Low-Cost Weed Identification System Using Drones

Abstract

Weeds compete with crops for resources such as light, nutrients, water and space. When mature, weeds can produce thousands to hundreds of thousands of seeds that can survive for a long time and posing a great threat to crops. The best way to avoid weed threats is to remove weeds before they bloom such that the chances of weed seeds falling into the soil can be reduced. Most existing drone-based weeds identification methods use additional equipments to enhance the identification ability. In addition to increasing the cost, such solutions also increase power consumption and load burden of drones. In this paper, we propose a low-cost Weed Identification System (WIS) using RGB images taken by drones as training data and applying Convolutional Neural Networks (CNN) to build the identification model. The result of the WIS can be used as a reference for agriculture researchers and can also be used to inform farmers to take necessary reactions. The WIS identifies weeds with an accuracy of up to 98.8%. Compared to other high-cost methods, the WIS does achieve similar identification accuracy at low cost.