Instrument maintenance assistance method and system based on visual analysis

Abstract

The invention relates to the technical field of image recognition, and discloses an instrument maintenance assistance method and system based on visual analysis. The method comprises the steps of constructing and utilizing a training sample set to obtain an identification model, collecting a surface image of a cleaned instrument to be detected through a high-resolution camera, analyzing the surface image through the identification model, and outputting abnormal prompt information. The system corresponds to the method. According to the invention, through combination of high-resolution visual acquisition and a deep learning algorithm, the precision and efficiency of instrument maintenance are significantly improved; compared with manual visual inspection, the method has the advantages that tiny residues and structural defects can be accurately identified by the aid of the trained identification model, the method is particularly suitable for fine instrument detection, missing detection risks are reduced, the problems of unstable quality, difficulty in fine detection and lack of standardization of a traditional mode are comprehensively solved, and instrument use safety and maintenance quality are guaranteed.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, specifically a method and system for assisting in the maintenance of equipment based on visual analysis. Background Technology

[0002] In the use of medical devices and precision instruments, the surface cleanliness and structural integrity of the instruments directly affect their safety and lifespan. Especially in scenarios such as hospital supply rooms, the removal of residues (such as bloodstains and tissue fragments) after instrument cleaning and the detection of structural defects (such as joint loosening and blade nicks) are key aspects of quality control.

[0003] In existing technologies, the aforementioned detection mainly relies on manual visual inspection or observation with a magnifying glass. This method has significant drawbacks: First, it is highly subjective, depending on the operator's sense of responsibility and experience, and is prone to missed detections due to fatigue or negligence; second, for instruments with delicate structures (such as microsurgical instruments and complex joint instruments), tiny residues or minor structural defects are difficult to identify with the naked eye, leading to potential quality risks; third, it lacks standardized testing procedures, making it difficult to achieve quality traceability and unified control, and failing to meet the application requirements of high cleanliness and high safety.

[0004] Chinese invention patent CN119314647B discloses a medical device use monitoring system based on image recognition, but the invention performs poorly in standardized testing procedures.

[0005] Therefore, there is an urgent need for an efficient, accurate, and automated equipment maintenance assistance solution to overcome the above-mentioned shortcomings. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for assisting in the maintenance of equipment based on visual analysis, so as to solve the technical problems mentioned in the background art.

[0007] To achieve the above objectives, this application discloses the following technical solutions: In a first aspect, this application discloses a visual analysis-based method for assisting with equipment maintenance, the method comprising: A training sample set is constructed, and a recognition model is obtained using the training sample set. The training sample set includes standard images of various instruments in normal form, first abnormal images with surface residues, and second abnormal images with structural integrity defects. The surface image of the instrument to be tested is acquired using a high-resolution camera after cleaning. The surface image of the instrument to be tested is input into the recognition model. The recognition model extracts and analyzes the features of the surface image to determine whether the instrument to be tested has surface residues and / or structural integrity defects. When the recognition model determines that the instrument to be tested has surface residues and / or structural integrity defects, it obtains and outputs an abnormal prompt message to prompt the instrument to be tested to be reprocessed.

[0008] Preferably, the step of training the deep learning algorithm using the training sample set to obtain the recognition model includes: Analyze the similarity between the standard image and the first abnormal image. Based on the detection sensitivity requirements of different types of instruments and the similarity, obtain a set of similarity thresholds. The set of similarity thresholds stores multiple similarity thresholds in response to different detection sensitivity requirements. The similarity thresholds are used to define the minimum value that the similarity must satisfy under a detection sensitivity requirement. By analyzing the standard image and the second abnormal image, a set of structural defect feature templates is obtained. The set of structural defect feature templates stores multiple structural defect feature templates, which are used to define structural defects. The similarity threshold set and the structural defect feature template set are stored in the recognition model, which is used to output the abnormality prompt information based on the input surface image, the similarity threshold set, and the structural defect feature template set.

[0009] Preferably, the recognition model performs feature extraction and analysis on the surface image, including: Enhance the feature extraction of key areas of the instrument to obtain regional features, wherein the key areas include at least the occlusal surface, cutting edge and joint of the instrument; By analyzing the surface images corresponding to the regional features at different sizes, the micro-features and macro-features of the region are obtained. Based on the micro-features and macro-features of the region, a feature vector for defect identification is obtained.

[0010] Preferably, determining whether the instrument under test has surface residues and / or structural integrity defects includes: The feature similarity between the feature vector of the device under test and the feature similarity between the standard image is analyzed to obtain a feature similarity score; the detection sensitivity requirement of the device under test is obtained, and based on the detection sensitivity requirement, the corresponding similarity threshold is selected from the similarity threshold set; when the feature similarity score is less than the similarity threshold, it is determined that there are surface residues. The feature vector of the instrument under test and the set of structural defect feature templates are analyzed. When the feature vector matches the structural defect feature template, it is determined that there is a structural integrity defect.

[0011] Preferably, when the identification model determines that the instrument under test has surface residues and / or structural integrity defects, it includes: Locate the defective areas and generate a defect heat map; The instruments under test are classified and graded based on the type and severity of defects to obtain classification and grading results, which include cleaning priority and maintenance priority. The defect heatmap and the classification and grading results are stored in a preset quality traceability database.

[0012] Preferably, obtaining and outputting the abnormality alert information to prompt reprocessing of the instrument under test includes: The defect heatmap and the classification and grading results are displayed through a preset human-computer interaction interface; Based on the defect heatmap and the classification and grading results, anomaly warning information is obtained, which includes defect type, location, and processing suggestions. The abnormality alert will be sent to the designated responsible persons. The abnormal notification information will be synchronized to the preset hospital management equipment.

[0013] Preferably, the step of training the deep learning algorithm using the training sample set to obtain the recognition model further includes: By analyzing the correlation between the similarity thresholds corresponding to surface residues in the similarity threshold set and the structural defect features corresponding to structural integrity defects in the structural defect feature template set, the correlation features of surface residue adhesion caused by structural integrity defects are obtained. Based on the aforementioned correlation features, structural residue features are obtained. These structural residue features are used to define the causal relationship between surface residues and structural integrity defects. The causal relationship includes the spatial correspondence between the location of structural defects and the distribution of residues, and the matching relationship between the morphology of residues and the type of defects. The structural residue features are integrated into the recognition model, which is used to identify independently existing surface residues, independently existing structural integrity defects, and surface residues caused by structural integrity defects based on the input surface image, the similarity threshold set, the structural defect feature template set, and the structural residue features. Based on the recognition results, the abnormal prompt information is output.

[0014] Preferably, the associated features obtained due to structural integrity defects leading to surface residue adhesion include: The samples with overlapping areas in the first and second abnormal images are labeled to determine the spatial correlation parameters between the structural defect area and the residue area. Based on the spatial correlation parameters, the residual morphological characteristics corresponding to different structural defect types are statistically analyzed to obtain a correlation feature library; Features with significant differences are selected from the associated feature library as the associated features, and the significant differences are determined based on a preset instrument maintenance knowledge base.

[0015] Preferably, when the identification model makes a judgment based on the structural residual features, it includes: When the identification model detects that the instrument under test has both surface residues and structural integrity defects, it calls the structural residue features to verify the correlation between the two. If the distribution location and morphology of the surface residue match the location and type of the structural integrity defect, then an association identifier is added to the abnormality warning information. The association identifier is used to identify that the structural defect caused the residue. Based on the association identifier, the processing priority in the classification and grading results is adjusted so that the processing priority of defects with association is higher than that of independent defects of the same type.

[0016] Secondly, this application discloses a vision-based equipment maintenance assistance system, which applies the vision-based equipment maintenance assistance method described above. The system includes: The model building module is used to build a training sample set, which includes standard images of various instruments in normal form, first abnormal images with surface residues, and second abnormal images with structural integrity defects. The deep learning algorithm is trained using the training sample set to obtain a recognition model. The image acquisition module is used to acquire surface images of the instrument to be inspected after cleaning using a high-resolution camera. The visual analysis module is used to input the surface image of the instrument to be inspected into the recognition model. The recognition model performs feature extraction and analysis on the surface image to determine whether the instrument to be inspected has surface residues and / or structural integrity defects. When the recognition model determines that the instrument to be inspected has surface residues and / or structural integrity defects, it obtains and outputs an abnormal prompt message to prompt the instrument to be inspected to be reprocessed.

[0017] Beneficial Effects: The vision-based instrument maintenance assistance method and system of this application significantly improves the accuracy and efficiency of instrument maintenance by combining high-resolution visual acquisition with deep learning algorithms. Compared with manual visual inspection, it can accurately identify minute residues and structural defects with the help of a trained recognition model, which is especially suitable for the inspection of fine instruments and reduces the risk of missed detection. By adapting dynamic thresholds and feature templates to different instrument needs, it achieves standardized inspection and reduces subjective differences. The introduction of structural residue feature analysis can distinguish between independent defects and related defects and optimize the processing priority. Combined with defect location, classification and quality traceability, it achieves full-process control and synchronizes with the management system to improve collaborative efficiency. It comprehensively solves the problems of unstable quality, difficulty in fine inspection and lack of standardization in traditional methods, and ensures the safety of instrument use and maintenance quality. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the vision analysis-based instrument maintenance assistance method provided in this application embodiment; Figure 2 This is a structural block diagram of a vision analysis-based instrument maintenance assistance system provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] In this document, the term "comprising" is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0022] In existing technologies, the detection of residues and structural integrity on instrument surfaces relies on manual visual inspection or magnifying glasses, which suffers from high subjectivity, high false negative rates, and difficulty in adapting to delicate instruments. To address this issue, this invention discloses a method and system for assisting instrument maintenance based on visual analysis. It optimizes existing deep learning technology for image recognition and combines it with existing image acquisition technologies to achieve efficient, accurate, and automated assistance in instrument maintenance.

[0023] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a vision analysis-based instrument maintenance assistance method, which includes: A training sample set is constructed, and a recognition model is obtained using the training sample set. The training sample set includes standard images of various instruments in their normal form, first abnormal images with surface residues, and second abnormal images with structural integrity defects. In this embodiment, the recognition model is based on an existing deep learning model, and the standard images, first abnormal images, and second abnormal images can be collected based on an existing image database. The surface image of the instrument to be tested is acquired using a high-resolution camera after cleaning. The surface image of the instrument to be tested is input into the recognition model. The recognition model extracts and analyzes the features of the surface image to determine whether the instrument to be tested has surface residues and / or structural integrity defects. When the recognition model determines that the instrument to be tested has surface residues and / or structural integrity defects, it obtains and outputs an abnormal prompt message to prompt the instrument to be tested to be reprocessed.

[0024] Based on the above, by constructing a training sample set to train the recognition model, and combining high-resolution image acquisition and intelligent analysis, automated identification and alerting of instrument defects can be achieved, effectively improving detection accuracy and standardization, and reducing reliance on manual labor. In existing deep learning model training processes, the judgment criteria for residues and structural defects are fixed, making it difficult to adapt to the detection requirements of different instruments. This embodiment constructs a similarity threshold set and a structural defect feature template set, enabling the recognition model to dynamically adjust the judgment criteria based on different sensitivity requirements. This enhances the model's adaptability to diverse instruments and detection scenarios, improving recognition flexibility and accuracy.

[0025] Specifically, deep learning algorithms are trained using a training sample set to obtain a recognition model, including: The similarity between the standard image and the first abnormal image is analyzed. Based on the detection sensitivity requirements and similarity of different types of instruments, a set of similarity thresholds is obtained. The set of similarity thresholds stores multiple similarity thresholds in response to different detection sensitivity requirements. The similarity thresholds are used to define the minimum value that the similarity must satisfy under a detection sensitivity requirement. In this embodiment, the detection sensitivity requirement can be set based on the actual instrument maintenance requirements. It can be understood that the detection sensitivity requirement corresponds to the instrument requirements of the hospital instrument supply room in handling the instrument requirements of different departments or different usage environments. By analyzing the standard image and the second anomalous image, a set of structural defect feature templates is obtained. The set of structural defect feature templates stores multiple structural defect feature templates, which are used to define structural defects. The similarity threshold set and the structural defect feature template set are stored in the recognition model. The recognition model is used to output abnormal prompt information based on the input surface image, similarity threshold set and structural defect feature template set.

[0026] It should be noted that, in this embodiment, apart from the processing of the similarity threshold set and structural defect feature templates mentioned above, the remaining training of the identification model can be performed using existing deep learning model techniques. For example, the standard image, the first abnormal image, and the second abnormal image are preprocessed, including image enhancement, normalization, and data augmentation operations; a deep learning algorithm is constructed using a convolutional neural network architecture, which includes a feature extraction layer, a convolutional layer, and a classification layer; and the performance metrics of the deep learning algorithm are optimized through cross-validation and hyperparameter tuning, including accuracy, recall, and F1 score.

[0027] Based on the optimization of similarity threshold sets and structural defect feature templates, and combined with existing deep learning model technology, this embodiment provides an effective identification model for visual analysis-based equipment maintenance assistance.

[0028] Traditional feature extraction often involves holistic image analysis, which can easily overlook subtle features in critical areas of the instrument, leading to insufficient recognition accuracy. This embodiment, based on existing attention mechanisms and feature extraction techniques, focuses feature extraction on key areas such as the instrument's interlocking surface and cutting edge. Through multi-scale analysis, it acquires microscopic and macroscopic features and fuses them into feature vectors, enhancing the ability to capture defects in critical areas and improving the targeting and comprehensiveness of feature extraction.

[0029] Specifically, the recognition model performs feature extraction and analysis on surface images, including: Enhance the feature extraction of key areas of the instrument to obtain regional features. The key areas include at least the occlusal surface, cutting edge and joint of the instrument. By analyzing surface images corresponding to regional features at different sizes, we can obtain the micro and macro features of the region. Based on the micro and macro features of the region, a feature vector for defect identification is obtained.

[0030] It should be noted that the selection of key areas in this embodiment can be set based on actual needs. Similarly, the dimensions of the micro and macro features of the area can also be set based on actual needs. Moreover, the above settings are all experiences known to those skilled in the art, and this text does not impose any restrictions on them.

[0031] Existing defect assessment methods lack unified quantitative standards, easily leading to inconsistent results due to ambiguous judgment logic. This embodiment assesses residues by comparing feature similarity scores with thresholds, and combines existing feature matching techniques to match feature templates to determine structural defects, establishing a standardized assessment process to ensure consistency and reliability of defect assessment across different instruments and scenarios. Specifically, determining whether the instrument under test has surface residues and / or structural integrity defects includes: The feature similarity between the feature vector of the instrument to be tested and the feature similarity between the standard image are analyzed to obtain the feature similarity score; the detection sensitivity requirement of the instrument to be tested is obtained, and the corresponding similarity threshold is selected from the similarity threshold set based on the detection sensitivity requirement. When the feature similarity score is less than the similarity threshold, it is determined that there are surface residues. The feature vector of the instrument under test and the set of structural defect feature templates are analyzed. When the feature vector matches the structural defect feature template, it is determined that there is a structural integrity defect.

[0032] It should be noted that the feature similarity analysis of the feature vector of the instrument under test and the feature similarity analysis of the standard image, as well as the similarity analysis of the standard image and the first abnormal image mentioned above, can all adopt existing similarity analysis techniques. It is only necessary to keep the similarity analysis strategy the same before and after to ensure that the analysis scale is the same.

[0033] Current detection methods only output defect results, lacking detailed location, classification, and recording of defects, which hinders subsequent processing and traceability. This embodiment generates defect heatmaps, classifies and grades them, and stores them in a database, enabling visualized location, priority ranking, and full-process traceability of defects, thus improving the targeting of maintenance and management efficiency.

[0034] Specifically, when the identification model determines that the instrument under test has surface residues and / or structural integrity defects, it includes: The defective areas are located and a defect heatmap is generated. In this embodiment, the defect heatmap can be generated by combining existing image generation techniques with the defective areas. The instruments to be tested are classified and graded based on the type and severity of defects, and the classification and grading results include cleaning priority and maintenance priority. The defect heat map and classification results are stored in a preset quality traceability database. In this embodiment, the quality traceability database is uniquely associated with the instrument to be tested, thereby achieving full-process traceability.

[0035] Traditional anomaly notification methods are limited and fail to quickly convey defect details and handling suggestions, impacting processing timeliness. This embodiment utilizes existing human-machine interface displays and text generation technologies to produce structured anomaly notification information, notifying responsible parties and synchronizing with the management system. This multi-dimensional output of anomaly information ensures relevant personnel receive detailed information promptly, accelerating the defect handling process.

[0036] Specifically, it obtains and outputs abnormal alerts to prompt the instruments to be tested to be reprocessed, including: The defect heatmap and classification results are displayed through a pre-designed human-computer interaction interface; Based on the defect heatmap and classification results, anomaly alert information is obtained, which includes the defect type, location, and handling suggestions. The abnormality alert will be sent to the designated responsible personnel. The abnormality alert information is synchronized to the preset hospital management equipment, which can be a hospital information system or a disinfection supply center management system.

[0037] In practical applications, surface residues can be caused by incomplete cleaning or by structural defects that prevent proper cleaning using the original cleaning process. However, existing technologies cannot distinguish between independently existing defects and residues caused by structural issues, easily leading to misjudgments in treatment plans. That is, if a structural defect prevents proper cleaning using the original process, even if it is detected and addressed in the current monitoring, it will still result in incomplete cleaning on the next cleaning attempt. This embodiment introduces structural residue features, enabling the model to identify all three types and provide targeted prompts, avoiding missed or incorrect judgments of related defects, improving the accuracy of defect cause analysis, and optimizing maintenance strategies.

[0038] Specifically, training deep learning algorithms using training sample sets to obtain recognition models also includes: The correlation between the similarity thresholds corresponding to surface residues in the similarity threshold set and the structural defect features corresponding to structural integrity defects in the structural defect feature template set is analyzed to obtain the correlation features of surface residue adhesion caused by structural integrity defects. Structural residue features are obtained based on correlation features. These features are used to define the causal relationship between surface residues and structural integrity defects. This causal relationship includes the spatial correspondence between the location of structural defects and the distribution of residues, as well as the matching relationship between the morphology of residues and the type of defects. Structural residue features are integrated into the recognition model. The recognition model is used to identify independent surface residues, independent structural integrity defects, and surface residues caused by structural integrity defects based on the input surface image, similarity threshold set, structural defect feature template set, and structural residue features. Based on the recognition results, anomaly prompts are output.

[0039] Based on the above, by analyzing correlations and extracting associated features, constructing structural residue features and integrating them into the identification model, it is possible to accurately distinguish between independent residues, independent structural defects, and residues caused by structural defects. This effectively solves the problem of ambiguous judgment of defect causes in existing technologies, avoids ineffective cleaning due to misjudgment, reduces the risk of repeated contamination, makes maintenance plans more targeted, improves the efficiency and quality stability of instrument processing, and ensures safety in subsequent use.

[0040] In a simple example, such as surgical scissors, the blade may have a minor nick (structural integrity defect), leaving bloodstains (surface residue) at the nick after cleaning. In this embodiment, the identification model analyzes the structural residue features, finding that the bloodstain distribution spatially corresponds to the nick location, and the bloodstain morphology matches the residue characteristics caused by the nick. It determines that the residue is caused by a structural defect and outputs a suggestion to prioritize repairing the nick before cleaning, avoiding repeated residue buildup from cleaning alone. It can be understood that based on the above determination, when the nick cannot be repaired, the instrument can be prematurely scrapped; that is, the above determination provides data reference for the reasonable scrapping of instruments.

[0041] Existing methods for extracting related features lack a systematic approach, resulting in insufficient feature reliability and impacting the accuracy of identifying related defects. This embodiment addresses this by annotating overlapping regions, constructing a related feature library, and filtering salient features. This standardizes the process of acquiring related features, ensuring that the extracted features are representative and discriminative, thus laying the foundation for accurate identification of related defects.

[0042] Specifically, the associated characteristics of surface residue adhesion caused by structural integrity defects are obtained, including: The samples with overlapping areas in the first and second anomalous images are labeled to determine the spatial correlation parameters between the structural defect area and the residue area. Based on spatial correlation parameters, the residual morphological characteristics corresponding to different structural defect types are statistically analyzed to obtain a correlation feature library; Features with significant differences are selected from the associated feature library as associated features. The significant differences are determined based on a preset instrument maintenance knowledge base.

[0043] Existing models lack correlation verification for coexisting residues and structural defects, which can easily lead to unreasonable processing priority settings. This embodiment verifies the correlation by calling structural residue features, adds correlation identifiers, and adjusts processing priorities to ensure that correlated defects are treated first, thereby improving the rational allocation and processing efficiency of maintenance resources.

[0044] Specifically, when the identification model makes judgments based on structural residual features, it includes: When the identification model detects that the instrument under test has both surface residues and structural integrity defects, it calls upon the structural residue features to verify the correlation between the two. If the distribution location, morphology, and type of surface residue match the structural residue characteristics of the structural integrity defect, an association identifier will be added to the abnormality warning message. The association identifier is used to indicate that the structural defect caused the residue. Adjust the processing priority in the classification and grading results based on the association identifier, so that the processing priority of defects with association relationship is higher than that of independent defects of the same type.

[0045] Example 2 like Figure 2 As shown, this embodiment discloses a vision-based instrument maintenance assistance system, which applies the vision-based instrument maintenance assistance method described above. The system includes: The model building module is used to build a training sample set, which includes standard images of various instruments in normal form, first abnormal images with surface residues, and second abnormal images with structural integrity defects. The deep learning algorithm is trained using the training sample set to obtain the recognition model. The image acquisition module is used to acquire surface images of the instrument to be inspected after cleaning using a high-resolution camera. The visual analysis module is used to input the surface image of the instrument to be inspected into the recognition model. The recognition model extracts and analyzes the features of the surface image to determine whether the instrument to be inspected has surface residues and / or structural integrity defects. When the recognition model determines that the instrument to be inspected has surface residues and / or structural integrity defects, it obtains and outputs an abnormal prompt message to prompt the instrument to be inspected to be reprocessed.

[0046] It should be noted that the vision-based instrument maintenance assistance system of this embodiment corresponds to the aforementioned vision-based instrument maintenance assistance method. Therefore, any content not specifically described in the vision-based instrument maintenance assistance system of this embodiment, including but not limited to functional definitions, working principles, and technical effects, can be referred to the description in the aforementioned vision-based instrument maintenance assistance method, and will not be repeated here.

[0047] In summary, the vision-based instrument maintenance assistance method and system of this embodiment significantly improves the accuracy and efficiency of instrument maintenance by combining high-resolution visual acquisition with deep learning algorithms. Compared with manual visual inspection, it can accurately identify minute residues and structural defects with the help of a trained recognition model, which is especially suitable for the inspection of fine instruments and reduces the risk of missed detection. By adapting dynamic thresholds and feature templates to different instrument needs, it achieves standardized inspection and reduces subjective differences. The introduction of structural residue feature analysis can distinguish between independent defects and related defects and optimize processing priorities. Combined with defect location, classification and quality traceability, it achieves full-process control and synchronizes with the management system to improve collaborative efficiency. It comprehensively solves the problems of unstable quality, difficulty in fine inspection and lack of standardization in traditional methods, and ensures the safety of instrument use and the quality of maintenance.

[0048] In the embodiments provided in this application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor may be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments may be performed by a computer program instructing the associated hardware. During implementation, the program may be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media may be any available medium accessible to a computer. Computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0049] Finally, it should be noted that the above description is only a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for assisting equipment maintenance based on visual analysis, characterized in that, The method includes: A training sample set is constructed, and a recognition model is obtained using the training sample set. The training sample set includes standard images of various instruments in normal form, first abnormal images with surface residues, and second abnormal images with structural integrity defects. The surface image of the instrument to be tested is acquired using a high-resolution camera after cleaning. The surface image of the instrument to be tested is input into the recognition model. The recognition model extracts and analyzes the features of the surface image to determine whether the instrument to be tested has surface residues and / or structural integrity defects. When the recognition model determines that the instrument to be tested has surface residues and / or structural integrity defects, it obtains and outputs an abnormal prompt message to prompt the instrument to be tested to be reprocessed.

2. The instrument maintenance assistance method based on visual analysis according to claim 1, characterized in that, The step of training a deep learning algorithm using the training sample set to obtain a recognition model includes: Analyze the similarity between the standard image and the first abnormal image. Based on the detection sensitivity requirements of different types of instruments and the similarity, obtain a set of similarity thresholds. The set of similarity thresholds stores multiple similarity thresholds in response to different detection sensitivity requirements. The similarity thresholds are used to define the minimum value that the similarity must satisfy under a detection sensitivity requirement. By analyzing the standard image and the second abnormal image, a set of structural defect feature templates is obtained. The set of structural defect feature templates stores multiple structural defect feature templates, which are used to define structural defects. The similarity threshold set and the structural defect feature template set are stored in the recognition model, which is used to output the abnormality prompt information based on the input surface image, the similarity threshold set, and the structural defect feature template set.

3. The instrument maintenance assistance method based on visual analysis according to claim 2, characterized in that, The recognition model performs feature extraction and analysis on the surface image, including: Enhance the feature extraction of key areas of the instrument to obtain regional features, wherein the key areas include at least the occlusal surface, cutting edge and joint of the instrument; By analyzing the surface images corresponding to the regional features at different sizes, the micro-features and macro-features of the region are obtained. Based on the micro-features and macro-features of the region, a feature vector for defect identification is obtained.

4. The instrument maintenance assistance method based on visual analysis according to claim 3, characterized in that, The determination of whether the instrument under test has surface residues and / or structural integrity defects includes: The feature similarity between the feature vector of the device under test and the feature similarity between the standard image is analyzed to obtain a feature similarity score; the detection sensitivity requirement of the device under test is obtained, and based on the detection sensitivity requirement, the corresponding similarity threshold is selected from the similarity threshold set; when the feature similarity score is less than the similarity threshold, it is determined that there are surface residues. The feature vector of the instrument under test and the set of structural defect feature templates are analyzed. When the feature vector matches the structural defect feature template, it is determined that there is a structural integrity defect.

5. The instrument maintenance assistance method based on visual analysis according to claim 4, characterized in that, When the identification model determines that the instrument under test has surface residues and / or structural integrity defects, it includes: Locate the defective areas and generate a defect heat map; The instruments under test are classified and graded based on the type and severity of defects to obtain classification and grading results, which include cleaning priority and maintenance priority. The defect heatmap and the classification and grading results are stored in a preset quality traceability database.

6. The instrument maintenance assistance method based on visual analysis according to claim 5, characterized in that, The step of obtaining and outputting anomaly alert information to prompt reprocessing of the instrument under test includes: The defect heatmap and the classification and grading results are displayed through a preset human-computer interaction interface; Based on the defect heatmap and the classification and grading results, anomaly warning information is obtained, which includes defect type, location, and processing suggestions. The abnormality alert will be sent to the designated responsible persons. The abnormal notification information will be synchronized to the preset hospital management equipment.

7. The instrument maintenance assistance method based on visual analysis according to claim 6, characterized in that, The step of training the deep learning algorithm using the training sample set to obtain the recognition model further includes: By analyzing the correlation between the similarity thresholds corresponding to surface residues in the similarity threshold set and the structural defect features corresponding to structural integrity defects in the structural defect feature template set, the correlation features of surface residue adhesion caused by structural integrity defects are obtained. Based on the aforementioned correlation features, structural residue features are obtained. These structural residue features are used to define the causal relationship between surface residues and structural integrity defects. The causal relationship includes the spatial correspondence between the location of structural defects and the distribution of residues, and the matching relationship between the morphology of residues and the type of defects. The structural residue features are integrated into the recognition model, which is used to identify independently existing surface residues, independently existing structural integrity defects, and surface residues caused by structural integrity defects based on the input surface image, the similarity threshold set, the structural defect feature template set, and the structural residue features. Based on the recognition results, the abnormal prompt information is output.

8. The instrument maintenance assistance method based on visual analysis according to claim 7, characterized in that, The associated features obtained due to structural integrity defects leading to surface residue adhesion include: The samples with overlapping areas in the first and second abnormal images are labeled to determine the spatial correlation parameters between the structural defect area and the residue area. Based on the spatial correlation parameters, the residual morphological characteristics corresponding to different structural defect types are statistically analyzed to obtain a correlation feature library; Features with significant differences are selected from the associated feature library as the associated features, and the significant differences are determined based on a preset instrument maintenance knowledge base.

9. The instrument maintenance assistance method based on visual analysis according to claim 7, characterized in that, When the identification model makes a judgment based on the structural residual features, it includes: When the identification model detects that the instrument under test has both surface residues and structural integrity defects, it calls the structural residue features to verify the correlation between the two. If the distribution location and morphology of the surface residue match the location and type of the structural integrity defect, then an association identifier is added to the abnormality warning information. The association identifier is used to identify that the structural defect caused the residue. Based on the association identifier, the processing priority in the classification and grading results is adjusted so that the processing priority of defects with association is higher than that of independent defects of the same type.

10. A vision-based instrument maintenance assistance system, employing the vision-based instrument maintenance assistance method as described in any one of claims 1-9, characterized in that, The system includes: The model building module is used to build a training sample set, which includes standard images of various instruments in normal form, first abnormal images with surface residues, and second abnormal images with structural integrity defects. The deep learning algorithm is trained using the training sample set to obtain a recognition model. The image acquisition module is used to acquire surface images of the instrument to be inspected after cleaning using a high-resolution camera. The visual analysis module is used to input the surface image of the instrument to be inspected into the recognition model. The recognition model performs feature extraction and analysis on the surface image to determine whether the instrument to be inspected has surface residues and / or structural integrity defects. When the recognition model determines that the instrument to be inspected has surface residues and / or structural integrity defects, it obtains and outputs an abnormal prompt message to prompt the instrument to be inspected to be reprocessed.

Images