CN112597904A - Method for identifying and classifying blast furnace charge level images - Google Patents
Method for identifying and classifying blast furnace charge level images Download PDFInfo
- Publication number
- CN112597904A CN112597904A CN202011556945.3A CN202011556945A CN112597904A CN 112597904 A CN112597904 A CN 112597904A CN 202011556945 A CN202011556945 A CN 202011556945A CN 112597904 A CN112597904 A CN 112597904A
- Authority
- CN
- China
- Prior art keywords
- image
- gradient
- images
- gas flow
- test set
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
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)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for identifying and classifying blast furnace burden surface images, which comprises the steps of denoising a burden surface data set image by adopting a denoising convolutional neural network algorithm; dividing the denoised charge level data set into a training set image and a test set image which both comprise 7 types; extracting features of each type of image in the training set by adopting a direction gradient histogram algorithm, and extracting features of each type of image in the test set by adopting a direction gradient histogram algorithm; designing a classifier by using the image characteristics of the training set through a multi-classification support vector machine algorithm to obtain a classifier belonging to the material surface data set image; adding the image characteristics of the test set into the trained classifier; classifying the images of the test set, and calculating the identification accuracy; the method for identifying and classifying the blast furnace charge level image has a good effect of removing the stripe noise in the charge level image, and has the advantages of simple calculation, high speed and capability of quickly and accurately identifying and classifying the furnace condition in the blast furnace charge level image.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a method for identifying and classifying blast furnace burden surface images.
Background
The inside of the blast furnace is a high-temperature, high-pressure and high-dust closed environment, so that the shot image details of the coal gas flow distribution of the blast furnace charge level are not clear, the defects of unobvious profile of the charge level and the like exist, and the on-site operating personnel of the blast furnace cannot obtain an accurate and clear coal gas flow distribution image, so that the accurate identification of the coal gas flow distribution image cannot be realized. Chinese patent publication No. CN104778432A, published 2015, 07, 15, entitled image recognition method, discloses an image recognition method for recognizing a picture in which characters are represented by using light and dark shadows interlaced with horizontal and vertical stripes, and has the disadvantage that the image recognition method cannot rapidly and accurately recognize blast furnace charge level coal airflow distribution images, and therefore, is lack of practicability in other image recognition except for pictures in which characters are represented by light and dark shadows interlaced with horizontal and vertical stripes.
Disclosure of Invention
The invention mainly aims to provide a method for identifying and classifying blast furnace charge level images.
A method for identifying and classifying blast furnace burden surface images is characterized by comprising the steps of denoising a burden surface data set image by adopting a denoising convolutional neural network algorithm; dividing the denoised charge level data set image into a training set image and a test set image which both comprise 7 types, wherein each type of the training set image comprises 600 pieces, and each type of the test set image comprises 100 pieces; extracting features of each type of image in the training set by adopting a direction gradient histogram algorithm, and extracting features of each type of image in the test set by adopting a direction gradient histogram algorithm; designing a classifier by using the image characteristics of the training set through a multi-classification support vector machine algorithm to obtain a classifier belonging to the material surface data set image; adding the image characteristics of the test set into the trained classifier; and classifying the images of the test set, and calculating the identification accuracy.
The method for identifying and classifying the blast furnace charge level images is characterized by comprising the following specific steps of:
(1) denoising the images of the charge level data set by adopting a denoising convolutional neural network algorithm, performing down-sampling processing on the original images at the input end, wherein the processed images are one fourth of the original images, and then synchronously inputting the four sub-images and a noise level image with a standard deviation of 50 into a convolutional neural network to obtain images with smooth surfaces;
the processed image well ensures the image details, removes the stripe noise caused by the video transmission process in the image, and avoids the following image feature extraction that the stripe noise is also used as the feature of the image;
(2) dividing the denoised charge level data set image into a training set image and a test set image which both comprise 7 types:
firstly, when the blast furnace charge level image is a region without edge gas flow, central gas flow and communication, the furnace condition is a full black charge level and the furnace is closed;
secondly, when the blast furnace charge level image is a gas flow with an edge and a central gas flow with a small area, the furnace condition is at the initial stage of normal combustion;
thirdly, when the blast furnace charge level image is of edge gas flow and large-area central gas flow, the furnace condition is in the middle of normal combustion;
fourthly, when the blast furnace charge level image is a non-edge gas flow, a large-area central gas flow and a communication area, the furnace condition is a material collapse state;
when the blast furnace charge level image is a non-edge gas flow and a small-area central gas flow, the furnace condition is that only the central gas flow is contained;
sixthly, when the blast furnace charge level image has edge gas flow and burr interference, the furnace condition is that the dust concentration is too high;
when the blast furnace burden surface image is of edge gas flow and no central gas flow, the furnace condition is that the ore amount is small;
wherein, each class of training set images comprises 600 images, and each class of testing set images comprises 100 images;
(3) extracting features of each type of image in the training set by adopting a directional gradient histogram algorithm, and extracting features of each type of image in the test set by adopting a directional gradient histogram algorithm, wherein the directional gradient histogram algorithm comprises the following steps:
firstly, reading an input training set image and an input test set image by using an image database store function carried by MATLAB software, considering that the classification of the coal gas flow distribution is not greatly influenced by areas except for edge coal gas flow in the image, and adding extra redundancy when extracting image features, so that the size of the data set image is processed in batches;
calculating gradient, namely calculating the gradient magnitude and gradient direction of each pixel of the training set image and the test set image by adopting [1,0, -1]Performing convolution operation on the original image by a gradient operator to obtain a horizontal gradient component, wherein the horizontal gradient component is the gradient component in the x direction and adopts [1,0, -1%]TThe gradient operator performs convolution operation on the original image to obtain a gradient component in the vertical direction, wherein the gradient component in the vertical direction is the gradient component in the y direction, and a formula is utilized
Solving gradient values in x-axis and y-axis directions, wherein H (x, y) and Gx(x,y)、Gy(x, y) respectively represent a pixel value at a pixel point (x, y) of the input image, a gradient value in a horizontal direction, and a gradient value in a vertical direction; reuse formula
Calculating the gradient direction and gradient magnitude of all pixel points of the original image;
constructing feature vectors of a training set image and a testing set image, wherein the sizes of all the images are cut, and then the images are divided into a plurality of 8 × 8 cell units;
fourthly, the cell blocks construct a gradient direction histogram, the gradient information of the pixels in each cell unit is counted by adopting a histogram mode, the gradient direction of the pixels in the cell units is divided into a plurality of direction blocks, namely the gradient direction of the pixel points in the image is divided into a plurality of regions; then, counting the gradient amplitude value in each cell unit according to the plurality of directions, and weighting and accumulating the gradient magnitude of the pixel points in the statistical area where the corresponding gradient direction is located, so that the feature vector of each cell unit can be formed;
combining and normalizing the cell blocks, combining the cell units into spatially communicated intervals, and combining each cell unit into a spatially communicated interval
Forming an interval by 2 cell units, connecting the feature vectors of all the cell units in the interval in series to obtain the feature vector of the directional gradient histogram of the interval, and taking the size of the interval as the size of a window of a scanning picture;
combining the characteristics of the intervals, and after the original image is scanned through a scanning window, connecting the directional gradient histogram characteristic vectors of all the intervals in series to obtain the training set image characteristics and the test set image characteristics;
(4) the classifier is designed by the image characteristics of the training set by adopting a multi-classification support vector machine algorithm, an SVM is designed between any two classes of image samples, so for 7 classes of image samples, 7(7-1)/2 classification functions need to be designed, specifically, extracting directional gradient histogram feature vectors of 7 kinds of images in a training set by a directional gradient histogram algorithm, giving 7 kinds of images (namely, S1, S2, S3, S4, S5, S6 and S7) corresponding classification labels, inputting the gradient histogram feature vector train Features of the training set image and the corresponding classification label train Labels to a support vector machine multi-classifier for training by using a support vector machine multi-classification function fitceoc of MATLAB software, and finally obtaining 21 trained support vector machine classifiers and storing the obtained result files;
(5) adding the image features of the test set into a trained classifier, specifically, adding the directional gradient histogram feature vectors of 7 types of images in the test set into 21 trained classifiers, respectively judging and voting all the feature vectors by the 21 classifiers, and determining which of the 7 types of the feature vectors of each picture has the largest vote and belongs to the 7 types of the images including (i), (ii), (iii), (iv), (v), (iv), and (iv) projected by the feature vectors of each picture;
(6) calculating the identification accuracy, firstly predicting the current test set image by adopting a predict function, wherein the predictIndex in the prediction result represents which category the current test set image belongs to, the score value represents the possibility that the current test set image belongs to 7 categories, the smaller the absolute value of a certain category value is, the more the current test set image belongs to,
[predictIndex,score] = predict(classifer, testFeature);
then by the formula
The accuracy of the prediction result is calculated, wherein M represents the number of correct identifications of the images in the test set, and M represents the total number of the images in the test set.
Compared with the prior art, the invention has the beneficial effects that:
the method for identifying and classifying the blast furnace burden surface image has the advantages that the noise reduction processing is carried out on the original image data, the surface of the processed image is smooth under the condition that the detail information in the image is kept, the stripe noise in the burden surface image is well removed, the calculation is simple, the speed is high, and the furnace condition in the blast furnace burden surface image can be identified and classified quickly and accurately.
Drawings
The present invention will be described in further detail with reference to the accompanying drawings.
FIG. 1 is a flow chart of a method for identifying and classifying blast furnace burden level images in accordance with the present invention.
FIG. 2 is a classification of blast furnace burden level images.
Fig. 3 is a flowchart of the method of step S101 in fig. 1.
Fig. 4 is a flowchart of the method of step S102 in fig. 1.
FIG. 5 is a segmentation of an original image for constructing feature vectors of a gas flow distribution image.
FIG. 6 is a flow diagram of training a support vector machine classifier.
In the figure: the method comprises the steps of S101, extracting features of each type of image of a training set by using a directional gradient histogram algorithm, S102, extracting features of each type of image of a testing set by using a directional gradient histogram algorithm, 201, fully black charge level, furnace closing state, 202, normal combustion initial stage, 203, normal combustion middle stage, 204, material caving state, 205, only containing central gas flow, 206, too high dust concentration, 207, less ore amount, 501, cell unit and 502 interval.
Detailed Description
The invention mainly aims to provide a method for identifying and classifying blast furnace charge level images.
As shown in FIG. 1, a method for identifying and classifying blast furnace burden level images comprises the following basic steps:
(1) denoising the charge level image data set by adopting a denoising convolutional neural network algorithm;
(2) dividing the denoised charge level data set image into a training set image and a test set image which both comprise 7 types, wherein each type of the training set image comprises 600 pieces, and each type of the test set image comprises 100 pieces;
(3) extracting characteristics of each type of image in the training set by adopting a direction gradient histogram algorithm S101, and extracting characteristics of each type of image in the test set by adopting a direction gradient histogram algorithm S102;
(4) training the image features of the training set by adopting a multi-classification support vector machine algorithm to design a classifier;
(5) adding the image characteristics of the test set into the trained classifier;
(6) and classifying the images of the test set, and calculating the identification accuracy.
The invention adopts the following technical scheme.
A method for identifying and classifying blast furnace charge level images comprises the following specific steps:
(1) denoising the charge level data set image by adopting a denoising convolutional neural network algorithm, segmenting an original noise image into four downsampling sub-images by adopting the denoising convolutional neural network algorithm through a downsampling operator mode, reducing the size of the sub-images to be one fourth of the original noise image, simultaneously inputting the original image containing noise and a noise level image with a set corresponding noise level into a convolutional neural network for training, improving the generalization capability by adopting an orthogonal regularization method, and finally obtaining a final denoised image through upsampling the output sub-images through a nonlinear hidden layer network;
the specific operation method comprises the steps of carrying out down-sampling treatment on the extracted blast furnace burden surface image at an input end, wherein the treated blast furnace burden surface image is one fourth of an original image, and then synchronously inputting the four sub-images and a noise level image with a standard deviation of 50 into a convolutional neural network to obtain an image with a smooth surface;
the processed image well ensures the image details, removes the stripe noise caused by the video transmission process in the image, and avoids the following image feature extraction that the stripe noise is also used as the feature of the image;
(2) as shown in fig. 2, the denoised charge level data set image is divided into a training set image and a test set image, both of which include 7 classes:
firstly, when the blast furnace charge level image is a region without edge gas flow, central gas flow and communication, the furnace condition is a full black level and the furnace is closed S201;
secondly, when the blast furnace charge level image is a gas flow with an edge and a central gas flow with a small area, the furnace condition is S202 at the initial stage of normal combustion;
thirdly, when the blast furnace charge level image is of edge gas flow and large-area central gas flow, the furnace condition is in the middle of normal combustion S203;
fourthly, when the blast furnace burden surface image is of a non-edge gas flow, a large-area central gas flow and a communication area, the furnace condition is a material collapse state S204;
when the blast furnace charge level image is a non-edge gas flow and a small-area central gas flow, the furnace condition is that only the central gas flow is contained S205;
sixthly, when the blast furnace charge level image has edge gas flow and burr interference, the furnace condition is that the dust concentration is too high S206;
seventhly, when the blast furnace burden surface image is of edge gas flow and no central gas flow, the furnace condition is that the ore amount is small S207;
wherein, each class of training set images comprises 600 images, and each class of testing set images comprises 100 images;
(3) as shown in fig. 3 and 4, a directional gradient histogram algorithm is adopted to extract features from each type of image in a training set S101, and a directional gradient histogram algorithm is adopted to extract features from each type of image in a testing set S102, in the actual implementation process of the directional gradient histogram algorithm, as shown in fig. 5, an image is firstly divided into a plurality of cell units 501, gradient magnitude and gradient direction are solved for all pixel points in each cell unit 501, and gradient information of the cell units 501 is counted by using a histogram in the gradient direction; then, adjacent cell units 501 form an interval 502, the feature vectors in the interval 502 are connected to obtain a multi-dimensional feature vector, all positions of the whole image are scanned through a detection window, and therefore the feature vectors of directional gradient histograms in all blocks are combined to form the feature vector of the whole image;
the gradient is mainly present at the edge, and the image can be accurately identified by counting the gradient information in the image and without knowing the gradient information and the edge information of the corresponding part in the image;
the histogram of oriented gradients algorithm in specific implementation comprises the following steps:
firstly, reading an input training set image and an input test set image by using an image database store function carried by MATLAB software, considering that the classification of the gas flow distribution is not greatly influenced by areas except for edge gas flow in the image, and when image features are extracted, extra redundancy is added, and firstly, carrying out large-small batch processing on the image size;
calculating gradient, i.e. large gradient for each pixel of training set image and test set imageThe direction of the minor sum gradient is calculated by using [1,0, -1%]Performing convolution operation on the original image by a gradient operator to obtain a horizontal gradient component, wherein the horizontal gradient component is the gradient component in the x direction and adopts [1,0, -1%]TThe gradient operator performs convolution operation on the original image to obtain a gradient component in the vertical direction, wherein the gradient component in the vertical direction is the gradient component in the y direction, and a formula is utilized
Solving gradient values in x-axis and y-axis directions, wherein H (x, y) and Gx(x,y)、Gy(x, y) respectively represent a pixel value at a pixel point (x, y) of the input image, a gradient value in a horizontal direction, and a gradient value in a vertical direction; reuse formula
Calculating the gradient direction and gradient magnitude of all pixel points of the original image;
constructing feature vectors of the training set image and the test set image, cutting the sizes of all the images in the step I, and dividing the input image into a plurality of 8 × 8 cell units 501, as shown in FIG. 5;
fourthly, constructing a gradient direction histogram by the cell blocks, and counting gradient information of pixels in each cell unit 501 in a histogram mode; dividing the gradient direction of pixels in the cell unit 501 into a plurality of direction blocks, namely dividing the gradient direction of pixel points in an image into a plurality of regions 502; then, the gradient amplitude in each cell unit 501 is counted according to the plurality of directions, and the gradient magnitude of the pixel point is weighted and accumulated in the counting area where the corresponding gradient direction is located. This forms a feature vector for each cell unit 501;
combining and normalizing cell blocks, combining all cell units 501 into a space communicated interval 502, combining every 2 x 2 cell units 501 into an interval 502, connecting feature descriptors of all cell units 501 in the interval 502 to obtain a directional gradient histogram feature vector of the interval 502, and taking the size of the interval 502 as the size of a window of a scanning picture;
merging the characteristics of the intervals, and after the original image is scanned through the scanning window, serially connecting the directional gradient histogram characteristic vectors of all the intervals 502 to obtain the training set image characteristics and the test set image characteristics;
(4) the image features of the training set are designed into a classifier by adopting a multi-classification support vector machine algorithm, an SVM is designed between any two classes of image samples, therefore, for 7 classes of image samples, 7(7-1)/2 classification functions are required to be designed, specifically, as shown in FIG. 6, directional gradient histogram feature vectors of 7 classes of images in the training set are extracted by a directional gradient histogram algorithm, 7 classes of images are endowed with classification labels corresponding to (S1), (S2), (S3), (S4), (S5), (S6) and (S7), the directional gradient histogram feature vectors and the corresponding classification labels of the training set images are respectively input into a support vector machine multi-classifier by utilizing a support vector machine multi-classification function carried by MATLAB software, and finally, 64 classifiers of the well-obtained support vector machine classifier are obtained, and storing the obtained result file;
(5) adding the image features of the test set into a trained classifier, specifically, adding the directional gradient histogram feature vectors of 7 types of images in the test set into 21 trained classifiers, respectively judging and voting all the feature vectors by the 21 classifiers, and determining which of the 7 types of the feature vectors of each picture has the largest vote and belongs to the 7 types of the images including (i), (ii), (iii), (iv), (v), (iv), and (iv) projected by the feature vectors of each picture;
(6) calculating the identification accuracy, firstly predicting the current test set image by adopting a predict function, wherein the predictIndex in the prediction result represents which category the current test set image belongs to, the score value represents the possibility that the current test set image belongs to 7 categories, the smaller the absolute value of a certain category value is, the more the current test set image belongs to,
[predictIndex,score] = predict(classifer, testFeature);
then by the formula
The accuracy of the prediction result is calculated, wherein M represents the number of correct identifications of the images in the test set, and M represents the total number of the images in the test set.
Claims (2)
1. A method for identifying and classifying blast furnace burden surface images is characterized by comprising the steps of denoising a burden surface data set image by adopting a denoising convolutional neural network algorithm; dividing the denoised charge level data set image into a training set image and a test set image which both comprise 7 types, wherein each type of the training set image comprises 600 pieces, and each type of the test set image comprises 100 pieces; extracting features of each type of image in the training set by adopting a direction gradient histogram algorithm, and extracting features of each type of image in the test set by adopting a direction gradient histogram algorithm; designing a classifier by using the image characteristics of the training set through a multi-classification support vector machine algorithm to obtain a classifier belonging to the material surface data set image; adding the image characteristics of the test set into the trained classifier; and classifying the images of the test set, and calculating the identification accuracy.
2. The method for identifying and classifying blast furnace burden level images as claimed in claim 1, comprising the following steps:
(1) denoising the images of the charge level data set by adopting a denoising convolutional neural network algorithm, performing down-sampling processing on the original images at the input end, wherein the processed images are one fourth of the original images, and then synchronously inputting the four sub-images and a noise level image with a standard deviation of 50 into a convolutional neural network to obtain images with smooth surfaces;
the processed image well ensures the image details, removes the stripe noise caused by the video transmission process in the image, and avoids the following image feature extraction that the stripe noise is also used as the feature of the image;
(2) dividing the denoised charge level data set image into a training set image and a test set image which both comprise 7 types:
firstly, when the blast furnace charge level image is a region without edge gas flow, central gas flow and communication, the furnace condition is a full black charge level and the furnace is closed;
secondly, when the blast furnace charge level image is a gas flow with an edge and a central gas flow with a small area, the furnace condition is at the initial stage of normal combustion;
thirdly, when the blast furnace charge level image is of edge gas flow and large-area central gas flow, the furnace condition is in the middle of normal combustion;
fourthly, when the blast furnace charge level image is a non-edge gas flow, a large-area central gas flow and a communication area, the furnace condition is a material collapse state;
when the blast furnace charge level image is a non-edge gas flow and a small-area central gas flow, the furnace condition is that only the central gas flow is contained;
sixthly, when the blast furnace charge level image has edge gas flow and burr interference, the furnace condition is that the dust concentration is too high;
when the blast furnace burden surface image is of edge gas flow and no central gas flow, the furnace condition is that the ore amount is small;
wherein, each class of training set images comprises 600 images, and each class of testing set images comprises 100 images;
(3) extracting features of each type of image in the training set by adopting a directional gradient histogram algorithm, and extracting features of each type of image in the test set by adopting a directional gradient histogram algorithm, wherein the flow of the directional gradient histogram algorithm is as follows:
firstly, reading an input training set image and an input test set image by using an image database store function carried by MATLAB software, considering that the classification of the coal gas flow distribution is not greatly influenced by areas except for edge coal gas flow in the image, and adding extra redundancy when extracting image features, so that the size of the data set image is processed in batches;
calculating gradient, namely calculating the gradient magnitude and gradient direction of each pixel of the training set image and the test set image by adopting [1,0, -1]The gradient operator performs convolution operation on the original image to obtain a horizontal gradient component, wherein the horizontal gradient component is the gradient in the x directionComponent by [1,0, -1%]TThe gradient operator performs convolution operation on the original image to obtain a gradient component in the vertical direction, wherein the gradient component in the vertical direction is the gradient component in the y direction, and a formula is utilized
Solving gradient values in x-axis and y-axis directions, wherein H (x, y) and Gx(x,y)、Gy(x, y) respectively represent a pixel value at a pixel point (x, y) of the input image, a gradient value in a horizontal direction, and a gradient value in a vertical direction; reuse formula
Calculating the gradient direction and gradient magnitude of all pixel points of the original image;
constructing feature vectors of a training set image and a testing set image, wherein the sizes of all the images are cut, and then the images are divided into a plurality of 8 × 8 cell units;
fourthly, the cell blocks construct a gradient direction histogram, the gradient information of the pixels in each cell unit is counted by adopting the histogram mode,
dividing the gradient direction of pixels in the cell unit into a plurality of direction blocks, namely dividing the gradient direction of pixel points in the image into a plurality of regions; then, counting the gradient amplitude value in each cell unit according to the plurality of directions, and weighting and accumulating the gradient magnitude of the pixel points in the statistical area where the corresponding gradient direction is located, so that the feature vector of each cell unit can be formed;
combining and normalizing the cell blocks, combining the cell units into spatially communicated intervals, and combining each cell unit into a spatially communicated interval
2 cell units constituting an intervalThe feature vectors of all cell units in one interval are connected in series to obtain the feature vector of the directional gradient histogram of the interval, and the size of the interval is used as the size of a window of a scanning picture;
combining the characteristics of the intervals, and after the original image is scanned through a scanning window, connecting the directional gradient histogram characteristic vectors of all the intervals in series to obtain the training set image characteristics and the test set image characteristics;
(4) the classifier is designed by the image characteristics of the training set by adopting a multi-classification support vector machine algorithm, an SVM is designed between any two classes of image samples, so for 7 classes of image samples, 7(7-1)/2 classification functions need to be designed, specifically, extracting directional gradient histogram feature vectors of 7 kinds of images in a training set by a directional gradient histogram algorithm, giving 7 kinds of images (namely, S1, S2, S3, S4, S5, S6 and S7) corresponding classification labels, inputting the gradient histogram feature vector train Features of the training set image and the corresponding classification label train Labels to a support vector machine multi-classifier for training by using a support vector machine multi-classification function fitceoc of MATLAB software, and finally obtaining 21 trained support vector machine classifiers and storing the obtained result files;
(5) adding the image features of the test set into a trained classifier, specifically, adding the directional gradient histogram feature vectors of 7 types of images in the test set into 21 trained classifiers, respectively judging and voting all the feature vectors by the 21 classifiers, and determining which of the 7 types of the feature vectors of each picture has the largest vote and belongs to the 7 types of the images including (i), (ii), (iii), (iv), (v), (iv), and (iv) projected by the feature vectors of each picture;
(6) calculating the identification accuracy, firstly predicting the current test set image by adopting a predict function, wherein the predictIndex in the prediction result represents which category the current test set image belongs to, the score value represents the possibility that the current test set image belongs to 7 categories, the smaller the absolute value of a certain category value is, the more the current test set image belongs to,
[predictIndex,score] = predict(classifer, testFeature);
then by the formula
The accuracy of the prediction result is calculated, wherein M represents the number of correct identifications of the images in the test set, and M represents the total number of the images in the test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011556945.3A CN112597904A (en) | 2020-12-25 | 2020-12-25 | Method for identifying and classifying blast furnace charge level images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011556945.3A CN112597904A (en) | 2020-12-25 | 2020-12-25 | Method for identifying and classifying blast furnace charge level images |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112597904A true CN112597904A (en) | 2021-04-02 |
Family
ID=75202018
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011556945.3A Pending CN112597904A (en) | 2020-12-25 | 2020-12-25 | Method for identifying and classifying blast furnace charge level images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112597904A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113095412A (en) * | 2021-04-14 | 2021-07-09 | 中北大学 | Mixed fine aggregate classification and identification method based on multi-feature fusion and support vector machine |
| CN114092819A (en) * | 2022-01-19 | 2022-02-25 | 成都四方伟业软件股份有限公司 | Image classification method and device |
| CN115841592A (en) * | 2022-11-29 | 2023-03-24 | 上海船舶运输科学研究所有限公司 | Ship image classification method and system based on SVM classifier and HOG features |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107066958A (en) * | 2017-03-29 | 2017-08-18 | 南京邮电大学 | A kind of face identification method based on HOG features and SVM multi-categorizers |
| CN109859181A (en) * | 2019-01-29 | 2019-06-07 | 桂林电子科技大学 | A kind of PCB welding point defect detection method |
| CN111563556A (en) * | 2020-05-11 | 2020-08-21 | 国网陕西省电力公司电力科学研究院 | Transformer substation cabinet equipment abnormity identification method and system based on color gradient weight |
-
2020
- 2020-12-25 CN CN202011556945.3A patent/CN112597904A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107066958A (en) * | 2017-03-29 | 2017-08-18 | 南京邮电大学 | A kind of face identification method based on HOG features and SVM multi-categorizers |
| CN109859181A (en) * | 2019-01-29 | 2019-06-07 | 桂林电子科技大学 | A kind of PCB welding point defect detection method |
| CN111563556A (en) * | 2020-05-11 | 2020-08-21 | 国网陕西省电力公司电力科学研究院 | Transformer substation cabinet equipment abnormity identification method and system based on color gradient weight |
Non-Patent Citations (2)
| Title |
|---|
| KAI ZHANG ET AL.: "FFDNet: Toward a Fast and Flexible Solution for CNN based Image Denoising", 《ARXIV》 * |
| 马蓓蓓: "基于HOG特征的车辆检测技术研究", 《中国优秀博硕士学位论文全文数据库（硕士） 信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113095412A (en) * | 2021-04-14 | 2021-07-09 | 中北大学 | Mixed fine aggregate classification and identification method based on multi-feature fusion and support vector machine |
| CN113095412B (en) * | 2021-04-14 | 2024-02-09 | 中北大学 | Mixed fine aggregate classification and identification method based on multi-feature fusion and support vector machine |
| CN114092819A (en) * | 2022-01-19 | 2022-02-25 | 成都四方伟业软件股份有限公司 | Image classification method and device |
| CN115841592A (en) * | 2022-11-29 | 2023-03-24 | 上海船舶运输科学研究所有限公司 | Ship image classification method and system based on SVM classifier and HOG features |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860499B (en) | Feature grouping-based bilinear convolutional neural network automobile brand identification method | |
| CN108596166B (en) | Container number identification method based on convolutional neural network classification | |
| CN107545239B (en) | Fake plate detection method based on license plate recognition and vehicle characteristic matching | |
| CN112597904A (en) | Method for identifying and classifying blast furnace charge level images | |
| Jiao et al. | A configurable method for multi-style license plate recognition | |
| CN111126115B (en) | Violent sorting behavior identification method and device | |
| CN104504669B (en) | A kind of medium filtering detection method based on local binary patterns | |
| CN110766017B (en) | Mobile terminal text recognition method and system based on deep learning | |
| CN106709530A (en) | License plate recognition method based on video | |
| CN103679187B (en) | Image-recognizing method and system | |
| Khalifa et al. | Malaysian Vehicle License Plate Recognition. | |
| EP2327044A2 (en) | Segmenting printed media pages into articles | |
| CN111738367B (en) | Part classification method based on image recognition | |
| CN111539330A (en) | Transformer substation digital display instrument identification method based on double-SVM multi-classifier | |
| CN111652846B (en) | Semiconductor defect identification method based on characteristic pyramid convolution neural network | |
| CN111340032A (en) | Character recognition method based on application scene in financial field | |
| CN111461002B (en) | Sample processing method for thermal imaging pedestrian detection | |
| CN116452506A (en) | Underground gangue intelligent visual identification and separation method based on machine learning | |
| CN108345835B (en) | Target identification method based on compound eye imitation perception | |
| CN111160107B (en) | Dynamic region detection method based on feature matching | |
| CN102129569B (en) | Based on body detection device and the method for multiple dimensioned contrast characteristic | |
| CN112348026A (en) | Magnetic hard disk sequence code identification method based on machine vision | |
| Sharma et al. | A deep cnn model for student learning pedagogy detection data collection using ocr | |
| Anggraeny et al. | Texture feature local binary pattern for handwritten character recognition | |
| Chandra et al. | An automated system to detect and recognize vehicle license plates of Bangladesh |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210402 |
|
| WD01 | Invention patent application deemed withdrawn after publication |