Vision-based model for maize leaf disease identification : a case study in Nyeri County

Abstract

Biotic stress which includes pest and diseases affect crop productivity due to either death of affected crops or reduced yield per crop. Abiotic stress such as water and temperature also contribute to lower yields. Maize is Kenya’s staple food with most households having limited choices of other foodstuffs thus increasing their reliance on maize. Diseases affecting maize in Kenya include: Maize Grey Leaf Spot disease, Maize stem borer, Maize Lethal Necrosis Disease, Ear Rot, Stem Borers, and Maize Streak Virus. Currently, the human visual examination is the most commonly used method for classifying diseases. The method gives room for a lot of errors as the diagnosis is based on the experience of the farmer or the extension worker. The method also takes a great deal of effort and time to identify crop diseases based on the visually observable characteristics. Different experts diagnose the same disease as a different disease due to their varied experiences leading to erroneous identification of diseases. Introduction of artificial intelligence in various aspects of agriculture has gained momentum in today’s world. Artificial intelligence has seen its application in predicting soil organic matter based on remote sensing data as well as in prediction of crop yield based on factors of production and in identification of crop diseases. The research sought to propose use of an artificial intelligence model for identification of maize leaf diseases. In the proposed model, images of maize leaves were acquired and extracted color features used to identify the specific disease. Artificial Neural Network was used to identify the disease by implementing a back propagation learning algorithm. The data obtained was segmented into training and test data for the model. The algorithm was preferred due to its strengths in adaptive learning, its fast processing speed and the accuracy of its output. The performance evaluation of the model was based on the accuracy of the classification, the precision, recall ratio and the F- Measure. The model was proven to be significantly accurate with an accuracy of 78.94 % while the precision obtained was 0.778. The recall ratio from the neural network was 1 and an F-measure of 0.875.