Proceeding of International Agriculture Innovation Conference

Abstract

Declining birthrate and the industrial transformation have caused the insufficiency supply of manpower in the farming field. Hence, more farms have been left uncultivated, and less farming know-how will be passed on to next generation. Therefore, agriculture digital twins are helpful to reduce human labor request and to help inexperienced new farmers. In this article, we build two agriculture digital twin prototypes for planting banana and dragon fruits in Taiwan by using AIoT and image processing machine learning algorithms. We install LoRaWAN/NB-IoT weather stations and soil sensors in the fields to collect weather data as well as soil data including temperature, humidity, and EC value, and also SpeedDome video camera to collect the video images of fruits. Then machine learning algorithms (based on convolution neural networks, CNN) are designed to train the fruit growth model, and then the model is used to analyze and predict the crop’s growth rate and yield estimation. This digital twin model can be referred by farmers to understand the key factors of crops growing and improve the processing of irrigation and fertilization, including the timing and volume of water and fertilizer.