A real-time detection method and system for the ash content of flotation tailings based on multispectral images

Abstract

The present invention discloses a real-time detection method and system for the ash content of flotation tail coal based on multi-spectral images, which relates to the technical field of prediction of key indicators in the coal preparation industry. The method includes: obtaining tail coal images of tail coal samples under different spectra by different industrial cameras; successively preprocessing and extracting features from the tail coal images to obtain tail coal image features; the multi-dimensional image features include grayscale image features, color image HSI space features, Bayer image features, and infrared image features; inputting the tail coal image features into a progressive random forest model, training the model with the features as inputs and the ash content values as labels, and determining the finally trained model as the tail coal ash content detection model; the tail coal ash content detection model is used for real-time detection of the ash content of flotation tail coal to determine the tail coal ash content detection result. The present invention can improve the detection accuracy and realize the detection of the ash content of tail coal in a relatively large range of ash content changes.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction of key indicators in the coal preparation industry, and particularly to a real-time detection method and system for the ash content of flotation tail coal based on multi-spectral images. Background Art

[0002] Coal is an important energy source and raw material for industrial production, and coal resources will be an important backup energy source in China for a long time to come. Flotation is the most efficient and economical method for treating fine coal, and it is also the starting point for the clean and efficient utilization of coal.

[0003] The ash content of flotation tail coal is an important feedback indicator in the flotation production process, which can reflect the grade and recovery rate of flotation clean coal. Measuring the ash content of tail coal is of great significance to the flotation process. Currently, the most commonly used ash detection method is the ignition gravimetric method, which uses a muffle furnace to burn the dried coal sample until the mass is constant, and then calculates the ratio of the mass of the residue to the mass of the original coal sample. This method is time-consuming and inefficient, unable to achieve real-time detection, and cannot meet the requirements of high-efficiency production. The main methods for real-time ash detection include radiation measurement method, natural radioactivity method, etc. However, the radiation measurement method has safety problems, and the natural radioactivity method has high costs and limited usage conditions, so it has not been widely applied. Summary of the Invention

[0004] The purpose of the present invention is to provide a real-time detection method and system for the ash content of flotation tail coal based on multi-spectral images, which can achieve real-time and rapid ash detection, thereby solving the deficiencies of traditional ash detection methods.

[0005] To achieve the above purpose, the present invention provides the following solutions:

[0006] A real-time detection method for the ash content of flotation tail coal based on multi-spectral images, comprising:

[0007] Obtaining tail coal images of the tail coal sample under multiple spectra; the tail coal images are images collected by an industrial camera under five light sources of white, red, green, blue, and infrared. The wavelength ranges selected for each light source are 440 - 760 nm, 630 - 680 nm, 520 - 560 nm, 420 - 480 nm, and 820 - 850 nm respectively. Grayscale images and color images are collected under the white light source, corresponding Bayer template images are collected under the red, green, and blue light sources respectively, and an infrared grayscale image is collected under the infrared light source; the industrial camera includes a color industrial camera, a grayscale industrial camera, and an infrared industrial camera;

[0008] Successively preprocessing and extracting features from the tail coal images to obtain tail coal image features; the tail coal image features include grayscale image features, color image HSI space features, Bayer image features, and infrared grayscale image features;

[0009] Input the image features of the tail coal into a progressive random forest model, train the model with the features as input and the ash content value as the label, and determine the finally trained model as the tail coal ash detection model; the tail coal ash detection model is used for real-time detection of the ash content of flotation tail coal to determine the tail coal ash detection result.

[0010] Optionally, the specific process of the preprocessing is as follows:

[0011] Screen the tail coal images to eliminate abnormal images;

[0012] Crop the images after eliminating abnormalities, remove the edge parts of the images, and retain the central region of the image with a size of 1100×600 pixels in the center of the image;

[0013] Randomly sample in the central region of the image, crop 5 images with a size of 512×512 pixels from the color image, and crop 5 images with a size of 256×256 pixels from the remaining images as the preprocessed image data.

[0014] Optionally, the extraction process of the grayscale image features is as follows:

[0015] Statistically analyze the distribution frequencies of each gray level in the grayscale image collected under a white light source to obtain a grayscale histogram;

[0016] Fit the grayscale histogram with an unbounded Johnson distribution to obtain a probability function, and use the four parameters ε, λ, γ, η in the probability function as the grayscale features of the image; where the probability function is expressed as:

[0017]

[0018] In the formula, f represents the probability function of x obtained by fitting, ε, λ, γ, η are the four parameters of the unbounded Johnson distribution, satisfying ε, γ∈(-∞, +∞), λ, η∈(0, +∞); x represents the gray value, and sinh -1 is the inverse hyperbolic sine function.

[0019] Optionally, the extraction process of the HSI space features of the color image is as follows:

[0020] Convert the tail coal color image collected under a white light source from the RGB space to the HSI space, statistically analyze the histograms of the three channel components of H, S, and I respectively, and calculate the mean, standard deviation, skewness, kurtosis, and median of the three channels of H, S, and I respectively, and use the 15 calculated data as the HSI space features; where H represents hue, S represents saturation, and I represents intensity.

[0021] Optionally, the extraction process of the Bayer image features is as follows:

[0022] Extract the R-channel information from the images collected under the red light source. Retain 25% of the pixels of the sensor that are red according to the Bayer array, discard the remaining 75% of the interpolated pixels, and extract the B-channel information matrix of the images collected under the blue light source and the Gr and Gb information matrices of the images collected under the green light source in the same way;

[0023] Calculate the features of the R, B, Gr, and Gb information matrices respectively to obtain the mean, variance, skewness, kurtosis, and median of each information matrix, and use the 20 calculated data as color features.

[0024] Optionally, the process of extracting the infrared grayscale image features is as follows:

[0025] Calculate the mean, variance, skewness, kurtosis, and median of the images collected under the infrared light source, and use the 5 calculated data as infrared features.

[0026] Optionally, the training and application process of the progressive random forest model is as follows:

[0027] Use the first-layer classification model of the progressive random forest model to make a preliminary judgment on the ash content of the sample to determine the preliminary judgment result; the preliminary judgment result includes: regarding the samples with ash content greater than 60% as high-ash samples, regarding the samples with ash content of 50%-60% as medium-ash samples, and determining the samples with ash content less than 50% as low-ash samples;

[0028] Use the second-layer regression model of the progressive random forest model to predict the features of the sample ash content to obtain the detection result of the tailings ash content; the second-layer regression model includes a high-ash model, a medium-ash model, and a low-ash model; among them, the high-ash model is trained with samples with ash content of 60%-80%, the medium-ash model is trained with samples with ash content of 50%-60%, and the low-ash model is trained with samples with ash content of 30%-50%.

[0029] The present invention also provides a real-time detection system for the ash content of flotation tailings based on multi-spectral images, including:

[0030] An image acquisition unit for obtaining tail coal images of tail coal samples under multiple spectra; the tail coal images are images acquired by an industrial camera under five light sources: white, red, green, blue, and infrared. The wavelength ranges selected for each light source are 440 - 760nm, 630 - 680nm, 520 - 560nm, 420 - 480nm, and 820 - 850nm respectively. A grayscale image and a color image are acquired under the white light source, corresponding Bayer template images are acquired under the red, green, and blue light sources, and an infrared grayscale image is acquired under the infrared light source; the industrial camera includes a color industrial camera, a grayscale industrial camera, and an infrared industrial camera;

[0031] A feature extraction unit for preprocessing and feature extraction of the tail coal images in sequence to obtain tail coal image features; the tail coal image features include grayscale image features, color image HSI space features, Bayer image features, and infrared grayscale image features;

[0032] An ash content detection unit for inputting the tail coal image features into a progressive random forest model, training the model with the features as input and the ash content value as the label, and determining the finally trained model as the tail coal ash content detection model; the tail coal ash content detection model is used for real - time detection of the ash content of flotation tail coal to determine the tail coal ash content detection result.

[0033] According to the specific embodiments provided by the present invention, the following technical effects are disclosed:

[0034] The present invention discloses a method and system for real - time detection of the ash content of flotation tail coal based on multi - spectral images. The method includes obtaining tail coal images of tail coal samples under different spectra by different industrial cameras; preprocessing and feature extraction of the tail coal images in sequence to obtain tail coal image features; the multi - dimensional image features include grayscale image features, color image HSI space features, Bayer image features, and infrared image features; inputting the tail coal image features into a progressive random forest model, training the model with the features as input and the ash content value as the label, and determining the finally trained model as the tail coal ash content detection model; the tail coal ash content detection model is used for real - time detection of the ash content of flotation tail coal to determine the tail coal ash content detection result. The present invention can improve the detection accuracy and realize the detection of the ash content of tail coal in a large range of ash content changes. The present invention can improve the detection accuracy and realize the detection of the ash content of tail coal in a large range of ash content changes (30% - 80%). Description of the Drawings

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0036] Figure 1 Schematic diagram of the real-time detection method for the ash content of flotation tail coal based on multi-spectral images of the present invention;

[0037] Figure 2 Schematic diagram of the structure of the progressive random forest model in this embodiment;

[0038] Figure 3 Comparison chart of the predicted ash content value and the actual test value in this embodiment; among them, part (a) is a schematic diagram of the model detection result; part (b) is a schematic diagram of the model detection error. Specific implementation manner

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0040] The purpose of the present invention is to provide a real-time detection method and system for the ash content of flotation tail coal based on multi-spectral images, which can improve the detection accuracy and realize the detection of the ash content of tail coal in a large range.

[0041] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific implementation manners.

[0042] As Figure 1 shown, the present invention provides a real-time detection method for the ash content of flotation tail coal based on multi-spectral images, including:

[0043] Step 100: Obtain tail coal images of the tail coal sample under multiple spectra; the tail coal images are images collected by an industrial camera under five light sources: white, red, green, blue, and infrared. The wavelength ranges selected for each light source are 440 - 760nm, 630 - 680nm, 520 - 560nm, 420 - 480nm, and 820 - 850nm respectively. A grayscale image and a color image are collected under the white light source, the corresponding Bayer template images are collected under the red, green, and blue light sources respectively, and an infrared grayscale image is collected under the infrared light source; the industrial camera includes a color industrial camera, a grayscale industrial camera, and an infrared industrial camera

[0044] Among them, the grayscale CMOS industrial camera is 110 mm away from the tail coal sample, the color CCD industrial camera is 110 mm away from the tail coal sample, and the infrared CMOS industrial camera is 200 mm away from the tail coal sample; the light sources used are all ring LED light sources to ensure balanced, non-flickering, and shadow-free lighting conditions.

[0045] Step 200: Preprocess and extract features from the tail coal image in sequence, and connect the extracted multi-dimensional features to obtain the tail coal image features; the tail coal image features include grayscale image features, color image HSI space features, Bayer image features, and infrared grayscale image features.

[0046] Step 300: Input the tail coal image features into a progressive random forest model, train the model with the features as the input and the ash content value as the label, and determine the finally trained model as the tail coal ash content detection model; the tail coal ash content detection model is used for real-time detection of the ash content of flotation tail coal to determine the tail coal ash content detection result.

[0047] As a more specific implementation manner, in this embodiment, a grayscale industrial camera, a color industrial camera, and an infrared industrial camera are used to collect tail coal images under multiple light sources, features in the images are extracted by multiple methods, and ash content detection is realized through a progressive random forest model. The ash content detection result is accurate and time-consuming is short, which can guide flotation production, save labor and time costs, and provide a detection basis for flotation intelligence.

[0048] Among them, during the training process of the tail coal ash content detection model, the samples for training the model are obtained by blending tail coal dry powder and clean coal dry powder in different proportions. The ash content of each sample is between 30% and 80%, and there are 12 samples with different ash contents in total. The blending method and ash content of the training set samples are shown in Table 1:

[0049] Table 1 Blending method and ash content

[0050] Sample Number 1 2 3 4 5 6 7 8 9 10 11 12 Quality of Clean Coal Dry Powder / g 14 13 12 11 10 9 8 7 6 5 4 2 Quality of Tail Coal Dry Powder / g 6 7 8 9 10 11 12 13 14 15 16 18 Ash Content of Sample / % 32.8 36.4 40.0 43.6 47.3 50.9 54.5 58.1 61.7 65.3 68.9 76.2

[0051] The tail coal image is collected by the built image acquisition device. The tail coal sample is placed on the platform. The grayscale CMOS industrial camera and the color CCD industrial camera are used to collect images under visible light; the infrared CMOS industrial camera is used to collect infrared grayscale images under infrared light sources. To avoid interference from ambient light on image acquisition, the entire image acquisition device is placed in a black light-shielding cover, and a ring light source with adjustable brightness is used to provide illumination to reduce the influence of shadows on the image.

[0052] After collecting the images, crop the images. To reduce the influence of spherical aberration, remove the edge part of the images and retain the area with a size of 1100×600 in the center of the images. Use the method of random cropping to crop 5 images with a size of 512×512 pixels from the color images in this area, and crop 5 images with a size of 256×256 pixels from the remaining images as the preprocessed image data.

[0053] Extract features from the preprocessed images, including grayscale features, HSI space features of color images, Bayer image features, and infrared features, and concatenate the image features.

[0054] When extracting grayscale features, for the grayscale images collected under white light sources, obtain the grayscale histogram by statistics, and fit the grayscale histogram with the unbounded Johnson distribution to obtain the probability function:

[0055]

[0056] Among them, f represents the probability function of x obtained by fitting, and ε, λ, γ, η are the four parameters of the unbounded Johnson distribution, satisfying ε, γ∈(-∞, +∞), λ, η∈(0, +∞). Take the four parameters ε, λ, γ, η of the fitting function as the grayscale features of the images.

[0057] When extracting the HSI space features of color images, since the components in the HSI space are more independent than those in the RGB space and can provide more extensive image information except for color information, based on the features of the three RGB channels after normalization, use the RGB-HSI space conversion formula to calculate the three channel components of H (hue), S (saturation), and I (intensity) respectively:

[0058]

[0059] Analyze the mean, standard deviation, skewness, kurtosis, and median of the three channels of H, S, and I respectively, and take the 15 calculated data as the HSI space features.

[0060] When extracting the features of the Bayer template images, to reduce the interference of interpolation on image features, only retain the information directly collected by the camera sensor and discard the interpolated information when extracting the color features of RGB images. Specifically, when extracting the R channel information of the images collected under red light sources, retain 25% of the pixels with the sensor being red according to the Bayer array, and discard the remaining 75% of the interpolated pixels. Extract the B channel information matrix of the images collected under blue light sources and the Gr, Gb information matrices of the images collected under green light sources in the same way.

[0061] Calculate the mean, variance, skewness, kurtosis, and median of the R, B, Gr, Gb information matrices, a total of 20 features, as the color features. The calculation formula for the mean is:

[0062]

[0063] Among them, H and W represent the height and width of the image, GS(i, j) represents the pixel value of the image at the corresponding position, and MEA is the calculated mean value. After calculating the mean value, the variance of each information matrix can be further obtained based on the mean value:

[0064]

[0065] VAR is the variance of the information matrix. Based on the mean value and variance, the skewness and kurtosis of each information matrix can be obtained:

[0066]

[0067] In the formula, SKE and KUR respectively represent the skewness and kurtosis of the information matrix. The color features extracted according to the Bayer array do not go through an interpolation algorithm. All information directly comes from the sensor of the industrial camera, and multiple monochromatic light sources are used, which can accurately reflect the reflection of the tail coal to light of various wavelengths, providing reliable features for predicting the ash content of the tail coal next.

[0068] When extracting infrared features, five features including the mean value, variance, skewness, kurtosis, and median of the image collected under infrared light are obtained as infrared features.

[0069] After completing feature extraction, a progressive random forest model as shown in Figure 2 is built. The concatenated features are used as the model input, and the actual ash content value is used as the label. The model is trained using the training set, and the performance of the model is tested through the test set. Finally, a tail coal ash content detection model with performance meeting the requirements is obtained. The progressive random forest model has fewer hyperparameters and a relatively simple structure. Compared with complex deep learning models, it does not require a large amount of hyperparameter tuning, has stronger interpretability, and a faster training speed.

[0070] The random forest classification model in the first layer of the progressive random forest model makes a rough judgment on the ash content of the sample, initially dividing it into three categories. Samples with an ash content greater than 60% are regarded as high-ash samples, samples with an ash content of 50%-60% are regarded as medium-ash samples, and samples with an ash content less than 50% are determined as low-ash samples.

[0071] The second layer of the progressive random forest model is the random forest regression models for high, medium, and low ash contents, which make predictions on the ash content of the sample through features. The high-ash model is trained with samples with an ash content of 60%-80%, the medium-ash model is trained with samples with an ash content of 50%-60%, and the low-ash model is trained with samples with an ash content of 30%-50%.

[0072] When detecting the ash content of flotation tailings, the first-layer model classifies the samples into three ash levels, high, medium, and low, according to the extracted features, and the second-layer model obtains the detection result of the ash content of the flotation tailings according to the classification.

[0073] The real-time detection of the ash content of flotation tailings can be achieved by the above method. As Figure 3 shown, the present invention can achieve the detection of the ash content of flotation tailings in a large range, and accurate detection results are obtained for flotation tailings with an ash content of 30%-80%. The detection error of 98% of the samples is less than 3%, and the root mean square error is 0.813. Moreover, the progressive random forest model used has the advantages of few parameters and strong interpretability. The ash content detection method of the present invention has a low complexity and is suitable for the real-time detection of flotation tailings with an ash content of 30%-80%.

[0074] Therefore, through the experiments in this embodiment, the present invention can achieve the detection of the ash content of flotation tailings in a large range. According to the prediction results, it can guide flotation production, save labor and time costs, and provide a solution for flotation intelligence.

[0075] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0076] Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the core idea of the present invention; at the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A real-time detection method for flotation tailing ash based on multispectral images, characterized in that: include: Obtain tail coal images of tail coal samples under multiple spectra; The tail coal image is an image collected by an industrial camera under five light sources: white, red, green, blue and infrared. The wavelength ranges of the light sources are 440-760nm, 630-680nm, 520-560nm, 420-480nm and 820-850nm respectively. Grayscale images and color images are collected under white light sources, corresponding Bayer template images are collected under red, green and blue light sources respectively, and infrared grayscale images are collected under infrared light sources. The industrial cameras include color industrial cameras, grayscale industrial cameras and infrared industrial cameras. Preprocessing and feature extraction are performed on the tail coal image in sequence to obtain tail coal image features; the tail coal image features include grayscale image features, color image HSI space features, Bayer image features and infrared grayscale image features; The tail coal image features are input into a progressive random forest model, the features are used as input and the ash value is used as a label for model training, and the finally trained model is determined as a tail coal ash detection model; the tail coal ash detection model is used for real-time detection of flotation tail coal ash to determine the tail coal ash detection result.

2. The real-time detection method of flotation tailing ash based on multispectral image according to claim 1 is characterized in that: The specific process of the pretreatment is: Screening the tail coal images and removing abnormal images; The image after the abnormalities are removed is cropped to remove the edge of the image and retain the central area of ​​the image with a size of 1100×600 pixels; Random sampling is performed in the central area of ​​the image, and the color image is cut into 5 images with a size of 512×512 pixels, and the remaining image is cut into 5 images with a size of 256×256 pixels as preprocessed image data.

3. The real-time detection method of flotation tailing ash based on multispectral image according to claim 1 is characterized in that: The extraction process of the grayscale image features is as follows: The distribution frequency of each gray level in the gray image collected under the white light source is counted to obtain a gray histogram; The grayscale histogram is fitted using an unbounded Johnson distribution to obtain a probability function, and the four parameters ε, λ, γ, and η in the probability function are used as the grayscale features of the image; wherein the probability function is expressed as: Where f represents the probability function of x obtained by fitting, ε, λ, γ, η are the four parameters of the unbounded Johnson distribution, satisfying ε, γ∈(-∞,+∞), λ, η∈(0,+∞); x represents the gray value, sinh -1 is the inverse hyperbolic sine function.

4. The real-time detection method of flotation tailing ash based on multispectral image according to claim 1 is characterized in that: The extraction process of the HSI spatial feature of the color image is as follows: The tail coal color image collected under white light source is converted from RGB space to HSI space, and the histograms of the three channel components H, S, and I are statistically obtained respectively. The mean, standard deviation, skewness, kurtosis, and median of the three channels H, S, and I are calculated respectively, and the 15 calculated data are used as HSI space features; among them, H represents hue, S represents saturation, and I represents intensity.

5. The real-time detection method of flotation tailing ash based on multispectral image according to claim 1 is characterized in that: The extraction process of the Bayer image features is as follows: Extract R channel information of the image collected under the red light source, retain 25% of the pixels of the sensor that are red according to the Bayer array, discard the remaining 75% of the pixels after interpolation, and extract the B channel information matrix of the image collected under the blue light source and the Gr and Gb information matrices of the image collected under the green light source in the same way; The features of the R, B, Gr, and Gb information matrices were calculated respectively to obtain the mean, variance, skewness, kurtosis, and median of each information matrix, and the 20 calculated data were used as color features.

6. The real-time detection method of flotation tailing ash based on multispectral image according to claim 1 is characterized in that: The extraction process of the infrared grayscale image features is as follows: The mean, variance, skewness, kurtosis and median of the images collected under infrared light source are calculated, and the five calculated data are used as infrared features.

7. The real-time detection method of flotation tailing ash based on multispectral image according to claim 1 is characterized in that: The training and application process of the progressive random forest model is as follows: Using the first-layer classification model of the progressive random forest model, a preliminary judgment is made on the ash content of the sample to determine a preliminary judgment result; The preliminary judgment results include: considering samples with ash content greater than 60% as high ash samples, considering samples with ash content of 50%-60% as medium ash samples, and determining samples with ash content less than 50% as low ash samples; The second-layer regression model of the progressive random forest model is used to perform feature prediction on the sample ash content to obtain the tail coal ash content detection result; the second-layer regression model includes a high ash content model, a medium ash content model and a low ash content model; wherein the high ash content model is trained with samples with 60%-80% ash content, the medium ash content model is trained with samples with 50%-60% ash content, and the low ash content model is trained with samples with 30%-50% ash content.

8. A real-time detection system for flotation tailing ash based on multispectral images, characterized in that: include: An image acquisition unit, used for acquiring tail coal images of tail coal samples under multiple spectra; The tail coal image is an image collected by an industrial camera under five light sources: white, red, green, blue and infrared. The wavelength ranges of the light sources are 440-760nm, 630-680nm, 520-560nm, 420-480nm and 820-850nm respectively. Grayscale images and color images are collected under white light sources, corresponding Bayer template images are collected under red, green and blue light sources respectively, and infrared grayscale images are collected under infrared light sources. The industrial cameras include color industrial cameras, grayscale industrial cameras and infrared industrial cameras. A feature extraction unit is used to perform preprocessing and feature extraction on the tail coal image in sequence to obtain tail coal image features; the tail coal image features include grayscale image features, color image HSI space features, Bayer image features and infrared grayscale image features; The ash detection unit is used to input the tail coal image features into a progressive random forest model, use the features as input and the ash values ​​as labels to train the model, and determine the final trained model as the tail coal ash detection model; the tail coal ash detection model is used for real-time detection of flotation tail coal ash to determine the tail coal ash detection results.

Images