Publication

Abstract : Plant diseases when not treated properly can pose a potential threat to food security. Thus, rapid detection of crop disease is essential to treat the disease early. Traditionally, leaves were tested in the laboratory and plant diseases were detected by observing the lesion appearance. However, nowadays, people are getting used to smartphones and computer devices. Farmers and every other individual are demanding new technologies which can bring an easier process to detect plant disease. Thus, Deep Learning (DL) algorithms are being incorporated inside smartphones to fasten image recognition and disease detection. Researchers are using various DL algorithms and providing training with datasets to detect plant disease from captured images by smartphones. In this study, primary research has been accomplished with a small training dataset (independent variable) to understand how the accuracy of “Convolutional Neural Network” or CNN (dependent variable) is affected. The linear regression analysis has been carried out to understand how independent variables impact the dependent variable. Independent variables are the “number of plant disease images”, their resolutions, features and categories. The dependent variable is the accuracy of the CNN algorithm. After linear regression, it has been found that the number of images in the dataset positively impacts the accuracy where a great number of images will require removal of background to improve accuracy. Image resolution does not have any positive impact on CNN accuracy when the features are clearly visible.