Development of Deep Learning Models on the Navigation System for Assistant Harvesting Robot

Abstract

Greenhouse farming can be more profitable if automation, computer and robotics technologies are applied to its environment. One of the robots specifically designed for the greenhouse environment is the robot for support harvesting process. The robot has the function of following the harvester and can carry the harvested product. This study has developed a navigation system for assistant robot using object detection based on deep learning. The image processing program uses the Python programming language and the deep learning models that have been tested on robots are SSD-MobileNet v2, Pednet, and Multiped. The deep learning model runs on the Jetson Nano device. The best detection results on SSD-MobileNet v2 with mean average precision of model is 72.7% and the sample detection accuracy is 88%.