Low-Altitude Unmanned Aerial Vehicle for Real-Time Greenhouse Plant Disease Monitoring Using Convolutional Neural Network

Abstract

This study presents a novel approach to real-time plant infection detection and automatic pesticide spraying using an unmanned aerial vehicle (UAV) inside greenhouses. Greenhouse maintains controlled environments for plant growth, which is presently achieved by using conventional labor-driven methods that are proven to be inefficient causing lower yields. To overcome such difficulties and improve the yields, a multipurpose unmanned aerial vehicle capable of capturing vision data has been developed to detect infected areas and automatically spray useful chemicals based on the detection. Onboard the UAV is an edge device connected to environmental parameters sensors continuously uploading data to ThingSpeak. The IOT cloud platform provides real-time temperature and humidity of the precise location. In this research work, a fast-semantic segmentation algorithm called LinkNet-34 is employed for real-time segmentation of the infected region. The experimental results during manual flights indicate a detection accuracy of 0.922 (MIoU) with LinkNet-34. The UAV can achieve 14 min of flight-time while spraying 500 mm of pesticide over 42 m2 area, during which time, a field map highlighting the infected regions is automatically generated and uploaded to the cloud for future analysis.