CN115228595B - A method for intelligent segmentation of mineral belts based on target detection - Google Patents

Abstract

The invention discloses an intelligent mining belt segmentation method based on target detection, which comprises the following steps: acquiring data of a cradle bed surface ore belt, performing feature calibration on the acquired bed surface ore belt information, and performing model training and evaluation on the data to obtain a target detection model algorithm; the camera collects data of the ore belt of the cradle bed surface in real time and transmits the collected data to the software central platform; the software center platform performs preliminary clear processing on the image; transmitting the pictures processed by the algorithm to a trained model, evaluating the picture information by the model, and outputting the pictures calibrated by the system; the algorithm obtains the moving direction and the moving distance through accurate calculation, the software center platform sends a moving direction and moving distance instruction to the motion control system, and the control system drives the motor to drive the ore receiving plate to move to the designated position. The intelligent shaking table ore dressing belt identification and segmentation method is a brand new algorithm and has a good effect.

Description

A method for intelligent segmentation of mineral belts based on target detection

Technical Field

The invention belongs to the technical field of re-selection and mineral processing, and in particular, relates to an intelligent segmentation method of a mineral belt based on target detection.

Background Art

my country has rich mineral resources, but mineral resources are always important basic resources that are indispensable to my country's economic development and military field. With the continuous progress of science and technology, industrial automation and intelligence are the irresistible development trend in the future. Most industries in my country are facing industrial upgrading, and the mining industry is also inevitably facing problems such as industrial structure adjustment and equipment upgrading. Under the general trend of the development of the times, only by keeping up with the pace of the development of the times can the mining industry provide strong support for my country's demand for mineral resources.

According to the composition and properties of minerals, the beneficiation methods are mainly divided into four categories: gravity separation, magnetic separation, flotation and electrostatic separation. Shaking table beneficiation is one of the important equipment in gravity separation and has been widely used at home and abroad. Minerals are separated on the shaking table in the form of slurry, which is mainly the result of the combined action of the bed bar type, the asymmetric movement of the bed surface and the horizontal flushing water on the bed surface. There are multiple fan-shaped mineral belts on the bed surface, such as concentrate belt, sub-concentrate belt, intermediate ore belt and tailing belt, so that minerals of different grades can be separated.

Nowadays, the shaking table beneficiation technology is very mature, but the level of beneficiation automation is still at a relatively low level. During the shaking table beneficiation process, due to the influence of the on-site feed amount, feed concentration, feed particle size and changes in the grade of the original ore, the ore belt position, ore belt width and ore belt color of the bed mineral will be affected to varying degrees, causing the ore belt position to change in real time. The on-site inspection operator must observe the characteristic changes of the bed ore belt with the naked eye, and then make corresponding adjustments to the position of the docking plate based on his work experience to achieve the purpose of sorting different ore belts, thereby achieving the concentrate grade required by the beneficiation plant. This traditional method of adjusting the docking plate has a high frequency of adjustment of the docking plate position, and the labor intensity of the inspection workers is high; due to the differences in the work of different personnel, the beneficiation indicators are also unstable, making it difficult to ensure the smooth development of the beneficiation work, affecting the work quality and work efficiency.

In view of this, the present invention is proposed.

Summary of the invention

The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method for intelligent segmentation of mineral belts based on target detection. To solve the above technical problem, the basic concept of the technical solution adopted by the present invention is:

A method for intelligent segmentation of mineral belts based on target detection comprises the following steps:

Step 1: collect the data of the mineral belt on the shaking table, calibrate the features of the collected mineral belt information, train and evaluate the model on these data, and obtain the target detection model algorithm. The model training includes the following steps:

Step 101, model building, input x, through the first CNN, get the distribution parameters of multiple latent variables q(t); input a set of t values obtained according to the distribution of q(t), through the second CNN, get the distribution parameters of the final image p(x);

Step 102, prove that the calculation of the latent variable model is unbiased;

Step 103, model optimization, input x, through the first CNN, obtain the distribution parameters sd and mean of multiple latent variables q(t);

Through the standard Gaussian distribution p(z), we get a set of z values; then through z×sd+mean, we get a set of t values;

Input the t value and pass it through the second CNN to get the distribution parameters of the final image p(x);

Step 2: The software central platform sends instructions to the on-site camera to collect data on the ore belt on the shaking table in real time, and transmits the collected data to the software central platform through the 4G gateway;

Step 3: The software central platform will perform corresponding algorithm processing on the transmitted bed surface mineral belt image to obtain a clear bed surface mineral belt image;

Step 4: The image processed by the algorithm is transmitted to the trained model, the model evaluates the image information, and then outputs the image after the system calibration;

Step 5: After the model algorithm calculates, determine whether the ore belt has changed compared to the previous round. If so, proceed to step 6; if not, proceed to step 7;

Step 6: The algorithm will obtain the moving direction and moving distance through precise calculation. The software central platform sends instructions of moving direction and moving distance to the motion control system, and the control system drives the motor to drive the receiving plate to move to the specified position.

Step 7: Keep the device in its original state.

Furthermore, the formula used in the model building of step 101 is:

q _i (t _i )=N(m( _xi ,φ),diag(s ² (x _i ,φ)))

Furthermore, the step 102 proves that the calculation of the latent variable model is unbiased and the formula used is:

Furthermore, the formula used for model optimization in step 103 is:

p(ε _i )＝N(0,I)

Furthermore, the information collected in step 2 includes the width of the ore belt, the color of the ore belt, and the location distribution characteristics of the ore belt.

Furthermore, the camera is installed above the rocking bed through a camera bracket, and the camera bracket is installed on a base.

Furthermore, the evaluation of the image information in step 4 includes comparing and identifying the on-site bed surface mineral zone images transmitted to the platform by the camera according to the model training data, and calibrating the identification points.

Furthermore, the model training also includes connecting multiple points of the identification points, and extending the line toward the end of the rocking bed according to the trend of the connection line, and the extension line is extended to the end of the bed.

Furthermore, the software central platform is set to collect a picture of the mineral belt on the on-site shaking table bed every 2 minutes and make corresponding adjustments to the ore receiving plate.

After adopting the above technical scheme, the present invention has the following beneficial effects compared with the prior art.

The present invention uses a model for target detection and recognition, and transmits the image information of the ore belt surface acquired in real time to the software central platform through a camera system. The algorithm calculates the information and determines whether to adjust the position of the receiving plate to achieve good ore selection. The present invention is a brand-new algorithm in the direction of intelligent shaking table ore selection belt identification and segmentation, which has a good effect; it can accurately segment the ore belts of the shaking table surface concentrate, sub-concentrate, intermediate ore and tailings, and will not be affected by on-site environmental factors such as light.

The present invention utilizes the rapid transmission of the network and the rapid processing of the image information by the algorithm, and has a fast calculation speed, which can almost achieve synchronization with the ore belt information on the shaking table surface on site, and meet the requirements of the ore dressing site. It reduces the error caused by human judgment, improves the working efficiency and work quality of the shaking table, and ensures the smooth progress of the ore dressing work.

The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings are part of this application and are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention, but do not constitute an improper limitation of the present invention. Obviously, the drawings described below are only some embodiments. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

FIG1 is a schematic diagram of a system logic framework of the present invention;

Fig. 2 is a schematic diagram of the workflow of the present invention;

FIG3 is a schematic front view of a rocking table according to the present invention;

FIG4 is a schematic top view of a shaking table of the present invention;

FIG5 is a schematic diagram of establishing an algorithm model of the present invention;

FIG6 is a schematic diagram showing the comparison of the grades of the ore collected by the machine and manually on the first day of the first embodiment of the present invention;

7 is a schematic diagram showing a comparison of machine and manual ore grades on the second day of Example 1 of the present invention;

FIG8 is a schematic diagram showing a comparison of the grades of machine and manual ore collection on the third day of Example 1 of the present invention.

In the figure: 1-bed head; 2-ore feeding trough; 3-bed surface; 4-water feeding trough; 5-slope adjustment machine; 6-lubrication system; 7-bed bar; 8-motor; 9-base; 10-camera bracket; 11-camera.

It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but rather to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments.

DETAILED DESCRIPTION

In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The following embodiments are used to illustrate the present invention but are not used to limit the scope of the present invention.

In the description of the present invention, it should be noted that the directions or positional relationships indicated by terms such as "upper", "lower", "front", "back", "left", "right", "vertical", "inside" and "outside" are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

As shown in FIGS. 1 to 8 , a method for intelligent segmentation of mineral belts based on target detection of the present invention comprises the following steps:

Step 1: collect the data of the mineral belt on the shaking table, calibrate the features of the collected mineral belt information, train and evaluate the model on these data, and obtain the target detection model algorithm. The model training includes the following steps:

Step 101, model building, input x, through the first CNN, get the distribution parameters of multiple latent variables q(t); input a set of t values obtained according to the distribution of q(t), through the second CNN, get the distribution parameters of the final image p(x);

q _i (t _i )=N(m( _xi ,φ),diag(s ² (x _i ,φ)))

If data is only passed through parameters between two CNN models, it is basically similar to the common autocoder neural network model, and the formula is:

q _i (t _i )=N(m( _xi ,φ),diag(s ² (x _i ,φ)))

If s(x)＝0then

Step 102, prove that the calculation of the latent variable model is an unbiased process:

Step 103, model optimization, input x, through the first CNN, obtain the distribution parameters sd and mean of multiple latent variables q(t);

Through the standard Gaussian distribution p(z), we get a set of z values; then through z×sd+mean, we get a set of t values;

Input the t value and pass it through the second CNN to get the distribution parameters of the final image p(x);

The advantage of model optimization is that the standard distribution p(z) that does not need to be trained is stripped out; the sd and mean that need to be trained are only in the linear expression:

p(ε _i )＝N(0,I)

The first CNN model is used to propose latent variables, and whether it is an outlier can be determined based on the latent variables; by transforming the input latent variable parameter t, the second CNN model is used to obtain a new image.

Step 2, the software central platform sends instructions to the on-site camera to collect real-time data of the mineral belt on the shaking table surface, including mineral belt bandwidth, mineral belt color, and mineral belt location distribution characteristics, and transmits the collected data to the software central platform through the 4G gateway; the camera is installed above the shaking table through a camera bracket, and the camera bracket is installed on the base.

Step 3: The software central platform will perform corresponding algorithm processing on the transmitted bed surface mineral belt image to obtain a clear bed surface mineral belt image;

Step 4: The image processed by the algorithm is transmitted to the trained model. The on-site bed surface ore belt image transmitted to the platform by the camera is compared and identified according to the model training data, and then the identification image is output and the identification point is calibrated. Because the ore-dressing table is a slurry fed to the bed surface, and water is added, it is inevitable that there will be reflections. Therefore, the algorithm is improved to solve this problem. The identification points are connected by multiple points, and an extension line is made toward the end of the bed according to the trend of the connection line. The extension line is extended to the end of the bed.

Step 5: After the model algorithm calculates, determine whether the ore belt has changed compared to the previous round. If so, proceed to step 6; if not, proceed to step 7;

Step 6, the algorithm will give the moving direction and moving distance after precise calculation, and the moving distance is accurate to millimeters. The software central platform will issue instructions to the motion control system, and the control system drives the motor to drive the ore receiving plate to move to the specified position. After understanding the on-site ore dressing situation and the frequency of ore belt changes, the system software central platform sets 2min to control the camera to collect the ore belt image of the on-site shaking table bed surface once, and the algorithm calculates once. If the ore belt changes relative to the upper wheel, the software central platform will issue corresponding instructions, and then adjust the direction and distance of the docking ore plate.

Step 7: Keep the device in its original state.

The intelligent segmentation method of the ore belt is the core of the patent of this invention. Image processing also plays an important role in the whole intelligent segmentation method. It mainly processes the image information sent back from the site and is responsible for extracting the ore belt information of the shaking table bed surface, such as the ore belt width, ore belt color, ore belt location distribution and other features. The processing accuracy of the bed surface ore belt image directly affects the ore belt segmentation accuracy. Because the shaking table is in a reciprocating and bumpy motion during normal operation, the camera is installed above the end of the shaking table, but the camera is fixed. This will result in that even if the ore belt does not change, the bed surface is always in reciprocating motion, and the position of the ore belt boundary photo extracted by the camera will also change accordingly, causing certain errors in the detection results.

In order to solve the problem that the camera cannot be relatively still with the bed surface, a machine learning method is used to solve it. Based on the training data, a machine learning algorithm is developed, which applies the learned information to the unseen data to make predictions or other types of decisions. The pattern that explains the data better than the original data itself is called the feature in the data. In the early stage, the bed surface mineral belt data is collected, the model is built and initialized. The collected bed surface mineral belt information is calibrated, and then the model is trained and evaluated for these data to obtain the target detection model, and finally applied.

Before the image segmentation works normally, it is necessary to collect the information of the on-site shaking table ore belt, and then calibrate the features of the collected shaking table ore belt information, form a data set with the calibrated images, and then build a model for model training. The trained model is tested and then adjusted to make the model more effective. The computer and 4G gateway can be connected via a network cable or a SIM card can be used for communication. The 4G gateway and the camera are connected via a network cable.

When the intelligent system starts working, the system first sends a command to the wireless gateway, and the gateway sends the command signal to the camera. The camera collects the image of the ore belt on the shaking table surface, and transmits the collected image to the system control platform through the 4G gateway. The system has an image preprocessing module, which first preprocesses the collected shaking table ore belt information image, and then transmits the preprocessed shaking table ore belt image to the trained data model. The model compares, analyzes, calibrates and draws lines on the image information, divides the concentrate belt, sub-concentrate belt, intermediate ore belt and tailings belt one by one, and then outputs the result image. The system will issue instructions to the motor control system according to the changes in the ore belt, and make corresponding adjustments to the docking ore board to complete the indicators set by the beneficiation plant. The system can set the image acquisition frequency according to requirements to achieve the purpose of real-time monitoring.

The shaking table for ore dressing is a common equipment for sorting fine-grained ores. It is usually composed of three parts: bed surface, frame and transmission mechanism. In addition, there are flushing troughs, ore feeding troughs, etc. The entire bed surface is supported or suspended by the frame, and the bed surface is installed on the frame. A slope adjustment machine is installed at the bottom of the frame. The bed head is equipped with a lubrication system and an electric motor as a transmission mechanism, and the bed surface is equipped with bed bars. The bed head is connected to one end of the bed head. Shaking tables are common equipment in the mining industry. This time, the existing shaking table equipment is also used. No technical improvements have been made to it. Its specific connection structure and working principle will not be repeated here.

Embodiment 1

As shown in Figures 1 to 8, the intelligent segmentation method of the ore belt based on target detection described in this embodiment is implemented in the mineral processing center of Yunnan Yunxi Datun Tin Mine for intelligent shaking table application research. The workshop has a daily processing capacity of 2000t/d of oxide ore and more than 200 shaking tables; the daily processing capacity of sulfide ore is 4000t/d, and there are 288 shaking tables; it is used for roughing, scavenging and concentrating of metal tin ore. Due to the large number of shaking tables and the fast frequency of ore belt changes, a large number of inspection workers are required to conduct continuous inspections to adjust the position of the ore receiving plate to achieve the desired indicators of the concentrator, which greatly increases the labor force of the inspection workers and the operating costs of the enterprise.

The industrial test site was selected in the Datun Yunxi Sulfide Mine Section. The 3-18 shaking table in Section 1, Building 3 was used as the test shaking table. A 220V power supply was selected to power the equipment. The camera was installed above the tail of the shaking table at a height of 2m. The real-time image of the shaking table surface was collected and the image was transmitted and processed. Then the driving motor drove the ore receiving plate to move to the target position, realizing the fully automatic ore receiving of the 3-18 shaking table.

1. Real-time monitoring efficiency

Through the early collection of real-time data from the on-site shaking table, it was found that the system was set to a 3-minute time interval for each calculation based on the changes in the on-site ore belt. Therefore, every 3 minutes, the camera automatically collects real-time images and transmits them to the software platform. After calculation, the position of the docking ore plate is adjusted to meet the ore connection requirements.

2. Mineral belt image processing

Because the shaking table selected for installing this equipment is the shaking table of the sulfide ore roughing section, it is only necessary to separate the secondary concentrate and the medium ore according to the on-site requirements, connect the concentrate and the secondary concentrate together, and then proceed to the next step of re-selection. There is a big difference in color difference between the secondary concentrate belt and the medium ore belt. The medium ore belt is white in color, and the secondary concentrate belt is black in color. Because tens of thousands of pictures were collected in the early stage for feature calibration, four points were calibrated between the medium ore and the secondary concentrate in each picture, and then the model was trained. Now, according to the settings, the on-site shaking table is used for image acquisition, the image is simply processed and input into the model, and the segmentation map of the secondary concentrate belt and the medium ore belt is output.

3. Comparison of mineral processing indicators

In order to compare the difference in mineral processing indicators between intelligent shaking table equipment and manual operation, two adjacent shaking tables in the roughing operation section were selected for the test. Shaking table 3-17 was manually operated, and shaking table 3-18 was installed with intelligent equipment. The two shaking tables were fed by the same rod mill, with the same ore grade, the same ore feed size, and the same slurry concentration. Under the condition of ensuring that all conditions were the same, industrial test sampling was carried out for 3 days, with an interval of 1 hour between each sampling, and 13 groups of samples were collected. The sampling comparison data is shown in Table 1. The test comparison results are shown in Figures 6, 7, and 8.

Table 1 Comparison of concentrate grades of 3-17 and 3-18

From the Sn grade line graph of 3-day data sampling, we can see that: ① From the three line graphs, we can see that under the same conditions, the fluctuation range of Sn grade of machine-connected ore is much smaller than that of manual ore-connected ore, and the fluctuation difference of data on the first day is the most obvious; ② From the three line graphs, we can see that the Sn grade of machine-connected ore is more stable than that of manual ore-connected ore; ③ From the overall three line graphs, we can see that the Sn grade of machine-connected ore is higher than that of manual ore-connected ore.

In order to further prove that the Sn grade of machine-wound ore is higher than that of manual-wound ore and that the Sn grade of machine-wound ore is more stable than that of manual-wound ore, the three-day sampling data is calculated by the mean value and variance. The calculation formula is as follows:

The calculation results of the data of 3-17 manual shaking table and 3-18 intelligent shaking table are shown in Table 2:

Table 2 Comparison of calculation results of mean and variance of sampling data 3-17 and 3-18

The calculation results of the mean and variance of the three-day sampling data show that: ① The mean calculation results of the three-day sampling data are equal on the first day, and on the second and third days, the 3-18 machine ore collection is greater than the 3-17 manual ore collection; ② The variance calculation results of the three-day sampling data are all 3-18 machine ore collection is less than 3-17 manual ore collection. Therefore, it can be seen from the calculation results that the machine ore collection is higher and more stable than the manual ore collection.

4. Conclusion

Three consecutive days of industrial test sampling show that under the same conditions, the 3-18 machine ore collection is more stable than the 3-17 manual ore collection, and the Sn grade of the machine ore collection is also higher than that of manual ore collection. It can be seen that the stability and Sn grade are higher than those of manual ore collection, and can meet the production indicators of the concentrator. The intelligent monitoring equipment of the shaking table can replace manual work, monitor the changes of the ore belt on the shaking table surface and data transmission in real time, and realize the "unmanned" mode. The boundary point of the ore belt is accurately identified through machine learning, and the identification accuracy meets the use requirements.

The industrial test and research on the application of intelligent shaking table was carried out in the shaking table of the roughing section of sulfide ore in the Yunnan Datun Tin Mine Mineral Processing Center. The grade of concentrate received by the automatic receiving plate of the intelligent shaking table can meet the process requirements of the beneficiation plant. The industrial application of the intelligent shaking table beneficiation system can liberate manual labor, realize the intelligentization of shaking table beneficiation, improve the production level and enterprise benefits of shaking table beneficiation, and play a strong role in promoting the development of gravity separation equipment.

The above description is only a preferred embodiment of the present invention, and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with this patent can make some changes or modify the technical contents suggested above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the solution of the present invention.

Claims (6)

1. A method for intelligent segmentation of mineral belts based on target detection, characterized in that it comprises the following steps: Step 1: Collect the data of the mineral belt on the shaking table bed, perform feature calibration on the collected mineral belt information on the bed, perform model training and evaluation on these data, and obtain the target detection model algorithm. The model training includes the following steps: Step 101, model building, input x, and obtain the distribution parameters of multiple latent variables q(t) through the first CNN; input a set of t values obtained according to the distribution of q(t), and obtain the distribution parameters of the final image p(x) through the second CNN; The formula used to build the model is: ; ; ; Step 102, prove that the calculation of the latent variable model is unbiased, the formula used is ; Step 103, model optimization, input x, through the first CNN, obtain the distribution parameters sd and mean of multiple latent variables q(t); Through the standard Gaussian distribution p(z), we get a set of z values; then through z×sd+mean, we get a set of t values; Input the t value and pass it through the second CNN to get the distribution parameters of the final image p(x); The formula used for model optimization is: ; ; ; Step 2: The software central platform sends instructions to the on-site camera to collect data of the ore belt on the shaking table in real time, and transmits the collected data to the software central platform through the 4G gateway; Step 3: The software central platform will perform corresponding algorithm processing on the transmitted bed surface mineral belt image to obtain a clear bed surface mineral belt image; Step 4: The image processed by the algorithm is transmitted to the trained model, the model evaluates the image information, and then outputs the image calibrated by the system; Step 5: After the model algorithm is calculated, determine whether the ore belt has changed compared to the previous round. If so, proceed to step 6; if not, proceed to step 7; Step 6: The algorithm will obtain the moving direction and moving distance through precise calculation. The software central platform sends instructions on the moving direction and moving distance to the motion control system, and the control system drives the motor to drive the receiving plate to move to the specified position. Step 7: Keep the device in its original state. 2. According to the target detection-based intelligent segmentation method of the ore belt in claim 1, it is characterized in that the information collected in step 2 includes the width of the ore belt, the color of the ore belt, and the location distribution characteristics of the ore belt. 3. According to a target detection-based intelligent segmentation method for mineral belts as described in claim 1, it is characterized in that: the camera is installed above the shaking bed through a camera bracket, and the camera bracket is installed on the base. 4. According to the target detection-based intelligent segmentation method of the mineral belt in claim 1, it is characterized in that: the evaluation of the image information in step 4 includes comparing and identifying the on-site bed mineral belt image transmitted to the platform by the camera according to the model training data, and calibrating the identification points. 5. According to the target detection-based intelligent segmentation method of the mineral belt in claim 4, it is characterized in that: the model training also includes connecting multiple points of the identification points, and extending the line toward the end of the shaking bed according to the trend of the connection line, and the extension line is extended to the end of the bed. 6. According to the target detection-based intelligent segmentation method of the ore belt in claim 4, it is characterized in that: the software central platform is set to collect the picture of the ore belt on the on-site shaking table bed once every 2 minutes, and make corresponding adjustments to the ore receiving plate.