阅读说明：本技术 耐张线夹压接缺陷的x射线图像识别方法及装置 (X-ray image identification method and device for strain clamp crimping defects ) 是由 李鹏吾 刘荣海 郭新良 蔡晓斌 郑欣 杨迎春 王达达 沈龙 许宏伟 周静波 焦宗 于 2021-08-16 设计创作，主要内容包括：本发明涉及一种耐张线夹压接缺陷的X射线图像识别方法及装置,该方法包括：获取耐张线夹压的X射线图像；通过CenterNet算法将所述X射线图像中的压接部位用检测框框出；将框出的压接部位图像从X射线图像中切割出；通过RetinaNet算法再进行数据增强处理后的图像中对压接缺陷进行识别并标注。本申请通过CenterNet算法计算压接缺陷在压接部位图像中平均占比提高到3.8％,再进行切割,大幅降低检测难度。对压接部位样本进行数据增强,有效避免了图片数量少导致训练中出现过拟合现象。最后利用RetinaNet算法对缺陷部位进行检测,以达到对耐张线夹压接缺陷的快速准确识别。(The invention relates to an X-ray image identification method and device for strain clamp crimping defects, wherein the method comprises the following steps: acquiring an X-ray image of tension line clamping; framing out the crimping part in the X-ray image by using a detection frame through a CenterNet algorithm; cutting out the framed image of the pressed part from the X-ray image; and identifying and marking the crimping defects in the image after data enhancement processing is carried out through a RetinaNet algorithm. This application calculates the crimping defect through the CenterNet algorithm and improves to 3.8% on average in crimping position image, cuts again, reduces the detection degree of difficulty by a wide margin. Data enhancement is carried out on the sample of the crimping part, and the phenomenon of overfitting in training caused by small number of pictures is effectively avoided. And finally, detecting the defect part by utilizing a RetinaNet algorithm so as to quickly and accurately identify the crimping defect of the strain clamp.)

1. An X-ray image identification method for a tension clamp crimping defect is characterized by comprising the following steps:

acquiring an X-ray image of tension line clamping;

framing out the crimping part in the X-ray image by using a detection frame through a CenterNet algorithm;

cutting out the framed image of the pressed part from the X-ray image;

and identifying and marking the crimping defects in the image after data enhancement processing is carried out through a RetinaNet algorithm.

2. The method for identifying the X-ray image of the tension clamp crimping defect according to claim 1, wherein the step of framing out the crimping part in the X-ray image by using a detection frame through a CenterNet algorithm comprises the following steps:

building a first network structure based on the Centeret, wherein the first network structure comprises a backbone network and three Head networks, and the three Head networks are Head-1, Head-2 and Head-3 respectively;

inputting the X-ray image into the backbone network for feature extraction to generate a first feature map;

the Head-1 estimates the position of the central point of the crimping position in the first characteristic diagram, and coarsely positions the crimping position;

the Head-2 can finely adjust the position of a central point of the roughly positioned crimping part and accurately position the crimping part;

the Head-3 estimates the width and the height of the accurately positioned crimping position and determines the size of the crimping position;

and framing the crimping position by using a detection frame according to the accurate position and size of the crimping position, wherein the size of the detection frame corresponds to the size of the crimping position.

3. The method for identifying the X-ray image of the tension clamp crimping defect of claim 2, wherein ResNet-50 is selected as the main network.

4. The method for identifying the tension clamp crimping defect according to claim 1 or 2, wherein after the framed crimping site image is cut out from the X-ray image, the method further comprises:

and performing data enhancement processing on the cut press connection part image.

5. The method for identifying the X-ray image of the tension clamp crimping defect according to claim 4, wherein the data enhancement processing mode comprises the following steps:

adjusting contrast, changing brightness, image rotation.

6. The method for identifying the X-ray image of the tension clamp crimping defect according to claim 4, wherein the identification and marking of the crimping defect are carried out in the image after the data enhancement processing is carried out through a RetinaNet algorithm, and the method comprises the following steps:

building a second network structure based on RetinaNet, wherein the second network structure comprises a ResNet main network and an FPN network;

inputting the image subjected to the data enhancement processing into the ResNet backbone network for feature extraction to generate a second feature map;

performing feature fusion processing on the second feature map through an FPN network;

and predicting the image subjected to the feature fusion processing, identifying the type and the position of the compression joint defect of the compression joint part, and marking.

7. The method for identifying the X-ray image of the tension clamp crimping defect according to claim 6, wherein the ResNet main network comprises five residual modules in different layers, the image subjected to the data enhancement processing is input into the ResNet main network for feature extraction, and a second feature map is generated, and the method comprises the following steps:

and inputting the images subjected to the data enhancement processing into five residual modules of different layers to generate five second feature maps with different resolutions, and recording the second feature maps as C1-C5 in sequence.

8. The method for identifying the X-ray image of the tension clamp crimping defect of claim 7, wherein the second feature map is subjected to feature fusion processing through an FPN network, and the method comprises the following steps:

the FPN network performs feature fusion on the second feature map input of C1-C5 to form the second feature map input of C₃、C₄、C₅Corresponding fused feature map P with identical resolution₃、P₄、P₅；

To P₅Obtaining P by convolution operation with convolution kernel of 3 x 3 and step length of 2₆；

To P₆Obtaining P by convolution operation with convolution kernel of 3 x 3 and step length of 2₇。

9. The method for identifying the X-ray image of the tension clamp crimping defect of claim 8, wherein the second network structure further comprises a classification network and a regression network, the image subjected to the feature fusion processing is predicted, and the type and the position of the crimping defect at the crimping part are identified, and the method comprises the following steps:

for the fusion feature map P₃-P₇Respectively generating a preselection frame;

the classification network pair fuses the feature map P₃-P₇Performing convolution operation to obtain the probability that each preselected frame contains the crimping defects;

the regression network pair fuses the feature map P₃-P₇Performing convolution operation to obtain the offset of a real area of each preselection frame, which contains the crimping defects;

and carrying out coordinate transformation by selecting the maximum value of the probability of the crimping defects contained in the preselected frame and the offset of the real area to obtain the target category and the accurate coordinate value contained in the preselected frame.

10. The utility model provides a strain clamp crimping defect's X ray image recognition device which characterized in that, the device includes:

the acquisition module is used for acquiring an X-ray image pressed by a tension-tolerant line;

the framing module is used for framing out the compression joint part in the X-ray image by using a detection frame through a CenterNet algorithm;

a cutting module for cutting the framed crimping part image out of the X-ray image;

and the identification module is used for identifying and marking the crimping defects in the image after data enhancement processing is carried out through a RetinaNet algorithm.

Technical Field

The application relates to the technical field of image recognition, in particular to an X-ray image recognition method for strain clamp crimping defects.

Background

In the stringing construction, the connection of the overhead conductor is a key project, and simultaneously, the connection is a concealed project. The quality of overhead conductor crimping is very important, and the method has very important significance for ensuring the reliable operation of a line and ensuring safe power supply.

In the operation process of the overhead transmission line, the strain clamp is used as an important component in the transmission line, and not only needs to bear the conductive function of the overhead conductor, but also needs to bear the whole tension of the overhead conductor. Under the conditions of overhead conductor galloping, long-term breeze vibration and the like, the wire clamp is easy to break, so that electric power accidents are caused, and social safety is damaged.

The strain clamp consists of an aluminum sleeve and a steel anchor, and the aluminum pipe and the steel anchor and the aluminum pipe and the lead are pressed and formed by a hydraulic method to integrate the lead and the strain clamp. The crimping positions of the aluminum pipe and the steel anchor and the aluminum pipe and the lead are called crimping positions. When the crimping is formed, the crimping defects are generated at the crimping position due to the fact that workers do not operate normally.

The traditional crimping defect inspection and detection of the crimping type strain clamp are mainly completed manually by workers, so that the workload is high, and the misjudgment, the misjudgment omission and the like of the defects are easily caused. The requirement of modern mechanical manufacturing on precision is higher and higher, and the requirement of production development cannot be met by manual operation.

In recent years, some X-ray image detection systems with application value have come into play. The X-ray image detection is to irradiate the object to be detected with a uniform intensity of radiation and to make the transmitted radiation sensitive on the photographic film, and to detect the defect position, type, size and number of the object to be detected on the imaged film. The X-ray image inspection system utilizes management of film information by using a computer, and realizes defect inspection and the like by digital image processing. The method can greatly improve the detection efficiency, but the defect detection technology is not mature, the targeted detection research on the crimping tension-resistant wire clamp is few, and other defect detection systems mostly need a large amount of sample data for support, so that the existing detection method still has a plurality of problems.

Disclosure of Invention

The application provides an X-ray image identification method and device for strain clamp crimping defects, and aims to solve the problem that manual inspection and detection cannot meet the production development requirements more and more.

The technical scheme adopted by the application is as follows:

the invention provides an X-ray image identification method for a strain clamp crimping defect, which comprises the following steps:

acquiring an X-ray image of tension line clamping;

framing out the crimping part in the X-ray image by using a detection frame through a CenterNet algorithm;

cutting out the framed image of the pressed part from the X-ray image;

and identifying and marking the crimping defects in the image after data enhancement processing is carried out through a RetinaNet algorithm.

Further, framing out the crimping portion in the X-ray image by a detection frame through a centrnet algorithm includes:

building a first network structure based on the Centeret, wherein the first network structure comprises a backbone network and three Head networks, and the three Head networks are Head-1, Head-2 and Head-3 respectively;

inputting the X-ray image into the backbone network for feature extraction to generate a first feature map;

the Head-1 estimates the position of the central point of the crimping position in the first characteristic diagram, and coarsely positions the crimping position;

the Head-2 can finely adjust the position of a central point of the roughly positioned crimping part and accurately position the crimping part;

the Head-3 estimates the width and the height of the accurately positioned crimping position and determines the size of the crimping position;

and framing the crimping position by using a detection frame according to the accurate position and size of the crimping position, wherein the size of the detection frame corresponds to the size of the crimping position.

Further, ResNet-50 is selected as the backbone network.

Further, after cutting out the framed pressure contact image from the X-ray image, the method further includes: and performing data enhancement processing on the cut press connection part image.

Further, the data enhancement processing mode comprises: adjusting contrast, changing brightness, image rotation.

Further, identifying and labeling the crimping defects in the image after data enhancement processing is performed through a RetinaNet algorithm, including:

building a second network structure based on RetinaNet, wherein the second network structure comprises a ResNet main network and an FPN network;

inputting the image subjected to the data enhancement processing into the ResNet backbone network for feature extraction to generate a second feature map;

performing feature fusion processing on the second feature map through an FPN network;

and predicting the image subjected to the feature fusion processing, identifying the type and the position of the compression joint defect of the compression joint part, and marking.

Further, the ResNet backbone network includes five residual modules of different layers, and the image after the data enhancement processing is input to the ResNet backbone network for feature extraction to generate a second feature map, including:

and inputting the images subjected to the data enhancement processing into five residual modules of different layers to generate five second feature maps with different resolutions, and recording the second feature maps as C1-C5 in sequence.

Further, performing feature fusion processing on the second feature map through an FPN network, including:

the FPN network performs feature fusion on the second feature map input of C1-C5 to form the second feature map input of C₃、C₄、C₅Corresponding fused feature map P with identical resolution₃、P₄、P₅；

To P₅Obtaining P by convolution operation with convolution kernel of 3 x 3 and step length of 2₆；

To P₆Obtaining P by convolution operation with convolution kernel of 3 x 3 and step length of 2₇。

Further, the second network structure further includes a classification network and a regression network, and predicts the image after the feature fusion processing, and identifies the category and the position of the crimp defect at the crimp position, including:

for the fusion feature map P₃-P₇Respectively generating a preselection frame;

the classification network pair fuses the feature map P₃-P₇Performing convolution operation to obtain the probability that each preselected frame contains the crimping defects;

the regression network pair fuses the feature map P₃-P₇Performing convolution operation to obtain the offset of a real area of each preselection frame, which contains the crimping defects;

and carrying out coordinate transformation by selecting the maximum value of the probability of the crimping defects contained in the preselected frame and the offset of the real area to obtain the target category and the accurate coordinate value contained in the preselected frame.

Further, this application still discloses strain clamp crimping defect's X ray image recognition device, the device includes:

the acquisition module is used for acquiring an X-ray image pressed by a tension-tolerant line;

the framing module is used for framing out the compression joint part in the X-ray image by using a detection frame through a CenterNet algorithm;

a cutting module for cutting the framed crimping part image out of the X-ray image;

and the identification module is used for identifying and marking the crimping defects in the image after data enhancement processing is carried out through a RetinaNet algorithm.

The technical scheme of the application has the following beneficial effects:

the invention relates to an X-ray image identification method and device for strain clamp crimping defects. This application calculates the crimping defect through the CenterNet algorithm and improves to 3.8% on average in crimping position image, cuts again, reduces the detection degree of difficulty by a wide margin. Data enhancement is carried out on the sample of the crimping part, and the phenomenon of overfitting in training caused by small number of pictures is effectively avoided. And finally, detecting the defect part by utilizing a RetinaNet algorithm so as to quickly and accurately identify the crimping defect of the strain clamp.

Drawings

In order to more clearly explain the technical solution of the present application, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.

FIG. 1a is a crimp defect image (the overall crimp defect image in the rectangular frame);

FIG. 1b is a crimp defect image (crimp defect image in crimp defect crimp site in rectangular frame);

FIG. 2 is a CenterNet target detection framework;

fig. 3 shows a network structure of the RetinaNet algorithm.

Detailed Description

Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following examples do not represent all embodiments consistent with the present application. But merely as exemplifications of systems and methods consistent with certain aspects of the application, as recited in the claims.

The tension clamp X-ray image in the application has the following characteristics as shown in fig. 1, wherein a rectangular frame in fig. 1a is an integral medium-pressure welding defect image, and a rectangular frame in fig. 1b is a pressure welding defect image in a pressure welding part with pressure welding defects: (1) the average size of each picture is 1660 multiplied by 1037, the average size of the target frame of the defect part is 45 multiplied by 45, the ratio of the defect part in the whole image is about 0.11%, and the direct detection of the target is very difficult. (2) The large area of the collected X-ray image is a black background, and the direct detection efficiency is not high. Due to the above features, hierarchical detection strategies are employed herein. Firstly, roughly positioning a strain clamp crimping part in an X-ray image by using CenterNet, cutting the roughly positioned crimping part from the X-ray image by using a cutting algorithm, performing data enhancement processing on the cut crimping part image, and identifying and marking crimping defects in the image after the data enhancement processing is performed by using a RetinaNet algorithm. According to the method and the device, the average proportion of the crimping defects in the crimping part image is increased to 3.8% through calculation, and the detection difficulty is greatly reduced. In order to effectively avoid the phenomenon of overfitting in training caused by the small number of pictures, data enhancement is carried out on the sample of the compression joint part. And finally, detecting the defect part by utilizing a RetinaNet algorithm so as to quickly and accurately identify the crimping defect of the strain clamp.

The core of strain clamp crimping defect detection is the positioning of a crimping part and the detection of the defective part, and a target detection algorithm is needed for improving the detection efficiency and the detection precision. The CenterNet and RetinaNet algorithms are employed herein. For the whole detection model, the positioning of the crimping position is a premise, so that the crimping position is detected by adopting a CenterNet algorithm with higher detection efficiency. The RetinaNet algorithm has more advantages in accuracy, and is used for final defect part detection.

Specifically, according to a first aspect of the present invention, the present application provides an X-ray image identification method of a strain clamp crimping defect, the method comprising:

s01, acquiring X-ray images of the tension line nip.

And S02, framing the crimping part in the X-ray image by using a detection frame through a CenterNet algorithm.

In S02, framing the crimping portion with a detection frame according to the precise position and size of the crimping portion, where the size of the detection frame corresponds to the size of the crimping portion, and the method specifically includes:

building a first network structure based on the Centernet, wherein the first network structure comprises a main network and three Head networks, the three Head networks are Head-1, Head-2 and Head-3 respectively, and as the scale of the strain clamp X-ray image data set is small, ResNet-50 is selected as the main network of the method in order to avoid the over-fitting phenomenon in network training;

inputting the X-ray image into the backbone network for feature extraction to generate a first feature map, wherein the backbone network in the CenterNet is used for extracting image features of an input image, the input image is the X-ray image of 512X 3, and a feature map of 128X 64 is output;

the Head-1 estimates the position of the central point of the crimping part in the first characteristic diagram, roughly positions the crimping part, and selects the Focal local to calculate the central point Loss L because the Focal local can better solve the unbalance problem of positive and negative samples_KAs shown in formula (1):

in the formulaTo predict pixel (x, y), i.e., the predicted position confidence,alpha and beta are hyper-parameters in the Focal local, and are respectively set to be 2 and 4;

the Head-2 can finely adjust the position of a center point of the roughly positioned crimping part and accurately position the crimping part, wherein the offset loss L of the fine adjustment center point of the crimping part_offSuch as (2)

In the formulaThe central point of the network prediction is biased, R is a downsampling multiple, and the value in the text is 4;to predict point correspondence profilesMarking information of the coordinate offset;

the Head-3 estimates the width and height of the accurately positioned crimping position, determines the size of the crimping position, and has a frame width and height loss function as shown in formula (3):

(3) in the formulaTo predict frame width height, i.e.S_KIn order to label the width and the height,

finally, the three loss functions are weighted and summed to obtain the overall loss function L_detAs shown in formula (4):

L_det＝L_K+λ_sizeL_size+λ_offL_off (4)

in the formula of_sizeAnd λ_offAs weight coefficient for balancing three loss functions, setting lambda in training process_size0.1 and λ_off＝1。

The basic flow of the cenernet for detecting the crimping position of the strain clamp is as follows:

(1) the size of the X-ray picture is adjusted to 512X 512 and then the X-ray picture is input into the detection network.

(2) As the scale of the strain clamp X-ray image data set is small, in order to avoid the overfitting phenomenon in network training, ResNet-50 is selected as the main network of the method.

(3) And performing feature extraction on the image to generate a first feature map. Extracting peak points of each class on the thermodynamic diagram: all response points on the thermodynamic diagram are compared with the 8 adjacent points to which they are connected, and if the point response value is greater than or equal to its eight adjacent point values, we retain all the first 100 peak points that meet the previous requirement. Finally, completing the regression process of the key points to the position frame through the formula (5), namely completing the detection of the crimping position of the strain clamp.

In the formula (5)Is the center point coordinate;is the center point offset;is the target frame width and height; non-maximum suppression (NMS) is not needed in the whole process, and the calculation amount can be effectively reduced.

S03, cutting the framed pressure contact image from the X-ray image.

Position information is calibrated by utilizing a prediction frame of the CenterNet, a strain clamp crimping position can be cut out from the strain clamp crimping position, and the cut strain clamp crimping position image can still have high resolution.

And S04, performing data enhancement processing on the cut press contact image, wherein the data enhancement processing comprises contrast adjustment, brightness change and image rotation.

Through cutting and data enhancement stage processing, a high-resolution strain clamp crimping position image is provided for the RetinaNet algorithm.

And S05, identifying and marking the crimping defects in the image after data enhancement processing is carried out through a RetinaNet algorithm.

The RetinaNet algorithm is a stage target detection algorithm which takes a backbone network ResNet and a Feature Pyramid (FPN) as main frames, and two Full Convolution (FCN) sub-Networks are responsible for classification and regression. The biggest contribution of the method is to propose the Focal local for solving the problem of class imbalance, so that RetinaNet is created. The accuracy of the method surpasses the target detection network of fast-RCNN of a classic two-stage detection algorithm.

The RetinaNet algorithm network structure is shown in FIG. 3, wherein an image is firstly subjected to feature extraction by a ResNet backbone network, and feature maps with different resolutions are output through 5 residual modules at different layers, and are sequentially marked as C₁-C₅. Subsequent feature fusion by FPN to form C₃、C₄、C₅Fused feature map P with identical resolution₃、P₄、P₅. Convolution operations with a convolution kernel of 3 × 3 and step size of 2 are applied to P5, resulting in P6. The same convolution operation is carried out on P6 in the same way to obtain P₇。

After feature fusion, for each fused feature map (P)₃-P₇) And respectively generating a pre-selection frame. Then, two sub-networks, i.e., a classification network and a regression network, are used for the prediction of the category and location. The classification network obtains the probability that each pre-selection frame contains the target object through a series of convolution operations on the fusion characteristic graph; similarly, the regression network obtains the offset of the real region containing the target for each preselected box through a series of convolution operations on the fused feature map. And finally, carrying out coordinate transformation by selecting the maximum value of the probability that the preselected frame contains the target object and the offset of the real area to obtain the target category and the accurate coordinate value contained in the preselected frame.

S05 specifically includes:

building a second network structure based on RetinaNet, wherein the second network structure comprises a ResNet main network and an FPN network;

inputting the image subjected to the data enhancement processing into the ResNet backbone network for feature extraction to generate a second feature map;

performing feature fusion processing on the second feature map through an FPN network;

and predicting the image subjected to the feature fusion processing, identifying the type and the position of the compression joint defect of the compression joint part, and marking.

The ResNet main network comprises five residual modules of different layers, the image subjected to the data enhancement processing is input into the ResNet main network for feature extraction, and a second feature map is generated, wherein the second feature map comprises:

and inputting the images subjected to the data enhancement processing into five residual modules of different layers to generate five second feature maps with different resolutions, and recording the second feature maps as C1-C5 in sequence.

Wherein, the feature fusion processing is carried out on the second feature map through the FPN network, and the processing method comprises the following steps:

the FPN network performs feature fusion on the second feature map input of C1-C5 to form the second feature map input of C₃、C₄、C₅Corresponding fused feature map P with identical resolution₃、P₄、P₅；

To P₅Obtaining P by convolution operation with convolution kernel of 3 x 3 and step length of 2₆；

To P₆Obtaining P by convolution operation with convolution kernel of 3 x 3 and step length of 2₇。

Further, the second network structure further includes a classification network and a regression network, and predicts the image after the feature fusion processing, and identifies the category and the position of the crimp defect at the crimp position, including:

for the fusion feature map P₃-P₇Respectively generating a preselection frame;

the classification network pair fuses the feature map P₃-P₇Performing convolution operation to obtain the probability that each preselected frame contains the crimping defects;

the regression network pair fuses the feature map P₃-P₇Performing convolution operation to obtain the offset of a real area of each preselection frame, which contains the crimping defects;

and carrying out coordinate transformation by selecting the maximum value of the probability of the crimping defects contained in the preselected frame and the offset of the real area to obtain the target category and the accurate coordinate value contained in the preselected frame.

And the data set used for training needs to be manually labeled, and the labeled data set can be used for identifying defects of other input pictures only after the model is learned and subsequently detected. The real area is the manually marked area, the offset of the real area is used for calculating a loss function in the model training process, and the loss function is reduced along with the increase of the number of model training rounds, so that the position of a frame predicted by the model is closer to the frame marked by the model.

Wherein, the loss function in the RetinaNet algorithm is defined as follows:

f_FL(p_t)＝-α_t(1-p_t)^γlogp_t (6)

wherein p is_tIs the confidence of the training sample class, α_tIs a balance factor; gamma is a regulating factor and takes the value of [0, 5%]In the meantime. Alpha is alpha_tAnd gamma is a fixed value, and the values are respectively 0.25 and 2.0 according to the experience of the author of the original text.

Detecting the crimping defects by utilizing a RetinaNet algorithm, wherein the specific process is as follows:

(1) the size of the cut press contact part image is adjusted to 512 x 512, and then the image is input into the detection network.

(2) The network frame adopts ResNet50+ FPN, the compressed part image generates a characteristic diagram through a main network, and P is obtained after the characteristics are fused₃-P₇The sizes are 64 × 64, 32 × 32, 16 × 16, 8 × 8 and 4 × 4, and the defect type and position of the pressure welding part are identified by predicting the feature fusion maps in different scales.

(3) Because the CenterNet algorithm-level detection determines the crimping part position with the crimping defect, the percentage of the crimping defect in the crimping part is greatly increased, and P in RetinaNet can be selectively removed₇Layer to improve detection efficiency.

According to the second aspect of the invention, an X-ray image recognition device for the strain clamp crimping defect is also disclosed, the device comprises:

the acquisition module is used for acquiring an X-ray image pressed by a tension-tolerant line;

the framing module is used for framing out the compression joint part in the X-ray image by using a detection frame through a CenterNet algorithm;

a cutting module for cutting the framed crimping part image out of the X-ray image;

and the identification module is used for identifying and marking the crimping defects in the image after data enhancement processing is carried out through a RetinaNet algorithm.

The embodiments provided in the present application are only a few examples of the general concept of the present application, and do not limit the scope of the present application. Any other embodiments extended according to the scheme of the present application without inventive efforts will be within the scope of protection of the present application for a person skilled in the art.