Estimation of Crop Yield using Deep Learning for Precision Agriculture

Abstract

Precision agriculture is the application of correct amount of fertilizers and water pesticide to achieve higher agricultural productivity. Furthermore, under the framework of precision agriculture is the automated estimation of yield with advanced technologies including Artificial Intelligence (AI) and Remote Sensing (RS). The use of RS has advanced crop yield estimations and predictions in recent years. However, to validate RS-based models it is important to perform in-situ exercises such as fruit counting, which is a time-consuming task that increases the production costs. Drones, robots, and in-situ cameras in combination with AI algorithms are widely used to efficiently address these issues. The recent advancement in computational resources and power available has enabled the utilization of Deep Learning AI models. One of the best-performing models for object detection is the You-Only-Look-Once (YOLO). In this study, the YOLOv5s is used for object detection, which is the second smallest and fastest YOLOv5 architecture, on two different benchmark datasets collected from AgML. The first dataset consists of 1730 images of mango trees in Australia during night, and the second dataset consists of 6512 images of wheat heads collected from different regions around the world. The main objective of this work is to demonstrate the capabilities of light AI models for object detection and to evaluate their performance, which will serve as a benchmark for future comparison with the on-board environment.