DEVELOPMENT OF AN IMAGE-BASED DETECTION OF MAIZE LEAF DISEASES USING CONVOLUTIONAL NEURAL NETWORK

Abstract

The area of agricultural information, automatic identification and classification of maize leaf diseases are highly desired and sought-after. One of the major problems facing the agricultural industry as a whole is the need to accurately and promptly detect diseases affecting plants and crops. Diseases have capability of affecting the quality and quantity of plants and crops by wiping out hectares of crop field resulting in low productivity and economic loss to farmers. The current diagnostic approach employed by farmers is mainly trial and error method which is time-consuming and requires physical presence of experts to analyze the affected crop, understand the symptoms, identify the disease, and suggest appropriate remedies. The limitation of the conventional approach has motivated the need to look for viable alternative approaches that can detect and classify maize leaf disease at an early stage. This study developed a Convolutional Neural Network-based (CNN-based) approach to identify and classify maize leaf disease. To improve the identification and classification accuracy of maize leaf disease model, a pre-trained Visual Graphic Group (VGG-16) architecture was employed as the base model for the newly developed CNN model. The model training was performed on a dataset with six distinct classes of maize leaf disease including healthy maize leaf images. The model classification was evaluated in terms of precision, recall and f1-score. The developed CNN model achieved an accuracy of 92.54% with 50 epochs with 0.001 learning rate. The maize leaf classification model also achieves a precision of 0.92, recall of 0.92, and f1-score of 0.92 all on scale of 1.00. The recorded performance metrics indicated that the model can identify and classify maize leaf disease, thereby meeting the need of creating a robust and effective maize leaf disease classification model. The results obtained proved that this research study can be further extended to form a comprehensive part of a general plant disease identification system capable of operating in real-time scenarios. This can as well empower the farming community by accurately identifying and knowing the exact type of disease they are battling with and initiate a timely treatment at the appropriate time.