DEEP LEARNING-BASED DETECTION AND CLASSIFICATION OF CARABAO MANGO LEAF DISEASES USING CONVOLUTIONAL NEURAL NETWORKS

Abstract

Leaf damages are mainly caused due to pests and diseases. Diseased leaves reduce crop production and affect the agricultural economy. The Philippines is one tropical country that produces carabao mango. However, its climate causes the variation of plant diseases that affect the yield of the agriculturist and the growth of mango trees. Agriculture plays a vital role in the economy; thus, image processing and machine learning technique are effective mechanisms to detect and recognize problem in an early stage. This study anal-yses the performance of two neural networks, a multilayer perceptron (MLP) an older type of network and convolutional neural network (CNN), a modern type of network as basis for the detection and recognition of carabao mango leaf diseases namely: anthrac-nose, sooty mold, red rust, and healthy leaf. There were 2856 disease and healthy leaf images captured using a mobile smartphone from the real cultivation conditions area. The 80% or 2285 was utilized for training and 20% or 571 were used for the validation tests of both healthy and disease leaves. Experimental results revealed that MLP achieved a performance rate of 86.30% and CNN generated 93.30% accuracy in the detection and recognition of carabao mango leaf diseases. From the result presented, the CNN and MLP neural network performance accuracy provide sufficient clues for the classification of a healthy and non-healthy carabao mango leaf