An ore particle size detection system

Abstract

This invention discloses an ore particle size detection system, relating to the field of ore identification technology. It includes a data acquisition module, a local feature extraction module, a local feature aggregation module, a particle size recognition module, and a control module. The data acquisition module acquires point cloud data of the ore, and the control module transmits the point cloud data to the local feature extraction module. The local feature extraction module extracts and groups key points from the point cloud data to obtain local feature vectors, which are then sent to the local feature aggregation module via the control module. The local feature aggregation module aggregates the local feature vectors to obtain global feature data, which is then sent to the particle size recognition module for particle size identification. This invention obtains global feature data by aggregating local feature data of the ore, better characterizing the shape and structural features of the ore, and then performing particle size recognition, thereby improving the performance and accuracy of ore detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ore identification, and more particularly to an ore granularity detection system. BACKGROUND

[0002] Intelligent sensing and identification of ore properties is the basis for realizing intelligentization of the beneficiation production process in the field of intelligent mining. The core development direction of intelligent beneficiation is ore property identification, intelligentization of flotation equipment, and genetic beneficiation technology. The grade, granularity, and crushing characteristics of valuable elements in ore are the most important three basic properties among various physical and chemical properties of ore. These three basic properties determine the theoretical basis and process measures for realizing the separation of each useful component, and the rational use or comprehensive use of products. In actual production, the fluctuations of the three directly affect the production stability and economic benefits of beneficiation operations, and have great significance for efficient and intelligent control of beneficiation production.

[0003] Among them, for the granularity identification of ore, with the rapid development of image processing and intelligent control methods, machine vision technology has begun to be applied to the online monitoring of ore granularity information. However, the ore image has the problems of ore adhesion and ineffective segmentation and positioning, resulting in low accuracy of granularity analysis. With the development of technology, commercial systems for online ore particle size measurement have gradually appeared. For example, patent application No. CN201810202367.X, entitled "A method for detecting ore strength based on deep learning", although it can predict the granularity of ore through machine learning, it uses two-dimensional picture learning, and the prediction accuracy is low. For example, the name is "Analysis of explosive pile granularity based on three-dimensional laser point cloud", which can identify the size and distribution of ore granularity, but is more focused on the identification of large ore blocks. In the field of sintering technology, the granularity of ore is generally below 10mm, and the analysis method of this paper cannot meet the precision requirements.

[0004] In summary, how to accurately identify the granularity data of ore is a technical problem that needs to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the present application provides an ore granularity detection system, which aggregates the local feature data of the extracted ore to obtain global feature data, better representing the shape and structure characteristics of the ore, and then performs granularity identification through a granularity identification module, improving the performance and accuracy of ore detection.

[0006] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0007] An ore granularity detection system, comprising: a data acquisition module, a key point extraction module, a local feature extraction module, a local feature aggregation module, a granularity identification module, a control module, the data acquisition module acquires point cloud data of the ore, and the point cloud data is transmitted to the key point extraction module through the control module; the key point extraction module extracts key point data from the point cloud data, and the key point data is transmitted to the local feature extraction module through the control module; the local feature extraction module groups the key point data and acquires local feature vectors, and the local feature vectors are sent to the local feature aggregation module through the control module; the local feature aggregation module aggregates the local feature vectors to obtain global feature data, and the global feature data is sent to the granularity identification module through the control module for granularity identification to obtain granularity data.

[0008] Preferably, the data acquisition module uses Keyence LJ-X8200 to acquire point cloud data.

[0009] Preferably, the local feature extraction module specifically comprises:

[0010] A first point cloud segmentation module, which uses a minimum cube to enclose the point cloud data of the ore, divides the minimum cube into eight uniformly sized primary cubes, judges whether each primary cube contains the point cloud data, if the primary cube contains the point cloud data and the preset segmentation number n is not reached, repeats the minimum cube segmentation step to segment the primary cube, until the segmented cube no longer contains the point cloud data to stop division, and obtains first point cloud segmentation data;

[0011] A second point cloud segmentation module, which uses a minimum cube to enclose the point cloud data of the ore, divides the minimum cube into eight uniformly sized primary cubes, judges whether each primary cube contains the point cloud data, if the primary cube contains the point cloud data and the preset segmentation number n-1 is not reached, repeats the minimum cube segmentation step to segment the primary cube, until the segmented cube no longer contains the point cloud data to stop division, and obtains second point cloud segmentation data;

[0012] A key point acquisition module, which calculates the midpoint coordinates of the point cloud data in the first point cloud segmentation data and the second point cloud segmentation data, and takes all the calculated midpoint coordinates as key point data;

[0013] An information acquisition module, which extracts shape information and structure information from the key point data through one-dimensional convolution and two-dimensional convolution, and the local feature data comprises shape information and structure information;

[0014] The feature dimension reduction module performs feature selection on the local feature data, reduces the feature dimension, and obtains a local feature vector.

[0015] Preferably, the system further comprises a data processing module configured to perform coordinate normalization processing on the key point data, and send the processed key point data to the local feature extraction module via the control module.

[0016] Preferably, the local feature aggregation module comprises:

[0017] The weight calculation module performs weight calculation on the local feature data by using orthogonal constraint weighting, and obtains feature weights.

[0018] The fusion module performs local feature fusion on the local feature data according to the feature weights, and obtains global feature data.

[0019] Preferably, the granularity recognition module comprises:

[0020] The data set establishment module obtains global feature samples after labeling the granularity of the ore, and establishes a label data set.

[0021] The model training module inputs the label data set into support vector machine classifiers of polynomial kernel and radial basis kernel respectively for model training, compares and discriminates the predicted labels output by the two trained support vector machine classifiers of the two kernel types, directly outputs the predicted label if the predicted labels of the two are consistent, and uses a multi-kernel collaboration recognition strategy to analyze the decision values of the kernels, formulates a voting mechanism to discriminate the final label of the global feature sample if the predicted labels of the two are inconsistent.

[0022] The granularity distinguishing module inputs the global feature data into the two trained support vector machine classifiers of the two kernel types for label discrimination.

[0023] Preferably, the system further comprises a patrol module, which is a wheel group robot carrying an ore granularity inspection system and performs inspection through a pre-set patrol path.

[0024] Compared with the prior art, the ore granularity detection system provided by the present disclosure can realize efficient feature extraction and classification of point cloud data through a deep learning method, improve the processing speed, realize fusion of multi-scale features and recovery of details through layered sampling and grouping operations, improve the accuracy of recognition, and perform double-kernel result discrimination on the recognized global features to improve the accuracy of granularity recognition. BRIEF DESCRIPTION OF DRAWINGS

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the structure provided by the present invention;

[0027] Figure 2 This is a schematic diagram of the local feature extraction module provided by the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] This invention discloses an ore particle size detection system, such as... Figure 1 As shown, it includes: a data acquisition module, a key point extraction module, a local feature extraction module, a local feature aggregation module, a granularity recognition module, and a control module. The data acquisition module acquires point cloud data of the ore and transmits the point cloud data to the key point extraction module through the control module. The key point extraction module extracts key point data from the point cloud data and transmits the key point data to the local feature extraction module through the control module. The local feature extraction module groups the key point data and obtains local feature vectors, which are then sent to the local feature aggregation module through the control module. The local feature aggregation module aggregates the local feature vectors to obtain global feature data, which is then sent to the granularity recognition module through the control module for granularity recognition to obtain granular data.

[0030] In one specific embodiment, the data acquisition module uses Keyence LJ-X8200 to acquire point cloud data.

[0031] In one specific embodiment, such as Figure 2 As shown, the local feature extraction module specifically includes:

[0032] The first point cloud segmentation module uses a minimum cube to enclose the point cloud data of the ore, divides the minimum cube into eight first-level cubes of uniform size, judges whether each first-level cube contains point cloud data, and if the first-level cube contains point cloud data and has not reached a preset segmentation number n, repeats the minimum cube segmentation step to segment the first-level cube, until the segmented cube no longer contains point cloud data to stop division, and obtains first point cloud segmentation data;

[0033] The second point cloud segmentation module uses a minimum cube to enclose the point cloud data of the ore, divides the minimum cube into eight first-level cubes of uniform size, judges whether each first-level cube contains point cloud data, and if the first-level cube contains point cloud data and has not reached a preset segmentation number n-1, repeats the minimum cube segmentation step to segment the first-level cube, until the segmented cube no longer contains point cloud data to stop division, and obtains second point cloud segmentation data;

[0034] In one specific embodiment, the preset segmentation number n = 5;

[0035] The key point acquisition module calculates the midpoint coordinates of the point cloud data in the first point cloud segmentation data and the second point cloud segmentation data, and takes all the calculated midpoint coordinates as key point data.

[0036] In one specific embodiment, the hierarchical sampling strategy of PointNet++ can be used to select key point data from the original point cloud.

[0037] In one specific embodiment, k points with smaller distances are selected as the neighboring points of the midpoint coordinates, and the formula is as follows:

[0038] f(p1,p2,...,p k )=Sort(d(p1,p),d(p2,p),...d(p k ,p));

[0039] Where Sort is a sorting function, d is a distance function, and p is a midpoint coordinate.

[0040] The information acquisition module extracts shape information and structure information from the key point data through one-dimensional convolution and two-dimensional convolution. The local feature data includes shape information and structure information, and the information acquisition module can use PointNet++ to extract the local feature data to capture shape and structure information.

[0041] When extracting information from the first point cloud segmentation data part in the key point data, the key point data is grouped, and the maximum number of neighboring points of the key point is 64 as a group.

[0042] When information extraction is performed on the second point cloud segmentation data part in the key point data, the key point data is grouped, and the maximum number of adjacent points of the key point is 128 as a group;

[0043] The local point cloud in the same group is convoluted, and the formula of one-dimensional convolution is as follows:

[0044]

[0045] Wherein, w is a filter, g1, g2,..., g q is a signal sequence, and Q is a filter length.

[0046] The formula of two-dimensional convolution is as follows:

[0047]

[0048] Wherein, X∈R M×N , filter W∈R U×V , U<<M, V<<N.

[0049] The feature dimension reduction module selects features and reduces feature dimensions of the local feature data to obtain a local feature vector.

[0050] In one specific embodiment, the feature dimension reduction module extracts a local feature vector by using a convolutional neural network, and for a local region, the maximum value of all neurons in the region is selected to represent the region, and the feature vector of the maximum value neuron region is extracted as the local feature vector.

[0051] In one specific embodiment, the data processing module is further included, and the data processing module performs coordinate normalization processing on the key point data, and sends the processed key point data to the local feature extraction module through the control module.

[0052] Wherein, the mean value of the center point coordinates is calculated first, as shown in the following formula:

[0053]

[0054] Wherein, μ x , μ y , μ z are the mean values of the coordinates in X, Y and Z directions.

[0055] Then, the mean values are brought into the variances in X, Y and Z directions:

[0056]

[0057] Wherein, σ x , σ y , σ z are the variances in X, Y and Z directions.

[0058] The normalized coordinates are calculated by mean and variance as follows:

[0059]

[0060] wherein, are normalized coordinates in three directions.

[0061] In one specific embodiment, the local feature aggregation module specifically comprises:

[0062] The weight calculation module calculates the weight of the local feature data by using orthogonal constraint weighting, to obtain the feature weight.

[0063] The specific weight calculation is as follows:

[0064]

[0065] wherein, h i represents the feature importance of the i-th local feature vector, h f represents the feature importance of the f-th local feature vector, β and are learnable parameters, <δ i , δ l represents the inner product between the output features of the first local feature vector corresponding to the first point cloud segmentation data and the second local feature vector corresponding to the second point cloud segmentation data, which is a measure of the correlation between their output features, and a is a regularization coefficient.

[0066] The fusion module fuses the local features according to the feature weight, to obtain the global feature data.

[0067] In one specific embodiment, the granularity recognition module specifically comprises:

[0068] The dataset establishment module obtains the global feature sample after the labeled ore granularity, and establishes a label dataset.

[0069] The model training module inputs the label dataset into the support vector machine classifiers of the polynomial kernel and the radial basis kernel respectively for model training, compares and discriminates the predicted labels output by the two support vector machine classifiers of the two trained kernel types, directly outputs the predicted label if the predicted labels of the two are consistent, and if the predicted labels of the two are inconsistent, uses a multi-kernel collaboration recognition strategy to analyze the decision values of each kernel, formulates a voting mechanism to discriminate the final label of the global feature sample.

[0070] The voting mechanism is used to find the test samples with different predicted labels, and the corresponding decision values are taken from each core. The decision value with the highest votes is considered, and the highest vote of each core is converted into a corresponding effective value. According to the effective value of each vote, if there is a vote with an effectiveness less than a preset threshold percentage, it is determined as an invalid vote, otherwise as a valid vote.

[0071] The voting environment of each core is evaluated, and the decision is made as follows:

[0072] When the number of valid votes of each core is different, the predicted label decided by the core function with more valid vote values is output.

[0073] When the number of valid votes of each core is equal, the mean square error of each core is calculated according to the decision value, and the predicted label decided by the core function with smaller difference is output, and the final predicted label is output.

[0074] In one specific embodiment, it further includes setting a label test set, evaluating the trained support vector machine classifier, calculating classification accuracy, recall rate and other indicators to evaluate the performance of the model.

[0075] According to the evaluation results, optimization is carried out, including adjusting the network structure, changing the sampling strategy, optimizing the training parameters, etc.

[0076] The granularity distinguishing module inputs the global feature data into the two kernel function type support vector machine classifiers trained to distinguish the labels.

[0077] In one specific embodiment, it further includes a patrol module, which is a wheel group robot carrying an ore granularity inspection system and performing inspection through an internally preset patrol path.

[0078] In the description, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. For the same and similar parts between the embodiments, refer to the description of the method part.

[0079] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An ore particle size detection system, characterised in that, include: The system comprises a data acquisition module, a local feature extraction module, a local feature aggregation module, a granularity recognition module, and a control module. The data acquisition module acquires point cloud data of the ore and transmits this point cloud data to the local feature extraction module via the control module. The local feature extraction module extracts and groups key points from the point cloud data to obtain local feature vectors, which are then sent to the local feature aggregation module via the control module. The local feature aggregation module aggregates the local feature vectors to obtain global feature data, which is then sent to the granularity recognition module via the control module for granularity recognition to obtain granularity data. The local feature extraction module specifically includes: The first point cloud segmentation module uses a minimum cube to enclose the point cloud data of the ore, and divides the minimum cube into 8 uniformly sized first-level cubes; it determines whether each of the first-level cubes contains the point cloud data. If it contains the point cloud data and the preset number of segmentation times n has not been reached, the minimum cube segmentation step is repeated to segment the first-level cubes until the segmented cubes no longer contain the point cloud data and the segmentation stops, thus obtaining the first point cloud segmentation data. The second point cloud segmentation module uses a minimum cube to enclose the point cloud data of the ore, and divides the minimum cube into 8 uniformly sized first-level cubes; it determines whether each of the first-level cubes contains the point cloud data. If it contains the point cloud data and the preset number of segmentation times n-1 has not been reached, the minimum cube segmentation step is repeated to segment the first-level cubes until the segmented cubes no longer contain the point cloud data and the segmentation stops, thus obtaining the second point cloud segmentation data. The key point acquisition module calculates the midpoint coordinates of the point cloud data in the first point cloud segmentation data and the second point cloud segmentation data, and uses all the calculated midpoint coordinates as key point data. The information acquisition module extracts shape and structural information from the key point data through one-dimensional and two-dimensional convolution, and uses the shape and structural information as local feature data. The feature dimensionality reduction module performs feature selection on the local feature data to reduce the feature dimension and obtain local feature vectors. The local feature aggregation module specifically includes: The weight calculation module applies orthogonal constraints to the local feature data to calculate the feature weights. The fusion module performs local feature fusion on the local feature data according to the feature weights to obtain global feature data.

2. An ore particle size detection system according to claim 1, characterised in that, The data acquisition module uses Keyence LJ-X8200 to acquire point cloud data.

3. An ore particle size detection system according to claim 1, characterised in that, It also includes a data processing module, which performs coordinate normalization processing on the key point data and sends the processed key point data to the local feature extraction module through the control module.

4. The ore particle size detection system of claim 1, wherein The granularity identification module specifically includes: The dataset creation module obtains global feature samples after labeling the ore granularity and creates a labeled dataset. The model training module inputs the label data set into support vector machine classifiers of polynomial kernel and radial basis kernel respectively for model training, compares and discriminates the predicted labels output by the two support vector machine classifiers of the two kernel types, and directly outputs the predicted label if the predicted labels are consistent, and utilizes a multi-kernel collaborative identification strategy to analyze the decision values of the kernels and formulate a voting mechanism to discriminate the final label of the global feature sample if the predicted labels are inconsistent. The granularity distinguishing module inputs the global feature data into the two support vector machine classifiers of the kernel types whose training is completed, and discriminates the labels.

5. The ore particle size detection system of claim 1, wherein The system further comprises a patrol module, which is a wheel group robot carrying an ore granularity inspection system and performs inspection through a preset internal patrol path.

Images