1. A method for detecting and counting carborundum particles is characterized by comprising the following steps:

acquiring a training picture;

manually marking the training picture, and marking the carborundum particles in the training picture as single or agglomerated carborundum particles;

establishing a detection and counting network model;

training the detection and counting network model by using the training picture after artificial marking to obtain a trained detection and counting network model;

acquiring a carborundum line picture to be detected and counted;

and determining the types of the diamond grains in the diamond line graph sheet to be detected and counted as single or aggregated by using the trained detection and counting network model, and counting the diamond grains in each type.

2. The method of claim 1, wherein the manually labeling the training picture and marking the emery particles in the training picture as single or agglomerated comprises:

and manually labeling the training picture by using a LabelImg tool, and marking the carborundum particles in the training picture as single or agglomerated carborundum particles.

3. The method of claim 1, wherein the establishing a detection and counting network model comprises:

and establishing a YOLOv3 network model as the detection and counting network model.

4. The method as claimed in claim 3, wherein the training of the detection and counting network model by using the training picture after artificial labeling to obtain the trained detection and counting network model comprises:

training the YOLOv3 network model by using the artificially marked training pictures, and modifying the connection weight of the YOLOv3 network model to obtain a new YOLOv3 network model;

and testing the new YOLOv3 network model, finishing training if the precision requirement is met, obtaining the trained detection and counting network model, and otherwise, training the new YOLOv3 network model again until the precision requirement is met.

5. The system for applying the method for detecting and counting diamond grains according to claim 1, comprising:

the image acquisition device is used for acquiring a training picture and a carborundum line picture to be detected and counted;

the data marking module is used for manually marking the training pictures and marking the carborundum particles in the training pictures as single or agglomerated carborundum particles;

the model training module is used for establishing a detection and counting network model, and training the detection and counting network model by utilizing the training pictures after artificial marking to obtain a trained detection and counting network model;

and the detection and counting module is used for identifying the categories of the diamond grains in the diamond line graph sheet to be detected and counted by using the trained detection and counting network model and counting the diamond grains in each category.

6. The system of claim 5, wherein the image capturing device comprises: industrial personal computers and industrial cameras;

the industrial personal computer is used for controlling the industrial camera to acquire the training picture and the diamond wire picture to be detected and counted.

7. The system of claim 6, wherein the image capturing device further comprises: a coaxial point light source;

the coaxial point light source is used for providing illumination for the diamond wire.

8. The system of claim 7, wherein the coaxial point light source is an LED light source.

Background

The diamond wire is widely used for cutting hard and brittle crystal materials, such as monocrystalline silicon, polycrystalline silicon, sapphire, hardened silicon and the like, and the quantity of diamond grains on the diamond wire directly influences the cutting quality, so that the quantity of diamond grains on the diamond wire needs to be determined before cutting.

It is now common for many companies to determine the amount of corundum by manual counting. Because the number of carborundum particles on the carborundum wire is large, manual counting is easy to fatigue and brings counting errors. Moreover, the manual counting efficiency is extremely low, and the labor cost is very high. Very few companies employ the method of machine vision counting, but conventional machine vision counting has limitations.

The diamond grains on the diamond wire exist singly, but a plurality of grains are clustered together, and the traditional machine vision counting can not distinguish the clustered diamond grains when detecting the grains, so that the counting is not accurate enough.

Disclosure of Invention

The embodiment of the invention provides a method and a system for detecting and counting carborundum particles, which are used for solving the problem of inaccurate counting caused by the fact that machine vision detection technology in the prior art cannot distinguish agglomerated carborundum particles.

In one aspect, an embodiment of the present invention provides a method for detecting and counting diamond grains, including:

acquiring a training picture;

manually marking the training picture, and marking the carborundum particles in the training picture as single or agglomerated carborundum particles;

establishing a detection and counting network model;

training the detection and counting network model by using the training picture after artificial marking to obtain a trained detection and counting network model;

acquiring a carborundum line picture to be detected and counted;

and determining the types of the diamond grains in the diamond line graph sheet to be detected and counted as single or aggregated by using the trained detection and counting network model, and counting the diamond grains in each type.

In one possible implementation, the manually labeling the training picture, and the marking of the diamond grains in the training picture as single or agglomerated may include: and manually marking the training picture by using a LabelImg tool, and marking the carborundum particles in the training picture as single or agglomerated.

In one possible implementation, establishing the detection and counting network model may include: and establishing a YOLOv3 network model as a detection and counting network model.

In a possible implementation manner, training the detection and counting network model by using the training picture after the artificial marking to obtain a trained detection and counting network model may include: training the YOLOv3 network model by using the artificially marked training pictures, and modifying the connection weight of the YOLOv3 network model to obtain a new YOLOv3 network model; and testing the new YOLOv3 network model, finishing training if the precision requirement is met, obtaining the trained detection and counting network model, or training the new YOLOv3 network model again until the precision requirement is met.

In another aspect, an embodiment of the present invention provides a system for detecting and counting diamond grains, which may include:

the image acquisition device is used for acquiring a training picture and a carborundum line picture to be detected and counted;

the data marking module is used for manually marking the training pictures and marking the carborundum particles in the training pictures as single or agglomerated carborundum particles;

the model training module is used for establishing a detection and counting network model, and training the detection and counting network model by utilizing the training pictures after artificial marking to obtain the trained detection and counting network model;

and the detection and counting module is used for identifying the types of the diamond grains in the diamond line graph sheet to be detected and counted by using the trained detection and counting network model and counting the diamond grains in each type.

In one possible implementation, the image capturing apparatus may include: industrial personal computers and industrial cameras; the industrial personal computer is used for controlling the industrial camera to obtain training pictures and diamond wire pictures to be detected and counted.

In one possible implementation, the image capturing apparatus may further include: a coaxial point light source; the coaxial point light source is used for providing illumination for the diamond wire.

In one possible implementation, the coaxial point light sources are LED light sources.

The method and the system for detecting and counting the carborundum particles have the following advantages that:

the method can detect whether the carborundum particles are single or aggregated in real time, and respectively display the number of the carborundum particles in each category so as to judge the quality of the carborundum wires.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow chart of a method for detecting and counting diamond grains according to an embodiment of the present invention;

fig. 2 is a schematic diagram of a system for detecting and counting diamond grains according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Fig. 1 is a flowchart of a method for detecting and counting diamond grains according to an embodiment of the present invention. The invention provides a method for detecting and counting carborundum particles, which comprises the following steps:

and S100, acquiring a training picture.

Illustratively, the training picture can be obtained by taking a picture of a large number of diamond grains by using an image acquisition system, when the picture is taken, a plurality of pictures of different shapes and containing diamond grains with different numbers need to be acquired, and each shape and diamond grain with different numbers need to be acquired in different directions, so as to acquire the diamond grains at each position, and improve the accuracy of subsequent detection and counting of the diamond grains. Meanwhile, when the picture is taken, the good illumination environment needs to be ensured, each carborundum particle can be cleaned and identified in the picture, and when the picture is taken for multiple times, the consistency of the illumination environment and the background needs to be ensured.

S101, manually marking the training pictures, and marking the carborundum particles in the training pictures as single or agglomerated carborundum particles.

Illustratively, the purpose of manual labeling is to label the category of each diamond grain in each training picture, so that subsequent model training is performed on the basis of the category labels, and the purpose that the trained model can be accurately identified is achieved.

And S102, establishing a detection and counting network model.

Illustratively, an Artificial Neural Network (ANN), a Convolutional Neural Network (CNN), a Radial Basis Function (RBF) Neural Network, a Back Propagation (BP) Neural Network, a yolo (you Look Only one) Network, or the like may be employed.

S103, training the detection and counting network model by using the training picture after the artificial marking to obtain the trained detection and counting network model.

Illustratively, since the marked training picture contains the image information of the diamond grains and the corresponding categories, when the network model is trained, the network model can gradually obtain the features of the diamond grains belonging to a single category and the features of the diamond grains belonging to an agglomeration category, and further adaptively adjust some parameters in the network model, so that the trained network model can determine whether the categories of the diamond grains in the picture are individuals or agglomerates.

And S104, acquiring a diamond wire picture to be detected and counted.

Illustratively, the picture of the diamond wire to be detected and counted can be obtained by shooting a picture of the diamond wire to be detected by an image acquisition system, and when the picture is shot, pictures of the diamond wire to be detected and counted in different directions need to be acquired, so that the diamond grains at each position are acquired, and the accuracy of subsequent detection and counting of the diamond grains is improved. Meanwhile, when the picture is taken, the good illumination environment needs to be ensured, each carborundum particle can be cleaned and identified in the picture, and when the picture is taken for multiple times, the consistency of the illumination environment and the background needs to be ensured.

S105, determining the types of the diamond grains in the diamond line graph sheet to be detected and counted to be single or aggregated by using the trained detection and counting network model, and counting the diamond grains in each type.

Illustratively, the trained detection and counting network model has been adapted to different types of diamond grains, so that after the diamond wire picture to be detected and counted is input into the trained detection and counting network model, it can obtain a result reflecting the type of each diamond grain. After the classification result is obtained, the number of the carborundum particles with the single classification and the number of the agglomerated carborundum particles with the agglomerated classification can be counted respectively.

In a possible embodiment, S101, manually labeling the training picture, and marking the diamond grains in the training picture as single or agglomerated, includes: and manually marking the training picture by using a LabelImg tool, and marking the carborundum particles in the training picture as single or agglomerated.

Exemplarily, LabelImg is a graphical image annotation tool written in Python and using Qt as its graphical interface. The annotations added on the pictures are saved as XML files in paschalloc format.

In one possible embodiment, S102, building a detection and counting network model includes: and establishing a YOLOv3 network model as a detection and counting network model.

Illustratively, YOLOV3 is a third version of the YOLO series target detection algorithm, and compared with V2 and V1 algorithms, the V3 algorithm has significantly improved detection accuracy on small targets. The Yolov3 backbone network is Darknet-53, which automatically generates 9 anchors by using a K-means clustering algorithm and predicts 3 boxes with different scales by using a pyramid-like network concept, so that the detection effect on small targets such as carborundum particles is better.

In a possible embodiment, S103, training the detection and counting network model by using the training picture after the artificial labeling, and obtaining a trained detection and counting network model includes: training the YOLOv3 network model by using the artificially marked training pictures, and modifying the connection weight of the YOLOv3 network model to obtain a new YOLOv3 network model; and testing the new YOLOv3 network model, finishing training if the precision requirement is met, obtaining the trained detection and counting network model, or training the new YOLOv3 network model again until the precision requirement is met.

Illustratively, the picture used in the test is a test picture, and the test picture and the training picture are obtained simultaneously. When the training pictures are obtained, a large number of pictures can be shot at the same time, and then all the pictures are randomly divided into the training pictures and the testing pictures according to a certain proportion.

The invention also provides a system for detecting and counting carborundum particles, as shown in figure 2, comprising:

the image acquisition device is used for acquiring a training picture and a carborundum line picture to be detected and counted;

the data marking module is used for manually marking the training pictures and marking the carborundum particles in the training pictures as single or agglomerated carborundum particles;

the model training module is used for establishing a detection and counting network model, and training the detection and counting network model by utilizing the training pictures after artificial marking to obtain the trained detection and counting network model;

and the detection and counting module is used for identifying the types of the diamond grains in the diamond line graph sheet to be detected and counted by using the trained detection and counting network model and counting the diamond grains in each type.

In a possible embodiment, the image acquisition device comprises: industrial personal computers and industrial cameras; the industrial personal computer is used for controlling the industrial camera to obtain training pictures and diamond wire pictures to be detected and counted.

For example, the industrial personal computer may use a PLC (Programmable Logic Component), an MPU (Micro Processing Unit), a single chip microcomputer, or the like. Industrial cameras have a mating lens.

In a possible embodiment, the image capturing device further comprises: a coaxial point light source; the coaxial point light source is used for providing illumination for the diamond wire.

Exemplarily, when a picture is taken by using the image acquisition device, it is necessary to ensure that the coaxial point light source is located at a fixed position of a shooting field, so that the illumination environment is fixed and unchanged, and the uniformity of the processing result is improved. And the number of the coaxial point light sources can be one or more.

In one possible embodiment, the coaxial point light sources are LED light sources.

For example, LEDs, i.e., light emitting diodes, have the characteristics of low power consumption and high brightness, and are widely used in lighting devices.

While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.