Image recognition-based water pump impeller cleaning control method and system

Abstract

This invention relates to the field of image recognition and processing technology, and discloses a water pump impeller cleaning control method and system based on image recognition. The method includes: acquiring an image sequence of the internal flow channel region of the water pump impeller; performing brightness nonlinearity suppression and curvature-aware texture enhancement; identifying impurity adhesion regions in the preprocessed image; obtaining a set of suggested center coordinates for secondary cleaning through image feature clustering; and finally generating cleaning control commands. Compared to existing technologies that rely on manual experience for identification, especially under conditions of strong reflection, complex impeller structure, and irregular impurity adhesion, which cannot achieve high-precision impurity adhesion identification, this application introduces a depth segmentation model and an area-path joint weight planning mechanism, achieving intelligent identification and automatic cleaning scheduling of adhesion regions in complex backgrounds. This avoids the problems of missed cleaning blind spots and wasted path resources, thereby improving the efficiency of water pump impeller internal flow channel cleaning operations.

Description

Technical Field

[0001] This invention relates to the field of image recognition and processing technology, and in particular to a water pump impeller cleaning control method and system based on image recognition. Background Technology

[0002] Currently, during the manufacturing, remanufacturing, or maintenance of water pump impellers, the internal flow channels often accumulate varying degrees of deposits due to casting residues, machining flash, or post-operation impurities, with sand residue being the most typical. If these deposits are not cleaned thoroughly and promptly, they can not only affect hydraulic performance and operating efficiency but may also lead to abnormal operating vibrations, increased energy consumption, or accelerated localized corrosion, thereby shortening the equipment's lifespan. Existing technologies commonly employ artificial lighting to photograph the impeller or perform direct visual inspection, combined with manual cleaning using sandblasting guns or mechanical brushes. However, manual identification is highly dependent on experience and visual fatigue, and when faced with various complex lighting conditions, reflective metallic backgrounds, or three-dimensional curved surfaces, it is prone to problems such as missed detection, misjudgment, incorrect cleaning, or over-cleaning of impurities.

[0003] Furthermore, when dealing with complex curved surfaces and highly reflective metallic materials, traditional image recognition methods (such as fixed threshold segmentation and edge detection) exhibit significant lack of robustness in sand adhesion detection, making it difficult to accurately extract key adhesion areas. For instance, in the presence of complex reflective interference and curved surface occlusion, traditional image enhancement methods based on single-channel images struggle to reconstruct the impurity boundary structure. Moreover, existing path control methods typically do not differentiate between adhesion levels, only ordering the robotic arm's cleaning path according to spatial distance, which can easily lead to delayed processing of some highly adhesive areas, affecting cleaning efficiency and quality.

[0004] Therefore, there is an urgent need for a water pump impeller cleaning control method based on image recognition that can effectively improve the automatic recognition accuracy and cleaning control efficiency under complex working conditions, thereby realizing the standardization and intelligentization of the impeller cleaning process. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the present invention aims to propose a water pump impeller cleaning control method based on image recognition. This method addresses the technical problem that existing technologies rely on manual experience for identification, particularly under conditions of strong reflection, complex impeller structure, and irregular impurity adhesion, making it impossible to achieve high-precision impurity adhesion identification.

[0006] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a water pump impeller cleaning control method based on image recognition.

[0007] The image recognition-based water pump impeller cleaning control method includes:

[0008] Step S10: Obtain image sequence I of the internal flow channel region of the water pump impeller workpiece through an image acquisition mechanism based on synchronous rotation and dual structured light;

[0009] Step S20: Based on the internal flow channel region image sequence I, a dual-channel brightness nonlinear suppression and curvature-aware texture enhancement mechanism is used to perform reflection suppression and texture extraction tasks, and a preprocessed image sequence is output. ;

[0010] Step S30: Based on the preprocessed image sequence A semantic segmentation deep neural network based on dual-branch attention is used to perform the task of identifying sand residue areas and output the impurity residue rate index R of the water pump impeller.

[0011] Step S40: Based on the impurity residue rate index R of the water pump impeller, an intelligent cleaning recommendation task is performed using a clustering mechanism based on map features, and the suggested set of center coordinates for secondary cleaning and the map index of the adhesion area are output.

[0012] Step S50: Based on the suggested set of center coordinates and the adhesion area map index, a path planning mechanism based on adhesion area priority is used to generate a set of water pump impeller cleaning control instructions.

[0013] Preferably, step S10, which involves acquiring the image sequence I of the internal flow channel region of the water pump impeller workpiece through an image acquisition mechanism based on synchronous rotation and dual structured light, specifically includes:

[0014] Step S101: Preset to include a two-degree-of-freedom industrial camera module equipped with a rotation axis Z-axis and a pitch axis X-axis;

[0015] Step S102: Based on the two-degree-of-freedom industrial camera module, a preset set of angle step values ​​is used to perform full-attitude collaborative acquisition and angle label encoding processing on the water pump impeller workpiece, and output a preliminary image sequence containing angle step values ​​and pose mapping relationship. ;

[0016] Step S103: Based on the preliminary image sequence containing angle step values ​​and pose mapping relationship A blind zone texture filling operation is performed using an image registration and fusion method based on gray-level gradient direction consistency constraints, and the internal flow channel region image sequence I is output.

[0017] Preferably, in step S20, a dual-channel brightness nonlinear suppression and curvature-aware texture enhancement mechanism is used to perform reflection suppression and texture extraction tasks based on the internal flow channel region image sequence I, and a preprocessed image sequence is output. The steps specifically include:

[0018] Step S201: Extract the red channel value and RGB grayscale channel value of each frame in the image sequence I of the internal flow channel region, and construct a brightness analysis map based on the red channel value and RGB grayscale channel value. With grayscale image Based on brightness analysis diagram The grayscale image G(x,y) is compressed using the power-law compression principle to dynamically darken the bright areas, outputting a dual-channel compressed image. Where x represents the horizontal coordinate of a pixel and y represents the vertical coordinate of a pixel;

[0019] Step S202: Based on dual-channel compressed image Local edge curvature estimation is performed using the multi-directional local gradient projection method, and the curvature field of the principal direction is output. ; with the curvature field in the principal direction A Gabor filter guided by the principal direction curvature field is set for the direction parameters, and the dual-channel compressed image is processed based on the Gabor filter guided by the principal direction curvature field. Perform texture detail response magnification processing to output an enhanced image. ;

[0020] Step S203: For the enhanced image Perform normalized brightness reconstruction and histogram equalization processing, and output a preprocessed image sequence. .

[0021] Preferably, in step S30, based on the preprocessed image sequence The steps for performing the task of identifying sand residue regions and outputting the impurity residue rate index R of the water pump impeller using a semantic segmentation deep neural network based on dual-branch attention guidance specifically include:

[0022] Step S301: Pre-construct a dual-branch attention-guided semantic segmentation deep neural network;

[0023] Step S302: Obtain the set of historical preprocessed image sequences and the corresponding set of historical water pump impeller impurity residue rate indicators. Based on the set of historical preprocessed image sequences and the corresponding set of historical water pump impeller impurity residue rate indicators, a weighted combination loss function consisting of cross-entropy loss term and Dice coefficient loss term is used to pre-train the semantic segmentation deep neural network.

[0024] Step S303: Preprocess the image sequence Input a pre-trained semantic segmentation deep neural network, and the semantic segmentation deep neural network outputs the impurity residue rate index R of the water pump impeller.

[0025] Preferably, in step S30, the semantic segmentation deep neural network includes:

[0026] Input layer, used to receive preprocessed image sequences ;

[0027] The backbone coding layer is used to extract preprocessed image sequences based on the ResNet-50 network. Global context features;

[0028] Edge enhancement branch layer is used to extract edge details of the sand-sticky region through a deep and shallow feature fusion structure fused by dilated convolutional network and dilated convolutional network;

[0029] A dual attention guidance layer is used to perform feature reweighting and region response enhancement operations based on global context features and edge details of the sand-sticky region using channel attention and position attention mechanisms, and outputs an attention fusion feature map.

[0030] The feature decoding layer is used to restore the attention fusion feature map to the preprocessed image sequence using upsampling and skip connection mechanisms. Same size;

[0031] The output layer is used to output the impurity residue rate index R of the water pump impeller.

[0032] Preferably, in step S40, the step of performing an intelligent cleaning recommendation task based on a map feature clustering mechanism using the water pump impeller impurity residue rate index R, and outputting the suggested set of center coordinates for secondary cleaning and the map index of the adhesion area, specifically includes:

[0033] Step S401: Construct a global cleanliness scoring function based on the impurity residue rate index R of the water pump impeller. Where n is the total number of image frames. The water pump impeller impurity residue rate index is set for the i-th image frame; a cleanliness qualification threshold is set. , ;when When the condition is met, it is determined that "the impeller impurity residue rate is not up to standard, and a secondary cleaning control task for the impeller needs to be performed"; the set of image indexes showing that the residue rate is not up to standard is output.

[0034] Step S402: Extract region map features from the image index set of images with substandard residual rates. The region map features include adhesion area features, gray-level mean features, gray-level standard deviation features, edge texture complexity features, and region dispersion features. Construct a first density clustering weight factor based on the adhesion area features and a second density clustering weight factor based on the region dispersion features. Perform cluster analysis using the DBSCAN algorithm driven by the first and second density clustering weight factors based on the region map features, and output the adhesion region map index.

[0035] Step S403: Extract the suggested set of center coordinates based on the adhesion region map index, and output the suggested set of center coordinates and the adhesion region map index that need to be cleaned up again.

[0036] Preferably, step S50, which involves generating a pump impeller cleaning control instruction set based on a path planning mechanism prioritizing adhesion area, according to the suggested set of center coordinates and the adhesion area map index, specifically includes:

[0037] Step S501: Extract the adhesion area features corresponding to the adhesion region map index, and construct an adhesion priority factor based on the adhesion area features;

[0038] Step S502: Use the adhesion priority factor as the weight input of the heuristic graph search algorithm, and perform path cost evaluation on the center coordinate suggestion set based on the heuristic graph search algorithm to output the comprehensive path cost score; and construct the target cleaning point order set based on the comprehensive path cost score.

[0039] Step S503: Generate a water pump impeller cleaning control instruction set based on the target cleaning point sequence set. The water pump impeller cleaning control instruction set includes impeller rotation control instructions and cleaning device start / stop control parameters.

[0040] The present invention also provides a water pump impeller cleaning control system based on image recognition, comprising:

[0041] The image acquisition module is used to acquire image sequence I of the internal flow channel region of the water pump impeller workpiece through an image acquisition mechanism based on synchronous rotation and dual structured light;

[0042] The texture enhancement preprocessing module is used to perform reflection suppression and texture extraction tasks based on the internal flow channel region image sequence I using a dual-channel brightness nonlinear suppression and curvature-aware texture enhancement mechanism, and outputs a preprocessed image sequence. ;

[0043] A semantic segmentation and recognition module is used to perform semantic segmentation based on preprocessed image sequences. A semantic segmentation deep neural network based on dual-branch attention is used to perform the task of identifying sand residue areas and output the impurity residue rate index R of the water pump impeller.

[0044] The cleaning strategy recommendation module is used to perform intelligent cleaning recommendation tasks based on the water pump impeller impurity residue rate index R and a graph feature clustering mechanism. It outputs a set of suggested center coordinates for secondary cleaning and a graph index of the adhesion area.

[0045] The path planning and control instruction generation module is used to generate a set of pump impeller cleaning control instructions based on the center coordinate suggestion set and the adhesion area map index using a path planning mechanism that prioritizes adhesion area.

[0046] The present invention also provides a water pump impeller cleaning control device based on image recognition, comprising: a memory, a processor, and an image recognition-based water pump impeller cleaning control program stored in the memory and executable on the processor. When the image recognition-based water pump impeller cleaning control program is executed by the processor, it implements an image recognition-based water pump impeller cleaning control method.

[0047] The present invention also provides a computer program product, including an image recognition-based water pump impeller cleaning control program, which, when executed by a processor, implements the image recognition-based water pump impeller cleaning control method.

[0048] The beneficial effects of this invention are as follows: By introducing a dual-structured light and synchronous rotation image acquisition mechanism, combined with brightness nonlinearity suppression and curvature-aware texture enhancement algorithms, this invention effectively suppresses reflective interference and enhances the texture features of subtle impurities in the image. This enables the deep semantic segmentation model to accurately extract the edges of sand-adhesive regions even in complex backgrounds, which is significantly better than traditional threshold segmentation and morphological algorithms, ensuring the reliability of subsequent cleaning scheduling decisions.

[0049] This invention introduces a path cost optimization mechanism based on adhesion area weighting during the path planning process, which takes into account both cleaning priority and path execution efficiency. It can prioritize the treatment of high residual areas and reduce redundant movement, thereby improving cleaning coverage and operation time efficiency. It is particularly suitable for complex cleaning scenarios inside water pump impellers with multiple areas and uneven adhesion. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart illustrating the first embodiment of a water pump impeller cleaning control method based on image recognition according to the present invention.

[0052] Figure 2 This is a schematic diagram of synchronous rotation and dual structured light image acquisition, representing a first embodiment of a water pump impeller cleaning control method based on image recognition according to the present invention.

[0053] Figure 3 This is a schematic diagram of a water pump impeller cleaning control method based on image recognition according to the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the water pump impeller cleaning control method based on image recognition of the present invention, which presents the first embodiment of the water pump impeller cleaning control method based on image recognition of the present invention.

[0056] In the first embodiment, the image recognition-based water pump impeller cleaning control method includes:

[0057] Step S10: Obtain image sequence I of the internal flow channel region of the water pump impeller workpiece through an image acquisition mechanism based on synchronous rotation and dual structured light;

[0058] It should be noted that the "synchronous rotation and dual structured light image acquisition mechanism" refers to using an image acquisition system that includes a stepper motor drive platform, a rotary encoder, a synchronous trigger controller, and a dual light source structure to perform uniform rotation scanning on the water pump impeller workpiece, and to obtain image frames under different viewing angles and lighting conditions by alternately or synchronously projecting dual structured light sources (such as blue light stripes and red light spots).

[0059] Understandably, because the inner wall of a water pump impeller is mostly curved metal, with strong specular reflection and local dead angles, images under natural light or single-angle illumination often exhibit overexposure, bright patches, and blurred textures, which are detrimental to subsequent image processing and recognition. This step introduces synchronous rotation control to ensure a uniform distribution of image acquisition angles throughout the impeller's rotation cycle. Simultaneously, the dual-structure light source projects onto the complex curved surface area from different positions, resulting in more complete image texture and morphological information and stronger contrast. This method yields high-quality image sequences that include enhanced illumination information and multi-view reconstruction characteristics, providing reliable input for subsequent sand residue detection.

[0060] For example, such as Figure 2As shown, an impeller rotates uniformly at a preset angular velocity on a rotary table. At fixed intervals, the impeller surface is alternately illuminated by blue and red structured light, with an image acquisition module simultaneously triggering image capture. The arrows of different colors in the image represent different structured light illumination directions, corresponding to one frame of image captured at each angle. This dual-structured light and rotation-synchronized acquisition mechanism effectively overcomes the uneven illumination and blind spot occlusion problems inherent in traditional single-angle imaging, enhancing the texture response of the impeller's internal flow channel area under different lighting conditions. It is particularly suitable for industrial scenarios with complex sand adhesion and severe irregular occlusion. The differential illumination of the structured light further improves the visibility of edge and curved surface residues, providing high-contrast image input for subsequent semantic segmentation and sand adhesion detection.

[0061] Step S20: Based on the internal flow channel region image sequence I, a dual-channel brightness nonlinear suppression and curvature-aware texture enhancement mechanism is used to perform reflection suppression and texture extraction tasks, and a preprocessed image sequence is output. ;

[0062] It should be noted that the "dual-channel brightness nonlinear suppression and curvature-aware texture enhancement mechanism" in this step refers to first converting each frame of the input image sequence into two channels: high-brightness and low-brightness. An adaptive threshold is then set based on the brightness histogram distribution, and a nonlinear compression function (such as logarithmic suppression or bilateral harmonic mapping) is applied to the high-brightness areas to reduce saturation interference caused by strong reflections. Simultaneously, a curvature-aware operator is introduced to enhance the response of abruptly curved areas in the image, extracting fine-grained texture features, especially targeting the grooves, corners, and surface residue boundary areas in the complex geometry of the impeller.

[0063] Understandably, within the flow channels of a water pump impeller, the polished metal surface and complex grooves and curved structures easily generate strong reflections and shadows under single lighting conditions, affecting image quality. However, by constructing dual brightness channels and combining them with a curvature-sensitive enhancement mechanism, multi-layered texture activation can be achieved, effectively solving the problem of traditional enhancement algorithms failing in highly reflective areas.

[0064] For example, when the image acquisition device is close to the internal surface of the water pump impeller (e.g., within 10cm), due to the large direct angle of the structured light, strong reflective areas are usually concentrated at the transition between the groove wall and the edge. If not effectively suppressed, this will result in a large area of ​​saturated pixels, blurring the boundaries of the adhesion area and causing loss of detail. The dual-channel brightness nonlinear suppression mechanism in this step can significantly reduce the interference of bright areas under such strong reflective conditions. Combined with the curvature-aware texture enhancement mechanism, it can effectively preserve real texture information even in high-curvature bending transition areas, improving subsequent recognition accuracy. In other acquisition scenarios at greater distances (e.g., above 20cm), the overall image brightness is uneven and the contrast is reduced, with low contrast between the adhesion area and the background. This mechanism, through dual-channel enhancement and local texture enhancement, can still extract weak but structurally characteristic texture changes, making the texture boundary between the subsequent sand residue area and the non-residue area clearer, thus exhibiting good universality under different distance and angle conditions.

[0065] Step S30: Based on the preprocessed image sequence A semantic segmentation deep neural network based on dual-branch attention is used to perform the task of identifying sand residue areas and output the impurity residue rate index R of the water pump impeller.

[0066] It should be noted that the "dual-branch attention-guided semantic segmentation deep neural network" in this step refers to a deep feature extraction structure that combines a backbone encoding branch and an edge enhancement branch. The backbone branch extracts global contextual semantic information based on the ResNet-50 architecture, while the edge branch fuses dilated and attenuated convolutions to extract details of the sand-adhesive edges. Subsequently, the dual attention module performs saliency guidance and fusion enhancement on the semantic and edge features respectively, forming an accurate feature map of the sand-adhesive region. Finally, a semantic segmentation map with the same size as the input image is generated through the feature decoding layer and the output layer, and the impurity residue rate index R is calculated based on the pixel ratio of the sand-adhesive region in the segmentation map.

[0067] Understandably, by introducing a dual-branch attention mechanism, this network structure not only enhances the sensitivity to sand adhering to the boundary region, but also effectively suppresses the misidentification of the background or reflective areas as impurities. Thus, it maintains good recognition performance under complex background and non-uniform lighting conditions, ensuring that the output residual rate index R has high stability and high reliability.

[0068] It should be understood that, compared to traditional single-branch semantic segmentation networks or threshold-based image processing methods, this network scheme has stronger generalization capabilities. In practical applications, the internal flow channels of water pump impellers are characterized by complex structures, large variations in lighting, and irregular sand adhesion morphologies. Traditional methods are easily limited by threshold selection and interference from blurred edge regions, leading to misidentification or missed detection. The dual-branch attention structure, however, can take into account both semantic information and edge details, exhibiting higher recognition accuracy, especially in the transition regions of adhesion boundaries, making it suitable for recognizing various impeller structures and diverse adhesion morphologies.

[0069] Step S40: Based on the impurity residue rate index R of the water pump impeller, an intelligent cleaning recommendation task is performed using a clustering mechanism based on map features, and the suggested set of center coordinates for secondary cleaning and the map index of the adhesion area are output.

[0070] It should be noted that the "graph feature-based clustering mechanism" in this step refers to mapping the adhesion regions identified in step S30 into a graph structure representation in space. Each adhesion region is regarded as a node in the graph, and the node attributes include features such as its center coordinates, adhesion area, residual rate, curvature, and illumination unevenness. The edge weights are defined based on the spatial distance and texture similarity between regions. Subsequently, an improved spectral clustering algorithm is used to group and cluster the adhesion regions based on the weighted graph structure, prioritizing the identification of densely distributed, severely adhered, and structurally complex regions. Finally, the proposed set of regions that need to be prioritized for secondary cleaning and their index numbers in the adhesion region graph are output.

[0071] It should be understood that, compared to traditional cleaning strategies based on a single threshold (such as cleaning if the area is greater than a certain value) or static rules (such as pixel-by-pixel sorting), this map feature clustering mechanism has stronger adaptability and context awareness. Especially in typical industrial application scenarios with complex flow channel structures, uneven impurity distribution, and high cleaning difficulty in some areas, this method can automatically discover potential secondary contamination risk areas based on the topological relationships and visual features between regions, avoiding blind cleaning or missed cleaning.

[0072] Step S50: Based on the suggested set of center coordinates and the adhesion area map index, a path planning mechanism based on adhesion area priority is used to generate a set of water pump impeller cleaning control instructions.

[0073] It should be noted that the "path planning mechanism based on adhesion area priority" in this step refers to: using the set of suggested center coordinates output in step S40 and the adhesion area map index as the set of cleaning target points, and combining the adhesion area weight of each target point to construct a cost function aimed at maximizing cleaning efficiency. In the specific path planning process, areas with large adhesion areas are prioritized as initial targets, and the selection priority of the next target point is dynamically adjusted during path iteration, ensuring that high-contamination key areas are covered first under the constraint of the shortest travel distance. The path planning algorithm used is an improved A* algorithm based on a weighted graph, where the heuristic function not only considers Euclidean distance but also introduces an inverse proportional weight of the adhesion area as a cost correction term, thereby achieving the dual optimization objectives of "area priority and shortest distance".

[0074] Understandably, introducing adhesion area as a weighting dimension for path scheduling not only improves the overall efficiency of the cleanup task but also significantly reduces the risk of missed detections in highly contaminated areas. Especially under conditions of limited cleanup resources or time constraints, this mechanism ensures that limited cleaning actions cover the most valuable areas.

[0075] It should be understood that, compared to traditional nearest neighbor priority traversal or static path planning methods, the path planning mechanism of this invention has the ability to dynamically adjust priorities, and the objective function comprehensively considers both the contaminated area and the path cost, realizing a strategy shift from "distance-driven" to "contamination value-driven". Especially in the context of uneven spatial distribution of multiple adhesion areas and high start-up and shutdown costs of cleaning equipment, this method can effectively avoid frequent skipping paths, shorten the total cleaning time, and reduce energy consumption.

[0076] Example 2: Furthermore, the present invention provides a water pump impeller cleaning control system based on image recognition, employing a water pump impeller cleaning control method based on image recognition from the above embodiments, which can solve the technical problem of water pump impeller cleaning control based on image recognition. Compared with the prior art, the beneficial effects of the water pump impeller cleaning control system based on image recognition provided by the present invention are the same as the beneficial effects of the water pump impeller cleaning control method based on image recognition provided in the above embodiments, and other technical features in the water pump impeller cleaning control system based on image recognition are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0077] Example 3: This invention provides a water pump impeller cleaning control device based on image recognition. Please refer to... Figure 3A water pump impeller cleaning control device based on image recognition includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the water pump impeller cleaning control method based on image recognition described in Embodiment 1 above. The water pump impeller cleaning control device based on image recognition in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. This water pump impeller cleaning control device based on image recognition is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. A water pump impeller cleaning control device based on image recognition may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the image recognition-based water pump impeller cleaning control device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows an image recognition-based pump impeller cleaning control device to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows an image recognition-based pump impeller cleaning control device with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0078] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the image recognition-based water pump impeller cleaning control method described above. The computer program product provided by this invention can solve the technical problem of image recognition-based water pump impeller cleaning control. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the image recognition-based water pump impeller cleaning control method provided in the above embodiments, and will not be repeated here.

[0079] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0080] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0081] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An image recognition-based water pump impeller cleaning control method, characterized by, The methods include: Step S10: Obtain image sequence I of the internal flow channel region of the water pump impeller workpiece through an image acquisition mechanism based on synchronous rotation and dual structured light; Step S20: Perform anti-reflection and texture extraction tasks based on the internal flow channel region image sequence I using a dual-channel brightness nonlinearity suppression and curvature-aware texture enhancement mechanism, output a preprocessed image sequence ; Step S30: based on the pre-processed image sequence The sand residue area recognition task is performed by using a semantic segmentation deep neural network based on a double-branch attention guide, and an impurity residue rate index R of the water pump impeller is output. Step S40: Based on the impurity residue rate index R of the water pump impeller, an intelligent cleaning recommendation task is performed using a clustering mechanism based on map features, and the suggested set of center coordinates for secondary cleaning and the map index of the adhesion area are output. Step S50: Based on the suggested set of center coordinates and the adhesion area map index, a path planning mechanism based on adhesion area priority is used to generate a set of water pump impeller cleaning control instructions.

2. The image recognition-based water pump impeller cleaning control method of claim 1, wherein, Step S10, which involves acquiring the image sequence I of the internal flow channel region of the water pump impeller workpiece through an image acquisition mechanism based on synchronous rotation and dual structured light, specifically includes: Step S101: Preset to include a two-degree-of-freedom industrial camera module equipped with a rotation axis Z-axis and a pitch axis X-axis; Step S102: Based on the two-degree-of-freedom industrial camera module, a preset angle step value set is used to perform full-pose cooperative collection and angle label coding processing on the water pump impeller workpiece, and a preliminary image sequence containing angle step values and pose mapping relationships is output ; Step S103: based on the preliminary image sequence containing angle step value and pose mapping relationship The blind area texture filling operation is performed by using the image registration fusion method based on the consistency constraint of gray gradient direction, and an internal flow channel region image sequence I is output.

3. The water pump impeller cleaning control method based on image recognition as described in claim 1, characterized in that, In step S20, based on the internal flow channel region image sequence I, a dual-channel brightness nonlinear suppression and curvature-aware texture enhancement mechanism is used to perform reflection suppression and texture extraction tasks, and a preprocessed image sequence is output. The steps specifically include: Step S201: Extract the red channel value and RGB grayscale channel value of each frame in the image sequence I of the internal flow channel region, and construct a brightness analysis map based on the red channel value and RGB grayscale channel value. With grayscale image Based on brightness analysis diagram The grayscale image G(x,y) is compressed using the power-law compression principle to dynamically darken the bright areas, outputting a dual-channel compressed image. Where x represents the horizontal coordinate of a pixel and y represents the vertical coordinate of a pixel; Step S202: Based on dual-channel compressed image Local edge curvature estimation is performed using the multi-directional local gradient projection method, and the curvature field of the principal direction is output. ; with the curvature field in the principal direction A Gabor filter guided by the principal direction curvature field is set for the direction parameters, and the dual-channel compressed image is processed based on the Gabor filter guided by the principal direction curvature field. Perform texture detail response magnification processing to output an enhanced image. ; Step S203: For the enhanced image Perform normalized brightness reconstruction and histogram equalization processing, and output a preprocessed image sequence. .

4. The water pump impeller cleaning control method based on image recognition as described in claim 1, characterized in that, In step S30, based on the preprocessed image sequence The steps for performing the task of identifying sand residue regions and outputting the impurity residue rate index R of the water pump impeller using a semantic segmentation deep neural network based on dual-branch attention guidance specifically include: Step S301: Pre-construct a dual-branch attention-guided semantic segmentation deep neural network; Step S302: Obtain the set of historical preprocessed image sequences and the corresponding set of historical water pump impeller impurity residue rate indicators. Based on the set of historical preprocessed image sequences and the corresponding set of historical water pump impeller impurity residue rate indicators, a weighted combination loss function consisting of cross-entropy loss term and Dice coefficient loss term is used to pre-train the semantic segmentation deep neural network. Step S303: Preprocess the image sequence Input a pre-trained semantic segmentation deep neural network, and the semantic segmentation deep neural network outputs the impurity residue rate index R of the water pump impeller.

5. The water pump impeller cleaning control method based on image recognition as described in claim 4, characterized in that, In step S30, the semantic segmentation deep neural network includes: Input layer, used to receive preprocessed image sequences ; The backbone coding layer is used to extract preprocessed image sequences based on the ResNet-50 network. Global context features; Edge enhancement branch layer is used to extract edge details of the sand-sticky region through a deep and shallow feature fusion structure fused by dilated convolutional network and dilated convolutional network; A dual attention guidance layer is used to perform feature reweighting and region response enhancement operations based on global context features and edge details of the sand-sticky region using channel attention and position attention mechanisms, and outputs an attention fusion feature map. The feature decoding layer is used to restore the attention fusion feature map to the preprocessed image sequence using upsampling and skip connection mechanisms. Same size; The output layer is used to output the impurity residue rate index R of the water pump impeller.

6. The water pump impeller cleaning control method based on image recognition as described in claim 1, characterized in that, In step S40, the intelligent cleaning recommendation task is performed based on the impeller impurity residue rate index R using a map feature clustering mechanism, outputting the suggested set of center coordinates requiring secondary cleaning and the map index of the adhesion area. This step specifically includes: Step S401: Construct a global cleanliness scoring function based on the impurity residue rate index R of the water pump impeller. Where n is the total number of image frames. The water pump impeller impurity residue rate index is set for the i-th image frame; a cleanliness qualification threshold is set. , ;when When the condition is met, it is determined that "the impeller impurity residue rate is not up to standard, and a secondary cleaning control task for the impeller needs to be performed"; the set of image indexes showing that the residue rate is not up to standard is output. Step S402: Extract region map features from the image index set of images with substandard residual rates. The region map features include adhesion area features, gray-level mean features, gray-level standard deviation features, edge texture complexity features, and region dispersion features. Construct a first density clustering weight factor based on the adhesion area features, and construct a second density clustering weight factor based on the region dispersion features. Perform cluster analysis using the DBSCAN algorithm driven by the first and second density clustering weight factors based on the region map features, and output the adhesion region map index. Step S403: Extract the suggested set of center coordinates based on the adhesion region map index, and output the suggested set of center coordinates and the adhesion region map index that need to be cleaned up again.

7. The water pump impeller cleaning control method based on image recognition as described in claim 1, characterized in that, Step S50, which involves generating a pump impeller cleaning control instruction set based on a path planning mechanism prioritizing adhesion area, using the suggested set of center coordinates and the adhesion area map index, specifically includes: Step S501: Extract the adhesion area features corresponding to the adhesion region map index, and construct an adhesion priority factor based on the adhesion area features; Step S502: Use the adhesion priority factor as the weight input of the heuristic graph search algorithm, and perform path cost evaluation on the center coordinate suggestion set based on the heuristic graph search algorithm to output the comprehensive path cost score; and construct the target cleaning point order set based on the comprehensive path cost score. Step S503: Generate a water pump impeller cleaning control instruction set based on the target cleaning point sequence set. The water pump impeller cleaning control instruction set includes impeller rotation control instructions and cleaning device start / stop control parameters.

8. A water pump impeller cleaning control system based on image recognition, applied to the water pump impeller cleaning control method based on image recognition as described in any one of claims 1 to 7, characterized in that, The image recognition-based water pump impeller cleaning control system includes: The image acquisition module is used to acquire image sequence I of the internal flow channel region of the water pump impeller workpiece through an image acquisition mechanism based on synchronous rotation and dual structured light; The texture enhancement preprocessing module is used to perform reflection suppression and texture extraction tasks based on the internal flow channel region image sequence I using a dual-channel brightness nonlinear suppression and curvature-aware texture enhancement mechanism, and outputs a preprocessed image sequence. ; A semantic segmentation and recognition module is used to perform semantic segmentation based on preprocessed image sequences. A semantic segmentation deep neural network based on dual-branch attention is used to perform the task of identifying sand residue areas and output the impurity residue rate index R of the water pump impeller. The cleaning strategy recommendation module is used to perform intelligent cleaning recommendation tasks based on the water pump impeller impurity residue rate index R and a graph feature clustering mechanism. It outputs a set of suggested center coordinates for secondary cleaning and a graph index of the adhesion area. The path planning and control instruction generation module is used to generate a set of pump impeller cleaning control instructions based on the center coordinate suggestion set and the adhesion area map index using a path planning mechanism that prioritizes adhesion area.

9. A water pump impeller cleaning control device based on image recognition, characterized in that, The image recognition-based water pump impeller cleaning control device includes: a memory, a processor, and an image recognition-based water pump impeller cleaning control program stored in the memory and executable on the processor. When the image recognition-based water pump impeller cleaning control program is executed by the processor, it implements the image recognition-based water pump impeller cleaning control method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes an image recognition-based water pump impeller cleaning control program, which, when executed by a processor, implements an image recognition-based water pump impeller cleaning control method according to any one of claims 1 to 7.

Images