Detection and Classification System for Cashew Plant Diseases using Convolutional Neural Network

Abstract

Cashew (Anacardium occidentale L.) is an important cash crop which serves as a major source of food, income and industrial raw materials to many African countries. However, cashew crops are commonly infected with various diseases that cause significant loss in yield. Existing automatic disease identification techniques are foliar identification-based while other infections appear on the nut and stem. In this work Convolutional Neural Network (CNN) was deployed for the detection and classification of diseases in leaves, stem and nut of cashew plant. Local dataset was sourced from LAUTECH teaching and research farm cashew plantation using Sony alpha A-330 digital camera with resolution of 10.20 Megapixels. The dataset comprises of 1050 sample images from three different parts of the various cashew plants that includes leaf, nut and stem. The database was divided into three sets for training, validation and testing in the ratio 50%, 30% and 20% respectively. Each of these sample images were preprocessed using histogram equalization and de-noised, using median filter. Feature extraction with classification and detection was done using transfer learning with CNN (ResNet-50) pre-trained deep learning network. The developed system was implemented in MATLAB R (2018a). It was evaluated using specificity, sensitivity, accuracy and error rate on the validation and test sets. The average performance was also computed. The system gave specificity, sensitivity, accuracy and error rate of 97.22, 98.19, 97.22, 2.78% on validations set and 97.56, 98.82, 98.29, 1.71% on test set, respectively. The average performance obtained for specificity, sensitivity, accuracy and error rate, were 97.39, 97.55, 97.76 and 2.25%, respectively. The system detected and classified four categories of cashew plant diseases that commonly occur on the leave, nut and stem.