Precision Irrigation Decision Model Analysis Using Deep Learning and Embedded Systems

This study innovatively uses deep learning-based image processing methods to make crop fertilization decisions. Herein, we examined the images of tomato leaves captured by an embedded system camera using Raspberry Pi. We also considered the on-site classification of tomato leaf deficiencies and the extraction of information after the leaf segmentation, using the expert knowledge base and the crop fertilizer model. The study provides appropriate fertilization recommendations to improve fertilizer utilization. The convolutional neural network model trained in this study for tomato leaf deficiency recognition achieved 95% verification accuracy. The average pixel accuracy of leaf segmentation and mean intersection-to-union ratio achieved were0.89 and 0.76, respectively.