CN113128126A - Modeling method of flotation dosing process based on generation of countermeasure network - Google Patents
Modeling method of flotation dosing process based on generation of countermeasure network Download PDFInfo
- Publication number
- CN113128126A CN113128126A CN202110455031.6A CN202110455031A CN113128126A CN 113128126 A CN113128126 A CN 113128126A CN 202110455031 A CN202110455031 A CN 202110455031A CN 113128126 A CN113128126 A CN 113128126A
- Authority
- CN
- China
- Prior art keywords
- image
- flotation
- loss
- foam
- dosing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of flotation dosing, in particular to a modeling method of a flotation dosing process based on a generation countermeasure network, which fully utilizes the strong image generation capacity of the generation countermeasure network, calculates the characteristic difference before and after the foam image changes in the dosing process through an image discrimination network, and simultaneously models the incidence relation between the dosing quantity and the foam image by utilizing the mutual information maximization between the flotation dosing quantity and the foam image after the dosing prediction; in the model training process, the fighting game between the foam image prediction network and the foam image discrimination network is judged by using the dosed foam image, so that the performances of the two networks are improved, and the final prediction network can accurately predict the dosed foam image based on the initial foam image and the addition amount of the flotation reagent before and after dosing adjustment.
Description
Technical Field
The invention belongs to the technical field of flotation dosing, and particularly relates to a modeling method of a flotation dosing process based on a generation countermeasure network.
Background
Mineral resources are important material bases for survival and development of human society, and the improvement of mineral recovery rate and comprehensive utilization rate is obvious in economic benefit. Mineral separation is an important link in mineral resource processing, and froth flotation is a mineral separation mode which is most widely applied.
The adjustment of the flotation dosing amount by the flotation operator is mainly dependent on the visual characteristics of the foam surface, such as foam size and color, which depends to a large extent on the operation experience; the manual medicine adding mode can cause mineral resource waste, increase environmental pollution and harm human health.
In the prior art, the traditional mode of judging the adding amount by depending on the experience of operators is limited to the operation experience, so that the automatic adding of the chemicals in the mineral flotation process has very important significance, but the relationship between the adding amount of the flotation chemicals and the foam image characteristics in the mineral flotation process is complex and the accurate modeling is difficult in the prior art.
Disclosure of Invention
In order to make up for the defects of the prior art and solve the problems that the relation between the addition amount of a flotation reagent and the characteristics of a froth image in the mineral flotation process is complex and accurate modeling is difficult, the invention provides a modeling method of a flotation dosing process based on generation of an antagonistic network.
The technical scheme adopted by the invention for solving the technical problems is as follows: a modeling method of a flotation dosing process based on a generation countermeasure network comprises the following steps:
s1: acquiring foam images before and after the addition amount of the flotation reagent is adjusted, and constructing a flotation dosing process data set by combining the adjustment record of the addition amount of the flotation reagent;
s2: establishing a flotation dosing network model based on a generated countermeasure network;
s3: inputting the flotation dosing process data into a flotation dosing network model, and training to obtain a parameter model of a flotation dosing network;
s4: inputting the initial foam image and the additive amount of the medicament before and after the medicament addition adjustment into a trained flotation medicament adding model, and predicting to obtain a foam image after medicament addition.
Preferably, in S1, the flotation dosing process data set is composed of flotation dosing condition records, and each flotation dosing condition record specifically includes dosing amounts before and after adjustment of the addition amount of the chemical and a corresponding foam image; the chemical addition amount is adjusted at time t, the foam image at the initial time of adjusting the chemical addition amount collected by the flotation monitoring system at this time is recorded as x (t) (for simplicity, referred to as initial foam image), and the chemical addition amounts before and after the adjustment of the chemical addition operation are respectively recorded as u (t)-) And u (t)+) (ii) a The time lag exists due to the action of the flotation reagent, and the reagent addition amount u (t)+) The resulting post-medicated foam image under the action of (a) is denoted as x (t + τ) (referred to simply as post-medicated foam image), where τ is the delay time for the full onset of the agent; thus, the dosing process can be formalized as a record of the flotation dosing regime for a segment of the flotation dosing processThe flotation dosing process data set records C from a large number of flotation dosing working conditionsiI is 1,2, …, M.
Preferably, in S1, the flotation dosing process data set is composed of flotation dosing condition records, and each flotation dosing condition record specifically includes dosing amounts before and after adjustment of the addition amount of the chemical and a corresponding foam image; adjusting the addition amount of the reagent at the moment t, and adjusting the initial addition amount collected by the flotation monitoring system at the momentThe initial foam image is denoted by x (t) (for simplicity, referred to as the initial foam image), and the amounts of the added drug before and after adjustment of the drug addition operation are denoted by u (t)-) And u (t)+) (ii) a The time lag exists due to the action of the flotation reagent, and the reagent addition amount u (t)+) The resulting post-medicated foam image under the action of (a) is denoted as x (t + τ) (referred to simply as post-medicated foam image), where τ is the delay time for the full onset of the agent; thus, the dosing process can be formalized as a record of the flotation dosing regime for a segment of the flotation dosing processThe flotation dosing process data set records C from a large number of flotation dosing working conditionsiI is 1,2, …, M.
Preferably, in S2, the network model of the flotation dosing process is composed of a dosed foam image prediction network and a foam image discrimination network;
the foam image prediction network after dosing is of an Encoder-Decoder structure, namely divided into two parts of coding and decoding; in the encoding part, continuous 6 convolutions of 4 multiplied by 4 are adopted to carry out feature extraction on the input initial foam image, and Batch Normalization (BN) and LeakyReLU activation functions are adopted after convolution; for the input flotation dosage, the flotation dosage is converted into a matrix with (dosage) information through pretreatment, and then the matrix is fused with a characteristic diagram output by the coding part and then input into the decoding part; the decoding part adopts continuous 6 deconvolution of 4 multiplied by 4 to reduce the characteristic diagram of the embedded medicament amount information into a foam image after medicine adding with the same size as the input foam image, a BN layer and a ReLU activation function are adopted after a deconvolution layer, the last layer of deconvolution is still combined with a Tanh activation function, and image data is output;
the foam image discrimination network consists of two different full convolution neural networks and shares weight parameters of 6 convolution layers, and example Normalization (IN) and LeakyReLU activation functions are adopted after the convolution layers; the two full convolution neural networks are respectively used for acquiring the characteristic distribution of the image and maximally predicting the mutual information between the foam image and the (dosage); in the process of obtaining the image characteristic distribution, firstly, 6 convolutions of 4 multiplied by 4 are adopted to respectively carry out characteristic extraction on the real dosed foam image and the forecast dosed foam image, and then, the convolution with the step length of 1 and the kernel of 4 multiplied by 4 is used to output the characteristic distribution of the image and is used for measuring the distance between the characteristic distributions of the two foam images; in the process of maximizing mutual information, two full convolution neural networks share most convolution layer weights, so that the calculation amount of repeated feature extraction is reduced, the predicted foam image features extracted by the previous convolution layer are directly subjected to down-sampling by using 4 x 4 convolution, and then the mean value and the variance meeting Gaussian distribution are respectively output through two different 1 x 1 convolutions.
Preferably, in S2, the loss function of the dosed foam image prediction network consists of confrontation loss, reconstruction loss, content loss, and mutual information loss, and the loss function of the foam image discrimination network consists of confrontation loss and gradient penalty loss;
the total loss of the foam image prediction network after dosing is specifically expressed as:
wherein λ isg-adv,λrec,λcontent,λinfoWeight coefficients corresponding to the countermeasures loss, reconstruction loss, content loss and mutual information loss, respectivelyLoss of reconstructionContent lossAnd mutual information lossRespectively expressed as:
the overall loss of the image discrimination network is:
wherein λ isd-adv,λgpWeight coefficients corresponding to the penalty loss and the penalty loss in gradient, respectively, to combat the lossAnd gradient penalty lossRespectively expressed as:
wherein the content of the first and second substances,sampling is performed for random interpolation between the foam image after adding the medicine and the prediction foam image.
Preferably, in S3, the training and updating of the flotation dosing process model based on generation of the antagonistic network is divided into three steps: firstly, fusing a high-dimensional characteristic image obtained by an initial foam image through a coding part of an image prediction network and a high-dimensional matrix with dosing amount information, then obtaining a predicted foam image through deconvolution operation of a decoding part, and calculating the countermeasure loss and reconstruction loss of the predicted foam image and the dosed foam image; secondly, inputting the predicted foam image and the medicated foam image into a coding part of an image prediction network again respectively, and obtaining content loss by calculating characteristic differences among the characteristic images after each convolution layer; thirdly, inputting the dosed foam image and the predicted foam image into an image discrimination network respectively to extract features, calculating the countermeasure loss and the gradient penalty loss after obtaining the image feature distribution, outputting the mean value and the variance related to the predicted image feature distribution in the process of maximizing mutual information, and finally calculating the mutual information loss between the foam image and the flotation dosing amount; the image prediction network performs back propagation and updates the network weight through the sum of the countermeasure loss, the reconstruction loss, the content loss and the mutual information loss, and the image discrimination network performs back propagation and updates the network weight through the sum of the countermeasure loss and the gradient penalty loss.
Preferably, in S4, a froth flotation monitoring system is used to collect a current froth image and a corresponding dosing amount, a planned adjustment dosing amount is given, and a predicted post-dosing froth image is obtained using a froth image prediction network; whether the predicted foam image after the medicine is added is an ideal foam image (flotation working condition) is further judged, and a basis can be provided for realizing optimal control of the medicine adding amount of the foam flotation based on machine vision.
The invention has the technical effects and advantages that:
the modeling method of the flotation dosing process based on the generation countermeasure network provided by the invention fully utilizes the strong image generation capacity of the generation countermeasure network, calculates the characteristic difference before and after the foam image changes in the dosing process through the image discrimination network, and simultaneously models the incidence relation between the dosing amount and the prediction image by utilizing the mutual information maximization between the flotation dosing amount and the prediction dosed foam image; in the model training process, the fighting game between the dosed foam image prediction network and the foam image prediction network is utilized to realize the improvement of the performances of the dosed foam image prediction network and the dosed foam image prediction network, and the final prediction network can realize the accurate prediction of the dosed foam image based on the initial foam image and the addition amount of the flotation reagent before and after dosing adjustment; the experimental verification and performance analysis are carried out by utilizing flotation dosing data and foam images acquired in an industrial field, and the results show that the flotation dosing process modeling method can effectively construct the relationship between the dosing amount and the foam images, and the characteristics of the dosed foam images obtained by models in different working conditions are consistent with those of the actual dosed foam images, so that the problems that the relationship between the addition amount of a flotation reagent and the characteristics of the foam images in the mineral flotation process is complex and the accurate modeling is difficult can be solved, the simulation of the continuous dosing amount regulation process in the flotation production field is realized, and the most key precondition is provided for the optimized control of the flotation dosing amount based on machine vision.
Drawings
The invention will be further explained with reference to the drawings.
FIG. 1 is a flow chart of a modeling method of the present invention;
FIG. 2 is a schematic diagram of an image prediction network in the present invention;
FIG. 3 is a schematic diagram of an image discrimination network in accordance with the present invention;
FIG. 4 is a schematic representation of a flotation dosing model of the present invention;
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further described with the specific embodiments.
Referring to fig. 1-4, the modeling method of the flotation dosing process based on the generation of the countermeasure network according to the present invention includes the following steps:
s1: acquiring foam images before and after the addition amount of the flotation reagent is adjusted, and constructing a flotation dosing process data set by combining the adjustment record of the addition amount of the flotation reagent;
s2: establishing a flotation dosing network model based on a generated countermeasure network;
s3: inputting the flotation dosing process data into a flotation dosing network model, and training to obtain a parameter model of a flotation dosing network;
s4: inputting the initial foam image and the additive amount of the medicament before and after the medicament addition adjustment into a trained flotation medicament adding model, and predicting to obtain a foam image after medicament addition.
Preferably, in S1, the flotation dosing process data set is composed of flotation dosing condition records, and each flotation dosing condition record specifically includes dosing amounts before and after adjustment of the addition amount of the chemical and a corresponding foam image; the chemical addition amount is adjusted at time t, the foam image at the initial time of adjusting the chemical addition amount collected by the flotation monitoring system at this time is recorded as x (t) (for simplicity, referred to as initial foam image), and the chemical addition amounts before and after the adjustment of the chemical addition operation are respectively recorded as u (t)-) And u (t)+) (ii) a The time lag exists due to the action of the flotation reagent, and the reagent addition amount u (t)+) The resulting post-medicated foam image under the action of (a) is denoted as x (t + τ) (referred to simply as post-medicated foam image), where τ is the delay time for the full onset of the agent; thus, the dosing process can be formalized as a record of the flotation dosing regime for a segment of the flotation dosing processThe flotation dosing process data set records C from a large number of flotation dosing working conditionsiI is 1,2, …, M.
Preferably, in S1, the flotation dosing process data set is composed of flotation dosing condition records, and each flotation dosing condition record specifically includes dosing amounts before and after adjustment of the addition amount of the chemical and a corresponding foam image; the chemical addition amount is adjusted at time t, the foam image at the initial time of adjusting the chemical addition amount collected by the flotation monitoring system at this time is recorded as x (t) (for simplicity, referred to as initial foam image), and the chemical addition amounts before and after the adjustment of the chemical addition operation are respectively recorded as u (t)-) And u (t)+) (ii) a The time lag exists due to the action of the flotation reagent, and the reagent addition amount u (t)+) Regulation of the resulting dosing under the action ofThe post-foam image is denoted x (t + τ) (referred to simply as post-medicated foam image), where τ is the delay time for adequate onset of the agent; thus, the dosing process can be formalized as a record of the flotation dosing regime for a segment of the flotation dosing processThe flotation dosing process data set records C from a large number of flotation dosing working conditionsiI is 1,2, …, M.
Preferably, in S2, the network model of the flotation dosing process is composed of a dosed foam image prediction network and a foam image discrimination network;
the foam image prediction network after dosing is of an Encoder-Decoder structure, namely divided into two parts of coding and decoding; in the encoding part, continuous 6 convolutions of 4 multiplied by 4 are adopted to carry out feature extraction on the input initial foam image, and Batch Normalization (BN) and LeakyReLU activation functions are adopted after convolution; for the input flotation dosage, the flotation dosage is converted into a matrix with (dosage) information through pretreatment, and then the matrix is fused with a characteristic diagram output by the coding part and then input into the decoding part; the decoding part adopts continuous 6 deconvolution of 4 multiplied by 4 to reduce the characteristic diagram of the embedded medicament amount information into a foam image after medicine adding with the same size as the input foam image, a BN layer and a ReLU activation function are adopted after a deconvolution layer, the last layer of deconvolution is still combined with a Tanh activation function, and image data is output;
the foam image discrimination network consists of two different full convolution neural networks and shares weight parameters of 6 convolution layers, and example Normalization (IN) and LeakyReLU activation functions are adopted after the convolution layers; the two full convolution neural networks are respectively used for acquiring the characteristic distribution of the image and maximally predicting the mutual information between the foam image and the (dosage); in the process of obtaining the image characteristic distribution, firstly, 6 convolutions of 4 multiplied by 4 are adopted to respectively carry out characteristic extraction on the real dosed foam image and the forecast dosed foam image, and then, the convolution with the step length of 1 and the kernel of 4 multiplied by 4 is used to output the characteristic distribution of the image and is used for measuring the distance between the characteristic distributions of the two foam images; in the process of maximizing mutual information, two full convolution neural networks share most convolution layer weights, so that the calculation amount of repeated feature extraction is reduced, the predicted foam image features extracted by the previous convolution layer are directly subjected to down-sampling by using 4 x 4 convolution, and then the mean value and the variance meeting Gaussian distribution are respectively output through two different 1 x 1 convolutions.
Preferably, in S2, the loss function of the dosed foam image prediction network consists of confrontation loss, reconstruction loss, content loss, and mutual information loss, and the loss function of the foam image discrimination network consists of confrontation loss and gradient penalty loss;
the total loss of the foam image prediction network after dosing is specifically expressed as:
wherein λ isg-adv,λrec,λcontent,λinfoWeight coefficients corresponding to the countermeasures loss, reconstruction loss, content loss and mutual information loss, respectivelyLoss of reconstructionContent lossAnd mutual information lossRespectively expressed as:
the overall loss of the image discrimination network is:
wherein λ isd-adv,λgpWeight coefficients corresponding to the penalty loss and the penalty loss in gradient, respectively, to combat the lossAnd gradient penalty lossRespectively expressed as:
wherein the content of the first and second substances,sampling is performed for random interpolation between the foam image after adding the medicine and the prediction foam image.
Preferably, in S3, the training and updating of the flotation dosing process model based on generation of the antagonistic network is divided into three steps: firstly, fusing a high-dimensional characteristic image obtained by an initial foam image through a coding part of an image prediction network and a high-dimensional matrix with dosing amount information, then obtaining a predicted foam image through deconvolution operation of a decoding part, and calculating the countermeasure loss and reconstruction loss of the predicted foam image and the dosed foam image; secondly, inputting the predicted foam image and the medicated foam image into a coding part of an image prediction network again respectively, and obtaining content loss by calculating characteristic differences among the characteristic images after each convolution layer; thirdly, inputting the dosed foam image and the predicted foam image into an image discrimination network respectively to extract features, calculating the countermeasure loss and the gradient penalty loss after obtaining the image feature distribution, outputting the mean value and the variance related to the predicted image feature distribution in the process of maximizing mutual information, and finally calculating the mutual information loss between the foam image and the flotation dosing amount; the image prediction network performs back propagation and updates the network weight through the sum of the countermeasure loss, the reconstruction loss, the content loss and the mutual information loss, and the image discrimination network performs back propagation and updates the network weight through the sum of the countermeasure loss and the gradient penalty loss.
As an embodiment of the present invention, in S4, a froth flotation monitoring system is used to collect a current froth image and a corresponding dosing amount, a planned adjustment dosing amount is given, and a predicted post-dosing froth image is obtained by using a froth image prediction network; whether the predicted foam image after the medicine is added is an ideal foam image (flotation working condition) is further judged, and a basis can be provided for realizing optimal control of the medicine adding amount of the foam flotation based on machine vision.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (6)
1. A modeling method of a flotation dosing process based on a generation countermeasure network is characterized by comprising the following steps: the modeling method comprises the following steps:
s1: acquiring foam images before and after the addition amount of the flotation reagent is adjusted, and constructing a flotation dosing process data set by combining the adjustment record of the addition amount of the flotation reagent;
s2: establishing a flotation dosing network model based on a generated countermeasure network;
s3: inputting the flotation dosing process data into a flotation dosing network model, and training to obtain a parameter model of a flotation dosing network;
s4: inputting the initial foam image and the additive amount of the medicament before and after the medicament addition adjustment into a trained flotation medicament adding model, and predicting to obtain a foam image after medicament addition.
2. The modeling method of the flotation dosing process based on generation of the countermeasure network according to claim 1, characterized in that: in the S1, the flotation dosing process data set consists of flotation dosing working condition records, and each flotation dosing working condition record specifically comprises dosing amount before and after the addition amount of the medicament is adjusted and a corresponding foam image; the chemical addition amount is adjusted at time t, the foam image at the initial time of adjusting the chemical addition amount collected by the flotation monitoring system at this time is recorded as x (t) (for simplicity, referred to as initial foam image), and the chemical addition amounts before and after the adjustment of the chemical addition operation are respectively recorded as u (t)-) And u (t)+) (ii) a The time lag exists due to the action of the flotation reagent, and the reagent addition amount u (t)+) The resulting post-medicated foam image under the action of (a) is denoted as x (t + τ) (referred to simply as post-medicated foam image), where τ is the delay time for the full onset of the agent; thus, the dosing process can be formalized as a record of the flotation dosing regime for a segment of the flotation dosing processThe flotation dosing process data set records C from a large number of flotation dosing working conditionsiI is 1,2, …, M.
3. The modeling method of the flotation dosing process based on generation of the countermeasure network according to claim 2, characterized in that: in the S2, the flotation dosing process network model is composed of a dosed foam image prediction network and a foam image discrimination network;
the foam image prediction network after dosing is of an Encoder-Decoder structure, namely divided into two parts of coding and decoding; in the encoding part, continuous 6 convolutions of 4 multiplied by 4 are adopted to carry out feature extraction on the input initial foam image, and Batch Normalization (BN) and LeakyReLU activation functions are adopted after convolution; for the input flotation dosage, the flotation dosage is converted into a matrix with (dosage) information through pretreatment, and then the matrix is fused with a characteristic diagram output by the coding part and then input into the decoding part; the decoding part adopts continuous 6 deconvolution of 4 multiplied by 4 to reduce the characteristic diagram of the embedded medicament amount information into a foam image after medicine adding with the same size as the input foam image, a BN layer and a ReLU activation function are adopted after a deconvolution layer, the last layer of deconvolution is still combined with a Tanh activation function, and image data is output;
the foam image discrimination network consists of two different full convolution neural networks and shares weight parameters of 6 convolution layers, and example Normalization (IN) and LeakyReLU activation functions are adopted after the convolution layers; the two full convolution neural networks are respectively used for acquiring the characteristic distribution of the image and maximally predicting the mutual information between the foam image and the (dosage); in the process of obtaining the image characteristic distribution, firstly, 6 convolutions of 4 multiplied by 4 are adopted to respectively carry out characteristic extraction on the real dosed foam image and the forecast dosed foam image, and then, the convolution with the step length of 1 and the kernel of 4 multiplied by 4 is used to output the characteristic distribution of the image and is used for measuring the distance between the characteristic distributions of the two foam images; in the process of maximizing mutual information, two full convolution neural networks share most convolution layer weights, so that the calculation amount of repeated feature extraction is reduced, the predicted foam image features extracted by the previous convolution layer are directly subjected to down-sampling by using 4 x 4 convolution, and then the mean value and the variance meeting Gaussian distribution are respectively output through two different 1 x 1 convolutions.
4. The modeling method of the flotation dosing process based on generation of the countermeasure network according to claim 3, characterized in that: in the step S2, the loss function of the foam image prediction network after dosing consists of confrontation loss, reconstruction loss, content loss and mutual information loss, and the loss function of the foam image discrimination network consists of confrontation loss and gradient penalty loss;
the total loss of the foam image prediction network after dosing is specifically expressed as:
wherein λ isg-adv,λrec,λcontent,λinfoWeight coefficients corresponding to the countermeasures loss, reconstruction loss, content loss and mutual information loss, respectivelyLoss of reconstructionContent lossAnd mutual information lossRespectively expressed as:
the overall loss of the image discrimination network is:
wherein λ isd-adv,λgpWeight coefficients corresponding to the penalty loss and the penalty loss in gradient, respectively, to combat the lossAnd gradient penalty lossRespectively expressed as:
5. The modeling method of the flotation dosing process based on generation of the countermeasure network according to claim 4, characterized in that: in the step S3, the training and updating of the flotation dosing process model based on the generation of the countermeasure network are divided into three steps: firstly, fusing a high-dimensional characteristic image obtained by an initial foam image through a coding part of an image prediction network and a high-dimensional matrix with dosing amount information, then obtaining a predicted foam image through deconvolution operation of a decoding part, and calculating the countermeasure loss and reconstruction loss of the predicted foam image and the dosed foam image; secondly, inputting the predicted foam image and the medicated foam image into a coding part of an image prediction network again respectively, and obtaining content loss by calculating characteristic differences among the characteristic images after each convolution layer; thirdly, inputting the dosed foam image and the predicted foam image into an image discrimination network respectively to extract features, calculating the countermeasure loss and the gradient penalty loss after obtaining the image feature distribution, outputting the mean value and the variance related to the predicted image feature distribution in the process of maximizing mutual information, and finally calculating the mutual information loss between the foam image and the flotation dosing amount; the image prediction network performs back propagation and updates the network weight through the sum of the countermeasure loss, the reconstruction loss, the content loss and the mutual information loss, and the image discrimination network performs back propagation and updates the network weight through the sum of the countermeasure loss and the gradient penalty loss.
6. The modeling method of the flotation dosing process based on generation of the countermeasure network according to claim 5, characterized in that: in the step S4, a froth flotation monitoring system is used for collecting a current froth image and a corresponding medicine adding amount, a planned adjustment medicine adding amount is given, and a predicted post-medicine adding froth image is obtained by using a froth image prediction network; whether the predicted foam image after the medicine is added is an ideal foam image (flotation working condition) is further judged, and a basis can be provided for realizing optimal control of the medicine adding amount of the foam flotation based on machine vision.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110455031.6A CN113128126B (en) | 2021-04-26 | 2021-04-26 | Modeling method of flotation dosing process based on generation of countermeasure network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110455031.6A CN113128126B (en) | 2021-04-26 | 2021-04-26 | Modeling method of flotation dosing process based on generation of countermeasure network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113128126A true CN113128126A (en) | 2021-07-16 |
CN113128126B CN113128126B (en) | 2022-06-10 |
Family
ID=76780030
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110455031.6A Active CN113128126B (en) | 2021-04-26 | 2021-04-26 | Modeling method of flotation dosing process based on generation of countermeasure network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113128126B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114832952A (en) * | 2022-05-09 | 2022-08-02 | 矿冶科技集团有限公司 | Intelligent chemical adding control method for flotation |
CN114897679A (en) * | 2022-04-12 | 2022-08-12 | 东北大学 | Floating foam lossy compression image restoration method based on generating type countermeasure network |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2008000036A1 (en) * | 2006-06-30 | 2008-01-03 | Newcastle Innovation Limited | Device and method for detecting the frothing ability of a fluid |
CN104657949A (en) * | 2015-02-12 | 2015-05-27 | 太原理工大学 | Method for optimizing structural elements during denoising of coal slime flotation froth image |
CN109671124A (en) * | 2018-12-27 | 2019-04-23 | 湖南师范大学 | A kind of mineral flotation foam image color calibration method and foam color detection system |
CN110288260A (en) * | 2019-07-02 | 2019-09-27 | 太原理工大学 | Coal slime flotation additive amount of medicament evaluation method based on semi-supervised clustering |
CN111598298A (en) * | 2020-04-16 | 2020-08-28 | 天津德通电气股份有限公司 | Method for predicting dosage of flotation reagent in coal preparation plant |
CN111709942A (en) * | 2020-06-29 | 2020-09-25 | 中南大学 | Zinc flotation dosing amount prediction control method based on texture degree optimization |
CN111753912A (en) * | 2020-06-28 | 2020-10-09 | 中国矿业大学 | Coal slime flotation clean coal ash content prediction method based on deep learning |
-
2021
- 2021-04-26 CN CN202110455031.6A patent/CN113128126B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2008000036A1 (en) * | 2006-06-30 | 2008-01-03 | Newcastle Innovation Limited | Device and method for detecting the frothing ability of a fluid |
CN104657949A (en) * | 2015-02-12 | 2015-05-27 | 太原理工大学 | Method for optimizing structural elements during denoising of coal slime flotation froth image |
CN109671124A (en) * | 2018-12-27 | 2019-04-23 | 湖南师范大学 | A kind of mineral flotation foam image color calibration method and foam color detection system |
CN110288260A (en) * | 2019-07-02 | 2019-09-27 | 太原理工大学 | Coal slime flotation additive amount of medicament evaluation method based on semi-supervised clustering |
CN111598298A (en) * | 2020-04-16 | 2020-08-28 | 天津德通电气股份有限公司 | Method for predicting dosage of flotation reagent in coal preparation plant |
CN111753912A (en) * | 2020-06-28 | 2020-10-09 | 中国矿业大学 | Coal slime flotation clean coal ash content prediction method based on deep learning |
CN111709942A (en) * | 2020-06-29 | 2020-09-25 | 中南大学 | Zinc flotation dosing amount prediction control method based on texture degree optimization |
Non-Patent Citations (3)
Title |
---|
LIU J 等: "Online monitoring of flotation froth bubble-size distributions via multiscale deblurring and multistage jumping feature-fused full convolutional networks", 《IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT》 * |
吴刚 等: "基于深度学习的浮选加药控制系统", 《山东煤炭科技》 * |
郭西进: "基于GA-BP神经网络的浮选加药量预测", 《煤炭工程》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114897679A (en) * | 2022-04-12 | 2022-08-12 | 东北大学 | Floating foam lossy compression image restoration method based on generating type countermeasure network |
CN114832952A (en) * | 2022-05-09 | 2022-08-02 | 矿冶科技集团有限公司 | Intelligent chemical adding control method for flotation |
CN114832952B (en) * | 2022-05-09 | 2023-10-03 | 矿冶科技集团有限公司 | Intelligent dosing control method for flotation |
Also Published As
Publication number | Publication date |
---|---|
CN113128126B (en) | 2022-06-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109492822B (en) | Air pollutant concentration time-space domain correlation prediction method | |
CN113128126B (en) | Modeling method of flotation dosing process based on generation of countermeasure network | |
CN108764072B (en) | Blood cell subtype image classification method based on multi-scale fusion | |
CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
CN108510280B (en) | Financial fraud behavior prediction method based on mobile equipment behavior data | |
CN109165664A (en) | A kind of attribute missing data collection completion and prediction technique based on generation confrontation network | |
CN107133436A (en) | A kind of multiple sample model training method and device | |
CN111126658A (en) | Coal mine gas prediction method based on deep learning | |
CN104091206A (en) | Social network information transmission prediction method based on evolutionary game theory | |
CN109218134B (en) | Test case generation system based on neural style migration | |
CN111523706B (en) | Section lane-level short-term traffic flow prediction method based on deep learning combination model | |
CN104408108A (en) | Hot topic group influence analysis system and method based on grey system theory | |
CN111105097B (en) | Dam deformation prediction system and method based on convolutional neural network | |
CN106407477A (en) | Multidimensional interconnection recommendation method and system | |
CN112307536B (en) | Dam seepage parameter inversion method | |
CN109567313B (en) | Intelligent insole with biological characteristic recognition function | |
CN112910711A (en) | Wireless service flow prediction method, device and medium based on self-attention convolutional network | |
CN114330101A (en) | Slope earthquake slip prediction method and system based on artificial neural network | |
CN115544873B (en) | Training efficiency and personalized effect quantitative evaluation method for personalized federal learning | |
CN110175247A (en) | A method of abnormality detection model of the optimization based on deep learning | |
CN110942182A (en) | Method for establishing typhoon prediction model based on support vector regression | |
CN108615360B (en) | Traffic demand day-to-day evolution prediction method based on neural network | |
CN114584406B (en) | Industrial big data privacy protection system and method for federated learning | |
CN114154672A (en) | Data mining method for customer churn prediction | |
CN111639882A (en) | Deep learning-based power utilization risk judgment method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |