Detection and Classification of Plant Leaf Disease Using Local Directional Pattern and CNN
Keywords:
Convolutional neural network, image classification, Kirsch edge detector, local binary pattern, local directional pattern, plant leaf disease detection, texture descriptorAbstract
The detection of plant leaf diseases plays a pivotal role in advancing agricultural productivity. Traditional methods
that rely on human expertise are being supplemented by artificial intelligence and computer vision. This paper applies
a descriptor called the Local Directional Pattern (LDP) to improve feature generation from plant leaves for disease
detection in apple crops. LDP uses Kirsch edge detectors to capture the textural and edge information of a leaf image,
which is what distinguishes healthy specimens from diseased ones. Because edge responses are more stable than raw
intensity values, the descriptor holds its pattern under noise and non-monotonic illumination change, which is a known
weakness of the Local Binary Pattern it builds on. The descriptor is benchmarked against features derived from
Convolutional Neural Networks (CNN) and the Histogram of Oriented Gradients (HOG). The system was
implemented in Python using TensorFlow and Keras in the Spyder environment, and was trained and tested on the
Plant Village dataset. It classifies apple leaves into four categories: healthy, cedar rust, scab and black rot. The results
indicate that LDP is a reliable feature extraction tool for the agricultural sector, supporting crop management and
apple yield prediction, and that combining edge-based descriptors with machine learning classifiers is a direction
worth further exploration.
