Detection of Leaf Diseases Based on Ensemble Convolutional Neural Networks

The appearance of infections in plantations is the most common cause of crop losses in a variety of crops. Due to lower product quality and a narrower offering of items, these usually result in reduced income for farmers and higher pricing for consumers. Infection symptoms are mostly evident in plant foliage, hence they are widely used to detect and identify diseases. Combined with computational intelligence or soft computing approaches, image-processing tools can help farmers diagnose diseases more effectively. The disease in plants can be identified using traits gleaned from its symptoms. A strategy using an ensemble approach for providing more precise predictions by merging results through different machine learning algorithms. A group of models cooperating to address a shared issue. Ensemble learning uses the strength of a variety of techniques to overcome the particular weaknesses of each model, as opposed to depending just on one model to provide the optimal answer. The set ought to be less error-prone than a single model. Combining many iterations of a projected model is the process of bagging. Before being used, each model is independently trained using an average model. Bagging's objective is to reduce variance below that of any model. In this work, images of leaves comprising healthy class and diseases obtained from the Plant Village dataset are fed into three Convolutional Neural Network architectures namely LeNet, Inception, and VGG16 net achieve an accuracy of 95%, 92%, and 95% respectively and the ensemble model obtains an improvement of 3.5% matched to the standard method.