Method and device for feature extraction of flotation froth, storage medium and computer equipment

Abstract

The application discloses a flotation froth feature extraction method and device, a storage medium and computer equipment, relates to the mineral flotation technology field, and mainly solves the problems of over-segmentation, noise sensitivity and large calculation of the existing flotation froth segmentation method, and comprises the following steps: acquiring flotation froth images of various mineral types in different flotation states, and extracting froth sample images from the flotation froth images to obtain a froth sample image dataset; performing model training operation on the improved visual large model by using the froth sample image dataset to obtain a trained froth image segmentation model; segmenting a to-be-segmented flotation froth image by using the froth image segmentation model to obtain a froth image segmentation result; and extracting bubble size distribution and froth flow velocity features of the flotation froth based on the froth image segmentation result.

Description

Technical Field

[0001] This invention relates to the field of mineral flotation technology, and in particular to a method and apparatus for extracting the characteristics of flotation foam, a storage medium, and a computer device. Background Technology

[0002] Flotation is a mineral processing method that utilizes the differences in the physicochemical properties of mineral particles to effectively separate different minerals. The most widely used industrial method is froth flotation. Flotation reagents are added to ground mineral pulp, and air is introduced to form a large number of bubbles. Hydrophobic mineral particles adhere to these bubbles and rise to the surface of the pulp, forming a mineralized froth layer. Hydrophilic mineral particles cannot adhere to the bubbles and remain in the pulp, thus achieving the separation of valuable minerals from gangue. During flotation, the size distribution and flow velocity characteristics of the flotation froth are closely related to flotation process variables and performance indicators.

[0003] The control of the flotation process generally relies on experienced flotation operators observing the size characteristics of the flotation froth and adjusting the operating conditions based on their experience. However, manual judgment is highly subjective and lacks universality. With the rapid development of digital image technology and computer vision technology, machine vision technology has been widely applied to mineral flotation processes. By segmenting flotation froth using computer vision technology, the size distribution and flow velocity of the flotation froth can be obtained. Current froth segmentation algorithms mostly use the watershed algorithm, but the watershed algorithm has drawbacks such as over-segmentation, sensitivity to noise, and high computational cost, making it difficult to achieve good segmentation results while meeting real-time segmentation requirements. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for feature extraction of flotation foam, a storage medium, and a computer device, the main purpose of which is to solve the problems of excessive segmentation, sensitivity to noise, and large computational load in existing flotation foam segmentation methods.

[0005] According to one aspect of the present invention, a method for feature extraction of flotation foam is provided, comprising:

[0006] Flotation foam images of various mineral types under different flotation states are obtained, and foam sample images are extracted from the flotation foam images to obtain a foam sample image dataset;

[0007] The improved visual large model was trained using the aforementioned foam sample image dataset to obtain a trained foam image segmentation model.

[0008] The foam image segmentation model described above is used to segment the flotation foam image to be segmented, and the foam image segmentation result is obtained.

[0009] Based on the foam image segmentation results, the bubble size distribution and foam flow velocity characteristics of the flotation foam are extracted.

[0010] Further, the step of extracting foam sample images from the flotation foam images to obtain a foam sample image dataset includes:

[0011] At least one foam region is randomly segmented from each of the flotation foam images using an image segmentation tool;

[0012] The flotation foam image is binarized based on the foam region to obtain the foam sample image; and the set of the flotation foam image and the foam sample image is determined as the foam sample image dataset.

[0013] Furthermore, before training the improved large visual model using the foam sample image dataset, the method further includes:

[0014] A parameter fine-tuning module is set in the image encoder structure of the large visual model;

[0015] A lightweight feature extraction module is provided outside the image encoder structure;

[0016] The output of the lightweight feature extraction module is used as the mask cue input of the cue encoder in the visual large model to obtain the improved visual large model.

[0017] Furthermore, the step of training the improved large visual model using the foam sample image dataset includes:

[0018] Obtain the center point of each foam region in the foam sample image dataset, and determine the center point as the segmentation cue point;

[0019] Based on the foam sample image and the segmentation cue points, the improved visual large model is used to segment the flotation foam image in the foam sample image dataset to obtain the foam classification prediction result corresponding to the flotation foam image.

[0020] The segmentation loss value of the improved visual large model is calculated based on the foam classification prediction results.

[0021] The improved visual large model is subjected to gradient training based on the segmentation loss value to obtain the foam image segmentation model.

[0022] Further, calculating the segmentation loss value of the improved visual large model based on the foam classification prediction result includes:

[0023] Based on the foam classification prediction results, the predicted probability that a pixel in the flotation foam image belongs to the bubble category is calculated, and a first loss value is determined based on the predicted probability.

[0024] The second loss value is determined based on the foam classification prediction results and the similarity coefficient loss function.

[0025] The first loss value and the second loss value are weighted and summed to obtain the segmentation loss value.

[0026] Further, the step of extracting the bubble size distribution and foam flow velocity features of the flotation foam based on the foam image segmentation results includes:

[0027] The area of ​​each bubble is calculated based on the bubble image segmentation results to obtain the bubble size distribution;

[0028] The foam image segmentation results of two consecutive frames of floating foam images in the time series are obtained and similarity matching is performed to obtain multiple successfully matched bubble pairs;

[0029] Obtain the centroids of the bubbles in the two frames before and after the bubble alignment, and calculate the displacement of each bubble based on the centroids;

[0030] The foam flow velocity characteristics are determined based on the displacement and the sampling time of the two consecutive frames of flotation foam images.

[0031] Further, determining the foam flow velocity characteristics based on the displacement and the sampling time of the two consecutive frames of flotation foam images includes:

[0032] The outliers in the displacement are removed according to the 3δ principle to obtain the displacement to be analyzed.

[0033] Calculate the average displacement based on the displacement to be analyzed;

[0034] The foam flow velocity characteristics are determined based on the average displacement and the sampling time.

[0035] According to another aspect of the present invention, a feature extraction apparatus for flotation foam is provided, comprising:

[0036] The sample extraction module is used to acquire flotation foam images of various minerals under different flotation states, and extract foam sample images from the flotation foam images to obtain a foam sample image dataset.

[0037] The model training module is used to train the improved visual large model using the foam sample image dataset to obtain a trained foam image segmentation model.

[0038] The segmentation processing module is used to segment the flotation foam image to be segmented using the foam image segmentation model to obtain the foam image segmentation result.

[0039] The feature extraction module is used to extract the bubble size distribution and foam flow velocity features of the flotation foam based on the foam image segmentation results.

[0040] Furthermore, the sample extraction module is also used for:

[0041] At least one foam region is randomly segmented from each of the flotation foam images using an image segmentation tool;

[0042] The flotation foam image is binarized based on the foam region to obtain the foam sample image; and the set of the flotation foam image and the foam sample image is determined as the foam sample image dataset.

[0043] Furthermore, the device also includes a model improvement module for:

[0044] A parameter fine-tuning module is set in the image encoder structure of the large visual model;

[0045] A lightweight feature extraction module is provided outside the image encoder structure;

[0046] The output of the lightweight feature extraction module is used as the mask cue input of the cue encoder in the visual large model to obtain the improved visual large model.

[0047] Furthermore, the model training module is also used for:

[0048] Obtain the center point of each foam region in the foam sample image dataset, and determine the center point as the segmentation cue point;

[0049] Based on the foam sample image and the segmentation cue points, the improved visual large model is used to segment the flotation foam image in the foam sample image dataset to obtain the foam classification prediction result corresponding to the flotation foam image.

[0050] The segmentation loss value of the improved visual large model is calculated based on the foam classification prediction results.

[0051] The improved visual large model is subjected to gradient training based on the segmentation loss value to obtain the foam image segmentation model.

[0052] Furthermore, the model training module is also used for:

[0053] Based on the foam classification prediction results, the predicted probability that a pixel in the flotation foam image belongs to the bubble category is calculated, and a first loss value is determined based on the predicted probability.

[0054] The second loss value is determined based on the foam classification prediction results and the similarity coefficient loss function.

[0055] The first loss value and the second loss value are weighted and summed to obtain the segmentation loss value.

[0056] Furthermore, the feature extraction module is also used for:

[0057] The area of ​​each bubble is calculated based on the bubble image segmentation results to obtain the bubble size distribution;

[0058] The foam image segmentation results of two consecutive frames of floating foam images in the time series are obtained and similarity matching is performed to obtain multiple successfully matched bubble pairs;

[0059] Obtain the centroids of the bubbles in the two frames before and after the bubble alignment, and calculate the displacement of each bubble based on the centroids;

[0060] The foam flow velocity characteristics are determined based on the displacement and the sampling time of the two consecutive frames of flotation foam images.

[0061] Furthermore, the feature extraction module is also used for:

[0062] The outliers in the displacement are removed according to the 3δ principle to obtain the displacement to be analyzed.

[0063] Calculate the average displacement based on the displacement to be analyzed;

[0064] The foam flow velocity characteristics are determined based on the average displacement and the sampling time.

[0065] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform an operation corresponding to the feature extraction method for flotation foam described above.

[0066] According to another aspect of the present invention, a computer device is provided, including a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0067] The memory is used to store at least one executable instruction that causes the processor to perform operations corresponding to the feature extraction method for flotation foam described above.

[0068] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages:

[0069] This invention provides a method, apparatus, storage medium, and computer equipment for feature extraction of flotation foam. Compared with existing technologies, this invention acquires flotation foam images of various mineral types under different flotation states, and extracts foam sample images from these images to obtain a foam sample image dataset. The improved visual large-scale model is trained using this dataset to obtain a trained foam image segmentation model. The foam image segmentation model is then used to segment the flotation foam image to be segmented, yielding a foam image segmentation result. Based on the segmentation result, the bubble size distribution and foam velocity features of the flotation foam are extracted, thus completing the feature extraction process for flotation foam in a timely and effective manner. This invention improves the segmentation performance of the visual large-scale model in the field of foam flotation by improving the visual large-scale model. The established sample dataset allows the model to extract the required foam features end-to-end without additional processing of the foam images, significantly improving the efficiency of the foam segmentation algorithm and meeting the real-time requirements of actual production processes.

[0070] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0071] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0072] Figure 1 A schematic flowchart of a method for feature extraction of flotation foam provided by an embodiment of the present invention is shown;

[0073] Figure 2 A schematic diagram of a foam sample image provided in an embodiment of the present invention is shown;

[0074] Figure 3 A schematic flowchart of another feature extraction method for flotation foam provided by an embodiment of the present invention is shown;

[0075] Figure 4 A schematic diagram of an improved visual large model provided by an embodiment of the present invention is shown;

[0076] Figure 5A schematic flowchart of another method for feature extraction of flotation foam provided by an embodiment of the present invention is shown;

[0077] Figure 6 A schematic flowchart of another method for feature extraction of flotation foam provided in an embodiment of the present invention is shown;

[0078] Figure 7 This diagram illustrates the structure of a feature extraction device for flotation foam provided in an embodiment of the present invention.

[0079] Figure 8 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0080] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure 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 disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0081] This invention provides a method for feature extraction of flotation foam, such as... Figure 1 As shown, the method includes:

[0082] 101. Obtain flotation foam images of various mineral types under different flotation states, and extract foam sample images from the flotation foam images to obtain a foam sample image dataset;

[0083] In this embodiment of the invention, the current execution terminal acquires flotation foam images of various mineral types under different flotation states. The mineral types include nickel ore and copper ore, etc., and this embodiment of the invention does not specifically limit the types. Next, the current execution terminal extracts foam sample images from the acquired flotation foam images to obtain a foam sample image dataset. Because the acquired flotation foam images correspond to different mineral types and flotation states, the extracted foam sample images have richer foam morphology features, which is more conducive to the learning of foam morphology by the large visual model.

[0084] 102. The improved visual large model is trained using the foam sample image dataset to obtain a trained foam image segmentation model.

[0085] In this embodiment of the invention, the current execution end uses a foam sample image dataset to train an improved visual large-scale model. The preferred visual large-scale model is the SAM (Segment Anything Model). The SAM model consists of three main parts: an image encoder, a cue encoder, and a mask decoder. The image encoder and mask decoder are composed of attention mechanism layers, MLP layers, and ReLU activation functions; the cue encoder is composed of CLIP and multi-layer convolutional modules. The process of the current execution end training the improved visual large-scale model using the foam sample image dataset can be viewed as the SAM model learning various foam morphologies; this embodiment of the invention does not impose specific limitations. After training is complete, a trained foam image segmentation model is obtained.

[0086] 103. The flotation froth image to be segmented is processed using the aforementioned froth image segmentation model to obtain the froth image segmentation result;

[0087] In this embodiment of the invention, the current execution end uses a trained foam image segmentation model to segment the flotation foam image to be segmented. For example, the flotation foam image to be segmented is used as the input of the SAM model, and the SAM model can output a mask image of the flotation foam image segmentation result to obtain the foam image segmentation result.

[0088] 104. Extract the bubble size distribution and foam flow velocity characteristics of the flotation foam based on the foam image segmentation results.

[0089] In this embodiment of the invention, the current execution end performs feature extraction operations based on the foam image segmentation results, including extracting the bubble size distribution features and foam flow velocity features of the flotation foam. The bubble size distribution features represent the probability distribution of flotation foam according to bubble size; the foam flow velocity features represent the flow speed of bubbles in the slurry during the flotation process.

[0090] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to quickly obtain foam samples of various forms, another feature extraction method for flotation foam is provided. The steps include extracting foam sample images from the flotation foam images to obtain a foam sample image dataset, including:

[0091] At least one foam region is randomly segmented from each of the flotation foam images using an image segmentation tool;

[0092] The flotation foam image is binarized based on the foam region to obtain the foam sample image; and the set of the flotation foam image and the foam sample image is determined as the foam sample image dataset.

[0093] In this embodiment of the invention, the current execution end randomly segments at least one bubble region from each of the flotation bubble images using an image segmentation tool. The preferred image segmentation tool is PaddleSeg, which allows for manual labeling of the bubble images, i.e., randomly segmenting at least one bubble region from each flotation bubble image. It should be noted that, to improve the efficiency of flotation bubble feature extraction, only one bubble region can be randomly segmented from each flotation bubble image. Since a random method is used, the diversity of the resulting bubbles remains unaffected.

[0094] Then, the current execution end performs binarization processing on the flotation foam image based on the segmented foam region to obtain a foam sample image, such as... Figure 2 As shown, Figure 2 The white area in the middle represents the segmented foam region after binarization. The current execution end defines the set of floated foam images and foam sample images as the foam sample image dataset, so as to obtain image data for model training from this dataset.

[0095] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to further improve the ability of the large visual model to segment bubbles in images, another feature extraction method for flotation bubbles is provided, such as... Figure 3 As shown, before training the improved large visual model using the foam sample image dataset, the method further includes:

[0096] 201. Set a parameter fine-tuning module in the image encoder structure of the large visual model;

[0097] 202. A lightweight feature extraction module is provided outside the image encoder structure;

[0098] 203. Use the output of the lightweight feature extraction module as the mask cue input of the cue encoder in the visual large model to obtain the improved visual large model.

[0099] In this embodiment of the invention, the current execution end sets a parameter fine-tuning module in the image encoder structure of the large visual model, such as setting a parameter fine-tuning module LoRA in the image encoder structure of the SAM model. The LoRA module allows the model to improve the segmentation effect by updating only some parameters during model training, thereby reducing computational costs. The current execution end sets a lightweight feature extraction module outside the image encoder structure of the large visual model, such as setting a lightweight feature extraction module ResNet outside the image encoder structure of the SAM model. The output of the lightweight feature extraction module ResNet is used as the mask cue input of the cue encoder in the large visual model, resulting in an improved large visual model, the structure of which is as follows. Figure 4As shown, the improved visual large model has a better segmentation effect on bubbles.

[0100] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to enable the improved visual large model to have a better segmentation effect on foam, another feature extraction method for flotation foam is provided, such as... Figure 5 As shown, the steps involve training the improved large visual model using the foam sample image dataset, including:

[0101] 301. Obtain the center point of each foam region in the foam sample image dataset, and determine the center point as the segmentation prompt point;

[0102] In this embodiment of the invention, the current execution end obtains the center point of each foam region in the foam sample image dataset, which can be represented by coordinates (x, y), and the obtained center point is determined as the segmentation prompt point.

[0103] It should be noted that, as Figure 4 As shown, in the improved visual large model SAM, the prompt encoder receives the segmentation prompt points or bounding boxes determined in this step as input, which guides the model to perform the segmentation task and improves the model's segmentation performance.

[0104] 302. Based on the foam sample image and the segmentation cue points, the improved visual large model is used to segment the flotation foam image in the foam sample image dataset to obtain the foam classification prediction result corresponding to the flotation foam image;

[0105] In this embodiment of the invention, the foam classification prediction result corresponding to the flotation foam image obtained by the current execution end includes the region predicted as foam after segmentation and the region predicted as background after segmentation. This embodiment of the invention does not make specific limitations.

[0106] 303. Calculate the segmentation loss value of the improved visual large model based on the foam classification prediction results;

[0107] In this embodiment of the invention, the current execution terminal calculates the segmentation loss value of the improved visual large model based on the foam classification prediction result. It should be noted that the steps for calculating the segmentation loss value include:

[0108] The current execution stage calculates the predicted probability that a pixel in the flotation foam image belongs to the bubble category based on the foam classification prediction results, and determines the first loss value based on the predicted probability. If the Focal Loss function is chosen as the first loss function, it adjusts the cross-entropy loss to reduce the weight of easily classified samples, causing the model to focus more on difficult-to-classify samples during training. Furthermore, in datasets with extremely imbalanced classes, Focal Loss can effectively reduce the influence of the dominant class on the loss function, enabling the model to better learn the features of the minority classes. Specifically, the predicted probability that a pixel in the foam classification prediction results belongs to the bubble category is first calculated, denoted as p. t Then, the first loss value is calculated using the following formula:

[0109] Loss(p t )=-a t (1-p t ) γ log(p t )

[0110] Among them, a t γ is the weight of the predicted class, and γ is an adjustment factor used to reduce the weight of simple samples.

[0111] Next, the current execution end determines the second loss value based on the bubble classification prediction result and the similarity coefficient loss function. The similarity coefficient loss function is the Dice Loss function, which performs stably with small sample sizes, especially in the early stages of training when the model's predictions are inaccurate, providing stable gradients. Specifically, the second loss value is calculated using the following formula:

[0112]

[0113] Where TP is the number of bubble pixels predicted as bubble pixels in the bubble classification prediction result, FP is the number of background pixels predicted as bubble pixels in the bubble classification prediction result, FN is the number of bubble pixels predicted as background pixels in the bubble classification prediction result, and ε is a smoothing factor used to avoid the case where the denominator is zero.

[0114] The current execution end performs a weighted summation of the first loss value and the second loss value calculated above to obtain the segmentation loss value. It should be noted that the weight coefficients of the two loss values ​​can be set as needed, such as setting the weight coefficient of the first loss value to 0.6 and the weight coefficient of the second loss value to 0.4, etc., and this embodiment of the invention does not make specific limitations.

[0115] 304. Based on the segmentation loss value, perform gradient training on the improved visual large model to obtain the foam image segmentation model.

[0116] In this embodiment of the invention, the current execution end uses the LoRA module and the lightweight ResNet module to fine-tune the SAM large model. During the training process, the parameters of the image encoder and mask decoder in the SAM model are frozen. Based on the segmentation loss value, only the parameters of the LoRA module, the lightweight ResNet module and the cue encoder module are trained and adjusted to obtain the foam image segmentation model.

[0117] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to quickly and accurately extract the characteristics of flotation foam, another method for extracting the characteristics of flotation foam is provided, such as... Figure 6 As shown, the steps for extracting the bubble size distribution and foam flow velocity features of flotation foam based on the foam image segmentation results include:

[0118] 401. Calculate the area of ​​each bubble based on the bubble image segmentation results to obtain the bubble size distribution;

[0119] In this embodiment of the invention, the current execution end calculates the area of ​​each bubble based on the bubble image segmentation result, then performs statistical processing on the calculated bubble area, and determines the distribution of bubble size based on the statistical data.

[0120] 402. Obtain the bubble image segmentation results of two consecutive frames of floating bubble images in the time series and perform similarity matching to obtain multiple successfully matched bubble pairs;

[0121] In this embodiment of the invention, the current execution terminal obtains the foam image segmentation results of two consecutive frames of floating foam images in the sampling time sequence, such as obtaining the foam image segmentation result A corresponding to sampling time t. t And the foam image segmentation result A corresponding to sampling time t+1. t+1 The embodiments of the present invention are not specifically limited. Next, based on the foam image segmentation result A... t And foam image segmentation result A t+1 Similarity matching is performed to identify bubbles that meet the matching criteria in the two segmentation results above as bubble pairs, thus obtaining multiple successfully matched bubble pairs.

[0122] 403. Obtain the centroid of the bubbles in the two frames before and after the bubble alignment, and calculate the displacement of each bubble based on the centroid;

[0123] In this embodiment of the invention, the current execution end obtains the centroid of the bubble in the two frames before and after bubble alignment, such as using coordinates (x... At y At A represents the foam image segmentation result. t The centroid of the bubble; using coordinates (x At+1 y At+1A represents the foam image segmentation result. t+1 The centroid of the bubble is not specifically limited in this embodiment of the invention. The current execution terminal is based on the above-mentioned centroid (x At y At ) and center of mass (x) At+1 y At+1 Calculate the displacement of each bubble.

[0124] 404. Based on the displacement and the sampling time of the two consecutive frames of flotation foam images, determine the foam flow velocity characteristics.

[0125] In this embodiment of the invention, the current execution terminal determines the foam velocity characteristics based on the multiple displacements calculated in step 403 and the sampling time of the two consecutive frames of flotation foam images. It should be noted that before calculating the foam velocity characteristics, outliers in the multiple displacements calculated in step 403 can be removed according to the 3δ principle to obtain the displacements to be analyzed, which can effectively avoid the influence of outliers on the foam velocity characteristics. Then, the current execution terminal calculates the average displacement based on the displacements to be analyzed after removing outliers; finally, the foam velocity characteristics are determined based on the average displacement and the sampling time.

[0126] This invention provides a method for feature extraction of flotation foam. Compared with existing technologies, this invention acquires flotation foam images of various mineral types under different flotation states, and extracts foam sample images from these images to obtain a foam sample image dataset. The improved visual large-scale model is trained using this dataset to obtain a trained foam image segmentation model. The foam image segmentation model is then used to segment the flotation foam image to be segmented, yielding a foam image segmentation result. Based on the segmentation result, the bubble size distribution and foam velocity features of the flotation foam are extracted, thus completing the feature extraction process for flotation foam in a timely and effective manner. This invention improves the segmentation performance of the visual large-scale model in the field of foam flotation by improving the visual large-scale model. The established sample dataset allows the model to extract the required foam features end-to-end without additional processing of the foam images, significantly improving the efficiency of the foam segmentation algorithm and meeting the real-time requirements of actual production processes.

[0127] As a response to the above Figure 1 The implementation of the method shown in this invention provides a feature extraction device for flotation foam, such as... Figure 7 As shown, the device includes:

[0128] Sample extraction module 51 is used to acquire flotation foam images of various mineral types under different flotation states, and extract foam sample images from the flotation foam images to obtain a foam sample image dataset;

[0129] The model training module 52 is used to train the improved visual large model using the foam sample image dataset to obtain a trained foam image segmentation model.

[0130] The segmentation processing module 53 is used to segment the flotation foam image to be segmented using the foam image segmentation model to obtain the foam image segmentation result.

[0131] The feature extraction module 54 is used to extract the bubble size distribution and foam flow velocity features of the flotation foam based on the foam image segmentation results.

[0132] Furthermore, the sample extraction module 51 is also used for:

[0133] At least one foam region is randomly segmented from each of the flotation foam images using an image segmentation tool;

[0134] The flotation foam image is binarized based on the foam region to obtain the foam sample image; and the set of the flotation foam image and the foam sample image is determined as the foam sample image dataset.

[0135] Furthermore, the device also includes a model improvement module for:

[0136] A parameter fine-tuning module is set in the image encoder structure of the large visual model;

[0137] A lightweight feature extraction module is provided outside the image encoder structure;

[0138] The output of the lightweight feature extraction module is used as the mask cue input of the cue encoder in the visual large model to obtain the improved visual large model.

[0139] Furthermore, the model training module 52 is also used for:

[0140] Obtain the center point of each foam region in the foam sample image dataset, and determine the center point as the segmentation cue point;

[0141] Based on the foam sample image and the segmentation cue points, the improved visual large model is used to segment the flotation foam image in the foam sample image dataset to obtain the foam classification prediction result corresponding to the flotation foam image.

[0142] The segmentation loss value of the improved visual large model is calculated based on the foam classification prediction results.

[0143] The improved visual large model is subjected to gradient training based on the segmentation loss value to obtain the foam image segmentation model.

[0144] Furthermore, the model training module 52 is also used for:

[0145] Based on the foam classification prediction results, the predicted probability that a pixel in the flotation foam image belongs to the bubble category is calculated, and a first loss value is determined based on the predicted probability.

[0146] The second loss value is determined based on the foam classification prediction results and the similarity coefficient loss function.

[0147] The first loss value and the second loss value are weighted and summed to obtain the segmentation loss value.

[0148] Furthermore, the feature extraction module 54 is also used for:

[0149] The area of ​​each bubble is calculated based on the bubble image segmentation results to obtain the bubble size distribution;

[0150] The foam image segmentation results of two consecutive frames of floating foam images in the time series are obtained and similarity matching is performed to obtain multiple successfully matched bubble pairs;

[0151] Obtain the centroids of the bubbles in the two frames before and after the bubble alignment, and calculate the displacement of each bubble based on the centroids;

[0152] The foam flow velocity characteristics are determined based on the displacement and the sampling time of the two consecutive frames of flotation foam images.

[0153] Furthermore, the feature extraction module 54 is also used for:

[0154] The outliers in the displacement are removed according to the 3δ principle to obtain the displacement to be analyzed.

[0155] Calculate the average displacement based on the displacement to be analyzed;

[0156] The foam flow velocity characteristics are determined based on the average displacement and the sampling time.

[0157] This invention provides a feature extraction device for flotation foam. Compared with existing technologies, this invention acquires flotation foam images of various mineral types under different flotation states, and extracts foam sample images from these images to obtain a foam sample image dataset. The improved visual large-scale model is trained using this dataset to obtain a trained foam image segmentation model. The foam image segmentation model is then used to segment the flotation foam image to be segmented, yielding a foam image segmentation result. Based on the segmentation result, the bubble size distribution and flow velocity features of the flotation foam are extracted, thus completing the feature extraction process for flotation foam in a timely and effective manner. This invention improves the segmentation performance of the visual large-scale model in the field of foam flotation by improving the visual large-scale model. The established sample dataset allows the model to extract the required foam features end-to-end without additional processing of the foam images, significantly improving the efficiency of the foam segmentation algorithm and meeting the real-time requirements of actual production processes.

[0158] According to one embodiment of the present invention, a storage medium is provided, the storage medium storing at least one executable instruction that can perform the feature extraction method for flotation foam in any of the above method embodiments.

[0159] Figure 8 The diagram illustrates a structural schematic of a computer device according to an embodiment of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0160] like Figure 8 As shown, the computer device may include: a processor 602, a communications interface 604, a memory 606, and a communications bus 608.

[0161] The processor 602, communication interface 604, and memory 606 communicate with each other via communication bus 608.

[0162] Communication interface 604 is used to communicate with other network elements such as clients or other servers.

[0163] The processor 602 is used to execute program 610, which can specifically perform the relevant steps of the above-mentioned feature extraction method for flotation foam.

[0164] Specifically, program 610 may include program code that includes computer operation instructions.

[0165] Processor 602 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0166] Memory 606 is used to store program 610. Memory 606 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0167] Specifically, program 610 can be used to cause processor 602 to perform the following operations:

[0168] Flotation foam images of various mineral types under different flotation states are obtained, and foam sample images are extracted from the flotation foam images to obtain a foam sample image dataset;

[0169] The improved visual large model was trained using the aforementioned foam sample image dataset to obtain a trained foam image segmentation model.

[0170] The foam image segmentation model described above is used to segment the flotation foam image to be segmented, and the foam image segmentation result is obtained.

[0171] Based on the foam image segmentation results, the bubble size distribution and foam flow velocity characteristics of the flotation foam are extracted.

[0172] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0173] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting features from flotation foam, characterized in that, include: Flotation foam images of various mineral types under different flotation states are obtained, and foam sample images are extracted from the flotation foam images to obtain a foam sample image dataset; The improved visual large model is trained using the aforementioned foam sample image dataset to obtain a trained foam image segmentation model, wherein the visual large model is the SAM model. The foam image segmentation model described above is used to segment the flotation foam image to be segmented, and the foam image segmentation result is obtained. Based on the foam image segmentation results, the bubble size distribution and foam flow velocity characteristics of the flotation foam are extracted; Before training the improved large visual model using the foam sample image dataset, the method further includes: A parameter fine-tuning module is set in the image encoder structure of the large visual model; A lightweight feature extraction module is provided outside the image encoder structure; The output of the lightweight feature extraction module is used as the mask cue input of the cue encoder in the visual large model to obtain the improved visual large model. The step of training the improved large visual model using the foam sample image dataset includes: Obtain the center point of each foam region in the foam sample image dataset, and determine the center point as the segmentation cue point; Based on the foam sample image and the segmentation cue points, the improved visual large model is used to segment the flotation foam image in the foam sample image dataset to obtain the foam classification prediction result corresponding to the flotation foam image. The segmentation loss value of the improved visual large model is calculated based on the foam classification prediction results. The improved visual large model is subjected to gradient training based on the segmentation loss value to obtain the foam image segmentation model. The calculation of the segmentation loss value of the improved visual large model based on the foam classification prediction results includes: Based on the foam classification prediction results, the predicted probability that a pixel in the flotation foam image belongs to the bubble category is calculated, and a first loss value is determined based on the predicted probability. The second loss value is determined based on the foam classification prediction results and the similarity coefficient loss function. The first loss value and the second loss value are weighted and summed to obtain the segmentation loss value.

2. The method according to claim 1, characterized in that, The step of extracting foam sample images from the flotation foam images to obtain a foam sample image dataset includes: At least one foam region is randomly segmented from each of the flotation foam images using an image segmentation tool; The flotation foam image is binarized based on the foam region to obtain the foam sample image; and the set of the flotation foam image and the foam sample image is determined as the foam sample image dataset.

3. The method according to claim 1, characterized in that, The step of extracting the bubble size distribution and foam flow velocity features of flotation foam based on the foam image segmentation results includes: The area of ​​each bubble is calculated based on the bubble image segmentation results to obtain the bubble size distribution; The foam image segmentation results of two consecutive frames of floating foam images in the time series are obtained and similarity matching is performed to obtain multiple successfully matched bubble pairs; Obtain the centroids of the bubbles in the two frames before and after the bubble alignment, and calculate the displacement of each bubble based on the centroids; The foam flow velocity characteristics are determined based on the displacement and the sampling time of the two consecutive frames of flotation foam images.

4. The method according to claim 3, characterized in that, The determination of the foam flow velocity characteristics based on the displacement and the sampling time of the two consecutive frames of flotation foam images includes: According to 3 The principle is to eliminate outliers from the displacement values ​​to obtain the displacement values ​​to be analyzed. Calculate the average displacement based on the displacement to be analyzed; The foam flow velocity characteristics are determined based on the average displacement and the sampling time.

5. A device for extracting the characteristics of flotation foam, characterized in that, The apparatus used in the feature extraction method for flotation foam as described in any one of claims 1-4 comprises: The sample extraction module is used to acquire flotation foam images of various minerals under different flotation states, and extract foam sample images from the flotation foam images to obtain a foam sample image dataset. The model training module is used to train the improved visual large model using the foam sample image dataset to obtain a trained foam image segmentation model. The segmentation processing module is used to segment the flotation foam image to be segmented using the foam image segmentation model to obtain the foam image segmentation result. The feature extraction module is used to extract the bubble size distribution and foam flow velocity features of the flotation foam based on the foam image segmentation results.

6. A storage medium storing at least one executable instruction that performs an operation corresponding to the feature extraction method for flotation foam as described in any one of claims 1-4.

7. A computer device, comprising a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform an operation corresponding to the feature extraction method for flotation foam as described in any one of claims 1-4.

Images