Deep neural network-driven precision agriculture multi-path multi-hop noisy plant image data transmission and plant disease detection

Abstract

Precision agriculture (PA) and plant disease detection (PDD) are essential for farm crops’ life quality and crop yield. Unfortunately, current PDD algorithms are trained and deployed with perfect plant images. This is impractical since PA sensor networks (PANs) transfer imperfect data due to wireless communication imperfections, such as channel estimation and noise, as well as hardware imperfections and noise. To capture the influence of channel imperfections and combat its effect, this work considers on- and/or offsite PDD implementation using plant image data transferred over multi-path imperfect PAN. Here, both traditional decode-and-forward (DF) data routing and channel effect considering machine learning data autoencoder multi-path routing are used for image data transmission. The multi-path DF data routing considers equal gain combining (EGC) and maximum ratio combining (MRC) techniques at the destination gateway for data decoding. In addition, a PDD deep learning algorithm is developed to predict whether or not a farm plant is diseased, using the noisy image data captured by the multi-path data routing PAN. From the PAN-PDD integrated system simulation, the proposed ML multi-path PAN-PDD algorithms (i.e., EGC and MRC) are compared to the ML single-path PAN-PDD algorithm and the traditional single-path PAN-PDD system. The simulation results showed that the multi-path approach performed fairly well over the other DF PAN-PDD systems. Incorporating the channel effects in designing an intelligent wireless data transfer solution/technique improves the communication system performance in PDD implementation.