Design and experiment of the truss-type platform to acquire high-throughput information from farmland using remote sensing

Crop phenotype collection is currently required for the supporting facilities of farmland. Manual collection is also lacking in the applicability and less collected data in fields. In this research, a remote collection platform was designed to acquire high-throughput information on farmland using a truss guide rail structure. The collection platform was composed of the movement rail with a truss rail structure, a camera, LiDAR, a Beidou module, and integrated environmental sensors. There were movements in the forward and backward (X-axis), left and right (Y-axis), and pitch (P-axis) on the rail. The remote multi-directional collection of crop phenotype and farmland environmental information was realized using 4G communication. The field test showed that the maximum delay of 4G remote end-to-end communication was no more than 60 ms. The data interaction between the terminals of remote monitoring and collection platforms was carried out in a very short period of time. The control commands of X, Y, and P axis step motion were released from the remote monitoring terminal. After that, the X, Y, and P axis motion speeds were 0.5, 0.0769 m/s and 1.81 °/s, respectively. The phenotype sensor also reached the target position for information acquisition in a relatively short period of time. The errors of response steady state were 0.04 m/s, 0.001 m and 0.5°, indicating the smooth and stable motion for the data quality of collection; The close-proximity capture was realized in the Y-axis from the leftmost end to the rightmost end of the span of 2 m in the process of the 10 rows of maize. The vertical, front and back sides of corn seedlings rows were also collected in the pitch shooting of the P-axis from -60° to 60°. The obtained images were consistent with the radar point cloud. There was less influence of the light on the camera rotation and pitch shooting; A three-dimensional point cloud with color information was generated from the collected images and radar data. The plant height of oilseed rape was extracted using the fused information. The R² value of plant height was 0.96, compared with the manual measurement. The collected data fully met the actual processing on the crop phenotypic parameters at later stages. The average time was 10 s for the collection of environmental information in a single area. There were no significant differences in the environmental parameters within the group. The accuracy was more than 0.95, compared with the measured data. The collection platform fully met the requirements of intelligent collection for the high-through phenotypic data in farmland.