A Cloud-Based Recognition Service for Agriculture During the COVID-19 Period in Taiwan

Abstract

The great popularity of cloud services, together with the increasingly important aim of providing internet context-aware services, has spurred interest in developing diverse agriculture applications. This paper presents a cloud-based service built by incrementally integrating state-of-the-art models of deep learning, photography, object recognition, and the multi-functionalities of cloud services. This study consists of an object detection phase with a convolutional neural network (CNN) model, which involves enabling simultaneous image data gathered from drones. The experimental results show 97% accurate watermelon recognition. The results also include a two-model comparison in the cloud-based service, with the main findings demonstrating the feasibility of developing accurate object recognition using a CNN model without the need for additional hardware. Finally, this study adopted a confusion matrix to validate the result with RetinaNet for recognizing images taken on the watermelon farm with an average precision in recognizing watermelon quantity of up to 98.8%.