A bulldozer tooth block surface defect identification method and system

Abstract

The application provides a bulldozer tooth block surface defect identification method and system, and relates to the technical field of bulldozer tooth block surface defect identification. The method comprises the following steps: determining a light mode sequence for bulldozer tooth block surface detection; based on the light mode sequence, acquiring multiple tooth block surface images under different light modes by image acquisition on the bulldozer tooth block surface; respectively extracting features of the multiple tooth block surface images to obtain feature information corresponding to each image; obtaining fused feature information by fusing the feature information corresponding to each image; and identifying the rust type and rust area of the bulldozer tooth block surface according to the fused feature information. The method aims to solve the problem that the rust type of the bulldozer tooth block surface is various, accompanied by pollutants, and single fixed light cannot effectively distinguish, overcome the limitations of information acquisition in the existing technology in complex surface defect detection, and significantly improve the accuracy and robustness of detection.

Description

Technical Field

[0001] This invention relates to the field of bulldozer tooth block surface defect identification technology, and more specifically, to a method and system for identifying bulldozer tooth block surface defects. Background Technology

[0002] Bulldozers, as heavy construction machinery, operate year-round in extremely harsh environments such as mines and construction sites. Their toothed blocks, as key working components that directly contact materials, endure immense mechanical stress and are continuously exposed to corrosive environments such as moisture, acids, or alkalis. This makes the toothed block surface highly susceptible to chemical corrosion, resulting in rust of varying forms, stages of development, and severity. This rust not only significantly reduces the material strength and wear resistance of the toothed blocks, accelerating their wear and failure, but also directly impacts the overall operating efficiency, safety, and service life of the bulldozer. Therefore, accurate and efficient detection of rust on the surface of bulldozer toothed blocks is an indispensable and crucial step in routine equipment maintenance and early warning inspections.

[0003] Traditional methods for detecting tooth block corrosion mainly rely on manual visual inspection. This method is highly subjective, and the results are easily affected by the experience and judgment of the inspectors. At the same time, it is inefficient and cannot meet the needs of large-scale, high-frequency inspections. In addition, the complex and variable ambient light (such as direct sunlight, shadows, dust, etc.) seriously interferes with manual visual judgment, further reducing the accuracy and reliability of the inspection.

[0004] To overcome the inherent limitations of manual inspection, the industry has widely adopted automated inspection solutions based on image data processing. These solutions typically consist of image acquisition equipment and image processing algorithm modules, enabling the identification and location of corrosion through automated analysis of images of the tooth block surface. To ensure the quality and consistency of image acquisition and effectively eliminate interference from complex ambient light, existing inspection systems generally employ an active, specific lighting scheme. This scheme typically uses a specialized light source device integrating multiple light-emitting units, which is placed close to the surface of the tooth block during inspection to create a locally enclosed lighting environment. By precisely controlling the brightness, direction, and even spectral characteristics of the internal light-emitting units, a stable and optimized lighting condition is provided for the image sensor, aiming to maximize the contrast between corrosion features and the background.

[0005] However, in practical applications, the complexity of rust on bulldozer tooth blocks far exceeds expectations. Rust is not simply reddish-brown iron oxide; it can manifest as various compounds, such as hydrated iron oxide (common red rust), magnetite (black rust), or sulfides formed in specific mining environments. These different types of rust exhibit significant differences in optical properties. Furthermore, rust develops in different stages, from initially microscopic oxide films that are difficult to detect with the naked eye, to dotted and flaky rust, and finally to severe layered flaking rust, each with varying surface roughness, color saturation, gloss, and other optical characteristics.

[0006] The limitation of existing specific lighting schemes is that they typically select lighting conditions that are considered effective in highlighting "typical" reddish-brown rust based on experience or preset parameters, such as using visible light of a specific wavelength or polarized light at a specific angle. Such single, fixed, and preset lighting conditions may perform well in detecting "typical" rust, but their detection capability is greatly reduced when faced with atypical rust (such as black rust, early oxide film) or surface contaminants with similar optical characteristics to rust.

[0007] Specifically, when black rust (Fe3O4) is present on the surface of the tooth block, its color is dark. Under normal or red rust-optimized lighting conditions, black rust may appear similarly dark in the image to oil, dirt, or wear marks on the tooth block surface, making it difficult for subsequent image processing algorithms to effectively distinguish, thus leading to missed or misjudged black rust. Similarly, for early-formed, extremely thin oxide films, their color changes are subtle, and their surface gloss may be similar to that of clean metal surfaces. Under specific lighting designed to highlight macroscopic corrosion, the contrast of these minute, early-stage corrosions is insufficient, making them easily ignored by image processing algorithms.

[0008] A further layer of complexity lies in the fact that bulldozer teeth inevitably accumulate various non-corrosive contaminants during operation, such as ore dust, oil stains, mud, or coating flaking. Under certain lighting conditions, the image features (such as color, brightness, and texture) of these contaminants may be highly similar to certain types of rust (e.g., laterite mud versus red rust, black oil versus black rust). Since existing specific lighting schemes are usually fixed and cannot be dynamically adjusted according to the actual surface condition of the object being measured or the types of defects that may exist, the image data acquired during the image acquisition stage itself is at risk of information confusion. This severely limits the ability of image processing algorithms to identify complex and varied types of rust. While algorithms may focus on the characteristics of typical rust under a specific lighting condition during training, when rust or interfering objects with different optical properties appear in actual detection, the robustness and accuracy of the algorithm will significantly decrease due to the lack of sufficient discriminative information in the image.

[0009] Therefore, designing a solution that can adapt to the complex and varied rust types and interferences on the surface of bulldozer tooth blocks, and dynamically adjust the lighting conditions according to the detection requirements to optimize image information acquisition, thereby effectively improving the comprehensiveness and accuracy of tooth block surface rust detection in extreme operating environments, is a key technical problem that urgently needs to be solved in the current bulldozer maintenance field. Summary of the Invention

[0010] The purpose of this invention is to provide a method and system for identifying surface defects on bulldozer tooth blocks, aiming to solve the problems of diverse types of rust on the surface of bulldozer tooth blocks, accompanied by contaminants, and the inability to effectively distinguish them under a single fixed light illumination. This invention overcomes the limitations of existing technologies in information acquisition for complex surface defect detection, and significantly improves the accuracy and robustness of detection.

[0011] In a first aspect, the present invention provides a method for identifying surface defects in bulldozer tooth blocks, comprising the following steps: Determine the lighting pattern sequence for bulldozer tooth block surface inspection; the lighting pattern sequence contains multiple sets of lighting parameters, each set of lighting parameters is used to highlight specific types of surface features or suppress specific types of interfering features; The image acquisition device controls the image acquisition of bulldozer teeth blocks based on a sequence of illumination patterns, and acquires multiple images of the teeth block surface under different illumination patterns; the image acquisition device adopts a partially enclosed illumination structure. Feature extraction is performed on multiple images of the tooth block surface to obtain the feature information corresponding to each image; By fusing the feature information corresponding to each image, fused feature information is obtained; Based on the fused feature information, the type and area of ​​rust on the surface of the bulldozer tooth block are identified.

[0012] The method for identifying surface defects on bulldozer tooth blocks provided by this invention forms a sequence of lighting modes by pre-setting different combinations of lighting parameters, and sequentially acquires images of the tooth block surface to obtain multiple images with different information emphases. Subsequently, feature extraction and fusion are performed on these multimodal images. By utilizing the complementarity of rust features under different lighting modes, a feature description with discriminative power for various types of rust and contaminants is constructed, thereby improving detection accuracy.

[0013] Secondly, the present invention provides a bulldozer tooth block surface defect identification system, comprising: The determination module is used to determine the sequence of illumination patterns for bulldozer tooth block surface detection; The acquisition module controls the image acquisition device, which acquires multiple images of the bulldozer tooth block surface under different lighting modes by acquiring images of the tooth block surface based on the lighting pattern sequence; the image acquisition device adopts a locally enclosed lighting structure. The extraction module is used to extract features from multiple images of the tooth block surface to obtain the feature information corresponding to each image; The fusion module is used to obtain fused feature information by fusing the feature information corresponding to each image; The identification module is used to identify the type and area of ​​rust on the surface of bulldozer tooth blocks based on fused feature information.

[0014] As can be seen from the above, the bulldozer tooth block surface defect identification method provided by this invention acquires optical response information of the bulldozer tooth block surface in different dimensions by designing and applying multispectral (visible light, near-infrared) and multi-angle (orthogonal, oblique) illumination pattern sequences. By extracting and fusing features from these multimodal images, a comprehensive feature description that can effectively distinguish between black rust and oil stains and accurately detect the initial oxide film is successfully constructed. Thanks to the acquisition of illumination pattern sequences and multimodal feature fusion, the comprehensiveness and accuracy of rust detection on the bulldozer tooth block surface are significantly improved, and it can effectively distinguish between different types of rust (such as black rust) and non-rust contaminants (such as oil stains and mud) with similar optical properties, greatly reducing false positives and false negatives. At the same time, by enhancing the contrast of weak features, early detection of rust (such as the initial oxide film) that is difficult to detect with the naked eye is achieved, providing key information for timely maintenance. It overcomes the limitations of information acquisition under traditional single fixed illumination conditions in complex and extreme environments, improves the robustness of the detection system, and ensures the safety and efficiency of bulldozer operation.

[0015] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for identifying surface defects in bulldozer tooth blocks provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of a bulldozer tooth block surface defect identification system provided in an embodiment of the present invention.

[0018] Label Explanation: 100. Determination module; 200. Acquisition module; 300. Extraction module; 400. Fusion module; 500. Recognition module. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Reference Appendix Figure 1 This invention provides a method for identifying surface defects in bulldozer tooth blocks, comprising the following steps: Determine the lighting pattern sequence for bulldozer tooth block surface inspection; the lighting pattern sequence contains multiple sets of lighting parameters, each set of lighting parameters is used to highlight specific types of surface features or suppress specific types of interfering features; The image acquisition device controls the image acquisition of bulldozer teeth blocks based on a sequence of illumination patterns, and acquires multiple images of the teeth block surface under different illumination patterns; the image acquisition device adopts a partially enclosed illumination structure. Feature extraction was performed on multiple images of the tooth block surface to obtain feature information corresponding to each image; the feature information includes color features, texture features, shape features, and optical response features; By fusing the feature information corresponding to each image, fused feature information is obtained; Based on the fused feature information, the type and area of ​​rust on the surface of the bulldozer tooth block are identified.

[0022] Illumination pattern sequence refers to a series of preset or dynamically adjusted illumination conditions. This can be achieved by adjusting parameters such as the wavelength, intensity, incident angle, polarization direction, or illumination area of ​​the light source. For example, this can be done by controlling LED light sources with different wavelengths or adjusting the relative position of the light source and the object being measured. Its main purpose is to specifically enhance or suppress specific types of surface features during the image acquisition stage to optimize the image data quality for subsequent identification. A locally enclosed illumination structure refers to a physical structure that isolates the area under test from external ambient light and provides controlled illumination. This can be achieved using a ring-shaped light source cover with integrated light-emitting units, a box structure with a diffuser plate, or a retractable light shield. Its main purpose is to ensure a stable and consistent illumination environment during image acquisition, effectively eliminating ambient light interference and improving the reliability of image data. Feature information refers to the set of data extracted from the image to describe the surface properties of an object. It can be represented in the form of color histograms, gray-level co-occurrence matrices, edge contour data, or spectral reflectance data. Its main purpose is to provide multi-dimensional and quantitative evidence for subsequent defect identification. Feature fusion refers to integrating feature data from different sources or different modalities to form a more comprehensive and discriminative feature representation. It can be achieved by feature concatenation, feature weighted averaging, feature fusion algorithms based on machine learning, or decision-level fusion. Its main purpose is to gather complementary information obtained under different lighting patterns to improve the robustness and discriminative ability of features in order to cope with complex and varied defect types and interference.

[0023] The working principle of this method lies in designing and applying a sequence of illumination patterns to effectively address the technical problem of insufficient image data discrimination under fixed illumination due to the diverse types of rust on bulldozer tooth blocks, their varying optical properties, and the presence of non-rust contaminants with similar optical characteristics. Specifically, recognizing the differentiated optical responses of different types of rust (such as black rust and initial oxide film) and contaminants under specific illumination, the scheme pre-defines a sequence of illumination patterns containing different wavelengths, incident angles, or polarization states. Multiple images with different information emphases are acquired sequentially; for example, one illumination pattern might highlight black rust, while another might enhance the contrast of the initial oxide film. Subsequent feature extraction and fusion steps integrate these complementary feature information from multi-dimensional illumination patterns to construct a more discriminative comprehensive feature description of various types of rust and contaminants. This enables the image processing system to more accurately identify and distinguish complex and varied rust types, reduce missed detections of atypical rust and misjudgments of contaminants, thereby comprehensively improving the accuracy of rust detection on bulldozer tooth blocks.

[0024] The core innovation of this application lies in integrating dynamically adjusted lighting pattern sequences with multi-dimensional feature information, thereby solving the problems of image information confusion and difficulty in comprehensively and accurately identifying complex rust types and interference caused by fixed lighting conditions in the prior art, and achieving the effect of improving the accuracy and comprehensiveness of bulldozer tooth block surface defect detection.

[0025] Specifically, the bulldozer tooth block surface defect identification method of this application works by constructing an intelligent image acquisition and analysis process that can adapt to complex surface conditions. First, considering the various types of rust and interference that may exist on the bulldozer tooth block surface, the system determines a sequence of illumination patterns containing multiple sets of illumination parameters. Each set of illumination parameters in this sequence is carefully designed to specifically highlight specific types of surface features or suppress specific types of interference features. Subsequently, under a partially enclosed illumination structure, the image acquisition device acquires images of the tooth block surface one by one according to this illumination pattern sequence, thereby obtaining multiple images of the tooth block surface taken under optimized illumination conditions. The partially enclosed illumination structure ensures that the acquisition process is not affected by external ambient light, guaranteeing the stability and consistency of image quality. Next, multi-dimensional feature extraction is performed on these images acquired under different illumination patterns, including features such as color, texture, shape, and optical response. Because each image captures different information emphases under specific illumination, this multi-modal feature extraction can fully exploit the unique discriminative information contained in each image. Based on this, the feature information extracted from multiple images is integrated to form a comprehensive and robust fused feature information. This integration process gathers complementary information from different lighting conditions, effectively compensating for the lack of information or confusion under single lighting conditions, and improving the distinguishing ability of features. Finally, based on this discriminative fused feature information, the system can accurately determine the specific type of rust on the surface of the bulldozer tooth block and its precise distribution area, thus effectively solving the problem of insufficient accuracy and comprehensiveness of traditional methods in complex environments.

[0026] As one embodiment, the solution of this application is specifically implemented as follows: In the process of identifying surface defects on bulldozer tooth blocks, a sequence of illumination patterns can be preset. This sequence may include, for example: a set of illumination patterns using 450nm blue light to highlight the initial oxide film, a set of illumination patterns using 660nm red light to enhance the contrast of red rust, and a set of illumination patterns using polarized light to suppress surface reflection and highlight texture features. The parameters of these illumination patterns, such as light intensity and incident angle, can be accurately controlled by a programmable LED array light source. The image acquisition device can use a high-resolution industrial camera, such as a GigE camera equipped with a CMOS sensor, and integrate it into a ring light source structure with a diffuser to form a locally enclosed lighting environment. During the acquisition process, the industrial camera will sequentially capture multiple images of the tooth block surface under the above illumination patterns. Subsequently, feature extraction is performed on each image, for example, color features are obtained by calculating the color histogram of the image, texture features are extracted by the gray-level co-occurrence matrix (GLCM), shape features are obtained by Canny edge detection and contour analysis, and optical response features are obtained by analyzing the reflectivity differences under different wavelength illumination. The extracted features can be organized into multi-dimensional vectors. These feature vectors from different lighting patterns can then be integrated through feature concatenation to form a unified fused feature vector. Finally, this fused feature vector can be input into a pre-trained deep learning model, such as a convolutional neural network (CNN), to identify the type of rust on the tooth block surface (e.g., red rust, black rust, initial oxide film) and its precise rust areas.

[0027] By employing the aforementioned solution, this application effectively addresses the problem of existing bulldozer tooth block surface defect identification methods suffering from image information confusion due to fixed lighting conditions when facing complex and varied rust types, rust at different development stages, and various interfering substances, thus affecting the accuracy and comprehensiveness of identification. This application enhances the discriminative features of different defect types and suppresses interfering information by dynamically adjusting the lighting mode, thereby improving data quality at the image acquisition source. Furthermore, by combining the extraction and integration of multimodal features, comprehensive defect information is gathered, improving the robustness and accuracy of the identification algorithm, making the detection of bulldozer tooth block surface rust comprehensive and reliable even in harsh operating environments.

[0028] In some embodiments, the step of determining the illumination pattern sequence for bulldozer tooth block surface detection includes: Obtain preliminary optical response information of the bulldozer tooth block surface; the preliminary optical response information is used to characterize the type of rust or contaminants currently present on the tooth block surface; Based on preliminary optical response information, the current dominant feature type on the tooth block surface is determined; the dominant feature type includes the dominant corrosion type or the dominant contaminant type. Based on the dominant feature type, a lighting pattern matching the dominant feature type is selected from the preset lighting pattern set to construct a lighting pattern sequence.

[0029] "Preliminary optical response information" refers to initial data on the light reflection, absorption, or scattering characteristics of the tooth block surface acquired through one or more rapid, non-invasive methods before detailed inspection. This information reflects the current material composition or state of the tooth block surface, such as a metallic substrate, different types of corrosion, or various contaminants. Acquisition methods may include: one-time or rapid scanning image acquisition under a broad spectrum or specific wavelength, analyzing its average brightness, color distribution, spectral reflectance curve, or preliminary texture features; or acquiring an overview image through low-intensity, non-focused floodlight illumination and extracting preliminary statistical features. "Dominant feature type" refers to the most significant or most concerning surface characteristic identified based on the preliminary optical response information analysis. This could be a dominant corrosion type, such as large areas of red rust or locally concentrated black rust, or a dominant contaminant type, such as oil or mud covering the tooth block surface. Identifying the dominant feature type aims to simplify complex surface conditions into actionable classifications for subsequent targeted adjustments to the illumination strategy. A "preset set of illumination patterns" refers to a database or library containing various predefined and optimized illumination conditions. Each illumination pattern is designed and optimized for a specific surface feature or detection target. These patterns can include combinations of light sources with different wavelengths, polarization directions, incident angles, intensities, or spatial distributions. "Selecting an illumination pattern that matches the dominant feature type" means intelligently choosing the illumination condition from the preset set of illumination patterns that best highlights the dominant feature or suppresses its interference, based on the identified dominant feature type. For example, if the dominant feature is black rust, an illumination pattern that enhances the contrast of dark areas is selected; if the dominant feature is a reflective contaminant, an illumination pattern that suppresses its reflective effects is selected. This selection process can be based on pre-established feature-illumination matching rules, lookup tables, or machine learning models.

[0030] This solution effectively addresses the technical problem of insufficient discrimination of fixed illumination pattern sequences caused by dynamic changes in the surface condition of bulldozer tooth blocks by introducing preliminary perception and analysis of the actual surface conditions. Specifically, when determining the illumination pattern sequence for detecting the bulldozer tooth block surface, the system first acquires preliminary optical response information of the tooth block surface. This preliminary optical response information is not used for fine identification, but rather as a rapid perception of the current macroscopic condition of the tooth block surface, aiming to characterize the types of rust or contaminants that may be present on the tooth block surface. It is precisely because of this preliminary perception that the system can avoid blindly applying preset illumination under unknown surface conditions, laying the foundation for subsequent adaptive adjustments. Secondly, based on the acquired preliminary optical response information, the system further determines the dominant feature type of the tooth block surface. This step is a refinement and summarization of the preliminary perception information, transforming the complex surface optical response into a clear and operable classification, such as a dominant type of rust or a dominant type of contaminant. By identifying the surface feature that requires the most attention, the system can focus its illumination adjustments on this feature, ensuring that subsequent illumination mode selections are highly targeted. Finally, based on the determined dominant feature type, the system intelligently selects an illumination mode from a preset set that matches the dominant feature type, thus constructing the final illumination mode sequence. It is precisely because of this dynamic, matching selection mechanism that the constructed illumination mode sequence can maximize the distinguishing information of the dominant feature or minimize its interference. For example, if the dominant feature is black rust, an illumination mode that enhances the contrast of the black rust is selected; if the dominant feature is a specific contaminant, an illumination mode that effectively suppresses the influence of that contaminant is selected. This adaptive determination of the illumination mode sequence, combined with the step in this application of controlling the image acquisition device to acquire multiple images of the tooth block surface under different illumination modes based on the illumination mode sequence, forms a close working relationship. By dynamically adjusting the illumination conditions to adapt to the actual condition of the tooth block surface, the image acquisition device can always acquire the most distinguishable image data. This allows for richer and more accurate feature information to be obtained when extracting and fusing features from multiple images of the bulldozer tooth block surface, thereby significantly improving the accuracy and robustness of identifying the type and area of ​​rust on the bulldozer tooth block surface. Due to this adaptive lighting pattern, this solution can effectively address the challenges of complex and variable surface rust types and contaminant conditions on bulldozer tooth blocks under extreme operating environments, ensuring the comprehensiveness and reliability of the detection.

[0031] As a specific implementation method, the following steps can be used to determine the illumination pattern sequence for bulldozer tooth block surface inspection. First, to obtain preliminary optical response information of the bulldozer tooth block surface, a broadband camera or a camera equipped with multiple narrowband filters can be used to acquire one or more preliminary images under low-intensity floodlight illumination during a single or rapid scan. For example, a visible light image and a near-infrared image can be acquired. By analyzing the average brightness, color histogram, or grayscale distribution of specific areas in these preliminary images, it can be preliminarily determined whether the tooth block surface mainly exhibits a metallic luster, reddish-brown rust, dark black material, or has significant soil or dust coverage. Further, based on this preliminary optical response information, the current dominant feature type of the tooth block surface can be determined. For example, if the initial image shows a large area of ​​reddish-brown with low brightness, the dominant feature type can be identified as "red rust"; if the image shows a large area of ​​dark black with low gloss, the dominant feature type can be identified as "black rust" or "oil stains"; if the image shows a bright and uniform area, it may be identified as "clean metal surface" or "initial oxide film"; if there are obvious granular or blocky deposits, it can be identified as "dirt / dust contaminants". This judgment process can be based on preset threshold rules or implemented through a lightweight classification model. Based on this, according to the determined dominant feature type, a matching illumination pattern can be selected from a preset illumination pattern set, and an illumination pattern sequence can be constructed. For example, if the dominant feature type is determined to be "red rust", the system can select a set of illumination patterns from the set that include specific wavelengths (such as blue or green light to enhance the contrast of red rust), specific incident angles (such as grazing light to highlight surface roughness), and specific polarization directions. If the dominant feature type is determined to be "black rust," a set of illumination patterns including high-intensity white light or a specific angled ring illumination pattern can be selected to enhance details in dark areas. If the dominant feature type is "oil stains," a set of illumination patterns including ultraviolet light or specifically polarized light can be selected to suppress the reflection of oil stains or to bring out their fluorescent properties. This preset set of illumination patterns can be a database storing optimized combinations of illumination parameters for different rust and contaminant types. In this way, the generated illumination pattern sequence can provide optimal image acquisition conditions for the specific condition of the current tooth block surface, thus providing high-quality input data for subsequent defect identification.

[0032] This scheme achieves adaptive determination of the illumination pattern sequence by introducing preliminary perception and analysis of the actual surface condition of the bulldozer tooth blocks. This allows for dynamic adjustment of illumination conditions based on the current type of rust or contaminants on the tooth block surface when acquiring multiple images of the tooth block surface under different illumination patterns. Therefore, even when the surface condition of the bulldozer tooth blocks is complex and variable, with atypical rust or multiple contaminants, the most discriminative image data can always be acquired, effectively enhancing the contrast of target features and suppressing interference information. This significantly improves the accuracy and robustness of subsequent rust type identification and area localization, overcoming the problem of insufficient discriminative power of fixed illumination pattern sequences in complex environments.

[0033] In some embodiments, the step of controlling the image acquisition device to acquire multiple images of the bulldozer tooth block surface under different lighting modes based on a sequence of lighting modes includes: Before image acquisition of the bulldozer tooth block surface, structured light is projected onto the tooth block surface through an image acquisition device and structured light images are acquired. Local morphological information of the tooth block surface is obtained based on the structured light images. The local morphological information is used to characterize pits, protrusions or scratches on the tooth block surface. For each set of illumination parameters in the illumination pattern sequence, based on local topography information, the illumination adjustment parameters used to enhance the display of specific surface features or suppress the display of interfering features are determined by analyzing the interaction characteristics between the illumination parameters and the local topography of the tooth block; the illumination adjustment parameters include illumination intensity or illumination incident angle. The image acquisition device controls the image acquisition device to acquire multiple images of the bulldozer tooth block surface under different lighting modes and with adjusted lighting parameters, based on each set of lighting parameters in the lighting mode sequence and the corresponding lighting adjustment parameters.

[0034] Structured light refers to an optical technique that projects specific patterns (such as stripes, dot matrix, or coded patterns) onto the surface of an object and calculates the object's three-dimensional shape information based on the deformation caused by the pattern on the surface. It can be implemented using methods such as laser stripe projection, coded light projection, or speckle projection. Local shape information refers to data acquired through structured light technology that characterizes the microscopic geometric features of the object's surface. This can include depth maps, normal maps, or point cloud data, used to describe pits, protrusions, or scratches on the surface of the tooth block. Illumination adjustment parameters refer to the light source parameters dynamically adjusted based on the local shape information of the object's surface during image acquisition. These can include light intensity, light incidence angle, or light color and illumination mode (such as diffuse or focused light). The interaction between illumination parameters and the local shape of the tooth block refers to the changes in optical behavior such as reflection, scattering, and absorption when light encounters different geometric structures (such as pits, protrusions, and scratches) on the tooth block surface. This can be analyzed using physical optics models, empirical models, or machine learning models.

[0035] This application's solution significantly improves the quality and relevance of image acquisition by introducing the perception of the local morphology of the bulldozer tooth block surface and adaptive illumination adjustment based on this morphology. Specifically, before formal image acquisition, the system first projects structured light onto the tooth block surface through an image acquisition device and acquires the corresponding structured light image. This preprocessing step enables the system to accurately acquire the three-dimensional geometric information of the tooth block surface, i.e., local morphology information, which can clearly characterize the pits, protrusions, or scratches present on the tooth block surface. It is precisely because these detailed local morphologies are acquired in advance that the system can provide accurate geometric basis for subsequent illumination adjustment, thereby overcoming the limitations of traditional fixed illumination or only based on macroscopic illumination pattern sequences, laying the foundation for achieving refined and adaptive image acquisition. On this basis, for each set of illumination parameters in the illumination pattern sequence, the system will deeply analyze the interaction characteristics between the illumination parameters and the local morphology of the tooth block based on the acquired local morphology information. This analysis aims to understand how the optical behavior of light, such as reflection, scattering, and shading, changes when encountering different local morphologies (such as deep pits or highly reflective protrusions). Through this analysis, the system can intelligently determine lighting adjustment parameters to enhance the display of specific surface features or suppress the display of interfering features, such as adjusting the light intensity or the incident angle. For example, for deep pits, it may be necessary to increase the light intensity or adjust the incident angle to reduce shadows and ensure that the defects within the pits are clearly visible; while for highly reflective protrusions, it may be necessary to adjust the incident angle to avoid overexposure or specular reflection, thereby suppressing them as interfering features. This dynamic and refined lighting adjustment ensures that the target defect features are maximized in complex surface morphologies, while interference caused by the morphology itself is effectively suppressed, greatly optimizing the signal-to-noise ratio of the image. Finally, the system controls the image acquisition device to acquire images of the bulldozer tooth block surface based on each set of lighting parameters and corresponding lighting adjustment parameters in the lighting mode sequence, thereby obtaining multiple images of the tooth block surface under different lighting modes with adjusted lighting parameters. By applying the intelligently adjusted lighting parameters to the actual image acquisition process, the images finally obtained by the system are no longer simply acquired under preset lighting modes, but are optimized for the specific local morphology of the tooth block surface. These images, adjusted for lighting parameters, are significantly higher quality than unadjusted images because they more clearly and comprehensively present the defect features of the tooth block surface, while reducing the risk of misjudgment or missed detection due to complex morphology. This scheme significantly improves the overall accuracy and robustness of identification. It involves determining the lighting pattern sequence, image acquisition, feature extraction, feature fusion, and finally identifying the type and region of corrosion. In the image acquisition stage, this scheme ensures higher quality and stronger discriminative power for the image data input to subsequent feature extraction and fusion steps through perception of the local morphology of the tooth block surface and adaptive lighting adjustment.This means that even when the surface of the tooth block has complex morphology (such as pits, protrusions, and scratches), the target corrosion features can be effectively captured and highlighted, avoiding information loss or confusion caused by improper lighting. Therefore, the subsequent feature extraction module can extract more accurate and representative feature information from higher-quality images, and the fusion module can also more effectively integrate these high-quality feature information, ultimately enabling the recognition module to more accurately determine the type and area of ​​corrosion. This pre-processing image optimization mechanism provides a solid data foundation for the entire defect recognition process, allowing the system to maintain a high level of recognition performance when facing complex real-world conditions.

[0036] In one specific embodiment, this method can be applied to an automated inspection production line for bulldozer tooth blocks. The image acquisition device may include a high-resolution industrial camera and a structured light projector with an integrated programmable LED array light source. Before acquiring images of the bulldozer tooth block surface, the structured light projector can project a set of preset laser stripe structured light patterns onto the tooth block surface, while the industrial camera simultaneously acquires multiple structured light images. Subsequently, an image processing unit (e.g., an embedded computer or industrial PC) can calculate and obtain local topographic information of the tooth block surface based on these structured light images using triangulation principles or phase measurement profilometry, for example, generating a depth map where the grayscale value of each pixel can characterize the depth of the corresponding surface point, thus clearly indicating the location and depth of pits, protrusions, or scratches. Next, for each set of illumination parameters in a preset illumination pattern sequence (e.g., a set of illumination combinations containing different incident angles and intensities), the image processing unit can determine illumination adjustment parameters based on the acquired local topographic information by analyzing the interaction characteristics between the illumination parameters and the local topography of the tooth block. For example, a lighting-morphology interaction model or lookup table can be pre-established. This model or lookup table, built based on physical optics principles or extensive experimental data, can predict the impact of different lighting intensities and incident angles on image contrast and brightness distribution under specific local morphologies. When a deep pit is detected, the system can determine whether to increase the local lighting intensity or adjust the incident angle to ensure sufficient light reaches the bottom of the pit, preventing excessively deep shadows that could lead to information loss. Conversely, when a highly reflective protrusion is detected, the system can determine whether to decrease the lighting intensity or adjust the incident angle to avoid local overexposure, thereby suppressing it as an interfering feature. Finally, the image processing unit can control the image acquisition device to acquire images of the bulldozer tooth block surface based on each set of lighting parameters in the lighting pattern sequence and the corresponding lighting adjustment parameters. For example, when acquiring the first image, if the preset lighting parameters are "45-degree incident angle, 80% intensity," but based on local topographic information, the system determines that a certain area has a deep pit and requires additional adjustment, then the actual lighting source is controlled to illuminate at "45-degree incident angle, 90% intensity." When acquiring the next image, if the preset lighting parameters are "90-degree incident angle, 60% intensity," but a certain area has a highly reflective protrusion, then the actual lighting source is controlled to illuminate at "90-degree incident angle, 50% intensity." In this way, the system can acquire multiple images of the tooth block surface under different lighting modes with adaptively adjusted lighting parameters. These images can more clearly and comprehensively reflect the corrosion characteristics of the tooth block surface, providing high-quality input data for subsequent defect identification.

[0037] This solution effectively addresses the problems of inconsistent lighting effects and difficulty in effectively capturing specific corrosion features caused by the complex surface morphology of bulldozer tooth blocks. By acquiring local topographic information of the tooth block surface before image acquisition and adaptively adjusting lighting parameters based on this information, the solution ensures optimized control of light reflection, scattering, and absorption characteristics on irregular surfaces, thus avoiding overexposure or underexposure in local areas during image acquisition. This allows target corrosion features to be clearly highlighted even in complex topographic regions, while suppressing interference caused by the topography itself. This significantly improves the quality and information integrity of the acquired image data, providing more reliable raw data for subsequent feature extraction and defect identification, thereby enhancing the accuracy and robustness of the entire defect identification method.

[0038] In some embodiments, the step of identifying the type and area of ​​rust on the surface of a bulldozer tooth block based on fused feature information includes: Based on the fused feature information, the surface of the bulldozer tooth block is initially identified, and the preliminary identification results and the identification confidence corresponding to the preliminary identification results are obtained. Based on the identification confidence level, determine whether the preliminary identification result is an uncertain identification result; uncertain identification results include identification results with an identification confidence level lower than a preset threshold, or identification results where the feature similarity between the preliminary identification result and multiple preset rust types or contaminant types is higher than a preset similarity threshold; When the initial identification result is uncertain, based on the fused feature information and the initial identification result, a feature enhancement strategy matching the uncertain identification result is selected from the preset feature enhancement strategy set, and the fused feature information is reconstructed according to the selected feature enhancement strategy to obtain enhanced fused feature information; the feature enhancement strategy is used to highlight the distinguishing features of the potential corrosion type or contaminant type corresponding to the uncertain identification result; Based on the enhanced fusion feature information, the type and area of ​​rust on the surface of the bulldozer tooth block are identified, or when the preliminary identification result is not an uncertain identification result, the type and area of ​​rust on the surface of the bulldozer tooth block are identified based on the fusion feature information.

[0039] In the aforementioned identification process, the preliminary identification result refers to the classification or segmentation output obtained by the system when it first attempts to identify the type and area of ​​rust on the surface of the bulldozer tooth block. This can be represented by classification labels based on machine learning models, pixel-level segmentation masks, or bounding box coordinates. Identification confidence refers to a quantitative indicator of the reliability or trustworthiness of the preliminary identification result. It can be represented by probability values, scores, or distance metrics, and is usually generated synchronously by the identification model when outputting the preliminary identification result. An uncertain identification result refers to a situation where the reliability of the preliminary identification result is insufficient and further processing is required. This can be judged by identification results with an identification confidence level lower than a preset threshold, or by identification results where the feature similarity between the preliminary identification result and multiple preset rust types or contaminant types is higher than a preset similarity threshold. The preset threshold refers to a reference value used to determine whether the identification confidence level is high enough to accept the preliminary identification result. It can be set using empirical values, statistical analysis results, or values ​​obtained through model training and optimization. A preset similarity threshold refers to a reference value used to judge the degree of similarity between the preliminary recognition result and different types of features. It can be set using metrics such as cosine similarity, Euclidean distance, or correlation coefficient, and is used to identify cases of highly confused features. A preset feature enhancement strategy set refers to a predefined set of methods or algorithms used to improve or highlight specific attributes in fused feature information. It can include various enhancement algorithms for different feature dimensions (such as color, texture, and spectrum), such as contrast enhancement, edge sharpening, specific frequency band filtering, or feature weighting. A feature enhancement strategy refers to a specific method or algorithm selected from the preset feature enhancement strategy set to process the fused feature information to highlight specific discriminative features. It can be implemented using deep learning-based feature transformation networks, combinations of traditional image processing algorithms, or multi-scale feature fusion methods. Feature reconstruction refers to the process of transforming or processing the original fused feature information according to the selected feature enhancement strategy to generate new, more discriminative feature representations. It can be implemented using techniques such as feature mapping, feature fusion, feature selection, or feature transformation. Enhanced fusion feature information refers to fusion feature information whose discriminative features are strengthened after feature reconstruction processing. It can be represented by feature vectors or feature maps with higher dimensions, greater information density, or higher weights for specific feature dimensions. Discriminative features refer to feature attributes that can effectively distinguish the differences between different types of rust or contaminants, and can include color saturation, texture roughness, spectral reflectance peaks, or shape edge sharpness, etc.

[0040] This solution introduces an adaptive, iterative identification mechanism to address the challenges posed by complex and varied rust types and contaminants in the identification of surface defects on bulldozer tooth blocks. The mechanism first utilizes rich feature information acquired and fused through multi-lighting modes to perform preliminary identification of the bulldozer tooth block surface. This preliminary identification process simultaneously outputs the preliminary identification result and its corresponding identification confidence level. The identification confidence level serves as a reliability indicator for the preliminary judgment, providing a basis for subsequent intelligent decision-making. Subsequently, the system intelligently determines whether the preliminary identification result is uncertain based on the identification confidence level. This uncertainty judgment mechanism considers two scenarios: first, the identification confidence level is lower than a preset threshold, indicating insufficient confidence in the current identification result; second, the feature similarity between the preliminary identification result and multiple preset rust or contaminant types is higher than a preset similarity threshold, indicating high confusion between different categories of current feature information, making effective differentiation difficult. Through this dual judgment, the system can accurately identify potential errors or uncertainties caused by feature ambiguity, high similarity, or insufficient model confidence. When the initial identification result is deemed uncertain, the system does not directly adopt the result but instead proceeds to a feature enhancement and reconstruction stage. At this point, the system intelligently selects the feature enhancement strategy that best matches the uncertain identification result from a pre-set set of feature enhancement strategies, based on the current fused feature information and potential confusion clues contained in the initial identification result. For example, if the initial identification result is confused between black rust and oil stains, the system might choose a strategy that highlights texture details or specific spectral responses. The selected feature enhancement strategy is then applied to the fused feature information for feature reconstruction, resulting in enhanced fused feature information. This reconstruction process aims to specifically highlight the distinguishing features of the potential rust type or contaminant type corresponding to the uncertain identification result, making previously blurred or overlapping features clearer. Finally, the system performs the final identification based on the processed feature information. If the initial identification result is uncertain, the enhanced fused feature information obtained through feature reconstruction is used to identify the rust type and rust area on the bulldozer tooth block surface, ensuring a judgment is made based on more discriminative data. If the initial identification result is sufficiently certain, the original fused feature information is used directly for identification, avoiding unnecessary computational overhead and demonstrating the efficiency of the method. This approach, combined with methods for acquiring fusion feature information through multiple illumination modes, can fully utilize multi-dimensional, high-quality input data. The fusion feature information acquired in the early stages by dynamically adjusting illumination conditions already contains rich surface details and optical responses, providing a solid foundation for the initial identification by this approach.When initial identification encounters challenges, this solution can further enhance the features based on the rich feature information, thereby effectively improving the ability to identify new, mixed, or fuzzy boundary corrosion under complex and ever-changing actual working conditions.

[0041] In a specific embodiment, the method for identifying surface defects on bulldozer tooth blocks can be implemented as follows: First, after acquiring the fused feature information, it can be input into a pre-trained deep learning model, such as a convolutional neural network based on the ResNet architecture or a U-Net segmentation network, to perform preliminary identification of the bulldozer tooth block surface. The model outputs a preliminary identification result, which can be a classification label and a pixel-level segmentation mask. Simultaneously, the model outputs a recognition confidence score corresponding to the preliminary identification result through its Softmax layer; this confidence score is a probability value between 0 and 1. Next, the system uses this recognition confidence score to determine whether the preliminary identification result is uncertain. For example, a preset threshold of 0.7 can be set. If the recognition confidence score is lower than 0.7, the preliminary identification result is considered uncertain. Furthermore, the system can also calculate the cosine similarity between the feature vector of the preliminary identification result and the typical feature vectors of multiple preset rust types or contaminant types. If the preliminary identification result has a similarity score higher than a preset similarity threshold (e.g., 0.85) with at least two different types of features, the preliminary identification result is considered uncertain, indicating high confusion between different categories. When the preliminary identification result is determined to be uncertain, the system initiates a feature enhancement process. At this point, the system selects a matching strategy from a preset set of feature enhancement strategies based on the potential confusion type indicated by the preliminary identification result. This set can include various strategies, such as: texture enhancement strategies, using Gabor filters or wavelet transforms to highlight subtle texture differences in the image and distinguish between rust or contaminants with different surface roughness; spectral enhancement strategies, adjusting the weights of different color channels or applying enhancement algorithms for specific bands to amplify subtle differences in color or spectral response between different rust types; contrast enhancement strategies, using adaptive histogram equalization or local contrast enhancement algorithms to improve the contrast of blurred areas; and shape feature enhancement strategies, utilizing edge detection algorithms or morphological operations to sharpen defect boundaries and distinguish shape features. The system selects the strategy that best highlights the discriminative features based on the confusion level of the preliminary identification result. For example, if the initial identification results are confused between black rust and oil stains, the system may choose a texture enhancement strategy or a spectral enhancement strategy. After selecting a strategy, the system will reconstruct the original fused feature information according to that strategy. For example, if a texture enhancement strategy is selected, the system can weight the texture feature dimension in the fused feature information or apply a specific convolution kernel to process it, thereby obtaining enhanced fused feature information. This enhanced fused feature information will more clearly show the key features that distinguish different types of defects. Finally, the system will use this enhanced fused feature information to input it again into the recognition model for the final identification of rust type and rust area.If the initial identification result is not uncertain, the original fused feature information is used directly for identification, avoiding unnecessary processing steps. Through this adaptive feature enhancement and reconstruction, the system can make accurate judgments even when faced with complex or ambiguous defect features.

[0042] This solution effectively addresses the problem of uncertain or misjudged identification results in bulldozer tooth block surface inspection due to the high similarity or blurred boundaries between the characteristics of novel, mixed, or composite corrosion defects and known types. By assessing the confidence level of preliminary identification results and intelligently determining whether further processing is needed, it avoids directly adopting potentially inaccurate preliminary results. When the identification result is uncertain, this solution can selectively select and apply feature enhancement strategies based on potential confusion types to reconstruct the fused feature information, thereby highlighting the distinguishing features between different defect types. This makes previously difficult-to-distinguish ambiguous features clearly discernible, significantly improving the accuracy of corrosion type and region identification in complex scenarios. Simultaneously, for cases where the preliminary identification result is certain, direct identification is performed, avoiding unnecessary computational overhead and improving detection efficiency.

[0043] In some embodiments, the step of reconstructing the fused feature information according to the selected feature enhancement strategy to obtain enhanced fused feature information includes: Based on the uncertainty identification results, obtain the characteristic evolution information of the potential corrosion type or contaminant type corresponding to the uncertainty identification results at different development stages; the characteristic evolution information is used to characterize the characteristic differences of the potential corrosion type or contaminant type at different development stages; Based on feature evolution information, assess the feature sensitivity of the selected feature enhancement strategy to potential corrosion types or contaminant types at different development stages; feature sensitivity is used to characterize the ability of the feature enhancement strategy to distinguish between different development stages. Based on the feature sensitivity assessment results, the reconstruction parameters for feature reconstruction are determined; the reconstruction parameters are used to refine the discriminative features related to the development stage of potential corrosion type or contaminant type in the feature information. Based on the selected feature enhancement strategy and reconstruction parameters, the fused feature information is reconstructed to obtain enhanced fused feature information.

[0044] Feature evolution information refers to the changing patterns of optical, physical, or chemical characteristics exhibited by potential corrosion types or contaminant types at different development stages. This information can be obtained using pre-established databases, expert knowledge bases, or by training and summarizing large amounts of historical data through machine learning models. Feature sensitivity refers to the effectiveness of a specific feature enhancement strategy in distinguishing features at different development stages of potential corrosion types or contaminant types. It can be characterized using quantitative indicators such as feature separation degree, classification accuracy, or information gain. Reconstruction parameters are control variables used to adjust feature weights, transformation methods, or fusion ratios during feature reconstruction. These parameters can be determined using rule-based adjustment, optimization algorithm calculations, or adaptive learning through deep learning networks.

[0045] This application aims to optimize the process of feature reconstruction based on fused feature information when the preliminary identification results are uncertain, so as to more accurately highlight the distinguishing features of potential defects, especially when considering the complexity of different development stages of defects, thereby solving the technical problem that the feature reconstruction distinguishing ability is insufficient due to the continuous changes in corrosion development.

[0046] First, when the system performs preliminary identification of the bulldozer tooth block surface based on fused feature information and determines the preliminary identification result to be uncertain, in order to handle this uncertainty more deeply, the system acquires the feature evolution information of the potential corrosion type or contaminant type corresponding to the uncertain identification result at different development stages. This step is crucial because it introduces consideration of the defect's "development stage." Traditional feature enhancement may only target a certain typical state, but the color, texture, shape, optical response, and other characteristics of corrosion or contaminants change at different development stages such as the initial, middle, and late stages. By acquiring this feature evolution information, the system can understand the characteristic differences of potential defects at different stages, providing basic data for subsequent refined processing.

[0047] Secondly, based on the acquired feature evolution information, the system evaluates the feature sensitivity of the previously selected feature enhancement strategy to different development stages of potential corrosion types or contaminant types. Feature sensitivity characterizes the ability of the feature enhancement strategy to distinguish between different development stages. This means that the system no longer applies a universal enhancement strategy, but rather evaluates the effectiveness of the current strategy in distinguishing defects at specific stages based on an understanding of the defect development stages. For example, a strategy effective for macro-corrosion may be insufficiently sensitive to early-stage micro-oxide films. Through this evaluation, the strengths and weaknesses of the current strategy in addressing defects at specific development stages can be identified.

[0048] Next, based on the feature sensitivity assessment results, the system determines the reconstruction parameters used for feature reconstruction. These parameters are used to refine the discriminative features related to the development stage of potential corrosion types or contaminant types within the fused feature information. This step is central to achieving refined reconstruction. Based on the assessment of the sensitivity of feature enhancement strategies at different development stages, the system can generate or adjust the reconstruction parameters. These parameters guide the feature reconstruction process, making it more focused on highlighting features with high discriminative power at specific development stages. For example, for early-stage corrosion, it may be necessary to emphasize subtle color changes or differences in surface gloss; for layered flaking corrosion, it may be necessary to highlight its three-dimensional morphological features. This parameter determination ensures that feature reconstruction is no longer a single approach, but rather a targeted enhancement of features most helpful in distinguishing currently uncertain defects.

[0049] Finally, the system reconstructs the fused feature information based on the selected feature enhancement strategy and determined reconstruction parameters, obtaining enhanced fused feature information. By combining the original feature enhancement strategy with reconstruction parameters finely adjusted according to the development stage, the system can generate more discriminative enhanced fused feature information. This enhanced information not only includes the features emphasized by the original strategy, but also, guided by the reconstruction parameters, finely highlights more identifiable features closely related to the development stage of potential defects. This enables subsequent identification modules to more accurately distinguish complex defects that appear uncertain in the initial identification, thereby improving the accuracy of identifying rust types in the intermediate stage of continuous change or those confused with similar contaminants.

[0050] Overall, when the initial identification results are uncertain, this solution no longer simply applies a pre-set feature enhancement strategy. Instead, it further analyzes the source of uncertainty, namely the development stage of potential defects. By acquiring feature evolution information, evaluating the sensitivity of existing strategies, and dynamically adjusting reconstruction parameters accordingly, this solution can perform more refined and targeted reconstruction of fused feature information. This adaptive feature reconstruction mechanism enables the system to effectively cope with the continuity and complexity of rust and contaminant features on the surface of bulldozer tooth blocks, improving the accuracy of identifying defects at ambiguous boundaries or intermediate stages, thereby overcoming the identification difficulties caused by feature confusion in existing technologies.

[0051] In a specific embodiment, when the initial identification result of the system is uncertain, such as being identified as "slight rust" but with low confidence, or having a high similarity to the features of both "initial oxide film" and "surface oil stains", the system can initiate a refined feature reconstruction process.

[0052] First, the system can access a pre-established "defect feature evolution database." This database stores typical feature evolution information from initial to late stages for various potential corrosion types (e.g., hydrated iron oxide, magnetite) and common contaminant types (e.g., ore dust, oil stains, mud). For example, for "initial oxide film," the database may record its subtle color changes under different lighting conditions, surface gloss decay curves, and microscopic texture features; for "point rust," it may record changes in color saturation, shape regularity, and edge sharpness over time. Based on the uncertain identification results, the system retrieves feature evolution information related to "minor corrosion," "initial oxide film," and "surface oil stains" from this database.

[0053] Secondly, the system can utilize a "strategy sensitivity evaluation module" to assess the sensitivity of the currently selected feature enhancement strategy (e.g., a wavelet transform-based texture enhancement strategy or a color space conversion-based contrast enhancement strategy) to the retrieved feature evolution information. This module can contain a series of pre-defined evaluation functions or a small classifier, generating a sensitivity score by simulating the application of the strategy and analyzing its performance in distinguishing features at different developmental stages. For example, if the evaluation results show that the current texture enhancement strategy is insufficiently sensitive to distinguishing the microscopic texture differences between "initial oxide film" and "clean metal," but highly sensitive to distinguishing the macroscopic texture differences between "punctate corrosion" and "flaky corrosion," then a corresponding sensitivity evaluation result will be generated.

[0054] Next, based on the feature sensitivity assessment results, the system can dynamically determine the reconstruction parameters used for feature reconstruction. This can be a parameter lookup table that directly maps the sensitivity assessment results to a set of predefined reconstruction parameters; or it can be an adaptive optimization algorithm that adjusts the weights, thresholds, or transformation functions in the feature reconstruction algorithm based on the assessment results. For example, if the assessment shows insufficient sensitivity to the texture of the initial oxide film, the system can determine a set of reconstruction parameters that, during feature reconstruction, significantly enhance the components related to micro-texture and gloss in the fused feature information and reduce the weight of macro-color features.

[0055] Finally, the system reconstructs the original fused feature information based on the previously selected feature enhancement strategy and these dynamically determined reconstruction parameters. For example, if the selected strategy is a feature mapping network based on deep learning, the reconstruction parameters can be used as input to the network or the weights of its internal layers can be adjusted. This allows the network to more precisely highlight features with high discriminative power at specific development stages when generating enhanced fused feature information. In this way, the resulting enhanced fused feature information will contain more explicit features that help distinguish uncertain defect development stages, thus providing a more reliable basis for subsequent final identification.

[0056] This solution incorporates information on the evolution of potential rust or contaminant types at different development stages when initial identification results are uncertain. Based on this, the sensitivity of the feature enhancement strategy is assessed, and reconstruction parameters are dynamically determined. This allows the reconstruction process, which integrates feature information, to refine the distinguishing features related to the defect development stage. This effectively solves the problem of insufficient feature reconstruction distinguishability caused by the continuous changes in rust development in bulldozer tooth blocks. It improves the system's accuracy in identifying rust types in the intermediate stages of continuous change or those confused with similar contaminants, thus avoiding missed detections or misjudgments due to feature confusion and enhancing the reliability of defect identification.

[0057] Reference Appendix Figure 2 This invention provides a bulldozer tooth block surface defect identification system, comprising: Module 100 is used to determine the lighting pattern sequence for bulldozer tooth block surface detection; The acquisition module 200 is used to control the image acquisition device. Based on the lighting pattern sequence, it acquires multiple images of the bulldozer tooth block surface under different lighting modes by acquiring images of the tooth block surface. The image acquisition device adopts a partially enclosed lighting structure. The extraction module 300 is used to extract features from multiple images of the tooth block surface to obtain feature information corresponding to each image. The fusion module 400 is used to obtain fused feature information by fusing the feature information corresponding to each image; The identification module 500 is used to identify the type and area of ​​rust on the surface of the bulldozer tooth block based on the fused feature information.

[0058] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0059] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying surface defects in bulldozer tooth blocks, characterized in that, Includes the following steps: Determine the lighting pattern sequence for bulldozer tooth block surface inspection; the lighting pattern sequence contains multiple sets of lighting parameters, each set of lighting parameters is used to highlight specific types of surface features or suppress specific types of interfering features; The image acquisition device controls the image acquisition of bulldozer teeth blocks based on a sequence of illumination patterns, and acquires multiple images of the teeth block surface under different illumination patterns; the image acquisition device adopts a partially enclosed illumination structure. Feature extraction is performed on multiple images of the tooth block surface to obtain the feature information corresponding to each image; By fusing the feature information corresponding to each image, fused feature information is obtained; Based on the fused feature information, identify the type and area of ​​rust on the surface of the bulldozer tooth block; The steps of controlling the image acquisition device to acquire multiple images of the bulldozer tooth block surface under different lighting modes based on a lighting pattern sequence include: Before image acquisition of the bulldozer tooth block surface, structured light is projected onto the tooth block surface through an image acquisition device and structured light images are acquired. Local morphological information of the tooth block surface is obtained based on the structured light images. The local morphological information is used to characterize pits, protrusions or scratches on the tooth block surface. For each set of illumination parameters in the illumination pattern sequence, based on local topography information, the illumination adjustment parameters used to enhance the display of specific surface features or suppress the display of interfering features are determined by analyzing the interaction characteristics between the illumination parameters and the local topography of the tooth block. The image acquisition device is controlled to acquire multiple images of the surface of the bulldozer tooth block under different lighting modes and with the lighting parameters adjusted, based on each set of lighting parameters in the lighting mode sequence and the corresponding lighting adjustment parameters. The steps for identifying the type and area of ​​rust on the surface of bulldozer tooth blocks based on fused feature information include: Based on the fused feature information, the surface of the bulldozer tooth block is initially identified, and the preliminary identification results and the identification confidence corresponding to the preliminary identification results are obtained. Based on the recognition confidence level, determine whether the preliminary recognition result is an uncertain recognition result; When the initial identification result is uncertain, based on the fused feature information and the initial identification result, a feature enhancement strategy matching the uncertain identification result is selected from the preset feature enhancement strategy set, and the fused feature information is reconstructed according to the selected feature enhancement strategy to obtain enhanced fused feature information; the feature enhancement strategy is used to highlight the distinguishing features of the potential corrosion type or contaminant type corresponding to the uncertain identification result; Based on the enhanced fusion feature information, identify the type and area of ​​rust on the surface of the bulldozer tooth block; or when the preliminary identification result is not an uncertain identification result, identify the type and area of ​​rust on the surface of the bulldozer tooth block based on the fusion feature information. The steps for reconstructing enhanced fused feature information based on the selected feature enhancement strategy to obtain enhanced fused feature information include: Based on the uncertainty identification results, obtain the characteristic evolution information of the potential corrosion type or contaminant type corresponding to the uncertainty identification results at different development stages; the characteristic evolution information is used to characterize the characteristic differences of the potential corrosion type or contaminant type at different development stages; Based on feature evolution information, assess the feature sensitivity of the selected feature enhancement strategy to potential corrosion types or contaminant types at different development stages; feature sensitivity is used to characterize the ability of the feature enhancement strategy to distinguish between different development stages. Based on the feature sensitivity assessment results, the reconstruction parameters for feature reconstruction are determined; the reconstruction parameters are used to refine the discriminative features related to the development stage of potential corrosion type or contaminant type in the feature information. Based on the selected feature enhancement strategy and reconstruction parameters, the fused feature information is reconstructed to obtain enhanced fused feature information.

2. The method for identifying surface defects in bulldozer tooth blocks according to claim 1, characterized in that, The feature information includes color features, texture features, shape features, and optical response features.

3. The method for identifying surface defects in bulldozer tooth blocks according to claim 1, characterized in that, The steps for determining the lighting pattern sequence for bulldozer tooth block surface inspection include: Obtain preliminary optical response information of the bulldozer tooth block surface; the preliminary optical response information is used to characterize the type of rust or contaminants currently present on the tooth block surface; Based on preliminary optical response information, the current dominant feature type on the tooth block surface is determined; Based on the dominant feature type, a lighting pattern matching the dominant feature type is selected from the preset lighting pattern set to construct a lighting pattern sequence.

4. The method for identifying surface defects in bulldozer tooth blocks according to claim 3, characterized in that, The dominant feature types include the dominant corrosion type or the dominant contaminant type.

5. The method for identifying surface defects in bulldozer tooth blocks according to claim 1, characterized in that, Lighting adjustment parameters include light intensity or light incident angle.

6. The method for identifying surface defects in bulldozer tooth blocks according to claim 1, characterized in that, Uncertain identification results include identification results with an identification confidence level lower than a preset threshold, or identification results with a preliminary identification result whose feature similarity to multiple preset rust types or contaminant types is higher than a preset similarity threshold.

7. A bulldozer tooth block surface defect identification system employing the bulldozer tooth block surface defect identification method as described in any one of claims 1-6, characterized in that, include: The determination module is used to determine the sequence of illumination patterns for bulldozer tooth block surface detection; The acquisition module is used to control the image acquisition device. Based on the lighting pattern sequence, it acquires multiple images of the bulldozer tooth block surface under different lighting patterns by acquiring images of the tooth block surface. The image acquisition device employs a partially enclosed lighting structure; The extraction module is used to extract features from multiple images of the tooth block surface to obtain the feature information corresponding to each image; The fusion module is used to obtain fused feature information by fusing the feature information corresponding to each image; The identification module is used to identify the type and area of ​​rust on the surface of bulldozer tooth blocks based on fused feature information.

Images