Plant disease identification and classification using Back-Propagation Neural Network with Particle Swarm Optimization

Abstract

Agriculture is the culture of land and rearing of plants to supply food to nourish and enhance life. Different types of plants are farmed every year based on environmental conditions and it is one of the main economic sources in India. These plants are prone to many diseases which hinders normal growth of the plants; these diseases are caused by seasonal changes, environmental variations, and cultivation procedures. To protect the plants from such damages, diseases need to be identified and properly diagnosed on time. Hence, innovation of feasible and powerful methods for identification and classification of plant diseases is an urgent need. There are lots of classifiers which are good in the classification of plant diseases: Back-propagation Neural Network (BPNN), Probabilistic Neural Network (PNN), Radial Basis Function Neural Network (RBFNN), Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) but only using these methods do not make the best tradeoff between time and accuracy. So to remove this constraint, in this paper we have given an image processing solution to distinguish and classify plant diseases efficiently and accurately. In our proposed method, for classification first, we use back-propagation algorithm to get the weights of neural network (NN) connections and then we optimize these weights using Particle Swarm Optimization (PSO) to come out of the problems like local optima and overfitting which are very common in conventional NN training methods. We have used images of leaves affected by different bacterial and fungal diseases: Alternaria Alternata, Anthracnose, Bacterial Blight and Cercospora Leaf Spot in our experiment and our proposed method achieves 96.2% accuracy.