Deep learning-based mineral flotation froth image classification method and system

Abstract

The invention belongs to the technical field of mineral flotation, and discloses a mineral flotation froth image classification method and system based on deep learning. The method comprises the following steps: constructing a mineral flotation froth primary selection model for classifying mineral flotation froth images based on a deep learning network; a laser camera is used for obtaining mineral flotation froth images in different time periods in the production condition, and extracted denoised parameters of standard texture features, color features and froth size distribution bottom layer features serve as an input set and are input into a mineral flotation froth primary selection model; and training the mineral flotation froth primary selection model to obtain clear features of mineral flotation froth images in different time periods, performing classification by using an improved image classification method, and obtaining mineral flotation froth working condition states in different time periods in a production condition according to a classification result. The concentrate grade is improved, and the mineral recovery rate is increased by about 20%.

Description

Technical Field

[0001] The present invention belongs to the technical field of mineral flotation, and particularly relates to a classification method and system for mineral flotation foam images based on deep learning. Background Art

[0002] The production process of mineral flotation is generally controlled by experienced workers observing the state of flotation foam. The uncertainty of this operation makes it difficult for the flotation process to operate in an optimal state. Using digital image processing technology to classify and interpret flotation foam images, obtain the working condition information of mineral flotation, and then perform optimized control is an effective method to improve economic benefits. At present, the research on the classification and recognition of flotation foam images mainly describes the foam images by extracting underlying feature parameters such as texture features, color features, and foam size distribution of the flotation foam images. Then, existing technologies use methods such as neural networks or support vector machines for classification and recognition to obtain the working conditions to guide production.

[0003] However, when the image is processed as a whole, the foam image described only by the underlying features of the image is not easy for people to understand, that is, there is a semantic gap problem in the classification results based on the underlying features of the image. In addition, since the local information of the foam image is not utilized, and there are noise interferences such as light and dust in the industrial field, for two completely different images, the extracted global underlying feature descriptions may be very close, which leads to low accuracy in the classification and recognition of neural networks or SVMs based on the underlying features of the foam image, resulting in frequent misoperations in production operations based on the working conditions and making it difficult for production to operate stably and optimally.

[0004] To solve the above problems, the existing technical invention patent on the classification method for mineral flotation foam images (authorization announcement number CN102855492A, authorization announcement date January 2, 2013) mainly includes three stages: (1) generation of a foam state vocabulary for the foam image based on texture features and color features; (2) bag-of-words description of the foam image; (3) classification of the foam image using a vector space model; this application abstracts the foam image into text, and then, like processing text, it can classify and recognize the foam image at the semantic level. Since the local information of the image is fully utilized, the foam image is processed in blocks and undergoes two abstraction processes to obtain a description of the foam image at the semantic level, solving the semantic gap problem.

[0005] However, the existing technology only theoretically analyzes that it is feasible, but in practical applications, the vector inner product method prefers linear solutions, but in actual production, the foam image may be random; the disadvantage of the K-nearest neighbor method is high computational complexity and high space complexity. This results in a relatively heavy computational process for foam classification.

[0006] Deep exploration provides a wider application space for the field of image processing. It can accurately classify some objects with unclear image source information and disturbed images to obtain accurate information. Summary of the Invention

[0007] To overcome the problems existing in the related technologies, the disclosed embodiments of the present invention provide a classification method and system for mineral flotation foam images based on deep learning.

[0008] The technical solution is as follows: A classification method for mineral flotation foam images based on deep learning includes the following steps:

[0009] S1, based on a deep learning network, construct a primary model for classifying mineral flotation foam images;

[0010] S2, use a laser camera to obtain mineral flotation foam images at different time periods during the production status, and take the parameters of the extracted and denoised standard texture features, color features, and underlying features of foam size distribution as the input set, and input them into the primary model for classifying mineral flotation foam images;

[0011] S3, train the primary model for classifying mineral flotation foam images containing the input set to obtain clear mineral flotation foam images at different time periods, classify them using an improved image classification method, and obtain the working conditions of mineral flotation foam at different time periods during the production status according to the classification results.

[0012] In step S1, based on a deep learning network, construct a primary model for classifying mineral flotation foam images, including:

[0013] S101, select mineral flotation foam images with abnormal foam production, set a mineral flotation foam image encoder and perform identification, and report to the mineral flotation foam image decoder, so that the mineral flotation foam image decoder cyclically detects the status of each mineral flotation foam image encoder;

[0014] S102, through adding a constraint on the total amount of mineral flotation foam generated, allocate the amount of mineral flotation foam generated under different production process conditions, update the local data of the mineral flotation foam production section, and transmit it to the mineral flotation foam image decoder.

[0015] In step S101, set a mineral flotation foam image encoder and perform identification, and report to the mineral flotation foam image decoder, so that the mineral flotation foam image decoder cyclically detects the status of each mineral flotation foam image encoder, including:

[0016] (1) The mineral flotation foam image encoder is connected to the mineral flotation foam image decoder. Find the parameters of the primary model of the mineral flotation foam to minimize the image classification offset function, and select the top Y mineral flotation foam image encoders with the largest offset to perform local updates. Among them, the expression of the image classification offset function is:

[0017]

[0018] In the formula, A(d0) is the image classification offset function, E is the number of mineral flotation foam image encoders, ξ is the sample batch, B e is the local data set, h(d, ξ) is the offset function of the sample batch ξ randomly selected from the local data set B e and the model parameter d0, O e is the data ratio of the e-th mineral flotation foam image encoder, A e (d) is the local offset function of the mineral flotation foam image e, d * is the model parameter;

[0019] (2) Select the current global primary model of the mineral flotation foam for global update, define the mineral flotation foam image selection strategy function λ gy , and map the global primary model of the mineral flotation foam d(g0) to the selected set of mineral flotation foam images C(λ gy , d0). The expression is:

[0020]

[0021] In the formula, Top means selected, is the updated set of active mineral flotation foam image encoders, A y is the offset function of the active mineral flotation foam image encoder, d(g0) is the global primary model of the mineral flotation foam, C(λ gy , d0) is the selected set of mineral flotation foam images, and λ gy is the mineral flotation foam image selection strategy function;

[0022] (3) Update the loop detection g selected by the mineral flotation foam image encoder. The expression is:

[0023]

[0024] In the formula, d e (g + 1, G) is the local primary model parameter of the mineral flotation foam image e at g + 1, Y is the size of the set of active mineral flotation foam image encoders, v is the size of the batch, is the random gradient on the batch ξ(g, G), is the global primary model parameter of the mineral flotation foam, dy (g, G) is the local mineral flotation foam primary selection model parameter of the mineral flotation foam image encoder y at g, and η(g, G) is the current local mineral flotation foam primary selection model parameter of the mineral flotation foam image encoder at g. is the downward gradient, and ξ(g, G) is the batch of the mineral flotation foam image e at g. is the global mineral flotation foam primary selection model parameter at g + 1.

[0025] In step (1), select the top Y mineral flotation foam image encoders with the largest offsets to perform local updates, including:

[0026] The mineral flotation foam production section sends the locally lost data in the production section to the mineral flotation foam image decoder, and the mineral flotation foam image decoder tracks each response; the mineral flotation foam image encoder is selected by the mineral flotation foam image decoder, processed according to the latest status of the mineral flotation foam production section and the channel status information, and broadcast to the mineral flotation foam production section before aggregation; set the latest offset value record of the unselected mineral flotation foam image encoder to 1, and the selected mineral flotation foam production section uploads the corresponding mineral flotation foam primary selection model parameters. The expression is:

[0027]

[0028] In the formula, are the global mineral flotation foam primary selection model parameters at g, respectively, and η g is the local mineral flotation foam primary selection model parameter at g, and d y (g) is the local mineral flotation foam primary selection model parameter at g, and ξ(g) is the mineral flotation batch at g.

[0029] In step S102, update the local data of the mineral flotation foam production section and transmit it to the mineral flotation foam image decoder, including:

[0030] (1) Model the system connecting the mineral flotation foam production section and the parameter mineral flotation foam image decoder through error control, and use orthogonal frequency division multiplexing technology for transmission.

[0031] (2) In the distributed classifier, the mineral flotation foam image decoder receives the superimposed signals from neighbors: in the b-th round of communication for classification, the local update is sent to the mineral flotation foam image decoder through error control, and there are S sub-channels with N different production process conditions; the sub-channels to Allocated to the mineral flotation foam production section \(y\in[Y]\), where \(y\) is a sub-mineral flotation foam image encoder of a certain production section in the mineral flotation foam image encoders for mineral flotation foam production, and \(Y\) is the set of mineral flotation foam image encoders in mineral flotation foam production; for the generated amount of mineral flotation foam Mineral flotation foam production section The capacity of the parallel wireless Gaussian channel between the mineral flotation foam production section and the mineral flotation foam image decoder is determined by the following formula:

[0032]

[0033] In the formula, \(U(g)\) is the amount of mineral flotation foam generated allocated to the mineral flotation foam production section \(y\) in communication round \(g\);

[0034] (3) Allocation of the amount of mineral flotation foam generated on the sub-channel \(o\) y,i [s] Adapts to the corresponding channel coefficient \(T\) y [s] To perform gradient aggregation through spatial compression, the mineral flotation foam image decoder aggregates messages according to the aggregation rule to obtain a new global model update.

[0035] In step (3), the constraints on the amount of mineral flotation foam generated on the sub-channel are as follows:

[0036]

[0037] In the formula, is the amount of mineral flotation foam generated on the sub-channel, \(V_0\) is the total initial amount of mineral flotation foam generated, is the number of constraints of the mineral flotation foam image encoder;

[0038] The channel coefficients are evenly distributed on the sub-channels, and the constraint on the amount of mineral flotation foam generated for each channel is reduced to:

[0039]

[0040] In the formula, \(S\) is the number of sub-channels;

[0041] Mineral flotation foam production section The amount of mineral flotation foam generated on the \(i\)-th sub-channel is as follows:

[0042]

[0043] In the formula, \(O_0\) is the proportionality factor for controlling the average released amount of mineral flotation foam generated, is the value under the constraint of the proportionality factor of the change value of the amount of mineral flotation foam generated, is the change value of the mineral flotation foam production volume on the i-th sub-channel at time g and in the mineral flotation foam production section of a certain mineral flotation foam image decoder;

[0044]

[0045] Each mineral flotation foam production section sends the processed signal in the χ iteration with the mineral flotation foam production volume V g V g satisfies:

[0046]

[0047] In the formula, is a linear decreasing function.

[0048] Furthermore, the linear decreasing function has the expression:

[0049]

[0050] In the formula, min_iter is the predefined minimum number of communication rounds, and mum_iter is the current number of communication rounds;

[0051] The exact value of the proportionality factor o0 for controlling the average released mineral flotation foam production volume is defined by the following formula:

[0052]

[0053] In the formula, s is the number of sub-channels with a linear trend, and Ei(θ) is the energy of the mineral flotation foam production volume at the θ angle on the i-th sub-channel.

[0054] Furthermore, the mineral flotation foam image decoder aggregates messages according to the aggregation rule to obtain a new global model update, including:

[0055] Find the geometric median of a set of points. Given point D (g) ∈{d y (g), y ∈ [Y]}, the fixed-point l mean is the point D that minimizes * , where D (g) is the node in the global model at time g, d y (g) is the local mineral flotation foam primary selection model parameter at time g, β y is the mineral flotation foam image decoder aggregation coefficient, and || || is the Euclidean norm;

[0056]

[0057] In the formula, The aggregation distance weight of the mineral flotation foam image decoder at time g, ω y The aggregation distance value of the mineral flotation foam image decoder at time g, is the weight of ω y (g);

[0058]

[0059] In the formula, min{} is the minimization set, and I is the smoothing factor of the distance;

[0060] The objective function is used to find the vector with the minimum distance from all updated messages, and is defined as:

[0061]

[0062] In the formula, is the fixed-point distance vector in the global model, and D is the fixed point in the global model;

[0063] To satisfy the average mineral flotation foam production constraint, the channel inversion application at the mineral flotation foam production section is:

[0064]

[0065] In the formula, x y (g) is the inversion parameter of the local mineral flotation foam primary selection model at time g, is the local mineral flotation foam production at time g, and k y (g) is the update information, and x' y (g) is the channel inversion of x y (g), is the mineral flotation foam production scaling factor.

[0066] In step S3, an improved image classification method is used for classification, including: calculating the correlation coefficient H between each standard texture feature, color feature, and the underlying feature nodes of the foam size distribution and the operating state of the mineral flotation foam f , and the calculation formula is as follows:

[0067]

[0068] In the formula, m is the overall state of the mineral flotation foam operating condition, n is the total number of categories of the standard texture feature, color feature, and the underlying feature nodes of the foam size distribution, and S ab is the corresponding classification count, and Q abThe classification count of the expected standard texture features, color features, and underlying features of the foam size distribution obtained from the primary mineral flotation foam selection model based on the original input set. Here, a represents the current state of a certain mineral flotation foam working condition, and b represents the category of the nodes of the standard texture features, color features, and underlying features of the foam size distribution.

[0069] Another object of the present invention is to provide a classification system for mineral flotation foam images based on deep learning. This system implements the classification method for mineral flotation foam images based on deep learning, and the system includes:

[0070] A primary mineral flotation foam model construction module for constructing a primary mineral flotation foam model for classifying mineral flotation foam images based on a deep learning network;

[0071] An image data input module for using a laser camera to obtain mineral flotation foam images at different time periods during the production status, and taking the parameters of the extracted denoised standard texture features, color features, and underlying features of the foam size distribution as an input set and inputting them into the primary mineral flotation foam model;

[0072] A mineral flotation foam classification module for training the primary mineral flotation foam model containing the input set to obtain clear mineral flotation foam images at different time periods, classifying them using an improved image classification method, and obtaining the mineral flotation foam working condition status at different time periods during the production status according to the classification results.

[0073] Combining all the above technical solutions, the beneficial effects of the present invention are as follows: The foam image classification method of the present invention can accurately classify and identify foam images, and the recognition rate is increased from 88% in the prior art to nearly 93% or so, thereby determining which working condition of excellent, good, medium, or poor the current production belongs to, and thus being used to guide and optimize production. Moreover, in-depth exploration is carried out on some objects with unclear image source information to obtain accurate information. After classifying the foam images, the current working condition is identified, and thus it is used to guide and optimize production, which can reduce the dosage of chemicals by 30 - 40% and increase the concentrate grade and mineral recovery rate by nearly 20% or so. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings herein are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure;

[0075] Figure 1 It is a flowchart of the classification method for mineral flotation foam images based on deep learning provided by an embodiment of the present invention;

[0076] Figure 2It is a flow chart of a primary mineral flotation foam selection model constructed based on a deep learning network for classifying mineral flotation foam images provided by an embodiment of the present invention;

[0077] Figure 3 It is a schematic diagram of a classification system for mineral flotation foam images based on deep learning provided by an embodiment of the present invention;

[0078] In the figure: 1. Primary mineral flotation foam selection model construction module; 2. Image data input module; 3. Mineral flotation foam classification module. Specific embodiments

[0079] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0080] Embodiment 1, as Figure 1 shown, the classification method for mineral flotation foam images based on deep learning provided by an embodiment of the present invention includes:

[0081] S1. Based on a deep learning network, construct a primary mineral flotation foam selection model for classifying mineral flotation foam images;

[0082] S2. Use a laser camera to obtain mineral flotation foam images at different time periods during the production process, and take the parameters of the extracted and denoised standard texture features, color features, and underlying features of foam size distribution as the input set, and input them into the primary mineral flotation foam selection model;

[0083] S3. Train the primary mineral flotation foam selection model containing the input set to obtain clear mineral flotation foam images at different time periods, classify them using an improved image classification method, and obtain the working conditions of mineral flotation foam at different time periods during the production process according to the classification results.

[0084] Exemplarily, as Figure 2 shown, in step S1, based on a deep learning network, construct a primary mineral flotation foam selection model for classifying mineral flotation foam images, including:

[0085] S101. Select mineral flotation foam images with abnormal foam production, set a mineral flotation foam image encoder for identification, and report to the mineral flotation foam image decoder, so that the mineral flotation foam image decoder circularly detects the status of each mineral flotation foam image encoder;

[0086] S102. By adding a constraint on the total amount of mineral flotation foam generated, allocate the amount of mineral flotation foam generated under different production process conditions, update the local data of the mineral flotation foam production section, and transmit it to the mineral flotation foam image decoder.

[0087] Exemplarily, the step S101 includes:

[0088] (1) Connect the mineral flotation foam image encoder to the mineral flotation foam image decoder, find the mineral flotation foam primary selection model parameters to minimize the image classification offset function, and select the top Y mineral flotation foam image encoders with the largest offset to perform local updates; where the expression of the image classification offset function is:

[0089]

[0090] In the formula, A(d0) is the image classification offset function, E is the number of mineral flotation foam image encoders, ξ is the sample batch, B e is the local data set, h(d, ξ) is the offset function of the sample batch ξ randomly selected from the local data set B e and the model parameter d0, O e is the data ratio of the e-th mineral flotation foam image encoder, A e (d) is the local offset function of the mineral flotation foam image e, d * is the model parameter;

[0091] (2) Select the current global mineral flotation foam primary selection model for global update, define the mineral flotation foam image selection strategy function λ gy , map the global mineral flotation foam primary selection model d(g0) to the selected mineral flotation foam image set C(λ gy , d0), and the expression is:

[0092]

[0093] In the formula, Top is selected, is the updated set of active mineral flotation foam image encoders, A y is the offset function of the active mineral flotation foam image encoder, d(g0) is the global mineral flotation foam primary selection model, C(λ gy , d0) is the selected mineral flotation foam image set, and λ gy is the mineral flotation foam image selection strategy function;

[0094] (3) Update the loop detection g selected by the mineral flotation foam image encoder, and the expression is:

[0095]

[0096] where d e (g + 1, G) are the local mineral flotation foam primary selection model parameters of the mineral flotation foam image e at g + 1, Y is the size of the active mineral flotation foam image encoder set, v is the batch size, is the stochastic gradient on batch ξ(g, G), are the global mineral flotation foam primary selection model parameters, d y (g, G) are the local mineral flotation foam primary selection model parameters of the mineral flotation foam image encoder y at g, and η(g, G) are the current local mineral flotation foam primary selection model parameters of the mineral flotation foam image encoder at g, is the downward gradient, ξ(g, G) is the batch of the mineral flotation foam image e at g, are the global mineral flotation foam primary selection model parameters at g + 1.

[0097] Exemplarily, in step (1), the top Y mineral flotation foam image encoders with the largest offsets are selected to perform local updates, including:

[0098] The mineral flotation foam production section sends the locally lost data in the production section to the mineral flotation foam image decoder, and the mineral flotation foam image decoder tracks each response; the mineral flotation foam image encoder is selected by the mineral flotation foam image decoder, processed according to the latest status of the mineral flotation foam production section and the channel status information, and broadcast to the mineral flotation foam production section before aggregation; the latest offset value records of the unselected mineral flotation foam image encoders are set to 1, and the corresponding mineral flotation foam primary selection model parameters are uploaded by the selected mineral flotation foam production section, and the expression is:

[0099]

[0100] where are the global mineral flotation foam primary selection model parameters at g, η g are the local mineral flotation foam primary selection model parameters at g, d y (g) are the local mineral flotation foam primary selection model parameters at g, and ξ(g) is the mineral flotation batch at g.

[0101] Exemplarily, in step S102, updating the local data of the mineral flotation foam production section and transmitting it to the mineral flotation foam image decoder includes:

[0102] (1) Modeling the system connecting the mineral flotation foam production section and the parameter mineral flotation foam image decoder through error control, and using orthogonal frequency division multiplexing technology for transmission;

[0103] (2) In the distributed classifier, the mineral flotation foam image decoder receives the superimposed signals from neighbors: In the b-th round of communication for classification, the local update is sent to the mineral flotation foam image decoder through error control, with S sub-channels having N different production process conditions; the sub-channels to are assigned to the mineral flotation foam production section y ∈ [Y], where y is the sub-mineral flotation foam image encoder of a certain production section in the mineral flotation foam image encoder set in mineral flotation foam production, and Y is the set of mineral flotation foam image encoders in mineral flotation foam production; for the generated mineral flotation foam production volume mineral flotation foam production section The parallel wireless Gaussian channel capacity between the mineral flotation foam production section and the mineral flotation foam image decoder is determined by the following formula:

[0104]

[0105] where U(g) is the mineral flotation foam production volume assigned to the mineral flotation foam production section y in communication round g;

[0106] (3) The mineral flotation foam production volume allocation o y,i [s] adapts to the corresponding channel coefficient T y [s] to perform gradient aggregation through spatial compression, and the mineral flotation foam image decoder aggregates the messages according to the aggregation rule to obtain a new global model update.

[0107] In step (3), the mineral flotation foam production volume on the sub-channel is constrained as follows:

[0108]

[0109] where is the mineral flotation foam production volume on the sub-channel, V0 is the total initial mineral flotation foam production volume, is the number of constraints of the mineral flotation foam image encoder;

[0110] The channel coefficients are evenly distributed on the sub-channels, and the mineral flotation foam production volume constraint for each channel is reduced to:

[0111]

[0112] where S is the number of sub-channels;

[0113] mineral flotation foam production section The mineral flotation foam production volume generated on the i-th sub-channel is as follows:

[0114]

[0115] where O0 is the proportionality factor for controlling the average released mineral flotation foam generation amount, is the value constrained by the proportionality factor of the change value of the mineral flotation foam generation amount, is the change value of the mineral flotation foam generation amount in the mineral flotation foam production section in the i-th sub-channel at time g and in a certain mineral flotation foam image decoder;

[0116]

[0117] Each mineral flotation foam production section transmits the processed signal with the mineral flotation foam generation amount V g in the χ iteration, V g satisfies:

[0118]

[0119] where, is a linear decreasing function.

[0120] Linear decreasing function has the expression:

[0121]

[0122] where min_iter is the predefined minimum number of communication rounds and mum_iter is the current number of communication rounds;

[0123] The exact value of the proportionality factor o0 for controlling the average released mineral flotation foam generation amount is defined by the following formula:

[0124]

[0125] where s is the number of sub-channels with a linear trend and Ei(θ) is the energy of the mineral flotation foam generation amount at the θ angle on the i-th sub-channel.

[0126] Exemplarily, for the model aggregation step, the mineral flotation foam image decoder aggregates messages according to certain aggregation rules to obtain a new global model update. Different from the standard method with one-step average aggregation, since the naive average aggregation rule is vulnerable to model or data attacks, the standard system for average aggregation requires a more powerful aggregation rule. The geometric median aggregation rule has good convergence guarantees for the mineral flotation foam production line, especially in the case where some mineral flotation foam production sections are under attack, which makes it an ideal tool for the standard system of average aggregation. Finding the geometric median of a set of points, given the point D (g) ∈{d y (g),y∈[Y]}, the fixed-point l mean is the point D that minimizes * , where D (g)is a node in the global model at time g, d y (g) are the local mineral flotation foam primary selection model parameters at time g, β y is the mineral flotation foam image decoder aggregation coefficient, || || is the Euclidean norm;

[0127]

[0128] In the formula, is the mineral flotation foam image decoder aggregation distance weight at time g, ω y (g) is the mineral flotation foam image decoder aggregation distance value at time g, is the weight of ω y (g);

[0129]

[0130] In the formula, min{} is the minimization set, and I is the smoothing factor of the distance;

[0131] The objective function is used to find the vector with the minimum distance from all updated messages, and is defined as:

[0132]

[0133] In the formula, is the fixed-point distance vector in the global model, and D is the fixed point in the global model;

[0134] To satisfy the average mineral flotation foam production constraint, the channel inversion application at the mineral flotation foam production section is:

[0135]

[0136] In the formula, x y (g) are the local mineral flotation foam primary selection model inversion parameters at time g, is the local mineral flotation foam production at time g, k y (g) is the update information, x' y (g) is the channel inversion of x y (g), is the mineral flotation foam production scaling factor.

[0137] Exemplarily, in step S3, an improved image classification method is used for classification, including: calculating the correlation coefficient H between each standard texture feature, color feature, and the underlying feature nodes of the foam size distribution and the mineral flotation foam working condition state f , and the calculation formula is as follows:

[0138]

[0139] In the formula, m represents the overall state of the mineral flotation foam working condition, n represents the total number of categories of the underlying feature nodes of the standard texture feature, color feature, and foam size distribution, S ab is the corresponding classification count, Q ab is the classification count of the expected standard texture feature, color feature, and underlying feature of the foam size distribution obtained based on the primary selection model of the mineral flotation foam that originally included the input set. a represents the current situation of a certain mineral flotation foam working condition, and b represents the category of the underlying feature nodes of the standard texture feature, color feature, and foam size distribution.

[0140] Example 2, as Figure 3 shown, the classification system for mineral flotation foam images based on deep learning provided by the embodiment of the present invention includes:

[0141] A mineral flotation foam primary selection model construction module 1, which is used to construct a mineral flotation foam primary selection model for classifying mineral flotation foam images based on a deep learning network;

[0142] An image data input module 2, which is used to obtain mineral flotation foam images at different time periods in the production situation by using a laser camera, and use the parameters of the extracted denoised standard texture feature, color feature, and underlying feature of the foam size distribution as an input set and input it into the mineral flotation foam primary selection model;

[0143] A mineral flotation foam classification module 3, which is used to train the mineral flotation foam primary selection model containing the input set to obtain clear mineral flotation foam images at different time periods, classify them by using an improved image classification method, and obtain the mineral flotation foam working condition states at different time periods in the production situation according to the classification results.

[0144] By applying the method proposed by the present invention to the actual scenario, it shows that the present invention avoids the problem of overcrowding in classification communication, can quickly classify images, and obtain the mineral flotation foam working condition states at different time periods in the production situation according to the classification results, providing a theoretical basis for actual production.

[0145] The above is only a relatively optimal specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A classification method for mineral flotation foam images based on deep learning, characterized in that, The method includes the following steps: S1. Based on a deep learning network, construct a primary model for classifying mineral flotation foam images; S2. Use a laser camera to obtain mineral flotation foam images at different time periods during the production status, and take the parameters of the extracted and denoised standard texture features, color features, and underlying features of the foam size distribution as the input set, and input them into the primary model for mineral flotation foam; S3. Train the primary model for mineral flotation foam containing the input set to obtain clear mineral flotation foam images at different time periods, classify them using an improved image classification method, and obtain the working conditions of mineral flotation foam at different time periods during the production status according to the classification results.

2. The classification method of mineral flotation foam images based on deep learning according to claim 1, characterized in that, In step S1, based on a deep learning network, constructing a primary model for classifying mineral flotation foam images includes: S101. Select mineral flotation foam images with abnormal foam production, set up and identify a mineral flotation foam image encoder, and report to the mineral flotation foam image decoder, so that the mineral flotation foam image decoder circularly detects the status of each mineral flotation foam image encoder; S102. Through adding a constraint on the total amount of mineral flotation foam generated, allocate the amount of mineral flotation foam generated under different production process conditions, update the local data of the mineral flotation foam production section, and transmit it to the mineral flotation foam image decoder.

3. The classification method of mineral flotation foam images based on deep learning according to claim 2, characterized in that, In step S101, setting up and identifying a mineral flotation foam image encoder and reporting to the mineral flotation foam image decoder so that the mineral flotation foam image decoder circularly detects the status of each mineral flotation foam image encoder includes: (1) Connect the mineral flotation foam image encoder to the mineral flotation foam image decoder, find the image classification offset function that minimizes the parameters of the primary model for mineral flotation foam, and select the top Y mineral flotation foam image encoders with the largest offset to perform local update; among them, the expression of the image classification offset function is: Wherein, A(d0) is the image classification offset function, E is the number of mineral flotation foam image encoders, ξ is the sample batch, and B e is the local dataset, and h(d,ξ) is the offset function of the sample batch ξ randomly selected from the local dataset B e and the model parameter d0, and O e is the data ratio of the e-th mineral flotation foam image encoder, and A e (d) is the local offset function of the mineral flotation foam image e, and d * is the model parameter; (2) Select the current global mineral flotation foam primary selection model for global update, and define the mineral flotation foam image selection strategy function λ gy , map the global mineral flotation foam primary selection model d(g0) to the selected mineral flotation foam image set C(λ gy , d0), and the expression is: Wherein, Top is selected, is the updated active mineral flotation foam image encoder set, A y is the offset function of the active mineral flotation foam image encoder, d(g0) is the global mineral flotation foam primary selection model, C(λ gy , d0) is the selected mineral flotation foam image set, λ gy is the mineral flotation foam image selection strategy function; (3) The update of the circular detection g selected by the mineral flotation foam image encoder for clustering, the expression is: where d e (g + 1, G) is the local mineral flotation foam preliminary selection model parameter of the mineral flotation foam image e at g + 1, Y is the size of the active mineral flotation foam image encoder set, v is the size of the batch, is the stochastic gradient on the batch ξ(g, G), is the global mineral flotation foam preliminary selection model parameter, d y (g, G) is the local mineral flotation foam preliminary selection model parameter of the mineral flotation foam image encoder y at g, and η(g, G) is the current local mineral flotation foam preliminary selection model parameter of the mineral flotation foam image encoder at g, is the downward gradient, ξ(g, G) is the batch of the mineral flotation foam image e at g, is the global mineral flotation foam preliminary selection model parameter at g + 1.

4. The classification method of mineral flotation foam images based on deep learning according to claim 3, wherein In step (1), selecting the top Y mineral flotation foam image encoders with the largest offset to perform local update includes: The mineral flotation foam production section sends the locally lost data in the production section to the mineral flotation foam image decoder, and the mineral flotation foam image decoder tracks each response; the mineral flotation foam image encoder is selected by the mineral flotation foam image decoder, processed according to the latest status of the mineral flotation foam production section and the channel status information, and broadcast to the mineral flotation foam production section before aggregation; set the latest offset value record of the unselected mineral flotation foam image encoder to 1, and the selected mineral flotation foam production section uploads the corresponding parameters of the primary model for mineral flotation foam, the expression is: In the formula, are the global mineral flotation foam primary selection model parameters at g, η g is the local mineral flotation foam primary selection model parameter at g, d y (g) is the local mineral flotation foam primary selection model parameter at g, and ξ(g) is the mineral flotation batch at g.

5. The classification method of mineral flotation foam images based on deep learning according to claim 2, characterized in that, In step S102, updating the local data of the mineral flotation foam production section and transmitting it to the mineral flotation foam image decoder includes: (1) Model the system connecting the mineral flotation foam production section and the parameter mineral flotation foam image decoder through error control, and use orthogonal frequency division multiplexing technology for transmission; (2) In the distributed classifier, the mineral flotation foam image decoder receives the superimposed signals from neighbors: in the b-th round of communication for classification, the local update is sent to the mineral flotation foam image decoder through error control, with S sub-channels under N different production process conditions; the sub-channels to are assigned to the mineral flotation foam production section y ∈ [Y], where y is the sub-mineral flotation foam image encoder of a certain production section in the mineral flotation foam image encoder set during mineral flotation foam production, and Y is the mineral flotation foam image encoder set during mineral flotation foam production; for the generated mineral flotation foam production volume mineral flotation foam production section and the parallel wireless Gaussian channel capacity between the mineral flotation foam image decoder are determined by the following formula: Where U(g) is the amount of mineral flotation foam generated in the mineral flotation foam production section y during communication round g; (3) Mineral flotation foam generation amount distribution on sub-channels y,i [s] Adapt to the corresponding channel coefficients T y [s] For gradient aggregation through spatial compression, the mineral flotation foam image decoder aggregates messages according to the aggregation rules to obtain a new global model update.

6. The classification method of mineral flotation foam images based on deep learning according to claim 5, characterized in that In step (3), the constraint on the amount of mineral flotation foam generated on the sub-channel is as follows: Wherein, is the mineral flotation foam generation amount on the sub-channel, and V0 is the total initial mineral flotation foam generation amount, is the number of constraints of the mineral flotation foam image encoder; The channel coefficients are evenly distributed on the sub-channels, and the constraint on the amount of mineral flotation foam generated on each channel is reduced to: Where S is the number of sub-channels; Mineral flotation foam production section The production of mineral flotation foam on the i-th sub-channel is as follows: Where, O0 is the proportionality factor for controlling the average generated amount of mineral flotation foam, is the value constrained by the proportionality factor of the change value of the mineral flotation foam generation amount, is the change value of the mineral flotation foam generation amount at the g-th moment on the i-th subchannel and in the mineral flotation foam production section of a certain mineral flotation foam image decoder; In each mineral flotation foam production section, the processed signal is sent in the χ iteration with the mineral flotation foam production volume V g where V g satisfies: In the formula, is a linearly decreasing function.

7. The classification method of mineral flotation foam images based on deep learning according to claim 6, characterized in that, Linear decline function The expression is as follows: Where min_iter is the predefined minimum number of communication rounds, and mum_iter is the current number of communication rounds; The exact value of the proportionality factor o0 for controlling the average released amount of mineral flotation foam is defined by the following formula: Where s is the number of sub-channels with a linear trend, and Ei(θ) is the energy of the amount of mineral flotation foam generated at angle θ on the i-th sub-channel.

8. The classification method of mineral flotation foam images based on deep learning according to claim 5, characterized in that The mineral flotation foam image decoder aggregates messages according to the aggregation rule to obtain a new global model update, including: Find the geometric median of a set of points, given point D (g) ∈{d y (g),y∈[Y]}, the fixed point l mean is minimized Point D * , where D (g) is the node in the global model at time g, d y (g) is the local mineral flotation foam primary selection model parameter at time g, β y is the aggregation coefficient of mineral flotation froth image decoder, || || is the Euclidean norm; In the formula, is the aggregation distance weight of the mineral flotation foam image decoder at time g, ω y (g) is the aggregation distance value of the mineral flotation foam image decoder at time g, is the weight of ω y (g); Where min{} is the minimization set, and I is the smoothing factor of the distance; The objective function is used to find the vector with the minimum distance from all updated messages, and is defined as: In the formula, is the fixed-point distance vector in the global model, and D is the fixed point in the global model; To meet the average amount of mineral flotation foam generation constraint, the channel inversion at the mineral flotation foam production section is applied as: where x y (g) is the inversion parameter of the local mineral flotation foam primary selection model at g, is the local mineral flotation foam generation amount at g, k y (g) is the updated information, x' y (g) is the channel inversion of x y (g), is the scaling factor of the mineral flotation foam generation amount.

9. The classification method of mineral flotation foam images based on deep learning according to claim 1, characterized in that, In step S3, classification is performed using an improved image classification method, including: calculating the correlation coefficient H between each standard texture feature, color feature, and underlying feature node of the foam size distribution and the mineral flotation foam condition state f , and the calculation formula is as follows: Wherein, m is the overall state of the mineral flotation foam condition, n is the total number of categories of the underlying feature nodes of the standard texture feature, color feature, and foam size distribution, S ab is the corresponding classification count, Q ab is the classification count of the expected standard texture feature, color feature, and foam size distribution underlying features obtained from the primary mineral flotation foam selection model based on the original input set, a is the current condition of a certain mineral flotation foam condition state, and b is the category of the underlying feature nodes of the standard texture feature, color feature, and foam size distribution.

10. A classification system for mineral flotation foam images based on deep learning, characterized in that, The system implements the classification method of the mineral flotation foam image based on deep learning according to any one of claims 1-9. The system includes: A mineral flotation foam primary selection model construction module (1) for constructing a mineral flotation foam primary selection model for classifying mineral flotation foam images based on a deep learning network; An image data input module (2) for using a laser camera to obtain mineral flotation foam images at different times during the production status, and taking the parameters of the denoised standard texture features, color features, and underlying features of the foam size distribution extracted as the input set and inputting them into the mineral flotation foam primary selection model; A mineral flotation foam classification module (3) for training the mineral flotation foam primary selection model containing the input set to obtain clear mineral flotation foam images at different times, classifying them using an improved image classification method, and obtaining the mineral flotation foam working conditions at different times during the production status according to the classification results.

Images