Flotation bubble size extraction method based on landform map optimization control mark watershed

Abstract

The invention relates to the technical field of image processing, in particular to a flotation bubble size extraction method based on landform map optimization control mark watershed, which comprises the following steps: converting an input bubble image into a grey-scale map, and preprocessing the grey-scale map; adopting an OSTU adaptive threshold method to extract a highlight area in the processed foam image as a watershed control mark; carrying out gray scale overturning on the preprocessed froth image to generate an initial landform map, extracting a transparent window region in the froth image by adopting a U-Net model, and minimizing partial numerical values of the corresponding transparent window region in the initial landform map to obtain an optimized flotation froth landform map; and segmenting the image by adopting a control mark watershed algorithm. According to the method, the initial landform map is optimized according to the transparent window region, a pseudo boundary caused by the fact that the gray scale of the transparent window region is similar to the gray scale of the boundary during segmentation is avoided, and the problem of bubble wrong segmentation caused by the transparent window region is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method for extracting the size of flotation bubbles by optimizing the control markers of the watershed based on a geomorphic map. Background Art

[0002] Mineral processing is an important and indispensable link in the processing of mineral resources, and froth flotation is one of the most widely used mineral processing methods. Currently, approximately 90% of metals such as lead, zinc, and antimony globally are separated by froth flotation. During the froth flotation process, the visual characteristics of the foam surface are closely related to the key production indicators. In the flotation operation, the operators mainly make manual adjustments based on the state of the surface foam. However, due to the lack of accurate and reliable judgment bases and objective quantification standards, it is difficult to dynamically optimize the process parameters, and ultimately the flotation process cannot continuously maintain the best operating state. The bubble size is the most critical visual characteristic of the foam surface. Studying the measurement method of the foam size has important practical significance for optimizing the flotation operation and reducing resource waste.

[0003] Currently, the method based on machine vision is the most common technical means for extracting the size of flotation bubbles, and mainly realizes the extraction of the bubble size through foam image segmentation. Among them, the watershed algorithm is one of the most common bubble segmentation methods. For example, Wang Weixing et al. proposed a watershed segmentation method combining fractional-order differential and level set. The specific method includes: extracting the image texture features based on fractal geometry, adaptively determining the order of fractional-order differential enhancement; using an improved level set algorithm to extract the bright spot area of the bubbles. This method effectively reduces the over-segmentation of bubbles and improves the accuracy of bubble size extraction. However, the existing research on the watershed algorithm mostly focuses on the extraction of control markers, and has not deeply explored the construction and optimization of the watershed geomorphic map. When the "transparent window" phenomenon appears on the surface of flotation bubbles, the traditional watershed algorithm is prone to significant segmentation errors in the "transparent window" area, seriously reducing the accuracy of bubble size measurement. Although the image segmentation method based on convolutional neural network can better segment the foam image, it requires a large number of data samples, and the segmentation training samples need to be labeled, with heavy workload and high computational complexity, making it difficult to apply. Summary of the Invention

[0004] In view of the deficiencies of the existing technology, the present invention provides the following technical solutions:

[0005] A method for extracting the size of flotation bubbles by optimizing the control markers of the watershed based on a geomorphic map, comprising the following steps:

[0006] S10: Convert the input foam image into a grayscale image and perform preprocessing.

[0007] S20: Use the OSTU adaptive threshold method to extract the highlighted area in the processed foam image as the watershed control marker.

[0008] S30: Perform grayscale flipping on the preprocessed foam image to generate an initial geomorphic map. Meanwhile, use the U-Net model to extract the transparent window area in the foam image, and set the values of the corresponding transparent window areas in the initial geomorphic map to the lowest to obtain an optimized flotation bubble geomorphic map.

[0009] S40: According to the control markers obtained in step S20 and the optimized geomorphic map obtained in step S30, use the control marker watershed algorithm to segment the image.

[0010] As an improvement of the above technical solution, the preprocessing of the foam image in step S10 at least includes:

[0011] Perform mean filtering and Gaussian filtering on the foam image to achieve image denoising.

[0012] Use the SSR algorithm to correct the unevenly illuminated parts of the foam image.

[0013] As an improvement of the above technical solution, the mean filtering image processing includes the following steps:

[0014] S11: Select a template for the current pixel to be processed, and the template consists of several adjacent pixels.

[0015] S12: Calculate the mean value of all pixels in the template, and then assign this mean value to the current pixel.

[0016] As an improvement of the above technical solution, the Gaussian filtering image processing includes the following steps:

[0017] S13: Place the Gaussian kernel on the target pixel so that the center of the kernel aligns with the target pixel.

[0018] S14: Calculate the weighted average of the target pixel and its neighboring pixels, and the weights are provided by the Gaussian kernel.

[0019] S15: Assign the calculated average value to the target pixel.

[0020] As an improvement of the above technical solution, the SSR algorithm includes the following steps:

[0021] S16: Perform logarithmic transformation on the image to convert the brightness information of the image to the logarithmic domain.

[0022] S17: Perform Gaussian blur processing on the logarithmically transformed image to estimate the components of the image brightness.

[0023] S18: Take the difference between the original logarithmic image and the Gaussian-blurred image to obtain the logarithmic representation of the reflection component.

[0024] S19: Perform antilogarithmic transformation to convert the reflection component back to the original brightness range.

[0025] As an improvement to the above technical solution, the step S30 further includes the following steps:

[0026] S31: Construct a flotation bubble transparent window area recognition dataset and establish a flotation bubble transparent window area recognition model based on U-Net.

[0027] S32: Perform grayscale flipping on the foam image processed in step S10 to generate an initial geomorphic map.

[0028] S33: According to the flotation bubble transparent window area recognition model obtained in step S31, recognize the foam image processed in step S10 to obtain a transparent window area marked map.

[0029] S34: In the initial geomorphic map extracted in step S32, set the pixel values of the pixel points with a value of 1 in the transparent window area marked map obtained in step S33 to 0, so that the values of the corresponding transparent window areas in the initial geomorphic map are the lowest, and then obtain an optimized flotation bubble geomorphic map.

[0030] As an improvement to the above technical solution, the flotation bubble transparent window area recognition model in step S31 is trained on the dataset with binary cross-entropy as the loss function. The input image of this model is the preprocessed foam image, and the output image is the flotation bubble transparent window area marked image.

[0031] As an improvement to the above technical solution, the U-Net model adopts an encoder-decoder structure. The U-Net model fuses the high-resolution features extracted in the encoding stage with the low-resolution features extracted in the decoding stage through skip connections to realize the recognition of the local features and global distribution of the transparent window area.

[0032] Advantages of the present invention:

[0033] By constructing a high-quality flotation bubble "transparent window" area recognition dataset and using the U-Net framework to construct a flotation bubble "transparent window" area recognition model, high-precision extraction of the "transparent window" area is realized. At the same time, according to the extracted "transparent window" area, the initial geomorphic map is optimized, so that the values of the corresponding "transparent window" area parts in the initial geomorphic map are the lowest, avoiding false boundaries caused by the similarity between the gray level of the "transparent window" area and the boundary gray level during watershed segmentation, effectively overcoming the problem of incorrect bubble segmentation caused by the "transparent window" area, and improving the accuracy of bubble size extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is the technical framework diagram of the present invention;

[0035] Figure 2 It is a schematic diagram of the present invention after extracting the control mark;

[0036] Figure 3 It is a state diagram before and after optimizing the geomorphological map of the flotation bubble model after the present invention completes flipping;

[0037] Figure 4 It is a bubble segmentation result diagram before and after optimizing the geomorphological map after the present invention completes segmentation. Detailed implementation manners

[0038] The following uses specific specific examples to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0039] Currently, the method based on machine vision is the most common technical means for extracting the size of flotation bubbles, and mainly realizes the extraction of bubble size through foam image segmentation. Among them, the watershed algorithm is one of the most common bubble segmentation methods. For example, Wang Weixing et al. proposed a watershed segmentation method combining fractional-order differential and level set. The specific method includes: extracting image texture features based on fractal geometry, and adaptively determining the order of fractional-order differential enhancement; using an improved level set algorithm to extract the bright spot area of the bubble. This method effectively reduces the over-segmentation of bubbles and improves the accuracy of bubble size extraction. However, the existing research on the watershed algorithm mostly focuses on the extraction of control marks, and has not deeply explored the construction and optimization of the watershed geomorphological map. When the "transparent window" phenomenon appears on the surface of flotation bubbles, the traditional watershed algorithm is prone to significant segmentation errors in the "transparent window" area, seriously reducing the accuracy of bubble size measurement. Although the image segmentation method based on convolutional neural network can better segment the foam image, it requires a large number of data samples, and the segmentation training samples need to be labeled, with heavy workload and high computational complexity, making it difficult to apply.

[0040] To solve the above problems, a method for extracting the size of flotation bubbles based on optimizing the control mark of the geomorphological map is provided, including the following steps:

[0041] S10: Convert the input foam image into a grayscale image and perform preprocessing.

[0042] Among them, the preprocessing of the geomorphological map in the step S10 at least includes:

[0043] Perform mean filtering processing and Gaussian filtering processing on the foam image to achieve image denoising. For the mean filtering image processing, the following steps are usually adopted:

[0044] S11: Select a template for the current pixel to be processed. The template consists of several neighboring pixels.

[0045] S12: Calculate the mean value of all pixels in the template and assign this mean value to the current pixel. In Python, the mean filtering function can be implemented by calling the OpenCV library.

[0046] For Gaussian filtering of image processing, the following method is usually adopted:

[0047] Define the Gaussian function as two-dimensional to generate a Gaussian kernel. The Gaussian kernel is a matrix, and each element value is obtained by calculating the Gaussian function. Gaussian filtering applies the Gaussian kernel to each pixel in the image and performs the following steps for each pixel in the image:

[0048] S13: Place the Gaussian kernel on the target pixel so that the center of the kernel aligns with the target pixel.

[0049] S14: Calculate the weighted average of the target pixel and its neighboring pixels, where the weights are provided by the Gaussian kernel.

[0050] S15: Assign the calculated average value to the target pixel.

[0051] Similarly, the Gaussian filtering function can also be implemented by calling the OpenCV library in Python.

[0052] After completing the processing of image noise, further processing of the image is required. Since the color perceived by the human eye depends not only on the intensity of the light reflected from the object surface but also on the light conditions of the surrounding environment. Therefore, the SSR algorithm is used to correct the unevenly illuminated parts in the geomorphological map.

[0053] Among them, the basic process of the SSR algorithm is as follows:

[0054] S16: Perform a logarithmic transformation on the image to convert the brightness information of the image to the logarithmic domain.

[0055] S17: Perform Gaussian blur processing on the logarithmically transformed image to estimate the component of the image brightness.

[0056] S18: Take the difference between the original logarithmic image and the Gaussian-blurred image to obtain the logarithmic representation of the reflection component.

[0057] S19: Perform an inverse logarithmic transformation to convert the reflection component back to the original brightness range.

[0058] Similarly, the relevant functions of the SSR algorithm can also be implemented by calling the OpenCV library in Python. After completing the preprocessing of the image, step S20 is executed.

[0059] S20: Use the OSTU adaptive threshold method to extract the highlighted area in the processed foam image as the watershed control marker.

[0060] The OSTU adaptive threshold method is used to calculate the optimal threshold of the image and segment the highlighted area and the dark part, as Figure 2 shown. In this embodiment, the segmented highlighted area is marked. Since the extraction of the high-brightness points is not affected by the transparent window area, in this embodiment, the high-brightness points are extracted first, then the transparent window area is extracted, and finally the optimized geomorphic map is reconstructed.

[0061] Based on this, after the extraction of the highlighted area is completed, step S30 needs to be executed to extract the transparent window area of the flotation bubble.

[0062] S30: Perform gray-scale flipping on the preprocessed foam image to generate an initial geomorphic map. At the same time, use the U-Net model to extract the transparent window area in the foam image, and set the values of the corresponding transparent window area in the initial geomorphic map to the lowest to obtain an optimized flotation bubble geomorphic map.

[0063] Specifically, the step S30 includes the following steps:

[0064] S31: Construct a dataset for identifying the transparent window area of the flotation bubble and establish a recognition model for the transparent window area of the flotation bubble based on U-Net;

[0065] The recognition model for the transparent window area of the flotation bubble in the step S31 is trained on the dataset with binary cross-entropy as the loss function. The input image of this model is the preprocessed foam image, and the output image is the marked image of the transparent window area of the flotation bubble.

[0066] S32: Perform gray-scale flipping on the foam image processed in step S10 to generate an initial geomorphic map.

[0067] The flipping in step S32 utilizes the characteristic that the gray value at the boundary of the foam image is low, so that the value of the bubble boundary is higher, thus enabling the generation of an initial geomorphic map that conforms to the characteristics of bubble boundary extraction.

[0068] S33: According to the recognition model for the transparent window area of the flotation bubble obtained in step S31, recognize the foam image processed in step S10 to obtain a marked map of the transparent window area;

[0069] When obtaining the marked map of the transparent window area, a mark of 1 indicates that the pixel point in the figure belongs to the "transparent window" area, and 0 indicates that the pixel point in the figure does not belong to the "transparent window" area.

[0070] S34: In the initial landform map extracted in step S32, set the values of the pixel points with a value of 1 in the marked map of the transparent window area obtained in step S33 to 0, so that the values of the corresponding transparent window area in the initial landform map are the lowest, and then obtain the optimized flotation bubble landform map.

[0071] Among them, the U-Net model adopts an encoder-decoder structure. The U-Net model fuses the high-resolution features extracted in the encoding stage with the low-resolution features extracted in the decoding stage through skip connections to realize the recognition of the local features and global distribution of the transparent window area.

[0072] After the landform map optimization process of S34, the values of the corresponding "transparent window" area of the flotation bubbles in the initial landform map are the lowest, ensuring that the terrain height of the water collection basin inside the bubbles is lower than the edge, and avoiding the pseudo-boundaries caused by the similar gray levels of the "transparent window" area and the boundary gray level during watershed segmentation. As Figure 3 shown, where Figure 3 a is the landform map of the flotation bubble model before optimization, Figure 3 b is the landform map of the flotation bubble model after optimization.

[0073] S40: According to the control marks obtained in step S20 and the optimized landform map obtained in step S30, use the control mark watershed algorithm to segment the image.

[0074] Among them, perform the watershed algorithm on the optimized landform map based on the control marks, use the control marks as the water injection starting points, simulate the "flooding" process, and the dividing boundaries are formed at the meeting places of the waters in different regions, and finally realize the segmentation of the foam image. The segmentation effect in this embodiment is as Figure 4 shown, where Figure 4 a is the image before the landform map is optimized, Figure 4 b is the image after the landform map is optimized.

[0075] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A method for extracting flotation bubble sizes by optimizing and controlling a marked watershed based on a geomorphic map, characterized in that It includes the following steps: S10: Convert the input foam image into a grayscale image and perform preprocessing; S20: Use the OSTU adaptive threshold method to extract the highlighted area in the processed foam image as the watershed control marker; S30: Perform grayscale flipping on the preprocessed foam image to generate an initial landform map. At the same time, use the U-Net model to extract the transparent window area in the foam image, and set the values of the corresponding transparent window areas in the initial landform map to the lowest to obtain an optimized flotation bubble landform map; S40: According to the control marker obtained in step S20 and the optimized landform map obtained in step S30, use the control marker watershed algorithm to segment the image.

2. The method for extracting the flotation bubble size by optimizing and controlling the marked watershed based on the geomorphic map according to claim 1, wherein: The preprocessing of the landform map in step S10 at least includes: Perform mean filtering and Gaussian filtering on the foam image to achieve image denoising; Use the SSR algorithm to correct the unevenly illuminated parts of the foam image.

3. The method for extracting the flotation bubble size by optimizing the control of the marked watershed based on the geomorphic map according to claim 2, wherein: The mean filtering image processing includes the following steps: S11: Select a template for the current pixel to be processed, and the template consists of several adjacent pixels; S12: Calculate the mean value of all pixels in the template, and then assign this mean value to the current pixel.

4. The method for extracting flotation bubble size by optimizing and controlling the marked watershed based on the geomorphic map according to claim 2, wherein: The Gaussian filtering image processing includes the following steps: S13: Place the Gaussian kernel on the target pixel so that the center of the kernel aligns with the target pixel; S14: Calculate the weighted average of the target pixel and its neighboring pixels, and the weights are provided by the Gaussian kernel; S15: Assign the calculated average value to the target pixel.

5. The method for extracting flotation bubble size by optimizing and controlling the marked watershed based on the geomorphic map according to claim 2, wherein: The SSR algorithm includes the following steps: S16: Perform logarithmic transformation on the image to convert the brightness information of the image to the logarithmic domain; S17: Perform Gaussian blur processing on the logarithmically transformed image to estimate the brightness component of the image; S18: Take the difference between the original logarithmic image and the Gaussian-blurred image to obtain the logarithmic representation of the reflection component; S19: Perform inverse logarithmic transformation to convert the reflection component back to the original brightness range.

6. The method for extracting flotation bubble size by optimizing and controlling the marked watershed based on the geomorphic map according to claim 1, wherein: The step S30 includes the following steps: S31: Construct a flotation bubble transparent window area recognition dataset and establish a flotation bubble transparent window area recognition model based on U-Net; S32: Perform grayscale flipping on the foam image processed in step S10 to generate an initial landform map; S33: According to the flotation bubble transparent window area recognition model obtained in step S31, recognize the foam image processed in step S10 to obtain a transparent window area marker map; S34: In the initial landform map extracted in step S32, set the values of the pixel points with a value of 1 in the transparent window area marker map obtained in step S33 to 0, so that the values of the corresponding transparent window areas in the initial landform map are the lowest, and then obtain an optimized flotation bubble landform map.

7. The method for extracting the flotation bubble size by optimizing the control marked watershed based on the geomorphic map according to claim 6, wherein: The flotation bubble transparent window area recognition model in step S31 is trained on the dataset with binary cross-entropy as the loss function. The input image of this model is the preprocessed foam image, and the output image is the flotation bubble transparent window area marker image.

8. The method for extracting the flotation bubble size by optimizing the control marker watershed based on the geomorphic map according to any one of claims 1-7, characterized in that: The U-Net model adopts an encoder-decoder structure. The U-Net model fuses the high-resolution features extracted in the encoding stage with the low-resolution features extracted in the decoding stage through skip connections to achieve the recognition of the local features and global distribution of the transparent window area.

Images