A Prototype of an Immature Citrus Fruit Yield Mapping System

Yield mapping is the first step for site-specific crop management. Many yield mapping systems have been developed for grain crops. However, it remains a difficult task for tree crops. In this study an autonomous yield mapping system for citrus crops was developed. The system was designed to detect fruit and create yield maps at early stages so that farmers could manage the grove site specifically based on the maps. It consisted of two major sub-systems, an autonomous navigation system and an imaging system. Robot Operating System (ROS) was used for developing the autonomous navigation system on top of an unmanned ground vehicle. An inertial measurement unit (IMU), wheel encoders and a GPS were integrated using an extended Kalman filter to provide reliable navigation solutions. In the imaging system, a high-resolution visible camera was carried by the vehicle for image acquisition. All the video frames were associated with latitude and longitude coordinates automatically. Detection of fruit from the video frames utilized a VGG16 model, which was trained with Faster-RCNN. Fruit detection was evaluated and an accuracy of 77% was achieved. The Lucas-Kanade optical flow method was used for tracking each detected fruit and counting the total number of fruit. The complete system was tested in a citrus grove in Florida.