In this study, a lightweight phenotyping system that combined the advantages of both deep learning-based panicle detection and the photogrammetry based on light consumer-level UAVs was proposed. A two-year experiment was conducted to perform data collection and accuracy validation. A deep learning model, named Mask Region-based Convolutional Neural Network (Mask R-CNN), was trained to detect panicles in complex scenes of paddy fields. A total of 13 857 images were fed into Mask R-CNN, with 80% used for training and 20% used for validation. Scores, precision, recall, Average Precision (AP), and F1-score of the Mask R-CNN, were 82.46%, 80.60%, 79.46%, and 79.66%, respectively. A complete workflow was proposed to preprocess flight trajectories and remove repeated detection and noises. Eventually, the evident changed in rice growth during the heading stage was visualized with geographic distributions, and the total number of panicles was predicted before harvest. The average error of the predicted amounts of panicles was 33.98%. Experimental results showed the feasibility of using the developed system as the high-throughput phenotyping approach.

The purpose of this study is to evaluate the performance of a robot combine harvester by comparing the Centimeter Level Augmentation Service (CLAS) and the Multi-Global Navigation Satellite System (GNSS) Advanced Demonstration tool for Orbit and Clock Analysis (MADOCA) from the Quasi-Zenith Satellite System (QZSS) by using the Real Time Kinematic (RTK) positioning technique as a reference. The first section of this study evaluates the availability and the precision under static conditions by measuring the activation time, the reconnection time, and obtaining a Twice Distance Root Mean Square (2DRMS) of 0.04 m and 0.10 m, a Circular Error Probability (CEP) of 0.03 m and 0.08 m, and a Root Mean Square Error (RMSE) of 0.57 m and 0.54 m for the CLAS and MADOCA, respectively. The second section evaluates the accuracy under dynamic conditions by using a GNSS navigation-based combine harvester running in an experimental field. The results show that the RMSE of the lateral deviation is between 0.04 m and 0.69 m for MADOCA and between 0.03 m and 0.31 m for CLAS; which suggest that the CLAS positioning augmentation system can be utilized for the robot combine harvester if the user considers these accuracy and dynamic characteristics.

The remote operation of multiple agricultural robotic vehicles (RVs) improves the efficiency of farming operations; however, there is still a lack of evidence for centralized monitoring systems capable of managing multiple tasks simultaneously in real-world applications. This study presents a GIS-based remote monitoring system for agricultural RVs implementing Level 3 automation, as defined by the Ministry of Agriculture, Forestry and Fisheries (MAFF) of Japan. The system was evaluated against Level 1 and Level 2 automation through theoretical estimations, simulations, and real-world experiments to assess improvements in efficiency for multi-RV operation. Experimental results demonstrated significant improvements in work-time efficiency during multi-RV operation, with Makespan reduced by up to 66%, Speedup reaching 1.66, and Parallel efficiency achieving 41%. Agreement between simulation and experimental results validated the proposed work-time formulation and simulator. Furthermore, integration of the simulator into the GIS-based monitoring system enables intuitive estimation of the optimal number of RVs and expected efficiency parameters for a given work plan. Although the current implementation is not yet cost-competitive with lower automation levels, it demonstrates the feasibility of a centralized GIS-based framework and provides practical validation of Level 3 automation in agricultural operations.