, including data dimension, feature dimension and decision dimension. Each dimension corresponds to a different processing level in the image fusion process, ensuring that the fused image can retain more key information, so that accurate analysis can be performed. The data dimension involves the fusion of the original image data level, fusing data from different bands. Through the fusion of data dimensions, data from different bands are spliced ​​or merged to form a comprehensive multi-band data set. The feature dimension is fused based on the features extracted from the image (such as edges, textures, morphology, etc.) to distinguish different types of surface wear or defects. The decision dimension determines the detection decision based on the fused feature data, such as whether there is wear, the location and degree of wear.

[0036] Select the appropriate fusion method based on the different fusion dimensions. A fusion method refers to the specific fusion method used within each fusion dimension, including weighted averaging, max pooling, concatenation, and contrast enhancement. Data dimension fusion primarily targets image data from different bands or viewpoints. Corresponding fusion methods include maximum, minimum, and weighted average fusion. The maximum method selects the maximum value across all images at each pixel location. This is used to enhance certain image features, such as edges and textures. The minimum method selects the minimum value across all images for each pixel. This is used to preserve darker areas and, under certain conditions, remove noise. Weighted average fusion assigns different weights to images of different bands and fuses them together using a weighted average. For each pixel, the pixel values ​​of each source image (different spectrum, viewpoint, etc.) are weighted averaged. For example, infrared images may provide more information about certain worn areas, while visible light images provide clearer details. Therefore, different weights can be assigned to different bands.

[0037] Feature fusion focuses on features extracted from the image, such as edges, textures, and morphology. Corresponding fusion methods include feature concatenation and max pooling. Feature concatenation combines features (such as textures and edges) extracted from various bands or viewpoints to generate a comprehensive feature vector. Max pooling selects the strongest eigenvalues ​​from image features across multiple bands or viewpoints, retaining the most representative features. This method is suitable for areas with significant wear and tear.

[0038] Decision dimension fusion primarily focuses on the process of making a final decision based on different features. Corresponding fusion methods include voting mechanisms and weighted decision-making. Voting mechanisms use multiple models or multiple dimensions (such as the characteristics of different bands) to determine the final outcome. For example, if a worn area is detected in all three bands, a voting mechanism is used to confirm that the area is indeed worn. Weighted decision-making uses different dimensions (such as the confidence level of different bands) to weight the decision output of each dimension.

[0039] Based on the fusion method corresponding to each dimension, a corresponding independent fusion layer is constructed. The purpose of each fusion layer is to process and fuse image information for a specific dimension. For the fusion layer of the data dimension, the input layer involves multiple channels and receives multi-source data (images of different perspectives, spectral bands, etc.); the processing layer processes the input image according to the selected fusion method; and the output layer outputs the processed data in the form of an image. Using the weighted averaging method, a dynamic weight layer is added to the fusion layer. This layer adjusts the fusion weights of different data sources based on image quality, signal-to-noise ratio, or other indicators. In the processing layer, a weighted averaging algorithm is used to first standardize the input image, calculate the weight coefficient of each image, and then perform pixel-level weighted summation.

[0040] For the feature-dimensional fusion layer, a corresponding layer is designed. Multiple input feature maps are concatenated along the feature dimension (i.e., the channel dimension) through a feature concatenation layer to generate a new feature map. In other words, all input feature maps are concatenated along the channel dimension to form a larger feature map. This concatenated feature map typically contains all features from different sources and can be fed into a convolutional layer for further feature extraction. A weighted summation layer applies weights to each input feature map based on its weight, and then performs a weighted sum. Each input feature map is pooled using a max pooling layer, and the pooled feature maps are merged as needed. Pooling typically reduces spatial dimensionality and retains the most significant features. The fused feature map is then output through the output layer. Furthermore, when constructing the feature fusion layer, consideration should be given to optimizing the expressiveness of the output feature map. Adding convolutional layers can further extract and fuse different features. Adding a batch normalization layer can help stabilize the training process and mitigate vanishing or exploding gradients. Finally, the constructed feature fusion layer will output a fused feature map, which contains the comprehensive information from multiple input feature maps.

[0041] Since the fusion task of feature dimensions is relatively complex, especially when dealing with modal differences, the representation of features may be different under different input modes. In order to enhance the effect of feature fusion, a cross-modal attention mechanism is introduced to allow the network to adaptively select important features and eliminate irrelevant features, thereby improving the quality of feature fusion. The cross-modal attention mechanism can selectively enhance the features of important modalities by calculating attention weights based on the feature importance of each modality, thereby improving the fusion effect. For example, in the detection of wear morphology on the surface of titanium rods, infrared images may be better able to capture surface temperature changes, while visible light images can provide clearer information. Therefore, the weight of important modalities can be increased through the cross-modal attention mechanism.

[0042] There are multiple options for integrating decision dimensions, and the specific method chosen generally depends on the application scenario and specific requirements. Voting is suitable for classification tasks and is generally implemented in two ways: majority voting and weighted voting. Majority voting selects the category with the most predictions from multiple models as the final decision. Weighted voting assigns different weights to the predictions of each model, making the votes of models with larger weights more influential. This reduces the impact of individual model errors and improves stability. If voting is used, a voting mechanism is created for each input model or task output. The outputs of all models are collected and then either majority voting or weighted voting is performed. Weighted voting dynamically assigns weights by learning the confidence level of each model. This is accomplished through simple counting or weighted accumulation operations, and the voting results are finally output.

[0043] When the predictions of multiple models are close and the differences are small, weighted voting can be used to adjust the influence of each model in the final decision based on its confidence level. Weighted voting not only improves the stability of the overall decision but also reduces the impact of occasional erroneous models. Especially when multiple models perform well, weighted voting can smooth and optimize the overall results. When the predictions between models differ significantly, majority voting is generally more effective. In this case, the judgments of multiple models that are relatively consistent are generally more reliable, while the judgments of a small number of models may have significant deviations. By allowing the majority model's choice to determine the final result, the majority voting mechanism can effectively avoid the influence of extreme predictions. In particular, when some models may overfit or misidentify, the influence of minority votes is effectively suppressed.

[0044] Weighted summation fuses the outputs of different models. Each model's output is multiplied by a weight and then summed to produce the final output. Based on the model's performance on the training set, specific tasks, or user input, each decision source (i.e., the output of each model) is assigned a weight. The output of each model is multiplied by the corresponding weight and summed to produce a fused decision result.

[0045] Fusion layer cascading involves sequentially connecting fusion layers of different dimensions (such as data, feature, and decision fusion layers). In other words, to combine these fusion layers into a comprehensive image fuser, the three fusion layers need to be cascaded, connecting the outputs of the fusion layers of different dimensions together to form a composite fusion structure. Lateral interaction channels establish information transfer channels between multiple dimensions, allowing different dimensions (such as data, feature, and decision dimensions) to communicate and share information with each other, strengthening the connection between each layer. This ensures that the output of each layer not only depends on the input of the current dimension but also reflects information from other dimensions, thereby improving the overall fusion effect.

[0046] Through lateral interaction channels, the fusion results from the data dimension can interact with the fusion results from the feature dimension and the decision dimension to improve the fusion effect of each layer. For example, the data dimension is adjusted by the features extracted from the feature dimension, the feature dimension optimizes the feature extraction strategy based on the output of the decision dimension, and the decision dimension uses information from both the data and feature dimensions to make more accurate decisions, ultimately improving the overall fusion effect. When implementing these interactions, methods such as skip connections or residual connections are used to make the information flow between different layers smoother, avoiding information loss or bottlenecks. After optimization through cascading and lateral interaction channels, the final output will be a multi-dimensional, multi-level fused image representation that can simultaneously contain information from the data, feature, and decision layers.

[0047] Through the design of hierarchical cascades and lateral interactive channels, an image fuser is obtained, which combines all information in the data dimension, feature dimension and decision dimension, and can generate high-quality fused images. It can not only efficiently fuse data from different sources, but also integrate information through hierarchical connections and interactive channels, making the image fuser have stronger expression and decision-making capabilities.

[0048] Multi-scale enhancement and confidence perception are used as processing objectives to enhance image information at different scales. In particular, when detecting wear on the surface of titanium rods, it is important to capture wear features of varying sizes and shapes. Confidence perception adjusts fusion weights based on the reliability of different regions in the image.

[0049] For shared features, the processing objectives are homogeneous enhancement and confidence weighting. Shared features refer to similar or consistent features, such as shape, texture, and structure, that can be found across different viewpoints and modalities (e.g., visible light and infrared images). Homogeneous enhancement of shared features means uniformly improving the performance of features at multiple scales, while confidence weighting weights these features based on the importance or reliability of different regions. This determines the contribution of each region to the final fusion result, ensuring that high-confidence regions dominate the fusion process. Confidence estimates are performed on different regions based on clarity, contrast, and texture information. Edge regions and regions with rich textures are considered to have higher confidence and are therefore assigned greater weights. In other words, multiscale enhancement enhances the performance of specific features in an image by processing the image at multiple scales. Homogeneous enhancement uniformly enhances features of the same type at multiple scales, making the performance of these shared features more consistent and clear across scales.

[0050] For complementary elements, the processing objectives are directional enhancement and confidence verification. Complementary elements refer to features in images from different perspectives or modalities that complement each other and provide different information. Unlike common elements, complementary elements are generally not fully captured through a single modality or perspective, providing multiple layers of information for image analysis. Directional enhancement employs different processing strategies based on the characteristics of each piece of information, performing directional enhancement on certain regions or features within the image. For example, this can enhance features highly correlated with wear areas in an image. Confidence verification verifies the reliability of each region after fusion to ensure the reliability of the final fusion result. Confidence verification is performed on image quality metrics (such as clarity and contrast). This involves evaluating image quality indicators, determining the imaging quality of each region, and estimating a confidence score. For example, regional clarity and contrast are calculated, and multiple metrics are normalized to the same scale (e.g., 0-1) using the maximum-minimum normalization method. A weight is assigned to each metric based on its importance, and a weighted calculation is performed on the weights and multiple metrics to obtain a confidence score. Typically, these weights are determined based on the importance of each metric in titanium rod surface wear detection, combined with historical experience. If the confidence score of a region is high, the region is considered reliable, while for regions with lower confidence, additional verification is required, such as multiple sampling or image comparison, to confirm whether wear is actually present in the region.

[0051] The developed image fuser is embedded in the detection platform. Whenever the detection platform collects new image data, it immediately fuses the image across the data, feature, and decision dimensions to generate the final fused image. By setting the fusion dimension and triggering the corresponding dimension branch according to different needs, it performs targeted processing, fusion, and spatial stitching of common and directional elements, improving image fusion results and ensuring more accurate final detection results.

[0052] S300: Based on the synchronization timestamp constraint, the multispectral image is imported into the detection platform, the image fusion device performs adaptive fusion processing to determine the fused image, and wear positioning and background assimilation processing based on corner window recognition are performed. As the wear morphology detection result, a pop-up window display is generated on the detection platform.

[0053] Specifically, the synchronization timestamp constraint means that when performing multispectral image acquisition, all image acquisition devices (such as spectrometers, cameras, sensors, etc.) are required to collect data at the same time point (timestamp), avoiding image data inconsistencies or misalignments caused by time differences, and ensuring the correct spatial alignment of multi-view or multispectral images. Multispectral images with the same timestamp are imported into the detection platform. The image fusion device in the detection platform automatically selects the appropriate fusion layer for activation based on the characteristics of the image and the fusion requirements, and performs the processing of common elements and complementary elements under the fusion dimension in parallel. Based on the distribution of elements in space, the spatial distribution of the fusion results of elements and the fusion results of complementary elements are spliced ​​to determine the fused image.

[0054] After obtaining the fused image, corner window recognition is performed to identify pixels with significant changes in the image, that is, defects or wear areas on the surface of the titanium rod. After identifying the corner window, background assimilation processing is performed to highlight the wear area. The pixels in the background area are replaced with a blank background to highlight the wear morphology area. Finally, according to the wear morphology area, the corresponding wear morphology distribution is matched in the titanium rod defect database to form the final detection result, including wear morphology, defect type, location, etc. For example, scratches and microcracks are identified on the surface of the titanium rod, and these areas are highlighted through background assimilation processing. A pop-up window will appear on the detection platform, including information such as the presence of scratches on the surface of the titanium rod (X axis = 100mm, Y axis = 50mm) and microcracks (X axis = 150mm, Y axis = 75mm).

[0055] By synchronizing timestamp constraints, image data from different sources are ensured to be synchronized in time. Through corner window recognition and background assimilation processing, the wear morphology of the titanium rod surface is accurately identified, and by highlighting the worn area, the visibility and accuracy of the detection results are improved.

[0056] Furthermore, the present application S300 includes:

[0057] A fusion requirement is received, wherein the fusion requirement includes at least one fusion dimension; and according to the fusion requirement, a fusion layer of the image fuser is activated, and common elements and complementary elements under the fusion dimensions are processed in parallel to determine the fused image.

[0058] Specifically, the system receives specific requirements for image fusion processing, including the dimensions to be fused and the fusion target. Based on the received fusion requirements, it determines the required fusion dimensions and activates the corresponding fusion layers in the image fuser, including the data fusion layer, feature fusion layer, and decision fusion layer. Each fusion layer processes different fusion dimensions. For example, if the requirements include enhancing worn areas and highlighting crack features, the system identifies the data dimension (processing different spectral data) and the feature dimension (processing surface wear and cracks).

[0059] After activating the corresponding fusion layer, the common and complementary elements within the fusion dimension are processed in parallel. Common elements refer to features or information shared between different modalities or data sources, while complementary elements refer to features with complementary relationships between different modalities. The fusion results of the common and complementary elements are spatially spliced ​​based on the distribution of image spatial elements to determine the fused image.

[0060] For example, suppose two different spectral images are used when inspecting titanium rod surface defects: one showing global wear characteristics, and the other highlighting crack areas. A fusion request is received to enhance the wear area and highlight the crack characteristics. Based on the request, the fusion layers of the data dimension and feature dimension are activated. The data dimension is responsible for processing different spectral information, while the feature dimension is responsible for highlighting wear and crack characteristics. Parallel processing is performed to enhance the wear area (common elements) and the crack area (complementary elements). After parallel processing, the two are fused into a final image. The result is that the final image clearly shows the surface wear and cracks of the titanium rod, while integrating the characteristic information under different spectra.

[0061] The fused image can fully display all key features (such as wear and cracks), help analyze the surface condition of the titanium rod, significantly improve image processing effects, and accurately identify and locate various defects.

[0062] Furthermore, the present application further comprises the following steps:

[0063] In parallel, the processing of common elements and complementary elements based on the fusion dimension is performed, and multi-scale homogeneity enhancement and confidence weighting of common elements are performed, including:

[0064] For any fusion dimension, the common elements of the multispectral image and the differential scale processing based on homogeneity are identified to determine a first enhancement result, wherein the differential scale is determined based on the difference between the homogeneity standard and the image standard; for the first enhancement result, N fusion groups are determined through multi-image mapping; the N fusion groups are traversed, and the confidence-weighted fusion processing is performed to determine the common element fusion result.

[0065] For each fusion group, the fusion weight of the features in each group is determined through confidence checking, and the fusion processing is performed with the fusion weight as a constraint.

[0066] Specifically, for multi-scale homogeneous enhancement and confidence weighting of shared elements: For each fusion dimension, shared elements in the multispectral image are identified, that is, similar or consistent parts in different spectral image modalities, such as surface texture or crack characteristics. Shared elements are typically basic structural features in the image that have similar appearances in different image modalities. Through differential scaling, these shared elements are homogeneously enhanced at different scales, and image enhancement is performed in areas with similar features, ensuring that image quality is improved in similar areas without introducing excessive noise or irrelevant information.

[0067] Homogeneity standards refer to determining the processing method based on similar or consistent areas in the image. They are used to identify the scales of similar areas in the image and perform different scale processing on these areas to extract features at different levels. Image standards are the characteristic expressions of the image itself at different scales. Based on the homogeneity standard, areas with similar features in the image are identified, and the image is processed at different scales. The differences between the images at larger scales (macro level) and smaller scales (micro level) are compared to obtain the difference, that is, the difference or change calculated between different scales. Differential scaling processing is a multi-scale image processing method that analyzes changes in the image and extracts important information by comparing and processing image features at different scales (such as large scale and small scale).

[0068] By processing the differences of the image at different scales, a preliminary image enhancement result, namely the first enhancement result, is obtained. It represents the state of the common elements after multi-scale enhancement and is the preliminary image enhancement effect obtained after differential scale processing. Through multi-image mapping, the first enhancement result is mapped with other image information to form N fusion groups. Through multi-image mapping, features at the same position (such as cracks, worn areas) can be aligned to avoid information loss or incorrect fusion. For example, there are two images from different perspectives, showing the same part at different angles. Through multi-image mapping, the two images are aligned so that the features at the same position in each image (such as cracks) can be aligned.

[0069] After multi-image mapping, multiple images are obtained, each of which may contain different spectral information, different perspective information, or information at different scales. Through multi-image mapping, N fusion groups are determined. Each group contains information from the same location or features across different images. The goal is to fuse these image features to obtain richer and more accurate image content. A fusion group refers to a subset of images combined from multiple images or different sources during the image fusion process.

[0070] For each fusion group, confidence checking is used to assign appropriate fusion weights to the features within each fusion group, reflecting the credibility of each information source in the fusion process. Confidence checking determines the weight of each feature in the final fusion based on its reliability. By evaluating the features within each fusion group, the confidence of each feature is calculated, and then the weights of each feature are adjusted based on this confidence. Information sources with higher confidence are assigned higher weights, while information sources with lower confidence are assigned lower weights. Based on the fusion weights corresponding to each fusion group, the fusion groups are fused to obtain a fusion result with shared elements. Features within each fusion group are weighted and fused according to their corresponding fusion weights, ensuring that high-weight features have a greater impact on the final fusion result and that the most important and reliable features dominate the final image.

[0071] For example, assume three images are collected from different spectral bands (infrared, ultraviolet, and visible light). Image 1 is from visible light and has an image quality assessment clarity of 0.8; Image 2 is from infrared light and has an image quality assessment clarity of 0.7; and Image 3 is from ultraviolet light and has an image quality assessment clarity of 0.6. Common features in these three images include wear areas, cracks, and microcracks on the titanium rod surface. Based on the difference between the image standard and the homogeneity standard, a differential scale is calculated. Image 1 has an edge contrast of 1.2, a homogeneity standard of 0.8, and a differential scale of 1.2-0.8=0.4. Image 2 has an edge contrast of 0.9, a homogeneity standard of 0.7, and a differential scale of 0.9-0.7=0.2. Image 2 has an edge contrast of 0.7, a homogeneity standard of 0.6, and a differential scale of 0.7-0.6=0.1. Based on the size of the differential scale, Image 1 has the strongest enhancement effect, followed by Image 2, and Image 3 has the weakest. Multi-image mapping is performed based on the different features of the three images. By observing the characteristics of different images, the following three fusion groups were determined: Group 1 is the fusion of image 1 and image 2, focusing on enhancing the display of surface wear areas and cracks; Group 2 is the fusion of image 2 and image 3, focusing on highlighting cracks and microcracks; Group 3 is the fusion of image 1 and image 3, strengthening the overall texture of the titanium rod surface and the display of microcracks.

[0072] According to the clarity score of each image, the fusion weight of the images in each fusion group is calculated. In group 1, the clarity of image 1 is 0.8, and the clarity of image 2 is 0.7. By calculating the total clarity to be 1.5, the calculated weight of image 1 is 0.8 / 1.5, which is approximately equal to 0.53, and the weight of image 2 is 0.47. Images 1 and 2 are weighted fused, and the weight of image 1 is higher. Therefore, in the final image, the details of the wear area of ​​image 1 are enhanced; in group 2, the clarity of image 2 is 0.70, and the clarity of image 3 is 0.60. The calculated weight of image 2 is 0.54, and the weight of image 3 is 0.46. After weighted fusion of images 2 and 3, the crack details of image 2 are better preserved, and the microcrack display of image 3 is enhanced; in group 3, the clarity of image 1 is 0.80, and the clarity of image 3 is 0.60. The calculated weight of image 1 is 0.57, and the weight of image 3 is 0.43. After the fusion of images 1 and 3, the overall texture and microcrack details are clearer.

[0073] Image fusion is performed through differential scaling and confidence correction, and the weight is determined at each stage according to the confidence of the features. The final fused image is obtained, which significantly reduces the impact of noise and enhances the effective information in the image, thereby achieving more accurate surface wear detection.

[0074] Furthermore, the present application further comprises the following steps:

[0075] Parallel execution of the processing of common elements and complementary elements based on the fusion dimension, performing directional enhancement and confidence verification of complementary elements, including:

[0076] An element recognition baseline is introduced, wherein the element recognition baseline is a quality standard; based on the element recognition baseline being directional, the complementary elements are enhanced to determine a second enhancement result; and confidence verification is performed on the second enhancement result to determine a complementary element fusion result.

[0077] Specifically, a feature recognition baseline is introduced, setting a quality standard to guide the enhancement of complementary features. This defines the quality and representation of certain features in an image (such as cracks, scratches, and surface unevenness). For example, cracks must be at least 5 pixels wide, their contrast must reach 20% of the image's average contrast, scratches must be at least 3 pixels wide, and background noise must be within ±10% of the image's standard deviation.

[0078] Based on the determined feature recognition baseline, complementary features in the image are enhanced in a targeted manner. Based on the requirements of the feature recognition baseline, different complementary features are amplified or strengthened to facilitate the extraction of these complementary features from the image. Features that appear in different spectral channels in the multispectral image are enhanced, especially those that can provide complementary information. The portions of the image containing complementary information are strengthened. For example, assuming a thermal defect is identified in the infrared image and a scratch is identified in the visible light image, both the thermal defect and the scratch are amplified based on predefined quality criteria (such as crack width and scratch contrast).

[0079] The second enhancement result is the image obtained after enhancing the complementary elements, highlighting the valuable complementary information more prominently than the initial image. After the second enhancement result is processed, confidence verification is performed. This checks whether the enhanced area meets expectations and whether the enhancement can provide real, valid feature information. Confidence verification detects whether the enhanced area exceeds the normal range and whether unreliable information has been introduced due to noise or processing errors. For example, if the crack area is over-enhanced, resulting in unnatural boundaries, or if too many false signals are introduced during the enhancement process, the confidence verification algorithm needs to be used to exclude these unqualified areas. After targeted enhancement and confidence verification, a reliable complementary element fusion result is obtained, which integrates the complementary information of multiple spectral channels and has been verified to ensure its quality. For example, a threshold is set based on the degree of enhancement of each complementary element to determine whether it should be included in the final fusion result. If the confidence level of the enhanced crack feature is greater than 80%, the crack feature can be considered valid and enter the final fusion process.

[0080] Through directional enhancement, the complementary features in different channels can be highlighted, making small wear or cracks more clearly visible in the fused image. Confidence verification is used to remove noise and irrelevant information, ensuring that the final image reaches a high level of quality and information accuracy.

[0081] Furthermore, the present application further comprises the following steps:

[0082] According to the distribution of image space elements, the common element fusion result and the complementary element fusion result are spatially distributed and spliced ​​to determine the fused image.

[0083] Specifically, based on the spatial distribution of image features (such as surface cracks and worn areas), including their distribution pattern and density, the shared and complementary fusion results are spatially spliced ​​to ensure proper spatial alignment. The shared and complementary fusion results each represent distinct feature regions within the image. First, it is necessary to determine which regions are shared and which are complementary.

[0084] Common elements usually represent common features in the image (such as overall wear or surface roughness), while complementary elements represent special or difficult-to-identify areas (such as cracks or tiny defects). Through spatial stitching, the information of common elements and complementary elements are combined to obtain an image that comprehensively reflects all features, ensuring that all important information in the image can be displayed while avoiding information loss or mis-stitching. If the boundary between the common elements and complementary elements is relatively clear, the two can be directly spliced ​​together without causing obvious transition problems. If the boundary between the common elements and complementary elements is fuzzy or has a certain overlap, a weighted approach is used to ensure that the two can transition smoothly during the stitching process, such as using Gaussian weighting or boundary blurring technology to achieve a smooth transition at the stitching point and avoid abrupt boundaries.

[0085] Spatial distribution stitching combines the features of two or more images based on their spatial location. In other words, it combines information from different regions of multiple images into a single image to form a cohesive whole. The resulting fused image, after spatial stitching and processing, incorporates information from both shared and complementary elements, providing a more complete and accurate representation of the target object (e.g., the surface of a titanium rod).

[0086] By spatially stitching common and complementary elements, an image is formed that displays multiple features such as wear and cracks on the titanium rod surface, identifying small cracks and wear areas and reducing the possibility of false detection and missed detection.

[0087] Furthermore, the present application further comprises the following steps:

[0088] Identify the fused image and determine a corner point window by performing corner point identification, wherein the pixel trend is used as the corner point and the corner point window is a preset pixel scale; use the corner point window to mark the fused image, wherein the corner point window marks the non-smooth points on the surface of the titanium rod.

[0089] According to the distribution of the identified corner point windows, the wear morphology area is framed; for the fused image, the background assimilation is performed on the non-wear morphology area to serve as the wear morphology distribution; the titanium rod defect database is connected, the wear morphology distribution is matched and marked accordingly, and determined as the wear morphology detection result, wherein the titanium rod defect database is built into the detection platform.

[0090] Specifically, corner recognition is performed on the fused image to detect significant feature points in the image, which are usually manifested as areas with large changes in the image. Corner points are a special type of image feature that represents the point with the most significant change in the image. Usually, the angle appears at the intersection of edges in two different directions or in a strong contrast area in the image. The surface of the titanium rod should be smooth under normal circumstances, which means that there are no obvious scratches, cracks or other defects on its surface. Therefore, when certain areas in the image show significant pixel changes, it usually means that there are flaws or defects in these areas.

[0091] Corner recognition methods are used to locate areas of significant change in images. Corner detection algorithms (such as Harris corner detection and Shi-Tomasi corner detection) are used to identify points in the image where pixels change dramatically. These points usually appear at the edges or corners of the image.

[0092] After the corner points are identified, a window with a preset pixel scale is set based on the position of each corner point. The corner point window is a fixed-size area defined around each corner point. It is used to capture local features around the corner point and analyze the changes in details in the image. For example, if the pixel scale is set to 5x5, the window size of each corner point is 5x5 pixels. The changes in pixel values ​​within these windows are further analyzed to check for surface unevenness, cracks, etc. For example, if the position of a corner point in the image is (100,150) and the pixel scale is 5x5, then the range of the corner point window will be the area from (98,148) to (102,152).

[0093] Based on the distribution of the corner point windows, the corner points are connected to form a rectangular or polygonal frame, which represents the location of the defect area. Based on the setting of the corner point windows, areas on the titanium rod surface that may contain defects such as cracks, wear, or dents are marked. By identifying multiple corner point windows and marking them on the image, a frame of the defect area is formed. Ultimately, the image of the titanium rod surface will display all defect areas located by corner point recognition.

[0094] Through corner point recognition and window calibration, we identify areas on the titanium rod surface where defects may exist, and define the areas corresponding to these corner points, which are known as wear or flaw points. The wear morphology area refers to the area on the titanium rod surface damaged by friction, pressure, and other factors, typically showing scratches, dents, cracks, or an uneven surface structure.

[0095] Background assimilation is performed on the non-worn regions of the fused image, changing the pixel values ​​in these regions to a uniform background color (such as white or transparent). This distinguishes the non-worn regions from the worn regions, thereby increasing the contrast of the worn regions. In the processed image, the non-worn regions may be set to a white or transparent background, while the worn regions retain their original texture and color, making the worn regions more prominent and easier to observe.

[0096] Compare and match the framed wear morphology area with the known defects in the titanium rod defect database to determine whether the wear area conforms to the known defect pattern and mark it. If similar features are detected, the corresponding defect type is extracted from the database and marked to obtain the wear morphology detection results, including the specific location of the wear area, defect type and other information. The wear morphology detection results are displayed on the detection platform in graphical or text form to understand the status of the titanium rod surface. The titanium rod defect database is a data set built into the detection platform, which records different types of titanium rod defect feature information, which is used to compare, match and mark the wear morphology of the titanium rod surface to help determine and identify specific defect types.

[0097] By framing the wear topography and performing background assimilation, the wear area is extracted, eliminating the influence of interference factors and improving the accuracy of wear detection. Background assimilation effectively highlights the wear area, significantly improving the contrast between the wear area and the background, and facilitating the automatic identification and labeling of defect types in the wear topography area, thereby improving detection efficiency.

[0098] In summary, the titanium rod surface wear morphology detection method based on multispectral image fusion provided in this application has the following beneficial effects:

[0099] By connecting a spectrometer, clamping the target titanium rod, and controlling a rotating stage, multispectral image acquisition under multi-view constraints is performed. Fusion dimensions are set, with multi-scale enhancement and confidence perception as processing goals. An image fusion device is developed and embedded in the detection platform. The fusion dimensions include at least data, feature, and decision dimensions, introducing cross-modal attention into the feature dimension. The processing goals are multi-scale homogeneity enhancement and confidence weighting of common elements, and directional enhancement and confidence verification of complementary elements. Based on synchronized timestamp constraints, the multispectral image is imported into the detection platform. Adaptive fusion processing is performed on the image fusion device to determine the fused image. Wear localization and background assimilation based on corner window recognition are performed, and the wear topography detection results are displayed in a pop-up window on the detection platform. In other words, through multispectral image acquisition, the titanium rod surface is scanned and analyzed from multiple dimensions. An image fusion device is developed to set multiple dimensional branches. Based on demand, the corresponding dimensional branches are triggered to perform targeted processing, fusion, and spatial splicing of common and directional elements, thereby improving the accuracy and reliability of surface wear topography detection results.

[0100] Example 2, based on the same inventive concept as the titanium rod surface wear morphology detection method based on multispectral image fusion in the above-mentioned Example 1, this application also provides a titanium rod surface wear morphology detection system based on multispectral image fusion, please refer to the attached Figure 2 The titanium rod surface wear morphology detection system based on multispectral image fusion includes:

[0101] The image acquisition module 11 is used to connect the spectrometer, clamp the target titanium rod and perform multispectral image acquisition under multi-view constraints by controlling the rotating table; the image fusion module 12 is used to set the fusion dimension, take multi-scale enhancement and confidence perception as the processing goals, develop an image fusion device and embed it in the detection platform, wherein the fusion dimension at least includes data dimension, feature dimension and decision dimension, introduces cross-modal attention into the feature dimension, and the processing goal is multi-scale homogeneous enhancement and confidence weighting of common elements, and directional enhancement and confidence verification of complementary elements; the wear morphology detection module 13 is used to import the multispectral image into the detection platform based on the synchronization timestamp constraint, perform adaptive fusion processing according to the image fusion device to determine the fused image, perform wear positioning and background assimilation processing based on corner window recognition, and generate a pop-up window display on the detection platform as the wear morphology detection result.

[0102] Furthermore, the image fusion module 12 in the titanium rod surface wear morphology detection system based on multispectral image fusion is also used for:

[0103] According to the fusion dimension, a fusion mode is determined; a fusion layer is constructed for each fusion mode corresponding to each fusion dimension, the fusion layer is cascaded and a lateral interaction channel is constructed to determine the image fuser.

[0104] Furthermore, the wear profile detection module 13 in the titanium rod surface wear profile detection system based on multispectral image fusion is also used for:

[0105] A fusion requirement is received, wherein the fusion requirement includes at least one fusion dimension; and according to the fusion requirement, a fusion layer of the image fuser is activated, and common elements and complementary elements under the fusion dimensions are processed in parallel to determine the fused image.

[0106] Furthermore, the wear profile detection module 13 in the titanium rod surface wear profile detection system based on multispectral image fusion is also used for:

[0107] In parallel, the processing of common elements and complementary elements based on the fusion dimension is performed, and multi-scale homogeneity enhancement and confidence weighting of common elements are performed, including:

[0108] For any fusion dimension, the common elements of the multispectral image and the differential scale processing based on homogeneity are identified to determine a first enhancement result, wherein the differential scale is determined based on the difference between the homogeneity standard and the image standard; for the first enhancement result, N fusion groups are determined through multi-image mapping; the N fusion groups are traversed, and the confidence-weighted fusion processing is performed to determine the common element fusion result.

[0109] Furthermore, the wear profile detection module 13 in the titanium rod surface wear profile detection system based on multispectral image fusion is also used for:

[0110] For each fusion group, the fusion weight of the features in each group is determined through confidence checking, and the fusion processing is performed with the fusion weight as a constraint.

[0111] Furthermore, the wear profile detection module 13 in the titanium rod surface wear profile detection system based on multispectral image fusion is also used for:

[0112] Parallel execution of the processing of common elements and complementary elements based on the fusion dimension, performing directional enhancement and confidence verification of complementary elements, including:

[0113] An element recognition baseline is introduced, wherein the element recognition baseline is a quality standard; based on the element recognition baseline being directional, the complementary elements are enhanced to determine a second enhancement result; and confidence verification is performed on the second enhancement result to determine a complementary element fusion result.

[0114] Furthermore, the wear profile detection module 13 in the titanium rod surface wear profile detection system based on multispectral image fusion is also used for:

[0115] According to the distribution of image space elements, the common element fusion result and the complementary element fusion result are spatially distributed and spliced ​​to determine the fused image.

[0116] Furthermore, the wear profile detection module 13 in the titanium rod surface wear profile detection system based on multispectral image fusion is also used for:

[0117] Identify the fused image and determine a corner point window by performing corner point identification, wherein the pixel trend is used as the corner point and the corner point window is a preset pixel scale; use the corner point window to mark the fused image, wherein the corner point window marks the non-smooth points on the surface of the titanium rod.

[0118] Furthermore, the wear profile detection module 13 in the titanium rod surface wear profile detection system based on multispectral image fusion is also used for:

[0119] According to the distribution of the identified corner point windows, the wear morphology area is framed; for the fused image, the background assimilation is performed on the non-wear morphology area to serve as the wear morphology distribution; the titanium rod defect database is connected, the wear morphology distribution is matched and marked accordingly, and determined as the wear morphology detection result, wherein the titanium rod defect database is built into the detection platform.

[0120] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1 The titanium rod surface wear morphology detection method based on multispectral image fusion and the specific examples in Example 1 are also applicable to the titanium rod surface wear morphology detection system based on multispectral image fusion in this embodiment. Through the above detailed description of the titanium rod surface wear morphology detection method based on multispectral image fusion, those skilled in the art can clearly understand the titanium rod surface wear morphology detection system based on multispectral image fusion in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here.

[0121] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0122] Obviously, for those skilled in the art, several improvements and modifications can be made to the present application without departing from the principles of the present application, and these improvements and modifications also fall within the scope of protection of the present application.

Claims

1. A titanium rod surface wear morphology detection method based on multispectral image fusion is characterized by: include: Connect the spectrometer, clamp the target titanium rod and perform multispectral image acquisition under multi-view constraints by controlling the rotation stage; Set fusion dimensions, with multi-scale enhancement and confidence perception as processing goals, develop an image fusion device and embed it in the detection platform. The fusion dimensions include at least data, feature, and decision dimensions. Introduce cross-modal attention into the feature dimension. The processing goals are multi-scale homogeneous enhancement and confidence weighting of common elements, and directional enhancement and confidence verification of complementary elements. Based on the synchronization timestamp constraint, the multispectral image is imported into the detection platform, and the image fusion device performs adaptive fusion processing to determine the fused image. Wear positioning and background assimilation processing based on corner window recognition are performed, and a pop-up window display is generated on the detection platform as the wear morphology detection result.

2. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 1 is characterized in that: Develop image fuser, including: Determining a fusion method according to the fusion dimension; A fusion layer is constructed for each fusion method corresponding to each fusion dimension, the fusion layer is cascaded and a lateral interaction channel is constructed to determine the image fuser.

3. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 1 is characterized in that: The method further comprises: performing adaptive fusion processing according to the image fuser to determine a fused image, comprising: receiving a fusion requirement, wherein the fusion requirement includes at least one fusion dimension; According to the fusion requirement, the fusion layer of the image fuser is activated, and processing based on common elements and complementary elements under the fusion dimension is performed in parallel to determine the fused image.

4. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 3 is characterized in that: In parallel, the processing of common elements and complementary elements based on the fusion dimension is performed, and multi-scale homogeneity enhancement and confidence weighting of common elements are performed, including: For any fusion dimension, identifying common elements of the multispectral image and performing differential scaling based on homogeneity to determine a first enhancement result, wherein the differential scaling is determined based on a difference between a homogeneity standard and an image standard; For the first enhancement result, determining N fusion groups through multi-image mapping; The N fusion groups are traversed, a confidence-weighted fusion process is performed, and a common element fusion result is determined.

5. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 4 is characterized in that: For each fusion group, the fusion weight of the features in each group is determined through confidence checking, and the fusion processing is performed with the fusion weight as a constraint.

6. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 4 is characterized in that: Parallel execution of the processing of common elements and complementary elements based on the fusion dimension, performing directional enhancement and confidence verification of complementary elements, including: Introducing an element identification baseline, wherein the element identification baseline is a quality standard; According to the element recognition baseline being directional, the complementary element is enhanced to determine a second enhancement result; A confidence verification is performed on the second enhancement result to determine a complementary element fusion result.

7. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 6 is characterized in that: According to the distribution of image space elements, the common element fusion result and the complementary element fusion result are spatially distributed and spliced ​​to determine the fused image.

8. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 7 is characterized in that: Perform wear location based on corner window recognition, including: Identify the fused image and determine a corner point window by performing corner point identification, wherein the corner point is defined as a pixel trend and the corner point window is a preset pixel scale; The fused image is marked using the corner point window, wherein the corner point window marks the non-smooth points on the surface of the titanium rod.

9. The titanium rod surface wear morphology detection method based on multispectral image fusion according to claim 8, characterized in that: After marking the fused image, the following steps are performed: According to the distribution of the marked corner windows, the wear morphology area is framed; For the fused image, background assimilation is performed on the non-wear morphology area to obtain the wear morphology distribution; Connecting to a titanium rod defect database, matching and correspondingly marking the wear morphology distribution, and determining it as the wear morphology detection result, wherein the titanium rod defect database is built into the detection platform.

10. Titanium rod surface wear morphology detection system based on multispectral image fusion, characterized in that: The method for detecting the surface wear and morphology of a titanium rod based on multispectral image fusion according to any one of claims 1 to 9 is implemented, wherein the system for detecting the surface wear and morphology of a titanium rod based on multispectral image fusion comprises: An image acquisition module is used to connect to the spectrometer, clamp the target titanium rod, and perform multispectral image acquisition under multi-view constraints by controlling the rotation stage; An image fusion module is used to set fusion dimensions, with multi-scale enhancement and confidence perception as processing goals. An image fuser is developed and embedded in the detection platform. The fusion dimensions include at least data, feature, and decision dimensions. Cross-modal attention is introduced into the feature dimension. The processing goals are multi-scale homogeneous enhancement and confidence weighting of common elements, and directional enhancement and confidence verification of complementary elements. The wear profile detection module is used to import the multispectral image into the detection platform based on the synchronization timestamp constraint, perform adaptive fusion processing according to the image fusion device to determine the fused image, perform wear positioning and background assimilation processing based on corner point window recognition, and generate a pop-up window display on the detection platform as the wear profile detection result.

Images