abstract

A flower bud detection system using image processing was developed to assist in picking Japanese pear buds. The prototype image acquisition system comprises a CCD camera and RTK-GNSS, which acquires moving images at 24fps and latitude and longitude at 4 Hz. YOLOv3 was applied to image processing, and 1800 images were used as training data and 360 images as test data. As a result, the accuracy of the detection system was 85.6 %. We integrated the estimated number of flower buds with GNSS information, interpolated by kriging, and created a 1mx1m grid square map. The results, using the grid square map, indicated that the maximum, average, and standard deviation of each grid were 84, 32, and 19 buds/m², respectively.

The driving environment and shape of the combine harvester of crawler causes changes in the posture and vibration, which leads to the overturn and comfort of the operator. Additionally, vibration is increased when driving on rough terrain as the Track-Roller of combine harvester is often directly fixed to the frame. To organize stable driving conditions, this study conducted simulations of driving on the rough terrain by changing road surface roughness and driving speed based on the ISO. Results revealed that the time leaving the road increased as the roughness and running speed increased. Moreover, we determined the running speed, which maintains operability on each roughness of the terrain, from the speed of rising the road.

This study proposes a guidance method using GNSS, IMU, and vision sensors to navigate a tractor-implement to align and hitch. The vision system, comprising a camera and marker, estimates the implement’s relative pose with respect to the tractor’s. The relative pose is fused with data from the GNSS and IMU to determine the implement’s hitching pose for planning the tractor-backing path accurately. The results of automatic hitching tests on a concrete surface using an autonomous tractor and broadcaster demonstrates that the proposed guidance method can accurately perform path planning and tracking. The calculated average tractor-implement alignment errors are 0.015±0.010 m, 0.011±0.010 m, and 2.3±1.3 ° in the lateral, longitudinal, and yaw directions, respectively.