Image recognition-based spray needle abrasion degree evaluation method and device

Abstract

The application provides a kind of based on image recognition's degree of needle erosion evaluation method and device, it is applied to image recognition technical field, above-mentioned method includes: obtaining the needle picture to be evaluated;The needle picture is input to the trained erosion area identification model, and the classification probability graph output by the erosion area identification model is obtained;The classification probability graph is carried out binary segmentation processing, and the erosion area is obtained;Determine the pixel point number of the needle cone area corresponding to the needle picture in the erosion area;The pixel point number is compared with the preset erosion pixel point number threshold, and the degree of needle erosion of the needle picture is determined. Through the application, the detection efficiency of needle erosion can be improved, and false judgment and missed judgment caused by human subjective factors can be avoided.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and apparatus for evaluating the degree of abrasion of a nozzle based on image recognition. Background Technology

[0002] Currently, the assessment and monitoring of the degree of nozzle abrasion mainly rely on visual inspection after periodic shutdowns, manual measurement, or simple sample comparison.

[0003] Traditional methods have significant limitations: they rely heavily on the experience of maintenance personnel, lack standardized and quantifiable evaluation criteria, and are prone to bias in assessment results. They require close-range manual inspection and measurement, a cumbersome and time-consuming process that increases unplanned downtime. Furthermore, for irregular and discontinuous erosion pits, precise location and area calculation are difficult, making it impossible to achieve a refined assessment of the degree of erosion.

[0004] It is evident that the methods for assessing the degree of nozzle abrasion in related technologies suffer from a lack of automation. Summary of the Invention

[0005] This invention provides a method and apparatus for evaluating the degree of nozzle abrasion based on image recognition, which solves the problem of low automation in the existing nozzle abrasion evaluation methods, greatly improves the detection efficiency of nozzle abrasion, and avoids misjudgment and missed judgment caused by human subjective factors.

[0006] This invention provides a method for evaluating the degree of nozzle abrasion based on image recognition, comprising the following steps: acquiring an image of the nozzle to be evaluated; inputting the nozzle image into a trained abrasion region recognition model to obtain a classification probability map output by the abrasion region recognition model; performing binary segmentation on the classification probability map to obtain abrasion regions; determining the number of pixels in the nozzle conical region corresponding to the nozzle image in the abrasion region; comparing the number of pixels with a preset abrasion pixel count threshold to determine the degree of nozzle abrasion in the nozzle image.

[0007] According to the present invention, a method for evaluating the degree of nozzle abrasion based on image recognition, before inputting the nozzle image into a trained abrasion region recognition model to obtain the classification probability map output by the abrasion region recognition model, the method further includes: acquiring a dataset of nozzle images under historical downtime conditions; annotating the original images in the nozzle image dataset with abrasion regions to generate an abrasion region mask, wherein each pixel is divided into an abrasion region or normal background; constructing a training dataset based on the abrasion region mask and the original images corresponding to the abrasion region mask; and training a preset semantic segmentation model based on the training dataset to obtain a trained abrasion region recognition model.

[0008] According to the image recognition-based method for evaluating the degree of nozzle abrasion provided by the present invention, determining the number of pixels in the nozzle conical region corresponding to the nozzle image in the abraded region includes: determining the nozzle conical region based on the nozzle image; and obtaining the number of pixels located within the nozzle conical region and identified as abraded based on the intersection of the nozzle conical region and the abraded region, wherein the number of pixels is calculated using the following formula: ;in, This indicates the number of pixels. This represents the set of pixels in the cone-shaped region of the nozzle. This represents the set of pixels in the eroded area.

[0009] According to the present invention, a method for evaluating the degree of nozzle abrasion based on image recognition is provided, wherein determining the nozzle conical region based on the nozzle image includes: responding to a user's annotation input on the nozzle conical region in the nozzle image received through an interactive tool, and dividing the nozzle conical region in the nozzle image based on the annotation input.

[0010] According to the present invention, a method for evaluating the degree of nozzle abrasion based on image recognition is provided. The step of performing binary segmentation on the classification probability map to obtain the abrasion region includes: comparing the probability value of each pixel in the classification probability map with a preset probability threshold to obtain a comparison result; generating a binary segmentation map based on the comparison result, wherein pixels in the binary segmentation map with probability values ​​greater than the probability threshold are classified as abrasion, and pixels with probability values ​​less than or equal to the probability threshold are classified as background; and extracting the abrasion region based on the binary segmentation map, wherein the abrasion region is composed of all pixels classified as abrasion.

[0011] According to the present invention, a method for evaluating the degree of nozzle abrasion based on image recognition is provided. The step of comparing the number of pixels with a preset threshold for the number of abraded pixels to determine the degree of nozzle abrasion in the nozzle image includes: determining at least one threshold for the number of abraded pixels, wherein the threshold for the number of abraded pixels is used to divide the degree of abrasion into different levels; comparing the number of pixels with the at least one threshold for the number of abraded pixels, and determining the degree of nozzle abrasion in the nozzle image according to the threshold range to which the number of pixels belongs.

[0012] This invention also provides an image recognition-based device for evaluating the degree of nozzle abrasion, comprising the following modules: an acquisition module for acquiring an image of the nozzle to be evaluated; a region recognition module for inputting the nozzle image into a trained abrasion region recognition model to obtain a classification probability map output by the abrasion region recognition model; a binary segmentation module for performing binary segmentation on the classification probability map to obtain abrasion regions; a determination module for determining the number of pixels in the nozzle cone region corresponding to the nozzle image in the abrasion region; and a comparison module for comparing the number of pixels with a preset abrasion pixel count threshold to determine the degree of nozzle abrasion in the nozzle image.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the image recognition-based method for evaluating the degree of nozzle abrasion as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the image recognition-based method for evaluating the degree of nozzle abrasion as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the image recognition-based method for evaluating the degree of nozzle abrasion as described above.

[0016] The present invention provides a method and apparatus for assessing the degree of nozzle abrasion based on image recognition. First, an image of the nozzle to be assessed is acquired, providing an accurate image data foundation for subsequent analysis. Second, the image is input into a trained abrasion region recognition model, which automatically outputs a classification probability map, effectively improving the accuracy and efficiency of abrasion region recognition. Next, the classification probability map is subjected to binary segmentation to clearly extract the abrasion region, avoiding the subjectivity of manual interpretation. Then, the number of pixels in the nozzle's conical region within the abrasion area is determined, converting the abrasion condition into a quantifiable numerical indicator. Finally, by comparing the number of pixels with a preset abrasion pixel count threshold, the degree of nozzle abrasion is objectively determined, thus providing a reliable basis for maintenance decisions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1This is a schematic flowchart of the image recognition-based method for evaluating the abrasion degree of a nozzle provided by the present invention.

[0019] Figure 2 This is an overall flowchart of the image recognition-based method for evaluating the abrasion degree of nozzles provided by the present invention.

[0020] Figure 3 This is a typical nozzle abrasion image provided by the present invention.

[0021] Figure 4 This is a schematic diagram of the typical needle abrasion area identification results provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the nozzle area marking results provided by the present invention.

[0023] Figure 6 This is a schematic diagram of the module of the nozzle abrasion assessment device based on image recognition provided by the present invention.

[0024] Figure 7 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] refer to Figure 1 , Figure 1 This is a flowchart illustrating the image recognition-based method for evaluating the abrasion degree of a nozzle provided by the present invention, which specifically includes the following steps.

[0027] Step 101: Obtain images of the nozzles to be evaluated.

[0028] In this embodiment of the invention, images of the nozzle to be evaluated (images of the nozzle surface) are acquired. In specific operation, a fixed high-resolution industrial camera is used, along with a standard lens, lighting device, and mechanical support, to ensure that the distance, angle, and lighting conditions between the camera and the nozzle are completely consistent for each shot, so as to eliminate the influence of environmental variables on image quality.

[0029] After the shooting was completed, the acquired nozzle images were screened, and only clear and typical nozzle surface images were retained as nozzle images to be evaluated.

[0030] Step 102: Input the nozzle image into the trained abrasion region recognition model to obtain the classification probability map output by the abrasion region recognition model.

[0031] In this embodiment of the invention, the acquired nozzle image is input into a pre-trained abrasion region recognition model. The abrasion region recognition model is built on the SegNet architecture and trained using historical datasets (composed of "original image-abrasion region mask" pairs), enabling it to automatically learn the abrasion features of the nozzle surface.

[0032] The nozzle image is input into the trained abrasion region recognition model. The abrasion region recognition model performs semantic segmentation on each pixel and outputs a category probability map, where each pixel value represents the probability that the point belongs to the abrasion region (ranging from 0 to 1).

[0033] Step 103: Perform binary segmentation on the classification probability map to obtain the abrasion region.

[0034] In this embodiment of the invention, after obtaining the classification probability map, binary segmentation is required to extract the clearly defined abrasion areas.

[0035] In practice, a fixed threshold (usually a probability of 0.5) is set. Pixels with a probability greater than 0.5 in the probability map are identified as abraded areas, while pixels with a probability less than or equal to 0.5 are identified as normal background areas. This thresholding operation generates a binary segmentation image, where abraded areas are represented by white pixels and background areas are represented by black pixels.

[0036] Step 104: Determine the number of pixels in the nozzle cone region corresponding to the nozzle image in the abrasive region.

[0037] In this embodiment of the invention, the aim is to quantify the distribution of the abrasion zone in the critical part of the nozzle (i.e., the conical region of the nozzle).

[0038] Using an interactive image annotation tool (such as Matlab's image annotation module), the conical region in the nozzle image is manually selected to obtain the set of pixels in that region. Simultaneously, the set of pixels representing the abrasion region is extracted from the binary segmentation image. Then, by calculating the intersection of the two sets, the number of pixels located within the nozzle's conical region and identified as abrasion-prone is obtained.

[0039] Step 105: Compare the number of pixels with the preset threshold for the number of abraded pixels to determine the degree of abrasion of the nozzle image.

[0040] In this embodiment of the invention, the degree of abrasion of the nozzle is evaluated by comparing the calculated number of pixels with a preset threshold for the number of abraded pixels. The threshold is set based on practical engineering experience; for example, it can be divided into multiple intervals: 0-100 pixels represent mild abrasion, 100-400 pixels represent moderate abrasion, and above 400 pixels represent severe abrasion. The abrasion level of the nozzle is directly determined based on the range in which the number of pixels falls. This method provides a unified and quantitative evaluation standard, avoids subjective human bias, and is suitable for automated monitoring.

[0041] Through this embodiment of the invention, firstly, images of the nozzle to be evaluated are acquired, providing an accurate image data foundation for subsequent analysis; secondly, the images are input into a trained abrasion region recognition model, which automatically outputs a classification probability map, effectively improving the accuracy and efficiency of abrasion region recognition; next, the classification probability map is subjected to binary segmentation to clearly extract the abrasion region, avoiding the subjectivity of manual interpretation; then, the number of pixels in the nozzle cone region within the abrasion region is determined, converting the abrasion condition into a quantifiable numerical indicator; finally, by comparing the number of pixels with a preset abrasion pixel count threshold, the degree of nozzle abrasion is objectively determined, thereby providing a reliable basis for maintenance decisions.

[0042] According to the image recognition-based method for evaluating the abrasion degree of a nozzle provided by the present invention, before inputting the nozzle image into a trained abrasion region recognition model to obtain the classification probability map output by the abrasion region recognition model, the method further includes: Obtain a dataset of nozzle images from historical downtime conditions; Abrasion regions are labeled in the original images of the nozzle image dataset to generate abrasion region masks, where each pixel is divided into an abrasion region or normal background. A training dataset is constructed based on the abrasion region mask and the original image corresponding to the abrasion region mask. The pre-defined semantic segmentation model is trained based on the training dataset to obtain a trained abrasion region recognition model.

[0043] In this embodiment of the invention, a fixed high-resolution industrial camera is used, and the distance, angle, and lighting conditions are completely consistent for each shot; clear and typical nozzle images are selected as samples.

[0044] Label the abrasion areas of the nozzle in the nozzle image, and divide each pixel in the nozzle image into two categories: abrasion area and normal background, which are used as abrasion area masks; construct "original image-abrasion area mask" pairs as training datasets; input them into the SegNet model for training to obtain the nozzle abrasion area recognition model.

[0045] In this embodiment of the invention, a standardized image acquisition environment must first be established. A fixed high-resolution industrial camera, along with a specific lens, lighting device, and mechanical support, is used to ensure that the relative distance between the camera and the nozzle, the shooting angle, and the lighting conditions remain completely consistent each time images are taken under historical downtime conditions. After obtaining a large number of nozzle surface images (i.e., nozzle pictures) through this standardized acquisition method, images with clear details and typical abrasion characteristics need to be further selected as valid samples to form the initial dataset.

[0046] Each original image in the initial dataset undergoes refined pixel-level annotation. Specifically, professional image annotation tools (such as Matlab's image annotation module) are used, with experts manually identifying and selecting the nozzle abrasion areas in the images. During this process, each pixel in the image is divided into two categories: pixels belonging to the abrasion area are labeled with a specific value (e.g., white), while pixels belonging to the normal background are labeled with another value (e.g., black), thus generating a binary abrasion area mask that corresponds one-to-one with the original image.

[0047] Each original image is paired with its generated abrasion mask, forming a structured "original image-abrasion mask" pair. All these paired data constitute the training dataset required for model training. To improve the efficiency of subsequent model training, the images in the dataset can be preprocessed uniformly, such as adjusting the image resolution to a size suitable for the model (e.g., 128x240 pixels). By constructing this paired training dataset, complete input and expected output are provided for supervised learning.

[0048] The constructed training dataset is input into a pre-defined semantic segmentation model for training. Specifically, this invention employs the SegNet model architecture, which includes an encoder-decoder structure. During training, the encoder can be configured with a two-layer structure: the first layer uses 64 filters, and the second layer uses 128 filters. The model automatically adjusts its internal parameters by continuously learning the feature mapping relationship between the "original image - abrasion region mask" pair, ultimately obtaining a well-trained abrasion region recognition model capable of accurately identifying the abrasion regions of the nozzle. This trained abrasion region recognition model has the ability to automatically convert new nozzle images into classification probability maps.

[0049] Through this invention, from standardized data collection and refined annotation to dataset construction, intelligent recognition is finally achieved through SegNet model training. The entire process ensures the accuracy and robustness of the model's learning of the nozzle abrasion features.

[0050] According to the present invention, a method for evaluating the degree of nozzle abrasion based on image recognition is provided, which determines the number of pixels in the nozzle conical region corresponding to the nozzle image in the abraded area, including: Based on the nozzle image, determine the nozzle cone region; Based on the intersection of the nozzle conical region and the abrasion region, the number of pixels located within the nozzle conical region and identified as abrasion-affected is obtained, where the number of pixels is calculated using the following formula: ; in, Indicates the number of pixels. This represents the set of pixels representing the cone-shaped region of the nozzle. This represents the set of pixels in the eroded area.

[0051] In this embodiment of the invention, an interactive tool is used to annotate the pixels in the nozzle conical working section region (i.e., the nozzle conical region) of the nozzle image. The set of pixels in the nozzle conical region is denoted as... .

[0052] The set of pixels in the abrasion region obtained in the above embodiments is denoted as . .

[0053] The total number of pixels located in the nozzle cone region and identified as abrasive (i.e., located in the abrasive region) is determined using the following formula: ; in, This represents the total number of pixels in the set.

[0054] In this embodiment of the invention, firstly, it is necessary to locate the conical working section region (i.e., the conical region of the nozzle) in the nozzle image. For example, this can be done manually by an operator using a professional image annotation tool (such as the image annotation module in Matlab) to manually outline the conical working section of the nozzle on the nozzle image. This conical region is the key part of the nozzle that directly participates in water flow regulation, and its annotation results form a complete set of pixels. To ensure consistency in subsequent calculations, the resolution generated by the annotation must be consistent with the output of the abrasion region recognition model; resolution adjustment may be necessary.

[0055] Next, the intersection of the abrasion region and the nozzle conical region is calculated. The pixel set of the nozzle conical region obtained in the previous steps is compared and analyzed with the pixel set of the abrasion region generated by model recognition to determine the pixels that belong to both regions simultaneously, and the intersection result is obtained.

[0056] Finally, the intersection results are quantitatively analyzed. The total number of pixels contained in the intersection results is directly calculated; this value represents the effective number of abraded pixels located within the nozzle's conical region. This quantitative result directly reflects the severity of abrasion on the nozzle's critical working surface.

[0057] Through the embodiments of the present invention, the erosion status of key areas of the nozzle is quantified by combining operations of region positioning, intersection calculation and pixel statistics. This effectively avoids the result deviation caused by unclear evaluation areas and significantly improves the accuracy and reliability of erosion degree judgment.

[0058] According to the present invention, a method for evaluating the abrasion degree of a nozzle based on image recognition is provided, which determines the conical region of the nozzle based on an image of the nozzle, including: In response to user annotation input on the nozzle cone region in the nozzle image received through the interactive tool, the nozzle cone region in the nozzle image is divided based on the annotation input.

[0059] In this embodiment of the invention, users are allowed to precisely identify specific areas by manually drawing or selecting boxes. In actual operation, users use a mouse or other input devices to meticulously trace the outer contour of the nozzle's conical working section to obtain the nozzle's conical area.

[0060] Since the conical region of the nozzle is a critical component directly involved in water flow regulation during turbine operation, its shape and location have clear engineering definitions and require accurate identification by personnel with specialized knowledge based on the nozzle's design features. When labeling, users need to clearly define the start and end boundaries of the conical region on the image, ensuring that all potentially abrasive working surfaces are included.

[0061] After the annotation input is completed, the corresponding region mask is automatically generated based on the annotation trajectory. During this process, pixels within the nozzle cone-shaped area are assigned a specific identifier (such as white), while pixels outside the area are treated as background (such as black), thus forming a complete binary region segmentation map.

[0062] To ensure the accuracy of subsequent calculations, the generated cone-shaped region mask needs to maintain the same image resolution as the identified abrasion region. If the abrasion region image output by the model has undergone downsampling, the cone-shaped region mask also needs to be adjusted to the same resolution to ensure that the two regions can be accurately compared at the same scale.

[0063] Through this invention, the interactive annotation method ensures the accuracy and consistency of the conical region division, laying a solid foundation for the precise quantification of subsequent abrasion levels. This method effectively overcomes the errors that may arise from fully automated identification and is particularly suitable for industrial components with fixed geometric features, such as spray needles.

[0064] According to the present invention, a method for evaluating the degree of nozzle abrasion based on image recognition is provided, which performs binary segmentation on a classification probability map to obtain the abrasion region, including: The probability value of each pixel in the classification probability map is compared with a preset probability threshold to obtain the comparison result; A binary segmentation map is generated based on the comparison results. Pixels with a probability value greater than the probability threshold in the binary segmentation map are classified as abrasion, and pixels with a probability value less than or equal to the probability threshold are classified as background. The abrasion region is extracted based on the binary segmentation map, where the abrasion region consists of all pixels that are classified as abrasion.

[0065] In this embodiment of the invention, the probability value of each pixel in the classification probability map is compared with a preset probability threshold. The classification probability map is generated by an abrasion region recognition model, where each pixel value represents the probability that the point belongs to an abrasion region, ranging from 0 to 1. All pixels in the probability map are automatically traversed, and the probability value of each point is compared with a fixed threshold. This threshold is typically based on a standard setting during model training and is used to distinguish between abrasion and non-abrasion states, ensuring the accuracy of segmentation.

[0066] Based on the comparison results, a binary segmentation map is generated. In the binary segmentation map, pixels with a probability value greater than a preset threshold are uniformly classified into the abrasion category, usually represented by white pixels; pixels with a probability value less than or equal to the preset threshold are classified into the background category, usually represented by black pixels. This binarization process simplifies the complex probability map into clear region segmentation, intuitively highlighting the outline of the abrasion area.

[0067] Based on the generated binary segmentation image, the abraded region is finally extracted. The abraded region consists of all the white pixels in the binary segmentation image that are classified as abraded; these pixels together form continuous abrasion patches or discrete abrasion points. In this way, the abraded parts of the nozzle surface in the nozzle image can be directly located and quantified.

[0068] Through this embodiment of the invention, a binary segmentation map is generated by comparing each pixel value in the classification probability map with a preset threshold to distinguish the abrasion area from the background, thereby accurately extracting the abrasion area. This process effectively improves the accuracy and reliability of abrasion detection, avoids interference from blurred boundaries or background noise, and achieves automated and refined assessment of the degree of nozzle abrasion.

[0069] According to the present invention, a method for evaluating the degree of nozzle abrasion based on image recognition compares the number of pixels with a preset threshold for the number of abraded pixels to determine the degree of nozzle abrasion in the nozzle image, including: Determine at least one threshold for the number of abraded pixels, which is used to classify the degree of abrasion into different levels; The number of pixels is compared with at least one threshold for the number of abraded pixels, and the degree of abrasion of the nozzle image is determined based on the threshold range to which the number of pixels belongs.

[0070] In this embodiment of the invention, based on actual engineering experience and historical data statistics, one or more threshold values ​​for the number of etched pixels are set as critical points for classifying the degree of etch. For example, two threshold points can be set to divide the degree of etch into three levels: mild etch, moderate etch, and severe etch.

[0071] The calculated number of abraded pixels is automatically compared with a preset threshold. This process is achieved through a simple numerical range judgment: the system automatically detects which preset threshold range the actual number of pixels falls into. For example, when the number of pixels is below the first threshold, it is judged as light abrasion; when the number of pixels is between the first and second thresholds, it is judged as moderate abrasion; and when the number of pixels exceeds the highest threshold, it is judged as severe abrasion.

[0072] The comparison results output a clear level of abrasion severity. Intuitive assessment conclusions are generated, such as qualitative descriptions like "mild abrasion," "moderate abrasion," or "severe abrasion," along with specific pixel counts as a quantitative reference. This method transforms consecutive pixel counts into discrete abrasion levels.

[0073] Through the embodiments of the present invention, by setting one or more threshold values ​​for the number of abrasive pixels, the number of pixels in the extracted abrasive region is mapped to a preset abrasive level range, thereby realizing the quantification and graded evaluation of the abrasion degree of the nozzle.

[0074] The following describes an example of the practical application of the image recognition-based method for evaluating the abrasion degree of a nozzle provided by the present invention.

[0075] refer to Figure 2 , Figure 2 This is an overall flowchart of the image recognition-based method for evaluating the abrasion degree of nozzles provided by the present invention.

[0076] First, an image of the nozzle is input, and the nozzle region is labeled and input into the abrasion region recognition model. The nozzle region labeling generates a binary image of the nozzle region, and the abrasion region recognition model outputs a classification probability map, which is then processed by binary segmentation to obtain a binary image of the abrasion region. Subsequently, the binary image of the nozzle region and the binary image of the abrasion region are combined to calculate the total number of abrasion pixels in the nozzle region. Finally, a threshold judgment is used to assess the degree of abrasion of the nozzle in the nozzle image.

[0077] Based on the method proposed in this invention, the abrasion images of the nozzles collected by the impact turbine testing platform are experimentally verified, specifically including the following steps.

[0078] Step 1: Obtain nozzle image data under historical shutdown conditions.

[0079] Sub-step A1: Use a fixed high-resolution industrial camera, lens, lighting device and mechanical support to ensure that the distance, angle and lighting conditions are completely consistent for each shot in order to obtain a high-definition image of the nozzle surface.

[0080] Sub-step A2: Select clear and typical images as samples.

[0081] refer to Figure 3 , Figure 3 This is a typical nozzle abrasion image provided by the present invention.

[0082] Several images of the jet nozzles of an impact turbine in a stopped state were captured by a high-speed camera, with a resolution of 1080×1920. Typical images are shown below. Figure 3 As shown.

[0083] Step 2: Train the SegNet model for recognizing the abrasion area of ​​the nozzle; Sub-step B1: Label the abrasion area on the back of the nozzle in the image. Use the Matlab image labeling tool to manually outline the abrasion area of ​​the nozzle as a mask for the abrasion area.

[0084] Sub-step B2: Construct "original image-eroded region mask" pairs as the training dataset.

[0085] Sub-step B3: Input the training dataset into the SegNet model for training to obtain the nozzle abrasion region recognition model. Here, to reduce the computational load, the resolution can be reduced, and the input layer size is set to 128×240. The SegNet model has a 2-layer encoder structure, with 64 filters in the first layer and 128 filters in the second layer.

[0086] Step 3: Input the newly acquired nozzle image into the SegNet model for abrasion region recognition to obtain the region classification probability map, and create a binary segmentation to obtain the abrasion region.

[0087] Sub-step C1: Input the newly acquired nozzle image into the SegNet model for abrasion region recognition. Obtain the category probability map.

[0088] refer to Figure 4 , Figure 4 This is a schematic diagram of the typical needle abrasion area identification results provided by the present invention.

[0089] Sub-step C2: Create a binary segmentation map. Regions with a probability greater than 0.5 are considered abrasion areas, and regions with a probability less than 0.5 are considered background areas. The processed binary image is as follows. Figure 4 As shown in the image. The white areas represent the abraded areas, and the black areas represent the background areas.

[0090] Step 4: Input the newly acquired images into the two models mentioned above to identify the functional areas and abrasion areas; refer to Figure 5 , Figure 5 This is a schematic diagram of the nozzle area marking results provided by the present invention.

[0091] Sub-step D1: Use interactive tools to annotate the pixels of the nozzle cone-shaped working section region in the nozzle image. The annotation result is as follows: Figure 5 As shown. The white area represents the nozzle area, and the black area represents the background area. The resolution and model results are kept consistent, downsampled to 128×240.

[0092] Sub-step D2: The total number of pixels in the abraded area obtained in step 3 is 187.

[0093] Sub-step D3: Calculate the number of pixels located within the aforementioned nozzle conical region and identified as abraded by the model. Here, the nozzle region completely encompasses the identified abraded area. Therefore, the total number of nozzle abraded pixels is 187.

[0094] Step 5: Set the threshold for the number of abraded pixels within the nozzle area and evaluate the degree of abrasion of the nozzle; Here, two thresholds (100 and 400 pixels) are set to evaluate the degree of abrasion: 0-100 is mild abrasion, 100-400 is moderate abrasion, and above 400 is severe abrasion. Therefore, it can be determined that the abrasion condition of this nozzle is moderate.

[0095] To address the erosion problem of nozzles in impact turbines, this invention proposes an image recognition-based method for assessing the degree of nozzle erosion. By automatically analyzing the acquired images using a trained SegNet model, the number of pixels in the eroded area on the nozzle cone surface can be accurately calculated, significantly improving detection efficiency and avoiding misjudgments and omissions caused by human subjectivity and fatigue. This provides a core technological foundation for online monitoring and periodic automatic assessment of the status of hydropower station units.

[0096] The image recognition-based nozzle abrasion assessment device provided by the present invention will be described below. The image recognition-based nozzle abrasion assessment device described below and the image recognition-based nozzle abrasion assessment method described above can be referred to in correspondence.

[0097] refer to Figure 6 , Figure 6 This is a schematic diagram of the module of the nozzle abrasion assessment device based on image recognition provided by the present invention.

[0098] The acquisition module 601 is used to acquire images of the nozzle to be evaluated; The region recognition module 602 is used to input the nozzle image into the trained abrasion region recognition model to obtain the classification probability map output by the abrasion region recognition model. Binary segmentation module 603 is used to perform binary segmentation on the classification probability map to obtain the abrasion region; The determination module 604 is used to determine the number of pixels in the nozzle cone region corresponding to the nozzle image in the abrasive region. The comparison module 605 is used to compare the number of pixels with a preset threshold for the number of abraded pixels to determine the degree of abrasion of the nozzle image.

[0099] Specifically, the image recognition-based nozzle abrasion assessment device provided by the present invention can realize all the method steps implemented in the above-mentioned image recognition-based nozzle abrasion assessment method embodiment, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0100] Figure 7 This is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as... Figure 7 As shown, the electronic device may include a processor 710, a communications interface 720, a memory 730, and a communication bus 740. The processor 710, communications interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute an image recognition-based method for evaluating the degree of nozzle abrasion. This method includes: acquiring an image of the nozzle to be evaluated; inputting the nozzle image into a trained abrasion region recognition model to obtain a classification probability map output by the abrasion region recognition model; performing binary segmentation on the classification probability map to obtain the abrasion region; determining the number of pixels in the nozzle cone region corresponding to the nozzle image within the abrasion region; and comparing the number of pixels with a preset threshold for the number of abrasion pixels to determine the degree of nozzle abrasion in the nozzle image.

[0101] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0102] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the image recognition-based method for evaluating the degree of nozzle abrasion provided by the above methods. The method includes: acquiring an image of the nozzle to be evaluated; inputting the nozzle image into a trained abrasion region recognition model to obtain a classification probability map output by the abrasion region recognition model; performing binary segmentation on the classification probability map to obtain the abrasion region; determining the number of pixels in the nozzle cone region corresponding to the nozzle image in the abrasion region; and comparing the number of pixels with a preset threshold for the number of abrasion pixels to determine the degree of nozzle abrasion in the nozzle image.

[0103] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the image recognition-based method for evaluating the degree of nozzle abrasion provided by the above methods. The method includes: acquiring an image of the nozzle to be evaluated; inputting the nozzle image into a trained abrasion region recognition model to obtain a classification probability map output by the abrasion region recognition model; performing binary segmentation on the classification probability map to obtain the abrasion region; determining the number of pixels in the nozzle cone region corresponding to the nozzle image in the abrasion region; and comparing the number of pixels with a preset threshold for the number of abrasion pixels to determine the degree of nozzle abrasion in the nozzle image.

[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the degree of nozzle abrasion based on image recognition, characterized in that, include: Obtain images of the nozzle to be evaluated; The nozzle image is input into the trained abrasion region recognition model to obtain the classification probability map output by the abrasion region recognition model. The classification probability map is subjected to binary segmentation to obtain the abrasion region; Determine the number of pixels in the nozzle cone region corresponding to the nozzle image in the abrasive region; The number of pixels is compared with a preset threshold for the number of abraded pixels to determine the degree of abrasion of the nozzle image.

2. The method for evaluating the degree of nozzle abrasion based on image recognition according to claim 1, characterized in that, Before inputting the nozzle image into the trained abrasion region recognition model to obtain the classification probability map output by the abrasion region recognition model, the method further includes: Obtain a dataset of nozzle images from historical downtime conditions; The original images in the nozzle image dataset are labeled with abrasion regions to generate an abrasion region mask, wherein each pixel is divided into an abrasion region or a normal background. A training dataset is constructed based on the abrasion region mask and the original image corresponding to the abrasion region mask; The preset semantic segmentation model is trained based on the training dataset to obtain a trained abrasion region recognition model.

3. The method for evaluating the degree of nozzle abrasion based on image recognition according to claim 1, characterized in that, Determining the number of pixels in the nozzle cone region corresponding to the nozzle image in the abraded region includes: Based on the image of the nozzle, the cone-shaped region of the nozzle is determined; Based on the intersection of the nozzle conical region and the abrasion region, the number of pixels located within the nozzle conical region and identified as abrasion points is obtained, wherein the number of pixels is calculated using the following formula: ； in, This indicates the number of pixels. This represents the set of pixels in the cone-shaped region of the nozzle. This represents the set of pixels in the eroded area.

4. The method for evaluating the degree of nozzle abrasion based on image recognition according to claim 3, characterized in that, The step of determining the cone-shaped region of the nozzle based on the nozzle image includes: In response to user annotation input received via an interactive tool regarding the nozzle cone region in the nozzle image, the nozzle cone region in the nozzle image is divided based on the annotation input.

5. The method for evaluating the degree of nozzle abrasion based on image recognition according to claim 1, characterized in that, The binary segmentation of the classification probability map yields the abrasion region, including: The probability value of each pixel in the classification probability map is compared with a preset probability threshold to obtain the comparison result; A binary segmentation map is generated based on the comparison result, wherein pixels with a probability value greater than the probability threshold in the binary segmentation map are classified as abrasion, and pixels with a probability value less than or equal to the probability threshold are classified as background. The abrasion region is extracted based on the binary segmentation map, wherein the abrasion region is composed of all pixels that are divided into abrasion regions.

6. The method for evaluating the degree of nozzle abrasion based on image recognition according to claim 1, characterized in that, The step of comparing the number of pixels with a preset threshold for the number of abraded pixels to determine the degree of abrasion of the nozzle image includes: Determine at least one threshold for the number of abraded pixels, the threshold being used to classify the degree of abrasion into different levels; The number of pixels is compared with the threshold number of at least one abraded pixel, and the degree of abrasion of the nozzle image is determined according to the threshold range to which the number of pixels belongs.

7. A device for evaluating the degree of nozzle abrasion based on image recognition, characterized in that, include: The acquisition module is used to acquire images of the nozzles to be evaluated. The region recognition module is used to input the spray needle image into the trained abrasion region recognition model to obtain the classification probability map output by the abrasion region recognition model; The binary segmentation module is used to perform binary segmentation on the classification probability map to obtain the abrasion region. The determination module is used to determine the number of pixels in the nozzle cone region corresponding to the nozzle image in the abrasive region; The comparison module is used to compare the number of pixels with a preset threshold for the number of abraded pixels to determine the degree of abrasion of the nozzle image.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the image recognition-based method for evaluating the degree of nozzle abrasion as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the image recognition-based method for evaluating the degree of nozzle abrasion as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the image recognition-based method for evaluating the degree of nozzle abrasion as described in any one of claims 1 to 6.

Images