Weed Identification and Removal using Machine Learning Techniques and Unmanned Ground Vehicles

摘要

This paper presents the use of unmanned ground vehicle (UGV) and machine learning techniques for the identification and removal of weeds in lettuce crop. In recent years, breakthroughs in deep learning, computer vision, and miniaturization of electronic devices have paved the way for use of unmanned systems and machine learning techniques for applications that are dull, dirty, and dangerous for humans including agricultural applications. Unmanned systems and machine learning techniques have potential to transform and modernize how the crops are grown and cared. One of the problems every farmer encounters is invasive weeds that can kill or hinder the growth of crops by stealing water, nutrients, and sunlight from the plants. Herbicides are used to kill and stop the growth of weeds. However, use of herbicides increases the cost of production, is labor intensive, and exposes human to dangerous chemicals. Manually removing the weeds is also very labor intensive. Using machine learning techniques and UGVs for the identification and removal of weeds will reduce the cost of production, human exposure to dangerous chemicals, and dependence on human labor. Models were trained using YOLO, Faster R-CNN, and SSD Mobile object detection techniques. For the training of machine learning models, images of the weeds in an experimental lettuce plot was collected throughout the growing season. Validation of the developed models was performed using different data sets than the training data sets in the same plot as well as a different plot. The identified weeds were then removed using the UGV through teleoperation.