Development of a High-Speed Time-Synchronized Crop Phenotyping System Based on Precision Time Protoco

Abstract

Aiming to address the problems of asynchronous acquisition time of multiple sensors in the crop phenotype acquisition system and high cost of the acquisition equipment, this paper developed a low-cost crop phenotype synchronous acquisition system based on the PTP synchronization protocol, realizing the synchronous acquisition of three types of crop data: visible light images, thermal infrared images, and laser point clouds. The paper innovatively proposed the Difference Structural Similarity Index Measure (DSSIM) index, combined with statistical indicators (average point number difference, average coordinate error), distribution characteristic indicators (Charm distance), and Hausdorff distance to characterize the stability of the system. After 72 consecutive hours of synchronization testing on the timing boards, it was verified that the root mean square error of the synchronization time for each timing board reached the ns level. The synchronous trigger acquisition time for crop parameters under time synchronization was controlled at the microsecond level. Using pepper as the crop sample, 133 consecutive acquisitions were conducted. The acquisition success rate for the three phenotypic data types of pepper samples was 100%, with a DSSIM of approximately 0.96. The average point number difference and average coordinate error were both about 3%, while the Charm distance and Hausdorff distance were only 1.14 mm and 5 mm. This system can provide hardware support for multi-parameter acquisition and data registration in the fast mobile crop phenotype platform, laying a reliable data foundation for crop growth monitoring, intelligent yield analysis, and prediction.

1. Introduction

In the field of high-throughput acquisition of crop phenotype information, synchronous acquisition of multi-source information (including visible light imaging, laser imaging, multispectral imaging, hyperspectral imaging, thermal imaging, etc.) has become an important trend [1,2,3]. By integrating multiple sensors to realize the synchronous acquisition of crop phenotype information, it can provide not only rich data for subsequent intelligent decision-making algorithms but also hardware support for multi-parameter acquisition and data registration in the fast mobile crop phenotype platform, improving the accuracy of agricultural production decisions [4,5,6,7].

Time synchronization of multi-sensor systems has always been a research hotspot in aerospace, environmental monitoring, ocean observation, industrial automation, energy and transportation, and other fields [8,9,10,11,12,13,14,15], among which the synchronization problem between each sensor is a challenging key issue. To address these challenges, the current multi-sensor relative time synchronization protocols mainly include the Network Time Protocol (NTP) and the Precision Time Protocol (PTP). The NTP technology is a protocol based on the User Datagram Protocol (UDP) and built on the application layer, realizing time synchronization through message interaction between multiple devices, with a synchronization time error ranging from 50 to 200 ms. The PTP can be built not only on the application layer but also between the Medium Access Control (MAC) and physical layers, enabling message interaction between multiple devices. Because it is at the bottom of the entire system, it is less affected by various interferences and instabilities of the system, so it can achieve higher-precision time synchronization, and the synchronization time error between the MAC and physical layers can reach the nanosecond level [16]. Li et al. achieved time synchronization between the timing center and the time receiving module in the transportation field based on the GPS satellite navigation system for timing and the Pulse Per Second (PPS) signal, with an error of less than 200 ns [17]. Venmani et al. realized high-precision time synchronization between 5G network and industrial equipment through intelligent PTP gateway (s-PTP gateway) in industrial IoT environment, which meets the synchronization requirements of Industry 4.0 and 5.0 application scenarios [18]. Xiaojiang Liu et al. effectively improved the clock synchronization performance of IEEE 1588 PTP in industrial network asymmetric delay scenarios by modeling unknown delays and optimizing filtering algorithms, providing a reliable clock parameter tracking solution for wired, wireless, and multi-hop networks [19]. Although multi-sensor time synchronization issues have been extensively studied across various industries, there are few reported cases of PTP applications in the agricultural sector. Furthermore, research reports on multi-sensor time synchronization in fast-moving crop phenology monitoring systems are even scarcer. Existing fast-moving crop phenology platforms (such as UAVs) typically only carry a single detection sensor, such as a multispectral camera or an infrared thermal imaging camera [20,21,22,23]. Some UAVs can carry two detection sensors simultaneously, but due to their high flight speeds (approximately 25 m/s) [24], without time synchronization technology, the data collection times of the two sensors are not synchronized, leading to significant differences in the samples collected by different sensors. This makes subsequent data processing and application extremely challenging, failing to adequately support the needs of smart agriculture and hindering its widespread adoption in agricultural production.

Finally, existing crop phenotyping systems are primarily provided by companies such as Zealquest (Zealquest Science and Technology Co., Ltd., Shanghai, China) and Phenospex (Phenospex B.V. Heerlen, the Netherlands), which are not only expensive but also fail to address the issue of synchronized data collection timing. For example, the minimum configuration price for the PhenoWatch imaging unit developed by Zealquest is 220,000 USD. Each parameter is collected independently, without achieving synchronized collection, making it unsuitable for application in fast-moving crop phenotyping monitoring systems. This severely hinders its development and application in the field of smart agriculture. Therefore, considering the issues of asynchronous data collection and high costs in current fast-moving crop phenotyping systems (such as UAVs), this paper selects relatively low-cost hardware and designs a low-cost, high-synchronization crop phenotyping data collection system based on the PTP between the MAC layer and physical layer of the network, laying the foundation for equipping fast-moving crop phenotyping platforms with multiple sensors and achieving synchronous collection of crop phenotyping parameters.

2. Materials and Methods

To achieve accurate crop growth monitoring with low cost and synchronous acquisition time, this paper presents a parameter acquisition system suitable for various fast mobile crop phenotype platforms based on the PTP high-precision time synchronization technology and the working principle of Ethernet [25]. This system innovatively introduced the time synchronization acquisition technology into the fast mobile crop phenotype platform for the first time, realizing the synchronous acquisition of visible light images, infrared thermal images, and laser point cloud images of crops, providing reliable data support for subsequent crop growth monitoring, yield intelligent analysis, and prediction.

2.1. Hardware Design

The hardware of the crop phenotype acquisition system mainly includes four parts: the main control component, the PTP switch, the timing module, and the acquisition module (Figure 1). The main control component selects the RK3588 main control board with an Ethernet card interface (iTOP-RK3588, Beijing Xunwei Electronic Co., Ltd., Beijing, China), which is responsible for data acquisition, storage, uploading to the server, sending control commands to the PTP switch, and controlling the switch. To ensure local debugging and subsequent function expansion, the main control component retains serial ports and HDMI; to ensure the security and reliability of data, data is saved in two ways: local IF card and remote server (LTE EC20-CE Mini PCIe, Shanghai Quectel Wireless Solutions Co., Ltd., Shanghai, China).

The switch selects an Ethernet switch with an Ethernet card interface and supporting the PTP (SYN2421 IEEE1588 industrial Ethernet switch, Xi’an Sync Electronic Technology Co., Ltd., Xi’an, China). It has the function of linking multiple acquisition devices in a local area network and forming a PTP time synchronization domain. It is networked with the timing module through the Gigabit Dual RJ45 Ethernet Card F902T (Gigabit Dual RJ45F902T, Beijing Guangruntong Technology Development Co., Ltd., Beijing, China) and follows the PTP, sending control commands and transmitting data.

The timing module selects three RK3568 timing boards (HZ-EVM-RK3568, Beijing Hezhong Hengyue Technology Co., Ltd., Beijing, China), which are connected to the PTP switch through the PCIE interface of the Gigabit Dual RJ45 Ethernet Card F902T and to the sensors through the network port. A serial port is designed for the rapidity and convenience of local debugging and burning. The Gigabit Dual RJ45 Ethernet Card F902T simultaneously supports the PTP and hardware timestamp synchronization, which can meet the PTP requirements of the three RK3568 timing boards using the hardware timestamp synchronization method.

With the goals of low cost and ease of use, the sensor module consists of a Hikvision visible light camera (MV-CS200-10GC, Hangzhou Hikvision Digital Technology Co., Ltd., Hangzhou, China), a Guide Infrared thermal sensor (IPT430M, Wuhan Guide Infrared Co., Ltd., Wuhan, China), and a LIVOX lidar (Mid-360, Shenzhen Livox Technology Co., Ltd., Shenzhen, China). The Hikvision MV-CS200-10GC visible light camera collects color information of the crop canopy, the Guide Infrared IPT430M thermal sensor continuously and automatically monitors temperature changes, and the LIVOX Mid-360 lidar is used to collect three-dimensional point cloud data of crops. Based on the multiple obtained information such as visible light images, thermal infrared images, and point cloud data, combined with deep learning methods and techniques, phenotype information such as crop branching structure, leaf inclination distribution, plant height, coverage, and texture are obtained. Detailed information on all equipment is shown in

Table 1. List of devices used in the developed crop phenotyping system.

No.	Item	Model	Main Parameters	Equipment Picture
1	Main Control Board	Xunwei iTOP-RK3588	CPU: ROCKCHIP RK3588 Octa-Core A76+A55 Main Frequency: 2 GHz Memory: 16 GB GPU: integrated Mali G610 3D quad-core GPU 2-way 1000 Mbps Ethernet, RJ45 port, etc.
2	PTP Switch	SYN2421 IEEE1588 Industrial Ethernet Switch	Ports: RJ45 x20, fiber optic interface×4, 10/100/1000M auto-detect, full/half duplex MDI/MDI-X adaptive Support IEEE 802.1×Support HTTP Support RADIUS
3	Timing Module	HZ-EVM-RK3568	CPU: ROCKCHIP RK3568 Quad-Core 64-bit Cortex-A55 Main Frequency: 2 GHz Memory: 4 GB GPU: ARM G52 2EE 2-way 1000 Mbps Ethernet, RJ45 port, etc.
4	Visible Light Camera	Hikvision MV-CS200-10GC	Sensor Type: CMOS, Rolling Shutter Resolution: 5472 × 3648 Maximum frame rate: 5.9 fps @5472 × 3648 Pixel format: Mono 8/10/12 Bayer GB 8/10/10Packed/12/12Packed YUV422Packed, YUV422_YUYV_Packed RGB 8, BGR 8
5	Infrared Thermal Sensor	Guide IPT430M	Wavelength range: 8 microns to 14 microns Resolution: 384 × 288 Thermal sensitivity: ≤50 mK Field of view: 46.0° × 34.1° Data interface: Ethernet interface Temperature range: −20~150 °C Temperature measurement accuracy: ±2 °C or ±2%
5	Lidar	LIVOX Mid-360	Laser wavelength: 905 nm FOV: 360° × 59 Distance random error (1σ): ≤2 cm (@10 m); ≤3 cm (@0.2 m) Angular random error (1σ): <0.15° Point cloud output: 200,000 points/second Point cloud frame rate: 10 Hz (typical) Data Port: 100 BASE-TX Ethernet
6	Ethernet Card	Guangruntong Gigabit Dual Port RJ45 Network Card F902T-V4.0	Ports: 2 ports Connector: RJ45 Speed: 5 GT/S Slot Width: ×4Lane

All technical specifications are taken from the product manual.

.

2.2. PTP Implementation

The data acquisition system designed in this paper is mainly applied to fast-moving crop phenotype acquisition platforms (such as UAV, greenhouse high-speed rail-guided platforms, etc.) [26]. To reduce costs, a relative time synchronization protocol that does not rely on an external clock and can synchronize each node to a time reference is selected to achieve time synchronization of multiple sensors. Therefore, this system selects inexpensive timing boards, gateways, and the LinuxPTP software package (linuxptp −4.0) with strong adaptability, universality, and stability. By building the PTP between the MAC and physical layers, synchronous acquisition and transmission of data between each component are realized.

The system requires three timing boards to ensure the synchronization of the acquisition time of the three acquisition devices: the visible light camera, the near-infrared camera, and the lidar. After the three timing boards are powered on and started, the ptp4l and phc2sys programs in the LinuxPTP software package are automatically run. The ptp4l program is used to achieve PTP clock synchronization between multiple devices, and the phc2sys program is responsible for synchronizing the PTP clock and the system clock within a single device. It has been verified that when the “hardware timestamp” mode is selected in the ptp4l program and IEEE 802.3 network transmission is used, the PTP clock synchronization accuracy between the timing boards is better than that in the “software timestamp” mode. The phc2sys program was set so that the system clock follows the PTP clock of the device.

During PTP initialization, the Best Master Clock Algorithm (BMCA) was used to determine the master clock. Its principle is to automatically select the master clock of the PTP clock of timing board 1 by comparing the priority, clock level, accuracy, offset, time source, and other attributes of the participating clock devices [27]. When the PTP enters the normal working stage, the PTP clocks of timing boards 2 and 3 follow and synchronize with the PTP clock of timing board 1 (the master clock), and the system clocks of each timing board follow and synchronize with the PTP clocks within their respective devices. The system clocks of each timing board use the PTP clocks within their respective devices as an intermediate medium, and finally, the system clocks of each timing board are synchronized with the PTP clock of timing board 1. The specific clock-following and synchronization relationship, along with the flowchart for establishing the clock relationship in the PTP domain, are shown in Figure 2.

2.3. Software Design

When the ptp4l and phc2sys programs in the LinuxPTP software package are running normally, the system establishes a complete clock synchronization relationship, and multi-sensor synchronous acquisition can be realized through software design. The overall design of the system software is shown in Figure 3.

The detailed process is as follows:

After the main controller receives the start instruction from the user, it sends the instruction to timing board 1 (i.e., the master clock) through the UDP, and timing board 1 serves as the reference source for clock synchronization.

After receiving the start instruction, timing board 1 immediately records the timestamp RT1 of the current system clock and broadcasts the timestamp RT1 and the timed acquisition command to timing boards 2 and 3 through UDP multicast. Timing board 1 then uses the nanosleep timer to set a 1 s delay. After the delay ends, the sensors start synchronous data acquisition. After the acquisition is completed, the data is preliminarily processed, and the result is sent back to the main controller.

When timing boards 2 and 3 receive the RT1 timestamp and the timed acquisition command, due to factors such as network delay and system interference, there may be time differences between different timing boards. To minimize this error, timing boards 2 and 3 record the timestamps RT2 and RT3 of their respective current system clocks when receiving the instructions. Then, timing boards 2 and 3 calculate and set the delay through the nanosleep timer as 1 s minus the time difference between each of them and RT1 (i.e., 1 − (RTi − RT1) seconds, where i = 2, 3). After the delay ends, the sensors start synchronous data acquisition. After the acquisition ends, each timing board preliminarily processes the acquired data and sends the result back to the main controller.

2.4. System Performance Evaluation

The performance evaluation of this crop phenotype acquisition system includes stability, dynamic performance, and the synchronization performance of each timing board. This system synchronously acquires the visible light image, thermal infrared image, and laser point cloud of the crop. Since the three have different formats, different evaluation methods are adopted.

2.4.1. Stability Evaluation of Visible and Thermal Image Acquisition

To quantitatively evaluate the stability of visible light images and thermal infrared images, this paper innovatively proposes the Difference Structural Similarity Index Measure (DSSIM) based on the structural similarity index to eliminate the influence of illumination brightness differences in different images [28,29,30]. In the actual calculation process, DSSIM replaces the brightness ${\mu }_y$ of image y with the brightness value ${\mu }_x$ of the reference image x, so that only the differences in contrast and structure are considered during the calculation, and the brightness difference between the images is not considered. DSSIM was calculated from

$$DSSIM\left(x,y\right)={\displaystyle \frac{\left(2{\mu }_x^2+c_1\right)\left(2{\sigma }_{xy}+c_2\right)}{\left({\mu }_x^2+{\mu }_x^2+c_1\right)\left({\sigma }_x^2+{\sigma }_y^2+c_2\right)}}={\displaystyle \frac{\left(2{\sigma }_{xy}+c_2\right)}{\left({\sigma }_x^2+{\sigma }_y^2+c_2\right)}}$$

(1)

where ${\sigma }_x$ and ${\sigma }_y$ are the variances of image x and image y, respectively, representing the estimate of contrast; ${\sigma }_{xy}$ is the covariance between image x and image y, reflecting the structural similarity between the two images; $c_2$ is a stability constant used to avoid a zero denominator. It is usually set as $c_2={(K_2\times L)}^2$, where L is the dynamic range (usually 255), and $K_2$ is a small constant (usually 0.03). The closer the value of DSSIM is to 1, the smaller the difference between the images.

In addition, the maximum error and the mean absolute error of the acquired data are also used to comprehensively measure the stability of the data acquired by the system developed in this paper.

2.4.2. Stability Evaluation of Point Cloud Acquisition

To verify the stability of the system in point cloud data acquisition, statistical indicators (average point number difference, average coordinate error), distribution characteristic indicators (Charm distance), and structural characteristic indicators (Hausdorff distance) are used for quantitative analysis. Among them, the smaller the Charm distance, the more similar the distribution characteristics of the two point cloud files. The Hausdorff distance calculates the maximum distance from each point in point set P to the nearest point in point set Q and the maximum distance from each point in point set Q to the nearest point in point set P, and finally takes the larger value of these two maximum distances. The smaller the value of the Hausdorff distance, the more similar the structural shapes of the two point cloud files, and vice versa.

2.5. Experimental Design

The fast mobile crop phenotype acquisition system in this experiment is shown in Figure 4a. On both sides of the top of the bracket are the Hikvision MV-CS200-10GC visible light camera, the Guide Infrared IPT430M thermal infrared sensor, and the LIVOX Mid-360 lidar. All sensors are positioned approximately 1.2 m away from the crop being measured, with an angle of approximately 12°. The experiment will be carried out in the greenhouse of the National Precision Agriculture Research Base in Changping District, Beijing (longitude 116.14 east, latitude 40.13 north). One hundred potted Guofu 909 peppers were selected as the test crops (as shown in Figure 4b). All peppers were in the seedling stage, with bright green leaves. Irrigation and lighting conditions were consistent, with generally similar growth conditions. The laser point cloud, thermal infrared image, and visible light image of each pepper were synchronously acquired, and this was repeated 133 times. In order to verify the steady-state and dynamic performance of the self-developed crop phenotype collection system, the stability of the collected data, the error of the collection time triggered by the synchronization of each timing board, and the response time of the system were recorded.

3. Results

The performance of the crop phenotype parameter acquisition system includes the synchronization performance of the timing boards and the acquisition performance (stability and dynamic performance) of the system.

3.1. Time Synchronization

This system has three timing boards, which are responsible for the time synchronization of laser point cloud, visible light image, and thermal infrared image acquisition, respectively, and are fundamental for ensuring that the system supports the fast mobile crop phenotype platform with a high acquisition frequency. By testing the following three types of synchronization, the root mean square value of the timing board synchronization time error is obtained.

After a 72 h test experiment, the clock errors of the three timing boards were recorded every 1 s, and a total of 2,073,600 error values were obtained. As shown in Figure 5, the data curves of 1 h were randomly intercepted (the time difference between the PTP master and slave clocks, the time difference between the system clock of each timing board and its own PTP clock, and the time difference between the system clock of each timing board and the PTP clock of timing board 1).

To further analyze and calculate the synchronization time differences in the different clocks of the three timing boards, a synchronization error bar chart was drawn (Figure 5). It can be seen from Figure 6 that the time errors between the PTP slave clocks of timing boards 2 and 3 and the PTP master clock of timing board 1, the time errors between the system real-time clocks of timing boards 1, 2, and 3 and their own PTP clocks, and the time errors between the system clocks of timing boards 1, 2, and 3 and the PTP clock of timing board 1 (the master clock) are mostly within (time error/ns). The root mean square values of the time errors are shown in

Table 2. Root mean square values of time errors.

Clock Under Test	Time Reference Clock	Clock Synchronization Time Error Root Mean Square (ns)
PTP clock of timing board 2	PTP clock of timing board 1	8.25
PTP clock of timing board 3	PTP clock of timing board 1	8.73
System clock of timing board 2	PTP clock of timing board 2	128.87
System clock of timing board 3	PTP clock of timing board 3	120.18
System clock of timing board 1	PTP clock of timing board 1	129.17
System clock of timing board 2	PTP clock of timing board 1	128.87
System clock of timing board 3	PTP clock of timing board 1	120.47

.

It can be concluded from Figure 5 and Table 2 that during the 72 h test period, the PTP slave clocks of timing boards 2 and 3 followed the PTP master clock of timing board 1 with high precision, and the root mean square values of the following time errors were 8.25 and 8.73 ns, respectively. The root mean square errors of the time differences between the system real-time clocks of each timing board and the PTP clock of timing board 1 (the master clock) were 129.17, 128.87, and 120.47 ns, respectively, all of which are less than 130 ns. This system exhibits high-speed time synchronization.

3.2. Crop Phenotype Parameter Acquisition Performance

3.2.1. System Stability

The independently developed crop phenotype acquisition device was used to continuously and synchronously acquire the crop phenotypes of Guofu 909 peppers, and the obtained visible light images, thermal images, and laser point cloud images (as shown in Figure 7) were stored in the system in categories. The equipment continuously acquired data for one hour. On the basis of no missed acquisitions, 133 successful acquisitions were achieved, and 133 sets of data were obtained. All images in this article were taken under natural light conditions without additional lighting.

(1) Stability of Visible and Thermal Image Acquisition

During the data acquisition process, due to the shift in sunlight, the brightness changes in the visible light images and thermal infrared images were relatively obvious in the early and late stages of shooting. As shown in Figure 8, the brightness value of the visible light image was 113.52 nits in the early stage of shooting and 139.37 nits in the late stage; the brightness value of the thermal infrared image was 101.94 nits in the early stage of shooting and 113.70 nits in the late stage. For the 133 sets of visible light images and thermal infrared images, the two indicators of SSIM and DSSIM were calculated to characterize the similarity between the images, and the maximum error and mean absolute error of the acquired data were calculated to measure the stability of the system acquisition. The results are shown in

Table 3. Similarity analysis of visible and thermal maps.

Data Type	Maximum Pixel Error (%)	Pixel Mean Absolute Error (%)	Pixel Mean Square Deviation	SSIM	DSSIM
Visible light image	3.52	2.8926	11.2633	0.8266	0.9688
Thermal infrared images	3.36	3.2549	23.5559	0.8887	0.9555

.

(2) Stability of Point Cloud Acquisition

To verify the stability of the system in point cloud data acquisition, statistical indicators, distribution characteristic indicators (Charm distance), and structural characteristic indicators (Hausdorff distance) were used for quantitative characterization. For 133 point cloud images (each image contains 3600 point cloud data), two statistical indicators (average point number difference and average coordinate error) were calculated. The average point number error of the point cloud files was about 2%, the average coordinate error was within 5%, the Charm distance was only 1.1448 mm, and the Hausdorff distance was only 5.54516 mm.

The Charm and Hausdorff distances between more than 130 point cloud images were small, with errors at the millimeter level, indicating that the point cloud data acquired by the crop phenotype acquisition system designed in this paper showed no obvious differences in distribution characteristics and structural characteristics. It showed that the point cloud files in this study were statistically very similar, further indicating that the system had good stability.

3.2.2. Dynamic Performance

The system dynamic response time includes the synchronous trigger time of the timing boards (the sum of the synchronization time between the system clock and the PTP clock and the relative error time between the PTP clocks), the data acquisition time, and the network transmission time. Among them, the network transmission time refers to the transmission time of public networks (such as 5G), and this part is not discussed in this study. As mentioned above, taking the PTP clock of timing board 1 as the reference (master clock), the root mean square values of the time differences between the system clocks of each timing board and the PTP clock of timing board 1 were all less than 130 ns. The length of the data acquisition time mainly depends on the internal interruption delay, software delay, and clock drift of the system, which are the main factors determining the system dynamic response.

To obtain the crop data acquisition time, 100 potted Guofu 909 peppers were used as test samples. The three timing boards were synchronously triggered to continuously acquire 100 groups of laser point clouds, visible light images, and thermal infrared images of the crops, and their respective acquisition times were recorded. The scatter plot of the acquisition times over 100 trials is shown in Figure 9.

The root mean square values of the acquisition times of the laser point cloud, visible light image, and thermal infrared image were calculated as 299.20, 255.70, and 223.22 μs, respectively. The dynamic response time of the fast mobile crop phenotype acquisition system is the sum of the synchronous trigger time of the timing boards and the data acquisition time, that is, 130 ns + 300 μs.

4. Discussion

As is known, data alignment has always been an important topic in crop phenotyping research, with many studies [31,32,33] focusing primarily on algorithmic research. However, there are fewer reports on improving alignment efficiency through hardware solutions. Aiming to address the problem of asynchronous acquisition time of parameters in the crop phenotyping platform, which hinders the application of multi-sensor technology, this study, with low cost as the goal, introduces PTP high-precision time synchronization technology into the parameter acquisition system of the crop phenotyping platform for the first time, realizing the synchronous acquisition of crop RGB images, infrared thermal images, and laser point cloud images, which provides a reliable data support for the subsequent data alignment and intelligent analysis. The paper analyzes the synchronous acquisition performance, system stability, and dynamic performance of the system in detail, and the results are as follows:

The synchronous triggering of each timing board is the core foundation of this study. In this paper, inexpensive timing boards and gateways were selected, and the PTP was applied at the physical layer. The root mean square value of the synchronization time error between each timing board was controlled within 130 ns. This synchronization error is much smaller than that of NTP technology (50–200 ms) and approaches the theoretical synchronization time error of PTP (approximately 20 ns) [16], reaching the ns level. Compared with Ferrari et al. [19], who built the PTP on the application layer and controlled the time synchronization error of the displacement sensor within 20 μs, although the hardware cost has increased slightly, the software cost has been greatly reduced, and the accuracy has been improved from the microsecond level to the nanosecond level, which can better serve the fast mobile crop phenotype platform in the agricultural field.

During the acquisition of visible light images and thermal infrared images, due to the shift in sunlight, the brightness changes in the visible light images and thermal infrared images in the early and late stages of acquisition were relatively obvious, resulting in a relatively low calculation result of the structural similarity SSIM. Therefore, we proposed the Difference Structural Similarity Index Measure (DSSIM) indicator, taking the first acquired file as the benchmark, and calculating the similarity between the first file and the remaining files. The DSSIM value between each two visible light images and thermal infrared images was as high as above 0.95, indicating that there was no obvious difference in structure and contrast between each two images, and the acquired data had good stability. In addition, the maximum value and mean absolute error of the pixel error between each two images were only about 3%, again indicating that the system had a high degree of stability in the acquisition of visible light and thermal infrared images.

5. Conclusions

Aiming at the problems of asynchronous multi-parameter acquisition and high cost in the crop phenotype platform, this paper introduced the PTP and built a low-cost and practical crop phenotype parameter acquisition system, realizing the time synchronization of multiple sensors at the ns level, providing an advanced technical means for crop phenotype research. The paper draws the following three conclusions:

The system developed in this paper built the PTP between the MAC and physical layers, which can realize the high-precision time synchronization of digital cameras, infrared cameras, and laser point cloud devices. The results of the 72 h time synchronization test experiment show that the root mean square errors of the time differences between the system real-time clocks of each timing board and the PTP clock of timing board 1 (the master clock) are 129.17, 128.87, and 120.47 ns, respectively, indicating that the system has high time synchronization accuracy.

The results of the stability test experiment (taking peppers as an example) show that the system achieved a 100% acquisition success rate in 133 consecutive acquisitions. The average point number difference and coordinate error of the acquired data were both about 3%, the DSSIM was about 0.96, the Charm distance was about 1.14 mm, and the Hausdorff distance was about 5 mm, proving that the system developed in this paper has good stability.

Without considering the public network transmission delay, the response time of the system developed in this paper reaches the microsecond level, which can not only meet the needs of the fast mobile crop phenotype system to carry multiple sensors for synchronous acquisition but also provide hardware support for improving the accuracy of data registration.