Classification of Leaf Images with CNN and RF

Abstract

Agricultural production in a country highly decreases by infecting pests to the agricultural plants.
Today, in general, an agricultural engineer or a farmer tries to detect plant diseases by checking the plant
leaves, but this process is very hard and time-consuming because of harsh environmental conditions. Instead
of this, leaf images of plants can be controlled by drones automatically, and diseased plants can be detected.
In this work, leaf images have been categorized as diseased and healthy leaves using Convolutional Neural
Networks and Random Forests. The leaf data set which consists of healthy and diseased RGB leaf images
has been divided into a train data set and a test data set. The systems with Convolutional Neural Networks
and Random Forest classifiers have been trained in the train set. Convolutional Neural Networks include
feature maps and classification operations, but in feature maps, convolution, batch normalization, ReLU,
and max pooling operations are performed. For the Random Forest classifier, the training features are
obtained and trained from the feature map of the Convolutional Neural Network. After the training stage,
the trained models detect the diseased and healthy leaf images in the leaf image test set. For the evaluation
of the systems, the accuracy and F1-score metric values of the models have been computed, and they have
been compared with each other.