Terminal sequence consistency verification method for small diameter abreast optical fibers based on computer vision

Abstract

In a kind of precision industrial equipment, small diameter abreast optical fibers are used for high-speed communication among functional nodes. The arrangement order at both terminals of the abreast optical fibers need to comply with communication protocols. In this paper, we propose an automatic terminal sequence consistency verification method based on computer vision. The Hue Saturation Value (HSV) color space is used for improving the image feature extraction capability. An abreast optical fiber sequence dictionary which converts the protocol logic into an input-output mapping table is provided to follow protocol confidentiality and improve inspecting speed. A light control baffle position adaptive algorithm is designed for improving the accuracy of optical fiber incident light control. The experimental results show that the method can achieve the conductivity inspection of 1 optical fiber every 50 seconds, and the inspection accuracy is over 96.5%, which generally improves the inspection efficiency by 45% compared with manual inspection.

Highlights

• •An open dictionary data structure for mapping the sequence relationship between terminals is proposed, which reduces the complexity of regularized operating procedure algorithms, and follows the confidentiality requirements of communication protocols.
• •Target recognition methods are implemented based on the HSV color space, which enhances the accuracy of feature extraction for images consisting of translucent turbid targets.
• •A baffle position adaptive algorithm is proposed, which overcomes the interference of core distance errors between optical fibers, and improves the fault tolerance of the method.

1. Introduction

In 1966, British-Chinese scientist Charles Kao proposed the method of optical fiber communication [1]. Since then, optical communication technology has continued to evolve. The invention of optical fibers has sparked a revolution in communication technology [2]. Optical fiber communication speeds have reached the Tb/s level and have very high reliability [3]. Optical fibers are not only used in general communication fields like 5G, but also in specialized communication fields for internal communication in many high-precision devices [4].

In a high-precision equipment, small diameter abreast optical fibers are used as high-speed and reliable communication channels among the function modules. At both terminals of the optical fibers, the order of arrangement is determined according to the requirements of the communication protocols between nodes. The order of arrangement poses challenges to the production and inspection of the optical fibers.

The abreast optical fibers have small outer diameters, typically around 0.5 mm, and the degree of abreast is relatively large, normally 12, 36, 64, etc. These make a certain error rate during the order arranging work which is executed manually. There must be a quality inspection before the abreast optical fibers are installed on the equipment. The quality inspection lights the abreast optical fibers sequentially at the input terminals and then observes whether their output sequence matches the protocol requirements, which is currently done manually.

There are obstacles to the efficiency and accuracy of the quality inspection. The optical fibers are translucent and small in diameter. The connectors on the optical fiber terminals are transparent. The difference between different communication protocols is not significant. The above obstacles make the order arranging work error-prone and the quality inspection inefficient. The workers have to repeatedly adjust the position of the baffle and the angle of the input light during the inspection process for better quality output patterns. And also, it is time-consuming to train experienced inspection operators, which usually costs 6-8 months. Therefore, there is an urgent need for a fast, accurate, and less manual inspection method.

Computer vision (CV) technology can simulate the human visual system, extract information from images or videos, and analyze and process it [5]. The application of computer vision technology in industrial inspection is becoming increasingly widespread, which can effectively solve the shortcomings of traditional inspection methods, improve detection efficiency and accuracy [6].

The inspection methods based on computer vision can be mainly divided into traditional vision based methods and deep learning based methods [7]. Traditional vision based methods perform better in the scenarios with easier quantified image features, relatively fixed detection process, requirements for lower computational complexity and better solvability. Deep learning based methods are more suitable for scenarios with difficult quantified image features, and also they have better generalization [8].

By comparing the characteristics of the abreast optical fibers quality inspection with those of the two computer vision methods, the optical fibers quality inspection is more suitable for the traditional vision based methods.

OpenCV is a set of open-source libraries of commonly used image processing [9]. With OpenCV, stable and high-performance quality inspection solutions using traditional vision based methods can be obtained, which are widely used in industrial inspection [10].

Wang et al. proposed an edge inspection method for solar panel defects based on OpenCV, which incorporates an adaptive threshold algorithm for better process results [11]. Wang et al. proposes a visual inspection algorithm based on OpenCV to address the issue of bubbles in polarizers during attachment in OLED manufacturing. Comparative experiments demonstrate that the proposed algorithm can achieve reliable detection results even under uneven illumination and severe reflection conditions [12].

Hanbay et al. proposed a real-time defect inspection method for industrial fabric production based on OpenCV, which can process 55 images in 1 second [13]. Vladimir et al. proposed a fabric defect inspection and classification algorithm base on OpenCV. Experimental results showed that its defect inspection success rate was 95%, and it was 50% faster than the human eye in fabric density calculation [14].

Pathinettampadian et al. provided a 3D printing quality monitoring system based on open-source computer vision. The printing defects that occur in the 3D printed part could be identified immediately through contours drawn over the defective part. The user could take respective action to control the defect formation [15]. Hicham et al. proposed a non-contact real-time defect detection method with the OpenCV library and a Raspberry Pi 4 board. The experimental results show that the defect on the part can be effectively inspected, located and recognized [16].

Raihan et al. proposed a printed circuit board (PCB) defect inspection solution based on OpenCV, which can quickly inspect PCBs and identify defective parts [17]. To address the inspection of low-contrast scratches on PCBs, Shen designed an inspection method based on OpenCV. Experiments show that this method can successfully detect low-contrast scratches on PCBs [18].

Based on the exploration of OpenCV-based traditional computer vision based methods [19] and the standard abreast optical fiber inspection operating procedure, we propose a terminal sequence consistency verification method for small diameter abreast optical fibers based on computer vision. Our method uses a stepper motor to drive the baffle to control the light transmission sequence of the optical fibers and an industrial camera to acquire the output patterns. Then, we use CV algorithms to extract image features and regularized operating procedure algorithms to verify the correctness of the terminal sequence.

The main contributions of this paper are as follows:

• 1.We propose an open dictionary data structure stored in Comma-Separated Values (CSV) format for mapping the sequence relationship between terminals, which reduces the complexity of regularized operating procedure algorithms, improves adaptation speed for switching between different protocols, and meets customer requirements for confidentiality of communication protocols.
• 2.We implement target recognition and noise filtering methods based on the HSV color space for images that consists of translucent turbid targets, which enhances the accuracy of feature extraction.
• 3.We provide an adaptive algorithm for the position adjustment of the baffle, which overcomes the interference of core distance errors between optical fibers, and improves the fault tolerance of the method.

The subsequent chapters are as follows. Section 2 introduces our proposed small diameter abreast optical fiber terminal sequence consistency verification method based on CV. Section 3 proposes the experiments and results. Section 4 is the conclusion and discussion.

2. Consistency verification method for the terminal sequence of small diameter abreast optical fibers

After thoroughly studying the existing consistency verification operating procedures, we propose our automatic inspection method, which consists of industrial camera, light source controlled by stepping motor and computer vision algorithms. Our method minimizes manual involvement and greatly improves inspection efficiency.

2.1. Overview of the terminal sequence consistency verification method

The overview of our proposed terminal sequence consistency verification method for small diameter abreast optical fibers based on computer vision is shown in Fig. 1.

Figure 1.

Overview of the terminal sequence consistency verification method for small diameter abreast optical fibers based on computer vision.

Our proposed method is implemented through 5 parts. (1) Information Collection. (2) Image Processing. (3) Baffle Position Adjustment. (4) Consistency Verification. (5) User Interaction.

The workflow is as follows:

• 1.The baffle blocks all the input terminals of the optical fibers, so that the light cannot get through the optical fibers. An image of the out terminals without lighting is captured by the industrial camera, which is taken as the baseline image during the inspection.
• 2.The baffle moves one optical fiber diameter distance, allowing the input terminals sequentially incident light.
• 3.The corresponding image of the output terminals partially lighting is captured. The information collection is completed.
• 4.The pattern of the light transmission is obtained by the image processing algorithm.
• 5.Search the input / output mapping dictionary according to the number of input optical fibers for the output terminal pattern.
• 6.Compare the number of lighting terminals in the output image and the number of lighting terminals in the searched pattern. If they are consistent, enable the consistence verification; otherwise, enable the baffle position adjustment to get the correct number of lighting terminals in the output image.
• 7.Verify the consistence between the lighting patterns of corresponding input terminals and output terminals.
• 8.Output the comparison result through user interaction.
• 9.Repeat the above steps 2 - 8 until all the input terminals are lightened.

2.2. Information collection

The principle of the information collection is shown in Fig. 2, which mainly consists of the parallel light source, the abreast optical fibers, the industrial camera, and the baffle.

Figure 2.

The principle of the information collection.

The baffle is driven by a step motor to move forward or backward, which controls the number of optical fibers receiving the parallel light. The industrial camera captures the light transmission of the output terminals for image processing step.

2.3. Image processing

Image processing is the difficulty of the consistency verification. Because the connectors are transparent, the images of output terminals are sensitive to ambient light. The small diameter of optical fibers makes feature extraction at the edges of the optical fibers susceptible to image noise. The number of images is small and not suitable for deep learning methods [20],[21].

In response to the above difficulties, we adopt OpenCV-based libraries, combine morphological operations [22] and filtering algorithms [23] to reduce noise and minimize interference in order to improve the quality of image feature extraction, as shown in Fig. 3.

Figure 3.

The procedure of image processing.

The optical fibers occupy a very small area in the original image, and will be overwhelmed by the background while processing the image in Red Green Blue (RGB) color space. Therefore, after many attempts, we transform the color space from RGB to HSV [24]. In HSV, extraction and control of different colors, color depth, and brightness can be achieved through 3 channels (Hue, Saturation, Value) and 6 thresholds ($(H_{min},H_{max})$, $(S_{min},S_{max})$, $(V_{min},V_{max})$). Image segmentation or specific area detection can be achieved by setting thresholds for different channels. With HSV, we can track the optical fibers easier according to their color characteristics.

The threshold values are sensitive to the environment, such as the intensity and color temperature of the light source, the reflection of light by the environment. We provide default values according to our experiment environment and directions on experimental environment arrangement for the users.

In the feature extraction, we first binarize the HSV converted image with the threshold setting method for the binarization [25], and then extract the optical fiber outlines based on the binarized image. In the binarized images, the optical fibers account for a small proportion, coupled with noise interference, the quality of optical fiber outlines extraction based on the operators, such as Sobel and Canny [26], is low. Therefore, we use connected components methods for the feature extraction [27], which brings two benefits: (1) The optical fibers can be located by the center of the components. (2) Parts of the noise interference can be filtered according to the area of the components.

There is noise in the images of output terminals, due to the transparency and uneven end face of the connectors. They cannot be filtered completely by the thresholds of the binarization. We filter the noise through two methods: (1) Increase the exposure value of the images to enhance the contrast of the original images, which improves the filter quality by thresholds. (2) Filter noise based on the area of connected components and distance between dots and the line consisting of the terminals of optical fibers. In the second method, we first set area threshold values, $(A_{min},A_{max})$, according to the area of the terminals of optical fibers. Then, the connected components whose area falls outside this range are filtered. Then we construct the linear equation of the terminals of the optical fibers with Least Squares [28] and set the minimum distance threshold, $D_{min}$. Any connected areas that are larger than $D_{min}$ from the line are filtered out.

2.4. Position feedback of the baffle

Fig. 4 is the image of the terminals, which shows the different spacings between optical fiber terminals.

Figure 4.

The image of the terminals with marked distances between terminals.

Due to the different distribution of the spacing values and manufacturing errors among connectors of different types, we cannot predetermine the moving distance of the baffle precisely, which will lead to error in image capturing. Therefore, we propose the adaptive control algorithm for the baffle position, as showed in Algorithm 1. If the algorithm returns false, the whole verification process breaks, waiting for manual intervention.

Algorithm 1.

Adaptive control algorithm for the baffle position.

2.5. Sequence consistency verification of the input and output terminals

Sequence consistency verification compares the sorting correspondence between input and output terminals, which has to confirm that the sequences of the optical fibers comply with the communication protocols.

Because of the confidentiality of communication protocols used by optical fibers manufacturers, we map the protocols as the sorting correspondence pattern between input and output terminals, as shown in Fig. 5.

Figure 5.

Sorting correspondence pattern between input and output terminals according to the protocols.

In order to store retrieve these patterns, we propose an abreast optical fiber sequence dictionary, as shown follows:

$$\mathbf{Dict}=\{(x_1:y_1),(x_2:y_2),\cdots ,(x_i:y_i),\cdots ,(x_n:y_n)\}$$ (1)

where the n is the number of the optical fibers. For any item $(x_i,y_i)$, $x_i$ and $y_i$ are the states of the input and output optical fibers respectively. We binarize these states. 1 represents light passing through, 2 represents no light passing through. The patterns in Fig. 5 are represented as:

$$\mathbf{Dic}{\mathbf{t}}_{\mathbf{a}}=\{(1:1),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0)\}$$ (2)

$$\mathbf{Dic}{\mathbf{t}}_{\mathbf{b}}=\{(1:1),(1:0),(0:1),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0),(0:0)\}$$ (3)

The sequences of each protocol are exhaustively listed and written into a CSV file. For the efficiency improvement of storage and read/write, the values of x and y are written separately, as shown in Table 1.

Table 1.

List of Dictionary Values for 12 Wire Abreast Optical Fibers.

Index	xi	yi
1	100000000000	100000000000
2	110000000000	101000000000
3	111000000000	111000000000
4	111100000000	111100000000
5	111110000000	111110000000
6	111111000000	111110100000
7	111111100000	111111100000
8	111111110000	111111110000
9	111111111000	111111111000
10	111111111100	111111111010
11	111111111110	111111111110
12	111111111111	111111111111

As an open interface, this CSV file can be modified by the users according to communication protocols, which achieves both protocol confidentiality and ease of use.

2.6. User interaction

We build a real-time human-machine interaction interface based on PyQT5 [29]. The interface provides interfaces for threshold setting, baffle driving motor controlling communication setting, inspection states real-time display, etc., as shown in the user interaction part in Fig. 1.

3. Experiments and results

We have verified the effectiveness of the proposed method through experiments and promoted its application in practical production of abreast optical fibers.

3.1. Experimental environment

Fully understanding the requirements for performance, stability, power consumption, cost, etc. in the quality inspection, we have determined the experimental environment configuration as shown in Table 2. These configurations have been validated in on-site applications and can fully meet the needs of industrial sites.

Table 2.

List of Dictionary Values for 12 Wire Abreast Optical Fibers.

No.	Devices or Software	Main Configuration
1	Host computer	AMD RyzenTM 7 5800H, 16 GB DDR4 3200 MHz RAM, 512GB SSD
2	Software development kits	Windows 11 operating system, Visual Studio Code 1.82, Python3.8.3, OpenCV3, PyQt5
3	Industrial camera	2K Pixels, 60 Hz output, 92 ms exposure time, 8-50 magnification, 12 mm × 8 m field of view, USB3.0 interface
4	Communication and driver for baffle	(1) Communication: USB to Serial Port, Support RS323, RS485 Protocols, Max baud rate 256000 b/s
		(2) Driver: Maximum drive current 100 mA, Maximum pulse output frequency 54 KHz
5	stepper motor and screw module for baffle driving	(1) 28 stepper motor small precision guideway slide, pitch 2 mm (2) maximum horizontal movement accuracy 0.1 mm
6	baffle	3 mm×8 mm×1 mm light-proof PP board
7	input light source	Microscope complementary LED coaxial light source, diameter 6 mm

3.2. Parameters setting of the experiments

3.2.1. Parameters of optical fibers

We tested our method with the most frequency types of abreast optical fibers, as listed in Table 3. The 3 main parameters of the abreast optical influence the quality inspection in our method.

Table 3.

Parameters of Abreast Optical Fiber Used in the Experiments.

No.	Number of optical fibers	Diameter of optical fiber	Length of optical fiber	Number of abreast optical fiber rows
1	12	0.5 mm	100 cm	10
2	36	0.5 mm	60 cm	10
3	64	0.5 mm	30 cm	10

The number of optical fibers is directly proportional to the stroke of the baffle, which affects the cumulative error of the baffle position and thus affects the accuracy of the input light. It requires more adjustment times in baffle position adjustment, which affects the accuracy and duration of the quality inspection.

The diameter of optical fiber affects the size of the light spots in the output side image, which further affects the parameter settings of the CV algorithms.

The length of the optical fiber affects the brightness of the output spot, which in turn affects the contrast of the image. This also needs to be addressed through the parameter settings of the CV algorithms.

3.2.2. Image processing threshold values

The Image binary segmentation reference thresholds setting in HSV space is as shown in Table 4, which are determined manually after multiple tests according to the intensity and color temperature of the light source and the reflection of light by the environment as depicted in Subsection 2.3.

Table 4.

Image Binary Segmentation Reference Thresholds in HSV Space.

Components in HSV space	Minimum Value	Maximum Value
Hue	0	180
Saturation	35	200
Value	56	205

3.2.3. Proficiency level of operators and workload of comparative manual experiments

The compared operators are the persons in charge of inspection, who have operating experience longer than 8 months. In the comparative manual experiments, the operator works 8-10 hours a day, with each shift lasting 40-60 minutes. 80-100 optical fibers are inspected per day, and it takes an average of 4 minutes for 1 optical fiber inspection (including inspecting, clamping and results recording time, excluding archive time of optical fibers and rest time between shifts).

3.3. Experimental results and comparisons

All the experiments were accomplished on a prototype made based on our proposed method. The comparative manual experiments were all conducted in the abreast optical fiber production plant that made the technical requests for our method.

The experiments were conducted from the following three aspects: (1) Accuracy verification. (2) Reliability verification. (3) Comparison of human-machine efficiency.

3.3.1. Accuracy verification

The standard manual inspection is carried out by two operators. The first operator conducts the inspection, and the second operator performs the review. If the deviation between the two inspections is significant, restart the second round. Referring to the standard, our accuracy experiments consist of the first-time inspection accuracy and the average accuracy of four repeated inspections.

The first-time inspection refers to the first inspection after the abreast optical fibers are clamped onto the equipment, of which the accuracy rate is crucial for the overall efficiency of our method.

As the description of the standard manual inspection process, there will be four inspections at most. Therefore, we also take the average accuracy of four repeated inspections as reference.

The results of the first-time inspection accuracy are listed in Table 5. Considering the impact of differences between optical fibers on inspection results, 8 sets of optical fibers were inspected in each type of optical fiber. The accuracy must be above 95% in order to apply our method in practical production.

Table 5.

The Results of the First-time Inspection Accuracy.

Number of optical fibers	1st Set	2nd Set	3rd Set	4th Set	5th Set	6th Set	7th Set	8th Set
12	100.0%	91.6%	100.0%	100.0%	100.0%	100.0%	91.6%	100.0%
36	100.0%	100.0%	97.2%	100.0%	100.0%	94.4%	100.0%	97.2%
64	100.0%	96.9%	100.0%	100.0%	98.4%	100.0%	100.0%	98.4%

According to the results in Table 5, our proposed method measures up the requirement. The result values less than 100% are mainly caused by the variation of image quality. The reasons for the variation of image quality mainly include the quality of the connector terminal face, and the variation of the angle between the optical fiber terminals and the input light due to manual clamping operation. The second reason was optimized with customized fixture in our second iteration version.

The 91.6% is caused by a broken optical fiber, which cannot be detected before the inspection begins. Optical fiber breaking is a common error that occurs, but cannot be detected during the manufacturing process. It can only be detected in quality inspection. Therefore, we did not exclude these data resulting in a decrease in accuracy due to broken optical fibers, but included them as part of the experimental results.

The results of the average accuracy of four repeated inspections are listed in Table 6. As shown in the list, all the average accuracy values are above 95%, but less than 100%. The repeated inspections can achieve the same effect of the standard manual inspection. Due to the exclusion of human interference factors in our method, we focus more on the accuracy of the first-time inspection. The average of four repeated inspections is only used as a reference for evaluating system stability and comparing with manual inspections during the experimental stage.

Table 6.

The Results of the Average Accuracy of Four Repeated Inspections.

Number of optical fibers	1st Set	2nd Set	3rd Set	4th Set	5th Set	6th Set	7th Set	8th Set
12	97.5%	95.8%	98.3%	99.2%	99.2%	99.2%	95.8%	99.2%
36	99.7%	98.9%	97.8%	99.2%	99.4%	96.6%	98.6%	97.5%
64	98.3%	98.1%	99.7%	99.5%	98.6%	99.2%	99.1%	98.9%

3.3.2. Reliability verification

In order to verify the reliability of the proposed method and the corresponding prototype equipment, we tested the change in accuracy of a single optical fiber clamped once and inspected repeatedly for 50 times. The test results are shown in Fig. 6.

Figure 6.

Reliability verification results.

According to Fig. 6, the overall distribution of the accuracy is stable, the percentage of 100% correct inspections is more than 70%, with 88% of the 12-wire, 80% of the 36-wire, and 72% of the 64-wire.

Because of the broken optical fiber, the 12-wire has an accuracy rate of less than 95%. Besides, the frequency of 12-wire errors is the lowest. The number of errors for the 36 and 64 lines increases in turn, mainly due to the increase in the number of optical fibers, which decreases the proportion of one single optical fiber terminal in terminal images, and makes the images interfered by noise easier. The problem can be solved subsequently by replacing the industrial camera with a larger magnification. However, this solution will also result in higher equipment costs.

3.3.3. Comparison of human-machine efficiency

The prime target of our work is to increase the accuracy and efficiency of the consistency verification by replacing the manual work with automation. For the comparison of human-machine efficiency, we arranged a manual inspection with the same workload with the reliability verification, and compared the accuracy and time costs between the two methods. The comparison of accuracy is shown in Fig. 7.

Figure 7.

The comparison of accuracy of 50 consecutive inspections.

We only compare the times of 100% accuracy. The compared operator is experienced, who achieved better results on 12-wire. However, along with the increase of wires, our proposed method achieves better results. The comparison of time cost is shown in Fig. 8.

Figure 8.

The comparison of time cost of 50 consecutive inspections.

The time data in Fig. 8 contains only inspection time, without clamping and unloading time. The figures show that the inspection time of our method is generally stable, on average 50 s per optical fiber. The fluctuation of time is mainly caused by the baffle feedback, which is caused by the gap errors between abreast optical fibers generated in the manufacturing process.

On the other side, the inspection time of manual work increases along with the inspection times due to fatigue. At the 50th time, the average inspection time and ratio are shown in Table 7.

Table 7.

Time Comparison of Man-Machine Efficiency at the 50th Repeated Inspection.

Number of optical fibers	Our method (s/optical fiber)	Manual work (s/optical fiber)	Ratio (manual work/our work)
12	51.67	71.75	1.39
36	50.56	75.03	1.48
64	50.14	90.00	1.79

As listed in the table, the more optical fibers there are, the more significant the efficiency improvement of our work becomes.

In the manual work, one operator operates one inspection equipment, which costs an average of 1.5 minutes for the inspection and calibration of 1 optical fiber, regardless of clamping and unloading time. The detection process is complex and highly repetitive, and it is hard to improve the work efficiency.

With our proposed method, the operators only do the clamping and unloading. The inspection is accomplished automatically. One operator can operate 2-5 devices at the same time. It costs an average of 50 s for the inspection and calibration of 1 optical fiber. Single optical fiber inspection time is improved by 45%, i.e., 1.45 per person per day.

According to the estimate that each operator can operate 4 devices simultaneously (the operation time varies for different numbers of optical fibers), our method saves 4 men compared to manual work. In addition, as the operator only needs to be responsible for loading and unloading the optical fibers and classifying them according to the inspection results, the complexity of the operation is greatly reduced, which in turn reduces the training cost for operators and shortens the training time.

4. Conclusions

In this paper, we present a small diameter abreast optical fiber terminal sequence consistency verification method based on CV, for improving the inspection accuracy, reliability, and speed by replacing the manual work with computer vision and automated control methods. The target of inspection is the optical fiber terminal sequence specified by the device communication protocol, which has strong regularity, and a small number of training samples. Therefore, we adopt CV-based algorithms.

We adopt the HSV space to improve the quality of image feature extraction. We propose the abreast optical fiber sequence dictionary converting the protocol logic into an input-output mapping table, which follows protocol confidentiality, reduces software complexity, and improves system operation speed. We design the adaptive control algorithm for the baffle position adjustment, to adapt the different distribution of the spacing values and manufacturing errors among connectors of different types.

As future work, we will focus on the automation of threshold values setting [30]. The threshold values are used for image binaryzation, which has a significant impact on the quality of image processing and are currently set manually with low efficiency. Swarm intelligence is planned to be involved, which can further improve the usability and efficiency of the inspection [31].

CRediT authorship contribution statement

Yan Wang: Conceptualization, Data curation, Methodology, Writing – original draft. Lei Wang: Data curation, Software. Dalin Li: Conceptualization, Methodology, Resources, Writing – review & editing. Yanchun Liang: Supervision, Validation. Lan Huang: Formal analysis, Funding acquisition, Resources, Supervision. Haoming Da: Data curation, Software. Hui Yang: Validation, Visualization.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.