阅读说明：本技术 一种基于机器视觉的矿石识别方法 (Ore identification method based on machine vision ) 是由 吴佳 李明宸 唐文妍 莫海红 郭宇豪 孟江南 于 2020-11-10 设计创作，主要内容包括：本发明公开了一种基于机器视觉的矿石识别方法,包括以下步骤：获取矿石图像并制作成数据集,将数据集中的图像进行标注,分为合格与不合格两类(合格标注为1,不合格标注为0)；将数据集进行预处理,包括去噪、裁剪等操作；对预处理后的图像进行颜色特征与纹理特征的提取,将两种特征进行组合形成新的特征集；采用特征约简算法对组合后的特征集进行特征选择；最后使用机器学习分类算法进行分类识别。本发明形成了一套完整的、基于机器视觉的矿石识别方法,算法简单高效,识别精度高。(The invention discloses an ore identification method based on machine vision, which comprises the following steps: acquiring an ore image, manufacturing the ore image into a data set, and labeling the image in the data set into a qualified type and an unqualified type (the qualified label is 1, and the unqualified label is 0); preprocessing a data set, including operations such as denoising and cutting; extracting color features and texture features of the preprocessed image, and combining the two features to form a new feature set; performing feature selection on the combined feature set by adopting a feature reduction algorithm; and finally, performing classification and identification by using a machine learning classification algorithm. The invention forms a set of complete ore identification method based on machine vision, and the algorithm is simple and efficient, and the identification precision is high.)

1. A machine vision-based ore identification method comprises the following steps:

the method comprises the following steps: acquiring an ore image, and marking the acquired image, wherein the qualified mark is 1, and the unqualified mark is 0;

step two: preprocessing images in the data set, wherein the preprocessing mainly comprises operations such as denoising and cutting;

step three: extracting color (RGB) features and gray level co-occurrence matrix (GLCM) features of ores in the image, and combining the two features to form a new feature set;

step four: using a feature reduction algorithm to the combined feature set to reduce the feature number in the feature set;

step five: inputting the simplified features and the labels thereof into a classifier for training.

2. The ore identification method based on machine vision according to claim 1, characterized in that the specific steps in the first step are as follows:

1-1) collecting an ore image sample set on the site of an ore processing plant;

1-2) segmenting and labeling the ore image, wherein the qualified mark is 1, and the unqualified mark is 0.

3. The ore identification method based on machine vision according to claim 2, wherein in the second step, the images in the sample set are preprocessed, and the influence of noise from external environment and imaging equipment is removed by using operations such as denoising.

4. The machine-vision-based ore identification method according to claim 3, wherein the third step specifically comprises the following steps:

i) extracting color (RGB) characteristics of each ore image to obtain an average value of R, G, B three-color components of each ore image;

II) extracting the gray level symbiosis (GLCM) characteristics of each ore image to obtain the contrast f of each ore image₁Correlation f₂Energy f₃Sum and inverse difference moment f₄4 texture features, and a 16-dimensional feature set is obtained by calculating corresponding values in four directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees;

in the formula (f)₁、f₂、f₃、f₄Contrast, correlation, energy and moment of dissimilarity, respectively; l is the number of gray levels; i and j respectively represent gray values of different positions in an image; d is a distance, typically 1; theta is the direction and takes the values of 0 degree, 45 degrees, 90 degrees and 135 degrees; u. of₁Is the mean value of the variable i and,u₂is the average of the variable j and is,σ₁is the variance of the variable i and is,σ₂is the variance of the variable j and,

III) combining the extracted color features with the gray level co-occurrence matrix features to form a feature set with 19 dimensions.

5. The ore identification method based on machine vision according to claim 4, characterized in that in the fourth step, the combined feature set is reduced by using an attribute reduction algorithm based on neighborhood granulation and inconsistency measurement, and an appropriate evaluation function is selected for reduction, so as to obtain a key feature set capable of identifying the ore image.

6. The machine vision-based ore identification method according to claim 5, wherein in the fifth step, the key feature set and the type label thereof are used as training samples in a classifier, and an ore classification model based on machine vision is obtained through training and is subjected to experimental verification.

Technical Field

The invention relates to an image identification method, in particular to an ore identification method based on machine vision.

Background

The ore screening is an important link in the ore process production, and the ore with high impurity content can influence the quality of the whole ore and is not suitable for producing building materials. The traditional ore sorting method is manual sorting, and the manual sorting method has the defects of poor operating environment, low production efficiency and the like. With the development of computer vision technology in recent years, the recognition technology based on image processing is widely applied, one of the most key points in image recognition is the extraction of image features, the image features mainly comprise three types of colors, textures and shapes, and the selection of the features has important influence on the recognition result. Since the ore has some special surface features, such as rough and mottled surface, adhesion between dust and ore particles, stacking and the like, the single texture and color features are not enough for effectively identifying complex images.

When a machine learning method is applied to classify ores, the selection of image features has a crucial influence on the classification and identification performance, redundant features not only cause the identification speed to be slow, but also reduce the classification accuracy.

Disclosure of Invention

In order to solve the technical problems, the invention provides the ore identification method based on the machine vision, which is simple and efficient in algorithm and high in identification precision.

The technical scheme for solving the problems is as follows: a machine vision-based ore identification method comprises the following steps:

the method comprises the following steps: obtaining an ore image, and marking the collected image, wherein the qualified image is marked as 1, and the unqualified image is marked as 0;

step two: preprocessing images in the data set, wherein the preprocessing mainly comprises operations such as denoising and the like;

step three: extracting color (RGB) features and gray level co-occurrence matrix (GLCM) features of the image, and combining the two features to form a new feature set;

step four: using a feature reduction algorithm to the combined feature set to reduce the feature number in the feature set;

step five: and inputting the key feature set and the corresponding type label into a classifier for training.

In the above ore identification method based on machine vision, in the first step, a plurality of ore image sample sets are collected at the site of an ore processing plant; and (5) segmenting and labeling the ore images in the sample set, wherein the qualified mark is 1, and the unqualified mark is 0.

In the second step, the images in the sample set are preprocessed to remove image quality damage caused in the imaging process, and the preprocessing method comprises operations such as median filtering and cutting.

In the above ore identification method based on machine vision, in the third step, the identification of the ore image includes the following steps:

i) extracting the color (RGB) characteristics of each ore image in a sample set to obtain the average value of R, G, B three color components of each ore image;

II) extracting the gray level co-occurrence matrix (GLCM) characteristics of each ore image in the sample set to obtain 4 texture characteristics of contrast, correlation, energy and inverse difference moment of each ore image, and obtaining a 16-dimensional characteristic set by calculating corresponding values in four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees;

in the formula (f)₁、f₂、f₃、f₄Contrast, correlation, energy and moment of dissimilarity, respectively; l is the number of gray levels; i and j respectively represent gray values of different positions in an image; d is a distance, typically 1; theta is the direction and takes the values of 0 degree, 45 degrees, 90 degrees and 135 degrees; u. of₁Is the mean value of the variable i and,u₂is the average of the variable j and is,σ₁is the variance of the variable i and is,σ₂is the variance of the variable j and,

III) combining the extracted color features with the gray level co-occurrence matrix features to form a feature set with 19 dimensions.

In the fourth step, the combined features are reduced by using an attribute reduction algorithm based on neighborhood granulation and inconsistency measurement, and a proper evaluation function is selected for reduction to obtain a key feature set capable of identifying the ore image.

In the fifth step, the key feature set and the corresponding type label are used as training samples in the classifier, an ore classification model based on machine vision is obtained through training, and experimental verification is performed.

The invention has the beneficial effects that: firstly, collecting ore images on the site of an ore processing factory, making the collected ore images into a sample set, and marking the ore images in the sample set after segmenting the ore images, wherein the ore images are divided into qualified and unqualified ore images; then, preprocessing the image, and eliminating the influence of noise from external environment, imaging equipment and the like to a certain extent when the image is obtained so as to obtain a high-quality clear image sample; extracting the features of the preprocessed images, extracting the color (RGB) features and the gray level co-occurrence matrix (GLCM) features in the texture features of each ore image, and combining the two features to form a new feature set; reducing the combined features by adopting a fast attribute reduction algorithm based on neighborhood granulation and inconsistency measurement, removing redundant features, reserving the most important features for identification, and forming a new feature set; and finally, taking the key feature set and the corresponding type label as training samples of the classifier, and obtaining a classification model through training. The whole method describes an ore identification method based on machine vision, different features are extracted and combined, and then a feature selection algorithm is used for selecting the features to obtain a key feature set, so that the simplification of feature numbers in the ore identification process is realized, the identification rate of a classifier is improved, the algorithm is simple, and the identification precision is high.

Drawings

FIG. 1 is a flow chart of the present invention.

FIG. 2 is a flow chart of feature selection in accordance with the present invention.

FIG. 3 is a result diagram of attribute reduction and classification accuracy in accordance with the present invention.

Detailed Description

The invention will be further described with reference to the accompanying drawings.

As shown in fig. 1, a machine vision-based ore identification method includes the following steps:

the method comprises the following steps: according to ores obtained in an ore processing plant site, acquiring an ore image, creating a data set, and segmenting and marking the image in the data set. The method is divided into two types, wherein the qualified ore image is marked as 1, and the unqualified ore image is marked as 0.

Step two: the images in the data set are pre-processed.

In the process of acquiring the image, the image is generally influenced by the external environment and the imaging equipment, noise interference is brought to a certain extent, and the quality of the image is reduced. The method has the advantages that the image preprocessing is needed before the characteristic extraction is carried out on the ore image, so that the high-quality clear image is obtained, the noise can be reduced by adopting median filtering according to the noise caused by the actual scene of ore mining and imaging, and the loss of the edge information of the image can be avoided while the noise is filtered.

Step three: and extracting color features and texture features of each ore image, and combining the two features. The method comprises the following steps:

i) extracting color (RGB) features of each ore image in a sample set to obtain an R, G, B three-component average value of each ore image as a 3-dimensional feature set;

II) extracting 4 texture features of contrast, correlation, energy and inverse difference moment of each ore image, and obtaining a 16-dimensional feature set by calculating corresponding values in four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees;

in the formula (f)₁、f₂、f₃、f₄Contrast, correlation, energy and moment of dissimilarity, respectively; l is the number of gray levels; i and j respectively represent gray values of different positions in an image; d is a distance, typically 1; theta is the direction and takes the values of 0 degree, 45 degrees, 90 degrees and 135 degrees; u. of₁Is the mean value of the variable i and,u₂is the average of the variable j and is,σ₁is the variance of the variable i and is,σ₂is the variance of the variable j and,

III) combining the extracted color features with the gray level co-occurrence matrix features to form a feature set with 19 dimensions.

Step four: as shown in fig. 2, for the features obtained in step three, a fast attribute reduction algorithm based on neighborhood granulation and inconsistency measurement is used for feature reduction, as shown in table 1, a key feature set capable of identifying an ore image is obtained through the reduction algorithm, feature dimensions are further reduced, and a large amount of original feature information is retained.

TABLE 1 Key feature set and Classification accuracy results

Step five: and regarding the key feature set obtained in the step four and the corresponding type label as a training sample in the classifier. Fig. 3 is a graph of attribute reduction and classification accuracy results, where the highest classification accuracy of 90.78% is achieved when the neighborhood radius σ is 0.28 and the number of reduced attributes is 11. Table 2 shows the influence of the neighborhood radius on the reduced feature number, and the partial data of which the classification accuracy varies with the reduced feature number.

TABLE 2 comparison of partial feature set number to Classification accuracy results