Regularization Proximity Censored Regressive Deep Reinforcement Learning for IoT aware Plant Disease Prediction

Agriculture is critical to food production worldwide and
is an important part of the global economy, but is faced with
significant threats to plants from diseases affected by climate
change. Early and accurate prediction of plant diseases is more
critical now than ever. The existing methods are not being able to
address the complicated problems of multi label disease
classification, we propose a new methodological approach to
address these limits by using a new Regularized Proximity Censor
Regressive Deep Reinforcement Learning (RPCRDRL) model.
The new proposed model contains five stages: image acquisition
using an IoT Framework, preprocessing using Hamann indexive
total regularization filtering in order to eliminate "noise" to help
provide accurate images, image segmentation using Moore
proximity censored regression to isolate segments that suggest
plant disease reference and isolate the ERGs (effectively removing
other plant parts), 4 - feature extraction from shape, color, and
texture features, and 5 - classification of plant diseases using Deep
Reinforcement Learning (DRL) with the Ruzicka similarity index
to improve accuracy and computing time. Our experimental
results indicate that the RPCRDRL model improved disease
identification of plant disease compared to traditional methods
while also improving computation time. The RPCRDRL model
significantly improved the model performance and real-time
organization process for a real-time plant disease prediction with
real-world data sets with accurate testing in agricultural scenarios
including accuracy rates, precision rates, recall rates, F1 scores
and prediction times.
Keywords: Plant Disease