CN113591689A - RGB (red, green and blue) image recognition method and system for coal and gangue - Google Patents
RGB (red, green and blue) image recognition method and system for coal and gangue Download PDFInfo
- Publication number
- CN113591689A CN113591689A CN202110862162.6A CN202110862162A CN113591689A CN 113591689 A CN113591689 A CN 113591689A CN 202110862162 A CN202110862162 A CN 202110862162A CN 113591689 A CN113591689 A CN 113591689A
- Authority
- CN
- China
- Prior art keywords
- coal
- gangue
- model
- image recognition
- identification
- 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.)
- Pending
Links
- 239000003245 coal Substances 0.000 title claims abstract description 126
- 238000000034 method Methods 0.000 title claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 34
- 238000005457 optimization Methods 0.000 claims abstract description 31
- 210000002569 neuron Anatomy 0.000 claims abstract description 29
- 238000012360 testing method Methods 0.000 claims abstract description 21
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 20
- 238000013508 migration Methods 0.000 claims abstract description 17
- 230000005012 migration Effects 0.000 claims abstract description 17
- 238000004364 calculation method Methods 0.000 claims abstract description 10
- 238000002474 experimental method Methods 0.000 claims abstract description 8
- 238000000926 separation method Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 239000010878 waste rock Substances 0.000 claims 1
- 238000005516 engineering process Methods 0.000 description 7
- 238000011156 evaluation Methods 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 239000010883 coal ash Substances 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 235000002566 Capsicum Nutrition 0.000 description 1
- 239000006002 Pepper Substances 0.000 description 1
- 235000016761 Piper aduncum Nutrition 0.000 description 1
- 235000017804 Piper guineense Nutrition 0.000 description 1
- 244000203593 Piper nigrum Species 0.000 description 1
- 235000008184 Piper nigrum Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 150000003839 salts Chemical class 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 238000005507 spraying Methods 0.000 description 1
- 239000002341 toxic gas Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a system for recognizing RGB (red, green and blue) images of coal and gangue, belonging to the technical field of coal and gangue recognition; s1, building a coal and gangue separation experiment platform, and acquiring coal and gangue images in real time; s2 expanding the coal and gangue data sets and carrying out regular naming; s3 dividing a training set and a test set of the recognition model; s4, designing a coal and gangue identification model optimization method; s5, optimizing a classical convolutional neural network structure, and constructing a coal and gangue identification model; s6 selecting the type of model optimizer and the setting mode of learning rate; s7 comprehensively contrasts and analyzes the performance of each network model, and determines an optimal coal and gangue identification method. The migration weight and simplified neuron model optimization method provided by the invention saves training time and parameter calculation amount, effectively avoids the problem of model overfitting, and simultaneously improves the accuracy and recognition rate of the recognition model.
Description
Technical Field
The invention relates to the technical field of coal and gangue identification, in particular to a method and a system for identifying RGB (red, green and blue) images of coal and gangue.
Background
At present, the popularization of mechanized mining leads the gangue inclusion rate in raw coal to be continuously increased, the gangue with different grain sizes and colors can not only reduce the combustion efficiency and the energy utilization rate of the coal, but also discharge a large amount of smoke dust and toxic gas, harm the health of human bodies and pollute the natural environment, thus being against the green development idea advocated in China. Therefore, sorting out gangue blocks mixed in raw coal is a prerequisite for developing clean coal technology.
X-ray transmission imaging methods are costly and pose a radiation hazard when acquiring target images. Compared with the ray method, the CCD and CMOS industrial cameras with high frame rate have no harm to human body and low cost, and can acquire the surface characteristics of clear targets.
The coal and gangue identification algorithm based on machine learning comprises the following steps: the SVM algorithm is improved, the superficial layer characteristics such as the surface texture, the gray value and the like of the coal and gangue images need to be manually extracted by utilizing image processing and pattern recognition technologies, and the process is complicated. And the extracted shallow layer characteristics cannot completely reflect the difference of the coal and the gangue, and the model identification accuracy is low. Since the image classification competition 2012(ILSVRC12) was won by the Convolutional Neural Network (CNN), the deep Convolutional neural network has been successfully applied to crop detection, unmanned driving, and the like. For the field of coal and Gangue identification, Alfarzaea M S and the like (Alfarzaea M S, NIU Qiang, ZHAO Jianqi, et al, coal/Gangue registration Using connected neural networks and Thermal Images [ J ]. IEEEAccess,2020,8:76780 and 76789) developed a CGR-CNN coal and Gangue identification model based on CNN and trained on two experimental hardware platforms in parallel. SU linking et al (Research on code door gap identification by Using volumetric neural network [ C ]//20182nd IEEEadvanced Information Management, communications, Electronic and evaluation Control reference (IMCEC). IEEE,2018: 810-.
The research methods all obtain better recognition results, but simultaneously show that: if a deep CNN classification model is designed and trained again, a large number of annotated gangue sample images are needed, and the requirement of the model training process on the performance of experimental hardware is high, so that the problem of designing a gangue identification model with high identification rate based on small sample data becomes a problem to be solved.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a method and a system for recognizing RGB (red, green and blue) images of coal and gangue, and solves the problems that in the prior art, small and medium sample data are difficult to construct a deep learning model, and the recognition rate of multi-scale forms and color coal and gangue is low under the actual working condition.
The purpose of the disclosure can be realized by the following technical scheme:
a coal and gangue RGB image recognition method is characterized in that: the method comprises the steps of establishing a migration weight and simplified neuron model optimization method, and establishing a coal and gangue identification model through a network structure of the migration weight and the simplified model.
Further, the method comprises the steps of:
s1, building a coal and gangue separation experiment platform, and acquiring coal and gangue images in real time;
s2 expanding the coal and gangue data sets and carrying out regular naming;
s3 dividing a training set and a test set of the recognition model;
s4, designing a coal and gangue identification model optimization method;
s5, optimizing a classical convolutional neural network structure, and constructing a coal and gangue identification model;
s6 selecting the type of model optimizer and the setting mode of learning rate;
s7 comprehensively contrasts and analyzes the performance of each network model, and determines an optimal coal and gangue identification method.
A coal and gangue RGB image recognition system, the system comprising the following modules:
an image acquisition module: acquiring images of coal and gangue;
an image expansion module: expanding a coal and gangue data set, and carrying out regular naming;
the first data processing module: randomly selecting a training set and a testing set of a data construction model;
the second data processing module: executing a coal and gangue identification model optimization method;
the first optimization calculation module: optimizing a classical convolutional neural network structure, and constructing a coal and gangue identification model;
the second optimization calculation module: selecting a model optimizer type and a learning rate setting mode;
a model selection module: and comprehensively comparing and analyzing the performance of each network model to determine an optimal coal and gangue identification method.
Further, in step S1, the RGB images of the coal and the gangue in the motion state are acquired in real time by the color CMOS industrial camera and stored in the industrial personal computer by the USB.
Further, in step S2, the data set is expanded in batch by image flipping, rotation, and noise adding, and the expanded coal and gangue images are named regularly.
Further, in step S3, 80% of the total data sets of coal and gangue are randomly selected as training sets, the remaining 20% are used as test sets, and there is no cross between the two data sets.
Further, in step S4, the weight parameter is migrated to the constructed coal and gangue identification network by using a method of migrating weights and simplifying neurons, and the parameters are fine-tuned by using the divided training set.
Further, in the step S5, the classical CNN network is improved by using the optimization method in the step S4, the number of neurons in the network full-link layer is manually reduced, and a coal and gangue identification model is constructed.
Further, in the step S6, according to the training sets of the coal and the gangue, the training results of each model are contrastively analyzed, and the parameter optimization mode and the learning rate setting mode in the coal and gangue identification network are optimized respectively, so as to determine the optimal hyper-parameter of each model.
Further, in step S7, an optimal coal and gangue identification model after CNN is improved based on the migration weight & simplified neuron optimization method is selected.
The beneficial effect of this disclosure:
(1) the high-frame-rate industrial camera is adopted to obtain the video or RGB images of the coal and the gangue, so that the equipment is safe, efficient and low in cost, the obtained images can be used without too many preprocessing steps, and a large amount of time cost of technicians is saved.
(2) According to the migration weight and simplified neuron model optimization method provided by the invention, model training time can be saved through weight migration, the number and the calculated amount of neurons can be greatly reduced through a GAP technology, and the spatial information of coal and gangue characteristics is protected.
Drawings
In order to more clearly illustrate the embodiments or technical solutions in the prior art of the present disclosure, the drawings used in the description of the embodiments or prior art will be briefly described below, and it is obvious for those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a flow chart of an RGB image recognition method for coal and gangue according to an embodiment of the invention;
FIG. 2 is an image enhancement method in an embodiment of the invention;
FIG. 3 is a flow chart of a migration weight & reduced neuron optimization method-based identification model construction in an embodiment of the present invention;
FIG. 4 is a graph of training accuracy and loss value of a coal and gangue identification model in an embodiment of the invention;
FIG. 5 is a test accuracy graph before and after the coal gangue identification model is improved in the embodiment of the invention.
Detailed Description
The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are only a part of the embodiments of the present disclosure, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.
Example 1: a coal and gangue RGB image recognition method comprises the following steps:
s1, building a coal and gangue separation experiment platform, and acquiring coal and gangue images in real time;
s2 expanding the coal and gangue data sets and carrying out regular naming;
s3 dividing a training set and a test set of the recognition model;
s4, designing a coal and gangue identification model optimization method;
s5, optimizing a classical convolutional neural network structure, and constructing a coal and gangue identification model;
s6 selecting the type of model optimizer and the setting mode of learning rate;
s7 comprehensively contrasts and analyzes the performance of each network model, and determines an optimal coal and gangue identification method.
Preferably, in step S1, the color CMOS industrial camera acquires RGB images of coal and gangue in motion in real time, stores the RGB images in the industrial personal computer through a USB, and mixes coal ash on the surface of the conveyor belt as environmental interference.
Preferably, in step S2, the data set is expanded in batch by image flipping, rotation, and noise adding, and the expanded coal and gangue images are named regularly.
Preferably, in step S3, 80% of the total data sets of coal and gangue are randomly selected as training sets, the remaining 20% are selected as test sets, and there is no crossover between the two data sets.
Preferably, in step S4, a model optimization method, migration weight & simplified neuron, is proposed, that is, weight parameters trained on the ImageNet data set are directly migrated to the constructed coal and gangue identification network, and then parameters are fine-tuned through the divided training sets. In addition, the Global Average Pooling (GAP) operation is directly performed on the N feature maps acquired by the feature extractor, so that N neuron input network full-link layers are obtained. Assuming that the size of each feature map is a × B, the GAP calculation principle is as follows:
c in formula (1)nRepresents the output result of the nth characteristic diagram passing through GAP, omegai,j (n)Representing the pixel values on the nth signature.
Preferably, in step S5, in order to verify the effectiveness of the migration weight & simplified neuron optimization method, the optimization method in step S4 is used to improve the classical CNN network, and the number of neurons in the network full-link layer is manually reduced to construct a coal and gangue identification model.
Preferably, in step S6, based on the coal and gangue training sets, through comparing and analyzing the training results of each model, optimizing the parameter optimization method and the learning rate setting method in the coal and gangue identification network, so as to determine the optimal hyper-parameter of each model.
Preferably, in step S7, on the coal and gangue test set, the performance of each model under the optimal hyper-parameter is analyzed, and after the training results of the models are comprehensively analyzed, an optimal coal and gangue identification method after CNN is improved based on the migration weight & simplified neuron optimization method is selected.
As shown in fig. 1 to 5, example 2: a coal and gangue RGB image recognition method comprises the following steps:
s1, building a coal and gangue separation experiment platform, and acquiring coal and gangue images in real time;
s2 expanding the coal and gangue data sets and carrying out regular naming;
s3 dividing a training set and a test set of the recognition model;
s4, designing a coal and gangue identification model optimization method;
s5, optimizing a classical convolutional neural network structure, and constructing a coal and gangue identification model;
s6 selecting the type of model optimizer and the setting mode of learning rate;
s7 comprehensively contrasts and analyzes the performance of each network model, and determines an optimal coal and gangue identification method.
In the concrete implementation, the method can accurately identify the coal gangue with various forms and particle sizes, has high generalization capability and strong robustness, and uses actual case simulation to illustrate the effectiveness of the technical scheme of the invention.
S1, simulating an underground actual production environment in a laboratory, building a coal and gangue separation experiment platform, and spraying coal ash on the surface of a conveyer belt to serve as environmental interference. The color CMOS industrial camera is manufactured by Haekwondo, and has model number MV-CA050-11UC, resolution 2448 × 2048 and frame rate 35 fps. The camera is matched with a WL1406-5MP type lens and an optical filter, RGB images of coal and gangue in a motion state are obtained in real time, and the RGB images are stored in an industrial personal computer through a USB.
S2, because of the problems of experiment material and time cost, the number of images of coal and gangue collected in the experiment is limited, so the sample image needs to be expanded, and the diversity of the sample is increased. The data sets are expanded in batch by adopting the image enhancement technology of turning around two central axes of the picture, rotating 180 degrees left and right, adding salt and pepper and Gaussian noise, and regular renaming is carried out on the data sets of the coal and the gangue through python codes.
And S3, 2905 expanded coal and gangue pictures, wherein the number of gangue pictures is 1519, and the number of coal pictures 1386. Firstly, in order to avoid the interference of the unbalanced sample problem and simultaneously eliminate redundant sample images, 1280 coal images and 1280 gangue images are selected respectively. Then, 80% of total data sets of coal and gangue are randomly selected as training sets, the rest 20% of the total data sets are used as testing sets, and the two data sets are independent and have no cross. Finally, the coal and gangue pictures are cut into 224 × 224 × 3 using a preprocessing packet transform () in the pyrtch, and normalization processing are performed on RGB three-channel pixels, wherein a pixel normalization formula is as follows:
in formula (2), x' is ∈ [0,1], and is the normalized pixel value, xmin is 0, xmax is 255,
the RGB three-channel normalized mean u is [0.485, 0.456, 0.406], the variance σ is [0.229, 0.224, 0.225], and the calculation formula is:
in formula (3), x "is the normalized pixel value.
S4, designing a coal and gangue recognition model optimization method, namely, migrating weights and simplified Neurons (Transfer Weight Neurons), namely, directly migrating Weight parameters trained on ImageNet data sets into the constructed coal and gangue recognition network, and then finely tuning the parameters through the divided training sets. In addition, the Global Average Pooling (GAP) operation is directly performed on the N feature maps acquired by the feature extractor, and N neurons are obtained and input to the network full-link layer. Assuming that the size of each feature map is a × B, the GAP calculation principle is as follows:
c in formula (1)nRepresents the output result of the nth characteristic diagram passing through GAP, omegai,j (n)Representing the pixel values on the nth signature.
S5, in order to verify the migration weight and simplify the effectiveness of the neuron optimization method, four classic CNN networks of Alexnet, VGG16, VGG19 and Resnet50 are improved according to the flow shown in the attached drawing 3, and 4 coal and gangue identification models are constructed: the method comprises the following steps of Im _ Alexnet, Im _ VGG16, Im _ VGG19 and Im _ Resnet50, wherein the specific processes of the improved method are as follows:
and (4) migrating the shallow network weights trained on the ImageNet data subset to a coal and gangue identification model by adopting a migration idea.
In order to relieve the problem that the parameters of the full connection layer are more, the overfitting problem of the model is easily caused, and the structure of the network full connection layer is optimized:
im _ Alexnet adds GAP after the convolution and pooling layer of the original Alexnet network, and directly converts the 256-dimensional feature matrix with size 6 × 6 into 256 × 1 column vectors. Thus, the number of neurons connected to the hidden layer is reduced from 9216 to 256; similarly, vggtet (Im _ VGG16, Im _ VGG19) network adopts GAP technology to reduce 25088 neurons originally input to the hidden layer to 512 neurons, and Resnet50 reduces 7 × 7 × 2048 feature maps to 2048 dimensional column vectors by using the GAP operation itself and inputs the vectors to the fully connected layer.
The number of the neurons of two hidden layers in the four models Im _ Alexnet, Im _ VGG16, Im _ VGG19 and Im _ Resnet50 is reduced from 2048 to 1024.
And inputting the image information extracted by the two hidden layers of the comprehensive shallow layer network into a softmax classifier, and outputting an image identification result in a probability manner.
And S6, performing optimization simulation experiments on the optimizer type and the learning rate of each model according to the expanded coal and gangue training samples with different forms. The optimizer types in the four recognition models are respectively set as an adaptive momentum estimation method adam (adaptive momentum estimation) and a random Gradient descent method SGD (stochastic Gradient decision) with driving momentum factors, the learning rate is respectively set to be constant at 0.001 and adaptively adjusted: namely, the learning rate is automatically reduced to 10 percent every 20 epochs, and each model has 4 groups of different parameter settings as shown in table 1;
TABLE 1
The training result curve is shown in figure 4, and after the comparison accuracy and loss curve graphs are analyzed, the hyper-parameters with the best performance of each model are selected, as shown in table 2;
TABLE 2
And S7, testing the performance of each model under the optimal hyper-parameter according to the coal and gangue test set, wherein the test accuracy rate curve chart is shown in the attached figure 5. It can be seen that the improved Im _ Alexnet model has the greatest improvement in test accuracy, and the performance of each model before and after improvement is quantitatively evaluated by combining three evaluation parameters of F1 score, model memory size and training time.
Table 3 shows the model evaluation result, and Im _ Resnet50 reduces the memory occupied and saves the training time on the basis of not reducing the test accuracy, compared with the original model. Therefore, for the coal and gangue secondary classification task, the number of neurons in the complex network is reduced, and the effective feature extraction of coal and gangue is not influenced. Due to the fact that the structure of the test accuracy curve is complex, even if the number of the neurons is reduced, the parameter calculation amount of the Im _ Resnet50 is still large, and the test accuracy curve is difficult to converge quickly.
The GAP technology is added to the improved Im _ Alexnet, Im _ VGG16 and Im _ VGG19 models, the generalization capability of the models to the characteristic space displacement of the coal and gangue images is improved, the recognition accuracy is further improved, and the memory occupied by the models is reduced. However, due to the deep series structure characteristics of the models, the training time is long, although the accuracy on the training set is high, the total parameter calculation amount is still large, and the overfitting condition appears on the test set;
TABLE 3
Therefore, the Im _ Alexnet coal and gangue identification model constructed based on the migration weight and simplified neuron optimization method is improved to the maximum extent in all aspects compared with the original model. The time for training 100 epochs by the model is only 10 minutes and 4 seconds, the memory occupied by the model is only 28115kb, the identification time of a single picture is 2.360 milliseconds, and the identification rate reaches 97.461 percent.
The working principle is as follows:
model training time is saved by using weight migration, the number and the calculated amount of neurons can be greatly reduced by using a GAP (good GAP search) technology, the spatial information of the characteristics of coal and gangue is protected, the problem that a deep learning model is difficult to construct by a small sample data set is solved, and the coal and gangue identification speed and accuracy are further improved.
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The foregoing shows and describes the general principles, essential features, and advantages of the 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.
Claims (10)
1. A coal and gangue RGB image recognition method is characterized in that: the method comprises the steps of establishing a migration weight and simplified neuron model optimization method, and establishing a coal and gangue identification model through a network structure of the migration weight and the simplified model.
2. The RGB image recognition method for coal and gangue as claimed in claim 1, wherein: the method comprises the following steps:
s1, building a coal and gangue separation experiment platform, and acquiring coal and gangue images in real time;
s2 expanding the coal and gangue data sets and carrying out regular naming;
s3 dividing a training set and a test set of the recognition model;
s4, designing a coal and gangue identification model optimization method;
s5, optimizing a classical convolutional neural network structure, and constructing a coal and gangue identification model;
s6 selecting the type of model optimizer and the setting mode of learning rate;
s7 comprehensively contrasts and analyzes the performance of each network model, and determines an optimal coal and gangue identification method.
3. The utility model provides a coal and waste rock RGB image recognition system which characterized in that: the system for performing the method of claim 2, the system comprising the following modules:
an image acquisition module: acquiring images of coal and gangue;
an image expansion module: expanding a coal and gangue data set, and carrying out regular naming;
the first data processing module: randomly selecting a training set and a testing set of a data construction model;
the second data processing module: executing a coal and gangue identification model optimization method;
the first optimization calculation module: optimizing a classical convolutional neural network structure, and constructing a coal and gangue identification model;
the second optimization calculation module: selecting a model optimizer type and a learning rate setting mode;
a model selection module: and comprehensively comparing and analyzing the performance of each network model to determine an optimal coal and gangue identification method.
4. The RGB image recognition method for coal and gangue as claimed in claim 2, wherein: in the step S1, RGB images of the coal and the gangue in the motion state are acquired in real time by the color CMOS industrial camera and stored in the industrial personal computer.
5. The RGB image recognition method for coal and gangue as claimed in claim 2, wherein: in the step S2, the data set is expanded in batch by image flipping, rotation, and noise adding, and the expanded coal and gangue images are named regularly.
6. The RGB image recognition method for coal and gangue as claimed in claim 2, wherein: in the step S3, 80% of the total data sets of coal and gangue are randomly selected as training sets, the remaining 20% are used as test sets, and the two data sets are independent and have no cross.
7. The RGB image recognition method for coal and gangue as claimed in claim 2, wherein: in the step S4, a method of migrating weights and simplifying neurons is adopted to migrate the weight parameters to the constructed coal and gangue identification network, and the parameters are fine-tuned through the divided training set.
8. The RGB image recognition method for coal and gangue as claimed in claim 2, wherein: in the step S5, the classical CNN network is improved by adopting the optimization method in the step S4, the number of neurons in the full connection layer of the network is manually reduced, and a coal and gangue identification model is constructed.
9. The RGB image recognition method for coal and gangue as claimed in claim 2, wherein: in the step S6, based on the coal and gangue training sets, the training results of each model are contrastively analyzed, and the parameter optimization mode and the learning rate setting mode in the coal and gangue identification network are optimized respectively, so as to determine the optimal hyper-parameter of each model.
10. The RGB image recognition method for coal and gangue as claimed in claim 2, wherein: in the step S7, an optimal coal and gangue identification model after CNN is improved based on the migration weight and simplified neuron optimization method is selected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110862162.6A CN113591689A (en) | 2021-07-29 | 2021-07-29 | RGB (red, green and blue) image recognition method and system for coal and gangue |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110862162.6A CN113591689A (en) | 2021-07-29 | 2021-07-29 | RGB (red, green and blue) image recognition method and system for coal and gangue |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113591689A true CN113591689A (en) | 2021-11-02 |
Family
ID=78251601
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110862162.6A Pending CN113591689A (en) | 2021-07-29 | 2021-07-29 | RGB (red, green and blue) image recognition method and system for coal and gangue |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113591689A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114120093A (en) * | 2021-12-01 | 2022-03-01 | 安徽理工大学 | Coal gangue target detection method based on improved YOLOv5 algorithm |
| CN114494696A (en) * | 2022-01-26 | 2022-05-13 | 安徽理工大学 | Method, system and device for rapidly detecting multi-scale coal gangue image |
| CN114882281A (en) * | 2022-05-16 | 2022-08-09 | 安徽理工大学 | Lightweight intelligent separation model, method, equipment and storage medium for coal and gangue |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596138A (en) * | 2018-05-03 | 2018-09-28 | 南京大学 | A kind of face identification method based on migration hierarchical network |
| AU2019100352A4 (en) * | 2019-04-04 | 2019-05-09 | Duan, Zicheng Mr | A Flower-Image Search System Based on Deep Learning |
| CN110135468A (en) * | 2019-04-24 | 2019-08-16 | 中国矿业大学(北京) | A kind of recognition methods of gangue |
| CN110245635A (en) * | 2019-06-20 | 2019-09-17 | 安徽理工大学 | A kind of Infrared image recognition of coal and spoil |
| CN111681210A (en) * | 2020-05-16 | 2020-09-18 | 浙江德尚韵兴医疗科技有限公司 | Method for identifying benign and malignant breast nodules by shear wave elastogram based on deep learning |
| CN111860689A (en) * | 2020-07-31 | 2020-10-30 | 中国矿业大学 | Coal and gangue identification method based on phase consistency and light-weight convolutional neural network |
| CN112006678A (en) * | 2020-09-10 | 2020-12-01 | 齐鲁工业大学 | Electrocardiogram abnormity identification method and system based on combination of AlexNet and transfer learning |
| CN112016461A (en) * | 2020-08-28 | 2020-12-01 | 深圳市信义科技有限公司 | Multi-target behavior identification method and system |
| CN112560941A (en) * | 2020-12-14 | 2021-03-26 | 中科院计算所西部高等技术研究院 | Coal and gangue identification method based on image detection |
-
2021
- 2021-07-29 CN CN202110862162.6A patent/CN113591689A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596138A (en) * | 2018-05-03 | 2018-09-28 | 南京大学 | A kind of face identification method based on migration hierarchical network |
| AU2019100352A4 (en) * | 2019-04-04 | 2019-05-09 | Duan, Zicheng Mr | A Flower-Image Search System Based on Deep Learning |
| CN110135468A (en) * | 2019-04-24 | 2019-08-16 | 中国矿业大学(北京) | A kind of recognition methods of gangue |
| CN110245635A (en) * | 2019-06-20 | 2019-09-17 | 安徽理工大学 | A kind of Infrared image recognition of coal and spoil |
| CN111681210A (en) * | 2020-05-16 | 2020-09-18 | 浙江德尚韵兴医疗科技有限公司 | Method for identifying benign and malignant breast nodules by shear wave elastogram based on deep learning |
| CN111860689A (en) * | 2020-07-31 | 2020-10-30 | 中国矿业大学 | Coal and gangue identification method based on phase consistency and light-weight convolutional neural network |
| CN112016461A (en) * | 2020-08-28 | 2020-12-01 | 深圳市信义科技有限公司 | Multi-target behavior identification method and system |
| CN112006678A (en) * | 2020-09-10 | 2020-12-01 | 齐鲁工业大学 | Electrocardiogram abnormity identification method and system based on combination of AlexNet and transfer learning |
| CN112560941A (en) * | 2020-12-14 | 2021-03-26 | 中科院计算所西部高等技术研究院 | Coal and gangue identification method based on image detection |
Non-Patent Citations (5)
| Title |
|---|
| QIANG LIU等: "Recognition Methods for Coal and Coal Gangue Based on Deep Learning", IEEE ACCESS, vol. 9, pages 77599 - 77610, XP011857589, DOI: 10.1109/ACCESS.2021.3081442 * |
| 曹现刚等: "煤矸石识别方法研究现状与展望", 工矿自动化, vol. 46, no. 1, pages 40 - 41 * |
| 苗雪菲: "基于深度学习的番茄病虫害诊断防治***", 《中国优秀硕士学位论文全文数据库农业科技辑》, no. 04, pages 046 - 115 * |
| 薛祯也: "基于深度学习的煤矸石识别方法研究", 《中国优秀硕士学位论文全文数据库工程科技I辑》, no. 01, pages 021 - 786 * |
| 郜亚松等: "基于深度学习的煤矸石识别技术与实现", 煤炭科学技术, vol. 49, no. 12, pages 203 - 206 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114120093A (en) * | 2021-12-01 | 2022-03-01 | 安徽理工大学 | Coal gangue target detection method based on improved YOLOv5 algorithm |
| CN114120093B (en) * | 2021-12-01 | 2024-04-16 | 安徽理工大学 | Coal gangue target detection method based on improved YOLOv algorithm |
| CN114494696A (en) * | 2022-01-26 | 2022-05-13 | 安徽理工大学 | Method, system and device for rapidly detecting multi-scale coal gangue image |
| CN114882281A (en) * | 2022-05-16 | 2022-08-09 | 安徽理工大学 | Lightweight intelligent separation model, method, equipment and storage medium for coal and gangue |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107133943B (en) | A kind of visible detection method of stockbridge damper defects detection | |
| CN113591689A (en) | RGB (red, green and blue) image recognition method and system for coal and gangue | |
| CN111767882B (en) | Multi-mode pedestrian detection method based on improved YOLO model | |
| WO2020244261A1 (en) | Scene recognition system for high-resolution remote sensing image, and model generation method | |
| WO2021043112A1 (en) | Image classification method and apparatus | |
| CN109215034B (en) | Weak supervision image semantic segmentation method based on spatial pyramid covering pooling | |
| CN113065558A (en) | Lightweight small target detection method combined with attention mechanism | |
| CN108009493B (en) | Human face anti-cheating recognition method based on motion enhancement | |
| KR20180004898A (en) | Image processing technology and method based on deep learning | |
| Lai et al. | Traffic Signs Recognition and Classification based on Deep Feature Learning. | |
| CN109858467B (en) | Face recognition method and device based on key point region feature fusion | |
| Liao et al. | Triplet-based deep similarity learning for person re-identification | |
| CN113221655B (en) | Face spoofing detection method based on feature space constraint | |
| WO2021175278A1 (en) | Model updating method and related device | |
| CN112465057B (en) | Target detection and identification method based on deep convolutional neural network | |
| CN112329771B (en) | Deep learning-based building material sample identification method | |
| Rangkuti et al. | A novel reliable approach for image batik classification that invariant with scale and rotation using MU2ECS-LBP algorithm | |
| CN108345835B (en) | Target identification method based on compound eye imitation perception | |
| Luan et al. | Sunflower seed sorting based on convolutional neural network | |
| CN114399018B (en) | Efficient ientNet ceramic fragment classification method based on sparrow optimization of rotary control strategy | |
| CN117333948A (en) | End-to-end multi-target broiler behavior identification method integrating space-time attention mechanism | |
| CN112907138A (en) | Power grid scene early warning classification method and system from local perception to overall perception | |
| Yuan et al. | An efficient attention based image adversarial attack algorithm with differential evolution on realistic high-resolution image | |
| Cai et al. | Betel nut classification method based on transfer learning | |
| CN112101084A (en) | Method for automatically extracting earthquake damage information of polarized SAR building based on convolutional neural network |
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 |