Vision-based multi-nozzle jet flow uniformity determination method

Abstract

The invention provides a multi-nozzle jet flow uniformity determination method based on vision, and relates to the technical field of thick plate hot rolling and computer vision. According to the method, key frame extraction and preprocessing are carried out on collected video data; jet region identification is carried out by using an image segmentation technology, and density reference values are calculated pixel by pixel for the identified region; designing an index for measuring the jet flow uniformity according to a density reference value and supporting thermodynamic diagram visualization and analysis; through a mode of combining visual image processing and linear calibration, efficient and accurate quantitative evaluation on the jet flow uniformity of the multiple nozzles is realized. Moreover, the method can be realized through laser jet visualization software developed on the basis of a Python language, necessary array acceleration operation is carried out in combination with a Cupy library, the calculation speed of a single 1080 * 720-pixel image is smaller than 800 ms, and the requirements of actual industrial deployment and real-time detection are met.

Description

Technical Field

[0001] This application relates to the fields of hot rolling of thick plates and computer vision technology, and in particular to a vision-based method for determining the uniformity of multi-nozzle jets. Background Technology

[0002] The uniformity of nozzle jets plays a crucial role in numerous industries and production processes, directly impacting the quality and performance of the final product. For example, in agricultural irrigation, the uniformity of water or pesticide sprayed from nozzles is vital for the collective growth of crops; uneven spraying can lead to areas being either too dry or too wet, affecting uniform growth and yield. In industrial production, the uniformity of nozzle arrays spraying material droplets is equally critical, especially in surface coating processes. Uneven spraying can result in inconsistent coating thickness, consequently affecting the performance and lifespan of parts. In the steel manufacturing industry, particularly in the hot rolling of thick plates, multi-nozzle cooling spray boxes use water mist for cooling. Uneven jet distribution in the nozzle array can lead to uneven cooling of the steel plate surface, affecting its internal metallographic structure, hardness, and other mechanical properties. Therefore, ensuring the uniformity of nozzle array jets is key to improving product quality, ensuring process stability, and increasing production efficiency across various fields.

[0003] Most existing online methods for detecting nozzle jet uniformity focus on analyzing the jet flow rate and spray state of a single nozzle. However, effective real-time online monitoring technologies are lacking for the jet distribution and uniformity of multi-nozzle arrays. Traditional nozzle detection methods primarily rely on physical sensors such as flow meters, temperature sensors, and pressure sensors. While these devices can provide some real-time data, their high installation and maintenance costs, complex equipment, and difficulty in effectively capturing the uniformity of the entire nozzle array limit their effectiveness in practical applications. Furthermore, measurement methods based on single-point sensors struggle to comprehensively monitor the entire cooling area and cannot accurately reflect the differences in spray state across different regions of the nozzle array. Currently, the perception and measurement of jet uniformity in multi-nozzle arrays often rely on human visual observation. However, this method lacks the necessary objectivity and quantitative standards in the complex cooling process of multi-nozzle arrays and cannot meet the demands of real-time changes in complex production environments. These issues, to some extent, limit the continuous optimization of nozzle array performance and the improvement of production efficiency. Summary of the Invention

[0004] In view of the above problems, this application is proposed to provide a vision-based method for determining the uniformity of multi-nozzle jets, which overcomes or at least partially solves the above problems. The technical solution is as follows: Firstly, a vision-based method for determining the uniformity of multi-nozzle jets is provided, the method comprising: For the acquired video stream of multiple nozzles in operation, key frames are selected as the first target image based on the sharpness index. ; For the first target image Each pixel in the first target image undergoes preprocessing. Each pixel in the image undergoes the same processing to obtain the corresponding weighted median as the new pixel value. This is applied to the first target image. The second target image is obtained by replacing the pixel value of all pixels with the corresponding weighted median. ; For the second target image Enhancement is performed on each pixel in the second target image. Each pixel in the image undergoes the same processing to obtain a new value for the corresponding enhanced pixel. This process is repeated for the second target image. The third target image is obtained by replacing the pixel values ​​of all pixels with the corresponding enhanced pixel values. ; Threshold segmentation method is used for the third target image Perform image segmentation to obtain the third target image. Based on the pixel matrix of the green channel, a binary result is generated. ; Based on the water flow density measurement method used by the nozzle manufacturer, a linear model is used for linear fitting calibration to obtain the linear parameter values ​​to be estimated. In practical applications, the uniformity of the jet region in the actual acquired image is calculated based on the linear parameter values ​​to be estimated.

[0005] In one possible implementation, for the acquired video stream of multiple nozzles operating, keyframes are selected as the first target image based on a sharpness index. ,include: Determine the set of n consecutive frames of images after time t: (1) In equation (1), For a moment The subsequent continuous collection A collection of frame images; from Select the first target image The formula is as follows: (2) In equation (2), The first target image selected from the set. , for A single frame image at time t satisfies equation (3): (3) In equation (3), It is a gradient-based sharpness operator that satisfies equation (4): (4) In equation (4), and These are the height and width of the image, respectively. and These represent the pixel gradient values ​​calculated along the image width and height directions, respectively.

[0006] In one possible implementation, the preprocessing operations include: Select the first target image A square window is defined by taking one of the pixels as the first target pixel. , Indicates the coordinates of the first target pixel. The x-axis is... The vertical axis is , This represents the neighborhood region centered at the coordinates of the first target pixel. For each pixel in the window, assign a corresponding weight. This represents the contribution of the pixel to the center pixel during the filtering process; the features defining the weights include pixel distance and similarity. Window based on corresponding weights The pixels in the array are paired and sorted to obtain an ordered pairing sequence. The target weighted median is calculated from the pairing sequence and used as the new value of the first target pixel.

[0007] In one possible implementation, a square window is defined for the first target pixel currently being processed. , The region of interest centered at the coordinates of the first target pixel includes: Define a size of square window The window satisfies equation (5): (5) In equation (5), It represents any pixel within the window and its pixel value, where (x', y') are the coordinates of that pixel. and These are the coordinate constraints for each pixel in the window, limiting the window's range along the horizontal and vertical axes. These are parameters that control the window size.

[0008] In one possible implementation, a corresponding weight is assigned to each pixel in the window. ,include: The weights are defined based on the Gaussian function, and the weight calculation formula is as follows: (6) In equation (6), It is an exponential function. It is the standard deviation of the Gaussian function.

[0009] In one possible implementation, the window is weighted according to its corresponding weight. The pixels in the array are paired and sorted to obtain an ordered pairing sequence. The target weighted median is calculated from this pairing sequence, and the target weighted median is used as the new value of the first target pixel, including: Window based on corresponding weights The pixels in the array are paired as follows: (7) The window is obtained according to equation (7). The set of pairs; The pairing sets are sorted sequentially according to their weight values ​​to obtain an ordered pairing sequence. The weighted median of each pixel in the paired sequence is calculated sequentially, and the pixel is selected when equation (8) is satisfied. The target weighted median is calculated using the following formula: (8) In equation (8), express The cumulative weight, window The total number of pixels.

[0010] In one possible implementation, the enhancement operations include: Select the second target image One of the pixels is used as the second target pixel , This represents the coordinates of the second target pixel, and defines a neighborhood for the second target pixel. , obtain The mean and standard deviation required for local contrast enhancement are calculated using the following formulas: (9) (10) In equation (9), Let represent the local mean within the neighborhood Ω(x,y). For the neighborhood Total number of pixels included. This represents the summation of all pixels within the neighborhood Ω(x,y); in equation (10), This represents the local standard deviation within the neighborhood Ω(x,y); The new value of the enhanced pixel is calculated based on the mean and standard deviation, using the following formula: (11) In equation (11), To enhance the new pixel value; It is the enhancement factor, used to control the intensity of contrast enhancement; It is a constant used to avoid numerical overflow problems.

[0011] In one possible implementation, a threshold segmentation method is used for the third target image. Perform image segmentation to obtain the third target image. Based on the pixel matrix of the green channel, a binary result is generated. ,include: The generated binary result Satisfying equation (12): (12) In equation (12), For the third target image The pixel matrix of the green channel, It is a manually set threshold used to exclude interference from the surrounding background in the image.

[0012] In one possible implementation, a linear model is used for linear fitting calibration based on the water flow density measurement method used by the nozzle manufacturer to obtain the linear parameter values ​​to be estimated, including: The fitting and calibration process satisfies equation (13): (13) In equation (13), For actual physical measurement of density, and Let the linear parameter values ​​be those to be estimated. The Hadamard product represents the multiplication of corresponding elements in matrices at the same latitude. Here, it is based on... Estimate and .

[0013] In one possible implementation, the uniformity of the jet region in the actual acquired image is calculated based on the linear parameter values ​​to be estimated, including: The calculation of the uniformity of the jet region in the actual acquired image satisfies equation (14): (14) In equation (14), This serves as the final index for regional evenness. The distribution rate of the actual acquired images, through Multiply by the pixel matrix of the green channel obtained from the actual captured image, and perform a Hadamard product on the binary result obtained from the actual captured image, then add... The binary result obtained from the actual acquired image is generated based on the pixel matrix of the green channel of the actual acquired image. The probability is an ideal uniform distribution with reference to this.

[0014] By employing the above technical solutions, the vision-based multi-nozzle jet uniformity determination method provided in this application embodiment extracts and preprocesses keyframes from the acquired video data; uses image segmentation technology to identify jet regions and calculates density reference values ​​pixel by pixel for the identified regions; designs an index to measure jet uniformity based on the density reference values ​​and supports heatmap visualization and analysis; thus, by combining visual image processing with linear calibration, efficient and accurate quantitative evaluation of multi-nozzle jet uniformity is achieved.

[0015] This application can be implemented using laser jet visualization software developed based on the Python language, and combined with the Cupy library (an open-source Python library that can significantly accelerate large-scale array operations by utilizing the parallel computing capabilities of the graphics processing unit, especially suitable for scenarios requiring high-performance computing such as deep learning, scientific computing, and image processing) to perform necessary array acceleration operations. The calculation speed for a single 1080×720 pixel image is less than 800ms, meeting the needs of actual industrial deployment and real-time detection. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0017] Figure 1 A flowchart illustrating the steps of a vision-based multi-nozzle jet uniformity determination method according to an embodiment of this application is shown. Figure 2 A flowchart illustrating the steps of a specific embodiment of this application is shown; Figure 3a and Figure 3b The image shows a keyframe selected by the vision-based multi-nozzle jet uniformity determination method in a specific embodiment of this application. Figure 4a and Figure 4b The image shown is an example of an embodiment of this application after preprocessing and enhancement operations; Figure 5a and Figure 5bThis illustrates an image of the jet region after image segmentation in a specific embodiment of this application; Figure 6a and Figure 6b This document shows a visualization diagram of the linear fitting calibration process in a specific embodiment of this application. Figure 7 A schematic diagram illustrating the effectiveness of the uniformity index in a specific embodiment of this application is shown. Detailed Implementation

[0018] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the term "comprising" and its variations should be interpreted as open-ended terms meaning "including but not limited to."

[0020] In the hot rolling process of thick plates, jet cooling technology uses multiple nozzles arranged in a matrix to spray water onto the moving strip steel for cooling. Its ultimate goal is to ensure the consistency of cooling rate and uniformity of temperature drop across the strip's width, thereby improving the internal metallographic structure of the steel and further enhancing its quality. While some domestic and international nozzle manufacturers currently specify the design parameters and water output status of individual nozzles in their product manuals, they lack methods for detecting and analyzing the jet uniformity of arrays composed of multiple nozzles. Currently, the perception of jet uniformity still relies primarily on visual observation, which is subjective, lacks theoretically supported quantitative standards, and is difficult to adapt to the optimization needs of complex process environments.

[0021] To address the aforementioned technical problems, embodiments of this application provide a vision-based method for determining the uniformity of multi-nozzle jets, such as... Figure 1 As shown, the vision-based multi-nozzle jet uniformity determination method may include the following steps S101 to S106: Step S101: For the acquired video stream of multiple nozzles operating, select keyframes as the first target image based on the sharpness index. .

[0022] In this step, by introducing a sharpness index as a screening criterion, the highest-sharpness single-frame image containing the richest key feature information of the jet is accurately located and selected from the video stream, providing reliable initial data support for subsequent image preprocessing, enhancement, segmentation and other series of operations.

[0023] Step S102, for the first target image Each pixel in the first target image undergoes preprocessing. Each pixel in the image undergoes the same processing to obtain the corresponding weighted median as the new pixel value. This is applied to the first target image. The second target image is obtained by replacing the pixel value of all pixels with the corresponding weighted median. .

[0024] In this step, by analyzing the first target image Preprocessing suppresses noise and highlight interference, effectively removes redundant interference components from the image, and optimizes the overall image quality. The processing also preserves details, retaining the texture features related to water droplets in the image, achieving a balance between noise reduction and detail preservation, thus meeting the core objectives of preprocessing.

[0025] Step S103, for the second target image Enhancement is performed on each pixel in the second target image. Each pixel in the image undergoes the same processing to obtain a new value for the corresponding enhanced pixel. This process is repeated for the second target image. The third target image is obtained by replacing the pixel values ​​of all pixels with the corresponding enhanced pixel values. .

[0026] In this step, by analyzing the second target image Enhancement can specifically strengthen the edge contours and texture features of water droplet particles in the image. By finely adjusting the local contrast of pixels, the originally blurry edges of water droplets become clearer and sharper, and the subtle texture details on the surface of the particles become more prominent and distinguishable. This effectively reduces the confusion between the background and the core object, greatly improves the recognition of the core observed object in the image, and lays a good foundation for subsequent accurate segmentation and density calculation.

[0027] Step S104: Use a threshold segmentation method to segment the third target image. Perform image segmentation to obtain the third target image. Based on the pixel matrix of the green channel, a binary result is generated. .

[0028] In this step, since the experimental platform uses green laser, the pixel matrix of the green channel of the image is used as the core basis for processing. The image is divided into regions by threshold segmentation method to clearly separate the jet-related region from the background region, thereby clearly marking the range of the jet region in the image.

[0029] Step S105: Based on the water flow density measurement method used by the nozzle manufacturer, a linear model is used for linear fitting calibration to obtain the linear parameter values ​​to be estimated.

[0030] In this step, the model parameters are adjusted based on the actual measurement data, and the values ​​of the linear parameters that need to be estimated in the linear model are finally determined, thereby establishing the correlation between image information and actual water droplet density, which is used to calculate the theoretical jet region uniformity of the actual acquired image.

[0031] Step S106: Calculate the uniformity of the jet region in the actual acquired image based on the linear parameter values ​​to be estimated.

[0032] In this step, after calculating the linear parameter value to be estimated by processing the relevant data of the image through the aforementioned steps, a uniformity analysis is performed on the actual acquired image, and finally a theoretical index that can characterize the uniformity of the jet in the region is obtained.

[0033] In this embodiment, the method extracts and preprocesses keyframes from the acquired video data; uses image segmentation technology to identify the jet region and calculates a density reference value pixel by pixel for the identified region; designs an index to measure the uniformity of the jet based on the density reference value and supports heat map visualization and analysis; thus, by combining visual image processing with linear calibration, it achieves efficient and accurate quantitative evaluation of the uniformity of multi-nozzle jets.

[0034] This application can be implemented using laser jet visualization software developed based on the Python language, and combined with the Cupy library for necessary array acceleration calculations. The calculation speed for a single 1080×720 pixel image is less than 800ms, which meets the needs of actual industrial deployment and real-time detection.

[0035] This application embodiment provides a possible implementation method. In step S101 above, for the acquired video stream of multiple nozzles working, key frames are selected as the first target image based on the sharpness index. Specifically, this may include the following steps: Determine the set of n consecutive frames of images after time t: (1) In equation (1), For a moment The subsequent continuous collection A collection of frame images; from Select the first target image The formula is as follows: (2) In equation (2), The first target image selected from the set. , for A single frame image at time t satisfies equation (3): (3) In equation (3), It is a gradient-based sharpness operator that satisfies equation (4): (4) In equation (4), and These are the height and width of the image, respectively. and These represent the pixel gradient values ​​calculated along the image width and height directions, respectively.

[0036] This embodiment introduces an objective image clarity metric, transforming the traditional blur assessment that relies on subjective human perception into precise numerical calculations based on mathematical models, and selecting the keyframes with the best clarity in the sequence.

[0037] This application embodiment provides a possible implementation method. In step S102 above, the preprocessing operation may specifically include the following steps: Select the first target image A square window is defined by taking one of the pixels as the first target pixel. , Indicates the coordinates of the first target pixel. The x-axis is... The vertical axis is , This represents the neighborhood region centered at the coordinates of the first target pixel. For each pixel in the window, assign a corresponding weight. This represents the contribution of the pixel to the center pixel during the filtering process; the features defining the weights include pixel distance and similarity. Window based on corresponding weights The pixels in the array are paired and sorted to obtain an ordered pairing sequence. The target weighted median is calculated from the pairing sequence and used as the new value of the first target pixel.

[0038] This embodiment transforms a single pixel value update into a refined processing that combines neighborhood features, thereby achieving a balance between image noise suppression and texture detail preservation at the pixel level. By defining a square window centered on the target pixel, the processing range is limited to the local neighborhood, avoiding the destruction of details caused by global processing. Furthermore, different weights are assigned to pixels within the window based on distance and similarity, making the update of the core pixel more closely aligned with the surrounding effective information and reducing interference from noisy pixels. Ultimately, the preprocessed image is both clean and retains the key features of the water droplet.

[0039] This application embodiment provides a possible implementation method, in which a square window is defined for the first target pixel currently being processed in step S103 above. , This represents the neighborhood region centered on the coordinates of the first target pixel, and may specifically include the following steps: Define a size of square window The window satisfies equation (5): (5) In equation (5), It represents any pixel within the window and its pixel value, where (x', y') are the coordinates of that pixel. and These are the coordinate constraints for each pixel in the window, limiting the window's range along the horizontal and vertical axes. These are parameters that control the window size.

[0040] In some embodiments, each pixel in the window is assigned a corresponding weight. , The weights are defined based on the Gaussian function, and the weight calculation formula is as follows: (6) In equation (6), It is an exponential function. It is the standard deviation of the Gaussian function.

[0041] In other embodiments, windows are weighted according to their corresponding values. The pixels in the array are paired and sorted to obtain an ordered pairing sequence. The target weighted median is calculated from this pairing sequence, and the target weighted median is used as the new value of the first target pixel. Specifically, this may include the following steps: Window based on corresponding weights The pixels in the array are paired as follows: (7) The window is obtained according to equation (7). The set of pairs; The pairing sets are sorted sequentially according to their weight values ​​to obtain an ordered pairing sequence. The weighted median of each pixel in the paired sequence is calculated sequentially, and the pixel is selected when equation (8) is satisfied. The target weighted median is calculated using the following formula: (8) In equation (8), express The cumulative weight, window The total number of pixels.

[0042] This embodiment can transform messy pixel information in an image into an ordered weighted median result, thereby suppressing noise and preserving texture details with the center pixel as the core. By allocating contribution based on the distance between the pixel and the center through Gaussian weights, the true information of neighboring pixels is preferentially preserved. Combined with the sorting and accumulation rules of the weighted median, extreme noise values ​​are further filtered, ultimately achieving a balance between noise reduction and detail preservation.

[0043] This application embodiment provides a possible implementation method. In step S103 above, the enhancement operation may specifically include: Select the second target image One of the pixels is used as the second target pixel , This represents the coordinates of the second target pixel, and defines a neighborhood for the second target pixel. , obtain The mean and standard deviation required for local contrast enhancement are calculated using the following formulas: (9) (10) In equation (9), Let represent the local mean within the neighborhood Ω(x,y). For the neighborhood Total number of pixels included. This represents the summation of all pixels within the neighborhood Ω(x,y); in equation (10), This represents the local standard deviation within the neighborhood Ω(x,y); The new value of the enhanced pixel is calculated based on the mean and standard deviation, using the following formula: (11) In equation (11), To enhance the new pixel value; It is the enhancement factor, used to control the intensity of contrast enhancement; It is a constant used to avoid numerical overflow problems.

[0044] This embodiment can transform bland local pixel features in an image into prominent detail information, thereby using a single pixel as a processing unit to achieve precise enhancement of the edges and texture features of water droplets.

[0045] This application embodiment provides a possible implementation method. In step S104 above, a threshold segmentation method is used to segment the third target image. Perform image segmentation to obtain the third target image. Based on the pixel matrix of the green channel, a binary result is generated. The binary result Satisfying equation (12): (12) In equation (12), For the third target image The pixel matrix of the green channel, It is a manually set threshold used to exclude interference from the surrounding background in the image.

[0046] This embodiment can transform complex enhanced image information into concise binary region markers, thereby accurately distinguishing the jet region from the background region at the pixel level. Because the experimental platform uses a green laser, the reflective properties of water droplets on green light allow the green channel pixel matrix to more clearly highlight the difference between the jet and the background, providing high-contrast raw data for segmentation; threshold It can flexibly adapt to the background interference intensity under different experimental environments, and through clear numerical judgment criteria, it can accurately eliminate interference, so that the binary results can accurately mark the jet range.

[0047] This application provides a possible implementation method. In step S105 above, a linear model is used to perform linear fitting calibration based on the water flow density measurement method used by the nozzle manufacturer to obtain the linear parameter value to be estimated. The fitting calibration process satisfies equation (13): (13) In equation (13), For actual physical measurement of density, and Let the linear parameter values ​​be those to be estimated. The Hadamard product represents the multiplication of corresponding elements in matrices at the same latitude. Here, it is based on... Estimate and .

[0048] This application embodiment provides a possible implementation method. In step S106 above, the uniformity of the jet region of the actual acquired image is calculated based on the linear parameter value to be estimated. The calculation of the uniformity of the jet region of the actual acquired image satisfies equation (14): (14) In equation (14), This serves as the final index for regional evenness. The distribution rate of the actual acquired images, through Multiply by the pixel matrix of the green channel obtained from the actual captured image, and perform a Hadamard product on the binary result obtained from the actual captured image, then add... The binary result obtained from the actual acquired image is generated based on the pixel matrix of the green channel of the actual acquired image. The probability is an ideal uniform distribution with reference to this.

[0049] The above introduces Figure 1 The embodiments shown have various implementation methods for each step. The vision-based multi-nozzle jet uniformity determination method of this application will be further explained below through specific embodiments.

[0050] This specific embodiment is based on a cooling spray box design and optimized hardware platform. In this platform, multiple nozzle arrays are fixed within a support grid. A camera is positioned above the grid and suspended (or mounted on a shooting bracket) in front of the nozzles, with a near-top-down shooting angle. The laser emitter is horizontally mounted; when the nozzle array is operating normally, the laser can intercept the ejected liquid water mist, which appears as a bright green area in the camera's view. The camera used has a resolution of 1920×1080. Figure 2 The flowchart illustrating the steps of this specific embodiment is shown below. Step 1: For continuously acquired video, follow the formula Keyframe saving and processing involves loading the next frame and calculating its sharpness. It then determines if the current frame has the highest sharpness; if not (N), it discards the frame. If yes (Y), it checks if all frames in the segment have been loaded; if not (N), it returns to loading the next frame and calculating its sharpness. If yes (Y), it performs subsequent frame-by-frame image preprocessing. Each selected video frame set is unique, and the image with the highest sharpness in the set is chosen as the keyframe. The number of elements in the selected set is typically... The selected keyframe images are as follows: Figure 3a and Figure 3b As shown.

[0051] Step 2: Preprocess the acquired images. First, perform image noise removal and highlight suppression algorithms. Specifically, for each image, pixel-by-pixel, apply the formula... A weighted median filter is calculated to smooth the image, reduce the loss of local information caused by water droplet reflection, and thus improve the quality of the image to be analyzed later.

[0052] Step 3: Perform image enhancement. According to the formula... The step of enhancing local contrast in an image involves increasing the contrast of specific areas to make image details stand out more, resulting in an image like... Figure 4a and Figure 4b As shown. After this step, the fine textures and high-frequency components of the surface structure of the water droplets are usually more apparent, thus providing clearer details for subsequent image segmentation and density calculation. The formula used in this step is... Involves two hyperparameters and In this invention, respectively select This is used as an empirical value for reference.

[0053] Step 4: Identify the jet region in the preprocessed image using the formula. Threshold segmentation is performed, and the identified jet region image is as follows: Figure 5a and Figure 5b As shown. During this process, the threshold... The empirical value range is selected to be between 180 and 230. The purpose of this step is to accurately separate the jet region and provide a clear region boundary for subsequent analysis.

[0054] Step 5: Calculate the visual density reference index of the jet. First, test the nozzle's flow rate, velocity, and jet pressure using physical measurement equipment (such as flow meters, pressure sensors, etc.). Then, use a high-speed camera to capture images of the jet flow and obtain the image pixel values ​​of the corresponding areas. Combining the physical measurement data, a data-driven method is used to fit a linear model and calculate the visual density reference index of the jet. The visualization diagram after density calibration is shown below. Figure 6a and Figure 6b As shown.

[0055] Step 6: After the jet density reference value is calculated, apply the formula... To calculate the jet uniformity index of a region, it is first necessary to design a reasonable uniform distribution for reference within the region's density reference range. Next, the KL divergence is calculated between the actual distribution and the reference distribution to measure the difference between the actual and uniform distributions. The smaller the calculated index value, the more consistent the actual and uniform distributions are, and the higher the jet uniformity of the region. A schematic diagram illustrating the effectiveness of the uniformity index in this specific embodiment is shown below. Figure 7 As shown, the uniformity of the water mist area is 0.24 and 1.34, respectively.

[0056] Step 7: Determine whether to terminate the measurement. If yes (Y), then end; if no (N), return to load the next frame and calculate the sharpness.

[0057] Considering the timeliness and convenience of on-site deployment, this method is developed based on Python and incorporates professional image processing libraries such as OpenCV, matrix calculation libraries such as NumPy, chart visualization libraries such as Matplotlib, and interactive operation libraries such as TKinter. Finally, the algorithm is packaged into an executable file, forming an easy-to-use GUI (Graphical User Interface). Users can input a video or an image, and the system automatically generates corresponding density visualization results and evenness index results. Experimental testing shows that the calculation speed for a single 1080×720 pixel image is less than 1 second, meeting the needs of practical industrial deployment and real-time detection.

[0058] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that within the spirit and principles of this application, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the corresponding technical solutions to leave the protection scope of this application.

Claims

1. A vision-based method for determining the uniformity of multi-nozzle jets, characterized in that, The method includes: For the acquired video stream of multiple nozzles in operation, key frames are selected as the first target image based on the sharpness index. ; For the first target image Each pixel in the first target image undergoes preprocessing. Each pixel in the image undergoes the same processing to obtain the corresponding weighted median as the new pixel value. This is applied to the first target image. The second target image is obtained by replacing the pixel value of all pixels with the corresponding weighted median. ; For the second target image Enhancement is performed on each pixel in the second target image. Each pixel in the image undergoes the same processing to obtain a new value for the corresponding enhanced pixel. This process is repeated for the second target image. The third target image is obtained by replacing the pixel values ​​of all pixels with the corresponding enhanced pixel values. ; Threshold segmentation method is used for the third target image Perform image segmentation to obtain the third target image. Based on the pixel matrix of the green channel, a binary result is generated. ; Based on the water flow density measurement method used by the nozzle manufacturer, a linear model is used for linear fitting calibration to obtain the linear parameter values ​​to be estimated. In practical applications, the uniformity of the jet region in the actual acquired image is calculated based on the linear parameter values ​​to be estimated.

2. The method according to claim 1, characterized in that, For the acquired video stream of multiple nozzles in operation, key frames are selected as the first target image based on the sharpness index. ,include: Determine the set of n consecutive frames of images after time t: (1) In equation (1), For a moment The subsequent continuous collection A collection of frame images; from Select the first target image The formula is as follows: (2) In equation (2), The first target image selected from the set. , for A single frame image at time t satisfies equation (3): (3) In equation (3), It is a gradient-based sharpness operator that satisfies equation (4): (4) In equation (4), and These are the height and width of the image, respectively. and These represent the pixel gradient values ​​calculated along the image width and height directions, respectively.

3. The method according to claim 1, characterized in that, Preprocessing operations include: Select the first target image A square window is defined by taking one of the pixels as the first target pixel. , Indicates the coordinates of the first target pixel. The x-axis is... The vertical axis is , This represents the neighborhood region centered at the coordinates of the first target pixel. For each pixel in the window, assign a corresponding weight. This represents the contribution of the pixel to the center pixel during the filtering process; the features defining the weights include pixel distance and similarity. Window based on corresponding weights The pixels in the array are paired and sorted to obtain an ordered pairing sequence. The target weighted median is calculated from the pairing sequence and used as the new value of the first target pixel.

4. The method according to claim 3, characterized in that, Define a square window for the first target pixel being processed. , The region of interest centered at the coordinates of the first target pixel includes: Define a size of square window The window satisfies equation (5): (5) In equation (5), It represents any pixel within the window and its pixel value, where (x', y') are the coordinates of that pixel. and These are the coordinate constraints for each pixel in the window, limiting the window's range along the horizontal and vertical axes. These are parameters that control the window size.

5. The method according to claim 4, characterized in that, For each pixel in the window, assign a corresponding weight. ,include: The weights are defined based on the Gaussian function, and the weight calculation formula is as follows: (6) In equation (6), It is an exponential function. It is the standard deviation of the Gaussian function.

6. The method according to claim 5, characterized in that, Window based on corresponding weights The pixels in the array are paired and sorted to obtain an ordered pairing sequence. The target weighted median is calculated from this pairing sequence, and the target weighted median is used as the new value of the first target pixel, including: Window based on corresponding weights The pixels in the array are paired as follows: (7) The window is obtained according to equation (7). The set of pairs; The pairing sets are sorted sequentially according to their weight values ​​to obtain an ordered pairing sequence. The weighted median of each pixel in the paired sequence is calculated sequentially, and the pixel is selected when equation (8) is satisfied. The target weighted median is calculated using the following formula: (8) In equation (8), express The cumulative weight, window The total number of pixels.

7. The method according to any one of claims 1 to 6, characterized in that, Enhanced operations include: Select the second target image One of the pixels is used as the second target pixel , This represents the coordinates of the second target pixel, and defines a neighborhood for the second target pixel. , obtain The mean and standard deviation required for local contrast enhancement are calculated using the following formulas: (9) (10) In equation (9), Let represent the local mean within the neighborhood Ω(x,y). For the neighborhood Total number of pixels included. This represents the summation of all pixels within the neighborhood Ω(x,y); in equation (10), This represents the local standard deviation within the neighborhood Ω(x,y); The new value of the enhanced pixel is calculated based on the mean and standard deviation, using the following formula: (11) In equation (11), To enhance the new pixel value; It is the enhancement factor, used to control the intensity of contrast enhancement; It is a constant used to avoid numerical overflow problems.

8. The method according to any one of claims 1 to 6, characterized in that, Threshold segmentation method is used for the third target image Perform image segmentation to obtain the third target image. Based on the pixel matrix of the green channel, a binary result is generated. ,include: The generated binary result Satisfying equation (12): (12) In equation (12), For the third target image The pixel matrix of the green channel, It is a manually set threshold used to exclude interference from the surrounding background in the image.

9. The method according to claim 8, characterized in that, Based on the water flow density measurement method used by the nozzle manufacturer, a linear model is used for linear fitting calibration to obtain the values ​​of the linear parameters to be estimated, including: The fitting and calibration process satisfies equation (13): (13) In equation (13), For actual physical measurement of density, and Let the linear parameter values ​​be those to be estimated. The Hadamard product represents the multiplication of corresponding elements in matrices at the same latitude. Here, it is based on... Estimate and .

10. The method according to claim 9, characterized in that, In practical applications, the uniformity of the jet region in the actual acquired image is calculated based on the linear parameter values ​​to be estimated, including: The calculation of the uniformity of the jet region in the actual acquired image satisfies equation (14): (14) In equation (14), This serves as the final index for regional evenness. The distribution rate of the actual acquired images, through Multiply by the pixel matrix of the green channel obtained from the actual captured image, and perform a Hadamard product on the binary result obtained from the actual captured image, then add... The binary result obtained from the actual acquired image is generated based on the pixel matrix of the green channel of the actual acquired image. The probability is an ideal uniform distribution with reference to this.

Images