CN111739108A - Iron ore powder grade rapid estimation method based on convolutional neural network - Google Patents
Iron ore powder grade rapid estimation method based on convolutional neural network Download PDFInfo
- Publication number
- CN111739108A CN111739108A CN202010533881.9A CN202010533881A CN111739108A CN 111739108 A CN111739108 A CN 111739108A CN 202010533881 A CN202010533881 A CN 202010533881A CN 111739108 A CN111739108 A CN 111739108A
- Authority
- CN
- China
- Prior art keywords
- grade
- training
- estimation
- interval
- model
- 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
- 239000000843 powder Substances 0.000 title claims abstract description 52
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 22
- 229910052742 iron Inorganic materials 0.000 title claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 101
- 238000012795 verification Methods 0.000 claims abstract description 14
- 230000003321 amplification Effects 0.000 claims abstract description 4
- 238000004364 calculation method Methods 0.000 claims abstract description 4
- 238000005520 cutting process Methods 0.000 claims abstract description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 4
- 238000012360 testing method Methods 0.000 claims description 27
- 229910052500 inorganic mineral Inorganic materials 0.000 claims description 13
- 239000011707 mineral Substances 0.000 claims description 13
- 238000005286 illumination Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000003892 spreading Methods 0.000 claims description 3
- 230000007480 spreading Effects 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000007670 refining Methods 0.000 claims description 2
- 238000003860 storage Methods 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 abstract description 6
- 238000013480 data collection Methods 0.000 abstract 1
- 230000008901 benefit Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000002156 mixing Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 3
- 239000002184 metal Substances 0.000 description 3
- 229910052751 metal Inorganic materials 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 238000004448 titration Methods 0.000 description 3
- 238000013098 chemical test method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005553 drilling Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000009614 chemical analysis method Methods 0.000 description 1
- 239000003153 chemical reaction reagent Substances 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000227 grinding Methods 0.000 description 1
- SZVJSHCCFOBDDC-UHFFFAOYSA-N iron(II,III) oxide Inorganic materials O=[Fe]O[Fe]O[Fe]=O SZVJSHCCFOBDDC-UHFFFAOYSA-N 0.000 description 1
- 230000001678 irradiating effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Mining & Mineral Resources (AREA)
- Marketing (AREA)
- Marine Sciences & Fisheries (AREA)
- Agronomy & Crop Science (AREA)
- Quality & Reliability (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Animal Husbandry (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a convolutional neural network-based iron ore powder grade rapid estimation method, which comprises the following steps of: the method comprises the following steps: data collection and data set production, wherein image information of the iron ore powder of each grade value is collected and is preprocessed, and the second step is as follows: establishing and training a model, performing data amplification on the sampled training and verification samples by using random cutting and rotating methods, inputting the amplified samples into a deep convolutional neural network to perform stage-by-stage segmental model training and whole model training of a whole grade interval, and obtaining a model after stage-by-stage segmental estimation training and a model after whole training estimation training of the whole grade interval; step three: model calling and grade estimation, and step four: and if the estimation of the grade interval is wrong, the midpoint value of the interval is taken as the final grade to be estimated for absolute error calculation, and if the absolute error is less than 1.5%, the estimation is correct.
Description
Technical Field
The invention belongs to the technical field of deep learning application, and particularly relates to a fast estimation method for iron ore powder grade based on a convolutional neural network.
Background
The metal mine enterprises need to drill the ore by the drilling machine, and the mine enterprises need to measure the grade of the drilled ore powder near the drill hole and use the grade of the ore powder to represent the ore grade of the drill hole area, so as to carry out corresponding ore proportioning and other production arrangement.
The ore blending work is a very important link in the actual production of metal mines, and the quality of the ore blending work directly influences the quality of sintered ores, so that the overall benefit of mine enterprises can be influenced. The ore blending work is made by the determined ore grade value, most of mine enterprises determine the ore grade by using the traditional chemical analysis method, namely, the ore powder collected near a drill hole is sent to a special chemical testing department, the secondary grinding is carried out by the chemical testing department, and then the chemical titration method is adopted to determine the ore grade. Although widely used, the chemical titration method has the disadvantages of troublesome sample taking and feeding, complicated measurement steps, expensive measurement reagents, long time consumption, poor timeliness and the like. Scholars at home and abroad propose various algorithms and estimation models aiming at the defects of the chemical titration method. However, the determination of the ore grade by the algorithms and estimation models still has the defects of complex flow, large calculation amount, long time consumption and the like.
Disclosure of Invention
The invention aims to provide a convolutional neural network-based iron ore powder grade rapid estimation method, and aims to automatically extract image chromaticity characteristics of iron ore powder of different grades by using a deep convolutional neural network increment-resnet-v 2, train a deep learning classification model and construct an iron ore powder grade estimation method, so that the rapid measurement of the grade is realized on a drilling site of a stope, an iron mine enterprise is helped to rapidly make a ore blending scheme, and the production efficiency is improved.
The purpose of the invention is realized by the following technical scheme:
the invention discloses a convolutional neural network-based iron ore powder grade rapid estimation method, which is characterized by comprising the following steps of:
the method comprises the following steps: data acquisition and data set generation
Collecting image information of the iron ore powder of each grade value, and preprocessing the image information;
finally, a plurality of mineral powder images of each grade interval are obtained, the collected digital images of different grade intervals are stored in different folders respectively, the folders are named by grade value intervals, the pictures are named as grade intervals and digital numbers, 80% -90% of the digital images of each grade interval are randomly selected as training data, one half of the rest digital images are used as verification data, and the other half of the rest digital images are used as test data; then storing the training data of each grade interval into the same folder as a training data set, storing the verification data of each grade interval into the same folder as a verification data set, and storing the test data of each grade interval into the same folder as a test data set;
step two: model building and training
2.1 under Windows or Linux system, using python language, designing and constructing an inclusion-Resnet-v 2 deep convolution neural network according to the principle of inclusion and Resnet;
2.2, performing data amplification on the sampled training and verification samples by using a random cutting and rotating method, setting a random number interval to be 1-3, and amplifying the image data of each grade interval to be more than 7000 pieces;
2.3 model training:
and (3) inputting the amplified sample into an increment-Resnet-v 2 deep convolutional neural network constructed in the step 2.1 to carry out stage-by-stage segmental model training and whole model training in a whole grade interval, so as to obtain a model after stage-by-stage segmental estimation training and a model after whole training estimation training in the whole grade interval.
Step three: model calling and grade estimation
3.1 drawing a Cartesian coordinate system by taking the central point of the image of the test data set as a coordinate origin, and dividing the image data in the test data set into 4 parts by using the x axis and the y axis of the coordinate system;
3.2, sequentially putting the test images in the test data set into the model after the stage-by-stage segmental estimation training and the model after the full-grade interval integral training estimation training for grade estimation, counting the correct number of the estimation interval, and calculating the ratio of the correct number to the total test number, namely the accuracy and the estimation result of the model;
step four: model evaluation
If the grade interval estimation is wrong, the midpoint value in the interval is taken as the final grade to be estimated for absolute error calculation, and if the absolute error is less than 1.5%, the estimation is correct.
As a further optimization of the present invention, the collecting of the image information of the iron ore powder of each grade value comprises the following steps:
the granularity of S1 is between 0 and 1.5mm, and the ore powder with different grades is separated by taking the fixed grade percentage as a unit,
s2, spreading the ore powder in different grade intervals into a round ore powder pile with the same area and smooth surface,
s3, under a laboratory environment, the LED lamp with controllable illumination intensity is adopted for illumination, the power of the LED lamp is gradually increased from small to large, random high-definition digital image shooting is carried out on different piles of mineral powder when the power is adjusted once, the number of pictures shot by each pile of mineral powder cannot be less than 200 when the illumination intensity is adjusted once, and finally the number of mineral powder images in each grade interval cannot be less than 1000.
As a further optimization of the invention, the inputting of the amplified samples into the inclusion-Resnet-v 2 deep convolutional neural network for model training comprises the following steps:
2.3.1 step-by-step segmental training: because the feature difference of the image is not obvious when the grade interval is small, in order to highlight the feature difference of the image, a step-by-step training mode is adopted for training,
2.3.1.1, firstly dividing a data set into two large regional data sets, inputting the two large regional data sets into a deep convolutional neural network for training, monitoring a loss value through a TensorBoar in the training process, outputting each neuron node parameter in the network once when a loss function value tends to be convergent, and storing the neuron node parameter as a ckpt file, namely an estimation model a after segmented training;
2.3.1.2, dividing two more refined grade intervals from the two grade intervals divided in step 2.3.1, respectively training, outputting each neuron node parameter in the whole network once when the loss function value tends to be convergent, and storing as a ckpt file, namely the estimation model b after the segmentation training;
2.3.1.3, according to the method in the step 2.3.1, refining the grade interval continuously, further performing model training and model storage after training on the refined data continuously, and so on until the grade interval is divided to the minimum, obtaining an estimation model n after segmentation training;
2.3.2 Whole training of all-grade interval:
in order to guarantee the accuracy of grade estimation, the model after the step 2.3.1 is segmented and trained step by step, and then the whole training of the full grade interval is adopted, wherein the step comprises the steps of separating the image data of all grade intervals, putting the image data into a deep convolutional neural network for training together, and outputting classification covering all grade intervals to obtain the estimation model after the whole training of the full grade interval.
As a further optimization of the invention, the estimation models a, b and n obtained in the step 2.3.1 after the step-by-step segmental training are loaded, and the estimation models a, b and n after the step-by-step segmental training are used for estimating the grade until the grade estimation is finished to the minimum interval, so as to obtain four grade interval values and respective probability scores;
loading the estimation model obtained in the step 2.3.2 after the whole training of the whole grade interval, and sequentially estimating the grade of the four parts by using the whole estimation model to obtain four grade interval values and respective probabilities;
and selecting one interval with the most interval values as a final estimation interval from the eight obtained grade interval values and the respective probability scores, respectively calculating the sum of the probability scores if the eight interval values have the same number and the most interval values, and selecting one interval with the highest sum of the probability scores as the final estimation interval so as to finish grade estimation.
Compared with the prior art, the invention has the advantages that:
(1) the invention applies the deep learning technology to the estimation of the grade of the iron ore powder for the first time;
(2) the grade of the ore powder can be rapidly estimated in the field construction period, the grade testing time is saved, and the production planning efficiency is improved;
(3) the method utilizes the different image (chromaticity) characteristics of the ore powder of different grades to construct the grade estimation model, and compared with other grade estimation methods, the method has the advantages that the cost is saved, and the operation is simpler and more convenient;
(4) the method has strong popularization, can be applied to the grade estimation of the iron ore powder, and only needs to replace training data with corresponding ore powder images when the method is applied to the grade estimation of other metal ore powder.
Drawings
Fig. 1 is a training flow chart.
FIG. 2 is a schematic view of test image data segmentation.
Detailed Description
The invention is further illustrated by the following examples in conjunction with the accompanying drawings:
in order to make the objects, technical solutions and advantages of the present invention more clearly apparent, the present invention will be described in further detail with reference to the accompanying drawings and specific examples, as shown in fig. 1 and 2, it is apparent that the described examples are some, but not all, of the examples of the present invention. All other examples, which can be obtained by a person skilled in the art without any inventive step based on the examples of the present invention, are within the scope of the present invention.
In order to realize the purpose of quickly estimating the grade of the iron ore powder, the method comprises the following specific steps:
establishing Tensorflow machine learning development environment
The convolutional neural network transfer learning takes Tensorflow as a deep learning framework, a Windows or Linux system is used as an experimental environment, the memory capacity of a CPU is more than 8GB, and the memory of a video card is more than 6G.
The method comprises the following steps: data acquisition and data set generation
1.1 collecting mineral powder with the magnetite grade value of more than 20 percent and the granularity of 1mm of the Qida mountain iron ore of Angang group mining company, and separating the mineral powder with different grades by taking 1 percent as an interval unit;
1.2, collecting ore powder with the grade between 20 and 21 percent, collecting ore powder with the grade value between 21 and 22 percent, and spreading the divided ore powder in different grade intervals into round ore powder piles with the same area and smooth surfaces by analogy;
1.3, under a laboratory environment, emptying background objects, irradiating by using an LED lamp with controllable illumination intensity (power), firstly adjusting the power of the LED lamp to 15W, then carrying out random high-definition digital image shooting (shooting angle and shooting distance are random) on different piles of mineral powder under the illumination intensity, randomly shooting 200 digital images on each pile of mineral powder under the illumination intensity, then gradually increasing the power of the LED lamp to change the illumination intensity, increasing 5W each time, repeating the random shooting process until the power of the LED lamp is adjusted to 100W, and finally enabling the number of mineral powder images in each grade interval to reach 1000;
1.4, respectively storing the acquired digital images of different grade intervals into different folders, naming the folders by grade value intervals, naming the pictures as grade intervals and numbering the numbers, for example: the ore powder pile digital images with the grade values of 20% -21% are stored in the folder named as '20% -21%', the picture names are 20% -21% -1, and the like;
1.5 randomly selecting 80% of digital images in each grade interval as training data, and using one half of the rest digital images as verification data and the other half as test data;
1.6 storing the training data of each grade interval into the same folder as a training data set, wherein 8000 sheets are used, and storing the verification data of each grade interval into the same folder as a verification data set, wherein 1000 sheets are used; storing the test data of each grade interval into the same folder as a test data set, wherein the total number of the test data set is 1000; and finishing the data set production.
Step two: model building and training
2.1 under the Windows10 system, the inclusion-Resnet-v 2 deep convolutional neural network is designed and constructed by the principle of inclusion and Resnet using python language. Specific computer hardware: a CPU: i7-8700, GPU GTX 108010 GB;
2.2, performing data amplification on the sampled training and verification samples by using a random cutting and rotating method, setting a random number interval to be 3, and amplifying the image data of each grade interval to 7200 images, wherein the training data set comprises 72000 digital images;
2.3 model training
Putting the sampled training and verification samples into the Inceposition-Resnet-v 2 deep convolution neural network constructed as above for training;
2.3.1 step-by-step segmental training:
taking a mineral powder image with the grade of 20-30%, and a training flow chart is shown in figure 1.
As shown in fig. 1, since the overall grade interval is 20% to 30%, it needs to train for 9 times step by step, the number of training steps is set to 50000, when the number of training steps reaches 50000, the neural node parameters, i.e. the training models, are saved, and 9 segmental training models are obtained.
2.3.2 Whole training of all-grade interval:
in order to guarantee the accuracy of grade estimation, the model after the step 2.3.1 is segmentally trained step by step, and then the whole training of the whole grade interval is adopted, wherein the step comprises the steps of separating the image data of all grade intervals, putting the image data into a deep convolutional neural network for training together, outputting the classification including all grade intervals, setting the training steps to be 50000, and obtaining the estimation model 10 after the whole training of the whole grade interval.
Step three: model calling and grade estimation
3.1 the test data set is not trained, a Cartesian coordinate system is drawn by taking the central point of the image as the origin of coordinates, and the image data in the test data set is divided into four parts by using the x and y axes of the coordinate system, which is shown in a schematic diagram 2.
3.2, firstly, sequentially inputting the four parts into a convolutional network, estimating the grade by using a segmented training model (model 1-model 9), and loading the training model stored in training, namely a ckpt file, according to a certain sequence, wherein the specific loading sequence is as follows: and (3) loading the ckpt file of the model 1 when the grade of the test image is between 20% and 30%, inputting the image into the network for grade estimation, if the grade interval is predicted to be between 20% and 25%, transferring the image to the input end, reloading the ckpt file of the model 2 for grade estimation, if the grade interval is predicted to be between 20% and 23%, transferring the image to the input end again, and loading the ckpt file of the model 4 for grade estimation. And repeating the steps until the grade is estimated to the end of the minimum interval, wherein the minimum interval is the grade estimation interval of the input image part.
3.3, estimating the grade intervals of the four parts divided in the step 3.1 by using a mode of a step 3.2, and obtaining four grade interval values and respective probability scores.
And 3.4, respectively inputting the four parts into the whole-region overall training model (model 10) to sequentially predict the grades of the four parts, so as to obtain four grade interval values and respective probability scores.
3.5, determining a final grade estimation interval, obtaining 8 grade interval values and respective probability scores through the step 3.3 and the step 3.4, and selecting an interval with the most interval values as the final estimation interval. If there are the same number of interval values with the maximum number, the probability score sums are calculated respectively, and the one with the highest probability score sum is selected as the final estimation interval.
Step four, model evaluation
According to the third step, 100 images of each grade interval in the test data set are sequentially placed into a segmental training model and an estimation model after full-grade interval training for grade estimation, and the estimation results are as follows:
and sequentially putting the image data of each grade interval in the test data set into the model for grade estimation, wherein when the estimation values with the absolute errors smaller than 1.5 percent are all correct, the estimation accuracy is 100 percent, namely the estimation grade absolute errors of the model are smaller than 1.5 percent, and the model is suitable for mine field application.
Claims (4)
1. A fast estimation method of iron ore powder grade based on a convolution neural network is characterized by comprising the following steps:
the method comprises the following steps: data acquisition and data set generation
Acquiring image information of iron ore powder of each grade value, preprocessing the image information to finally obtain a plurality of ore powder images of each grade interval, respectively storing the acquired digital images of different grade intervals into different folders, naming the folders by grade value intervals, naming the images as grade intervals and number numbers, randomly selecting 80-90% of the digital images of each grade interval as training data, and using one half of the rest digital images as verification data and the other half of the rest digital images as test data; then storing the training data of each grade interval into the same folder as a training data set, storing the verification data of each grade interval into the same folder as a verification data set, and storing the test data of each grade interval into the same folder as a test data set;
step two: model building and training
2.1 under Windows or Linux system, using python language, designing and constructing an inclusion-Resnet-v 2 deep convolution neural network according to the principle of inclusion and Resnet;
2.2, performing data amplification on the sampled training and verification samples by using a random cutting and rotating method, setting a random number interval to be 1-3, and amplifying the image data of each grade interval to be more than 7000 pieces;
2.3 model training:
and (3) inputting the amplified sample into an increment-Resnet-v 2 deep convolutional neural network constructed in the step 2.1 to carry out stage-by-stage segmental model training and whole model training in a whole grade interval, so as to obtain a model after stage-by-stage segmental estimation training and a model after whole training estimation training in the whole grade interval.
Step three: model calling and grade estimation
3.1 drawing a Cartesian coordinate system by taking the central point of the image of the test data set as the origin of coordinates, and dividing the image data in the test data set into four parts by using the x and y axes of the coordinate system;
3.2, sequentially putting the test images in the test data set into the model after the stage-by-stage segmental estimation training and the model after the full-grade interval integral training estimation training for grade estimation, counting the correct number of the estimation interval, and calculating the ratio of the correct number to the total test number, namely the accuracy and the estimation result of the model;
step four: model evaluation
If the grade interval estimation is wrong, the midpoint value in the interval is taken as the final grade to be estimated for absolute error calculation, and if the absolute error is less than 1.5%, the estimation is correct.
2. The convolutional neural network-based method for rapidly estimating the grade of iron ore powder according to claim 1, wherein the step of collecting image information of iron ore powder at each grade value comprises the following steps:
the granularity of S1 is between 0 and 1.5mm, and the ore powder with different grades is separated by taking the fixed grade percentage as a unit,
s2, spreading the ore powder in different grade intervals into a round ore powder pile with the same area and smooth surface,
s3, under a laboratory environment, the LED lamp with controllable illumination intensity is adopted for illumination, the power of the LED lamp is gradually increased from small to large, random high-definition digital image shooting is carried out on different piles of mineral powder when the power is adjusted once, the number of pictures shot by each pile of mineral powder cannot be less than 200 when the illumination intensity is adjusted once, and finally the number of mineral powder images in each grade interval cannot be less than 1000.
3. The convolutional neural network-based method for rapidly estimating the iron ore powder grade according to claim 1, wherein the step of inputting the amplified sample into an inclusion-Resnet-v 2 deep convolutional neural network for model training comprises the following steps:
2.3.1 step-by-step segmental training: because the feature difference of the image is not obvious when the grade interval is small, in order to highlight the feature difference of the image, a step-by-step training mode is adopted for training,
2.3.1.1, firstly dividing a data set into two large regional data sets, inputting the two large regional data sets into a deep convolutional neural network for training, monitoring a loss value through a TensorBoar in the training process, outputting each neuron node parameter in the network once when a loss function value tends to be convergent, and storing the neuron node parameter as a ckpt file, namely an estimation model a after segmented training;
2.3.1.2, dividing two more refined grade intervals from the two grade intervals divided in step 2.3.1, respectively training, outputting each neuron node parameter in the whole network once when the loss function value tends to be convergent, and storing as a ckpt file, namely the estimation model b after the segmentation training;
2.3.1.3, according to the method in the step 2.3.1, refining the grade interval continuously, further performing model training and model storage after training on the refined data continuously, and so on until the grade interval is divided to the minimum, obtaining an estimation model n after segmentation training;
2.3.2 Whole training of all-grade interval:
in order to guarantee the accuracy of grade estimation, the model after the step 2.3.1 is segmented and trained step by step, and then the whole training of the full grade interval is adopted, wherein the step comprises the steps of separating the image data of all grade intervals, putting the image data into a deep convolutional neural network for training together, and outputting classification covering all grade intervals to obtain the estimation model after the whole training of the full grade interval.
4. The convolutional neural network-based iron ore powder grade rapid estimation method according to claim 3, wherein the estimation models a, b and n obtained in the step 2.3.1 are loaded after stage-by-stage segmental training, and the estimation models a, b and n after stage-by-stage segmental training are used for grade estimation until the grade estimation is finished to a minimum interval, so that four grade interval values and respective probability scores are obtained;
loading the estimation model obtained in the step 2.3.2 after the whole training of the whole grade interval, and sequentially estimating the grade of the four parts by using the whole estimation model to obtain four grade interval values and respective probabilities;
and selecting one interval with the most interval values as a final estimation interval from the eight obtained grade interval values and the respective probability scores, respectively calculating the sum of the probability scores if the eight interval values have the same number and the most interval values, and selecting one interval with the highest sum of the probability scores as the final estimation interval so as to finish grade estimation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010533881.9A CN111739108B (en) | 2020-06-12 | 2020-06-12 | Rapid iron ore powder grade estimation method based on convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010533881.9A CN111739108B (en) | 2020-06-12 | 2020-06-12 | Rapid iron ore powder grade estimation method based on convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111739108A true CN111739108A (en) | 2020-10-02 |
CN111739108B CN111739108B (en) | 2023-09-15 |
Family
ID=72648956
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010533881.9A Active CN111739108B (en) | 2020-06-12 | 2020-06-12 | Rapid iron ore powder grade estimation method based on convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111739108B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112916432A (en) * | 2021-03-24 | 2021-06-08 | 江西理工大学 | Intelligent magnetic ore sorting method and equipment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109508650A (en) * | 2018-10-23 | 2019-03-22 | 浙江农林大学 | A kind of wood recognition method based on transfer learning |
CN110516954A (en) * | 2019-08-23 | 2019-11-29 | 昆明理工大学 | One kind referring to calibration method based on GA-BP neural network algorithm optimization mineral processing production |
CN110852395A (en) * | 2019-11-15 | 2020-02-28 | 鞍钢集团矿业有限公司 | Ore granularity detection method and system based on autonomous learning and deep learning |
US20200074270A1 (en) * | 2018-09-05 | 2020-03-05 | International Business Machines Corporation | Estimate ore content based on spatial geological data through 3d convolutional neural networks |
-
2020
- 2020-06-12 CN CN202010533881.9A patent/CN111739108B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200074270A1 (en) * | 2018-09-05 | 2020-03-05 | International Business Machines Corporation | Estimate ore content based on spatial geological data through 3d convolutional neural networks |
CN109508650A (en) * | 2018-10-23 | 2019-03-22 | 浙江农林大学 | A kind of wood recognition method based on transfer learning |
CN110516954A (en) * | 2019-08-23 | 2019-11-29 | 昆明理工大学 | One kind referring to calibration method based on GA-BP neural network algorithm optimization mineral processing production |
CN110852395A (en) * | 2019-11-15 | 2020-02-28 | 鞍钢集团矿业有限公司 | Ore granularity detection method and system based on autonomous learning and deep learning |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112916432A (en) * | 2021-03-24 | 2021-06-08 | 江西理工大学 | Intelligent magnetic ore sorting method and equipment |
Also Published As
Publication number | Publication date |
---|---|
CN111739108B (en) | 2023-09-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112465748B (en) | Crack identification method, device, equipment and storage medium based on neural network | |
CN108416686A (en) | A kind of Eco-Geo-Environment Type division method based on Coal Resource Development | |
CN104318109B (en) | Shale Gas Reservoir Identification Method Based on Support Vector Machine | |
CN108229811B (en) | Method for evaluating water injection effect of fractured-vuggy carbonate reservoir | |
CN109214026A (en) | Shale gas horizontal well initial-stage productivity prediction method | |
CN109872311A (en) | A kind of Rock Mass Integrality sentences knowledge method | |
CN108346144A (en) | Bridge Crack based on computer vision monitoring and recognition methods automatically | |
CN113487533B (en) | Part assembly quality digital detection system and method based on machine learning | |
CN101140216A (en) | Gas-liquid two-phase flow type recognition method based on digital graphic processing technique | |
CN111833310B (en) | Surface defect classification method based on neural network architecture search | |
CN111724358A (en) | Concrete quality detection method and system based on image and convolutional neural network | |
CN111160389A (en) | Lithology identification method based on fusion of VGG | |
CN111079783B (en) | Method for identifying stratum lithology parameters based on multi-core ensemble learning | |
CN111739108A (en) | Iron ore powder grade rapid estimation method based on convolutional neural network | |
CN116309155A (en) | Image restoration method, model and device based on convolution and converter hybrid network | |
CN115655612A (en) | Tunnel blasting charging structure identification recording method, device, equipment and medium | |
CN111626358A (en) | Tunnel surrounding rock grading method based on BIM picture recognition | |
CN108229692B (en) | Machine learning identification method based on dual contrast learning | |
CN113283467A (en) | Weak supervision picture classification method based on average loss and category-by-category selection | |
CN110222981B (en) | Reservoir classification evaluation method based on parameter secondary selection | |
CN117372144A (en) | Wind control strategy intelligent method and system applied to small sample scene | |
CN115081485B (en) | AI-based magnetic flux leakage internal detection data automatic analysis method | |
CN115082713B (en) | Method, system and equipment for extracting target detection frame by introducing space contrast information | |
CN108345943B (en) | Machine learning identification method based on embedded coding and contrast learning | |
CN116206169A (en) | Intelligent gangue target detection 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 |