Multi Label Leaf Disease Classification Using Enhanced Deep Convolutional Neural Network

K. Subhadra and Dr.N. Kavitha

Agriculture, the major cause of economic resource and a big society is occupied cultivation of several crops depending on criteria of environment. Farmers are facing huge challenges in cultivation, such as illnesses. Automated identification of disease in leaf and system of preventing the crops from the disease are much supportive. The previous work designed a system for recognition of disease in rice blast via Deep Convolutional Neural Network (CNN). However, it provides low performance since the infection area is not accurately segmented and also it does not support multi-label prediction. In this proposed research work, Region of Interest (ROI) based Bat Algorithm (BA) and prediction of multi-label leaf diseases are done by Enhanced Deep Convolutional Neural Network (ECNN) approach. Initially, dataset for both plant village and rice leaf diseases are considered input data. The preprocessing of leaf images is performed by means of median filtering approach. ROI based bat optimization performs segmentation of the preprocessed image to extract the exact disease regions. Segmentation accuracy is major objective of the bat algorithm. The Gray Level Co-occurrence Matrix (GLCM) and shape features are extracted on region of segmented area. Based on these features, Enhanced deep Convolutional Neural Network (ECNN) is used for multi-label leaf diseases prediction. To minimize the error value, weighted mean is computed in ECNN. ECNN proposed classifier results are evaluated with current approaches for various metrics like accuracy and sensitivity. Multi-label leaf diseases prediction is implemented by proposed models and the performance is measured by using real-world datasets.

Volume 12 | 04-Special Issue

Pages: 97-108

DOI: 10.5373/JARDCS/V12SP4/20201470