A rational method for assessing biotic stresses in plants using machine learning

Abstract

One of the ways to replenish antioxidants in the body is to add a share of berry crops rich in antioxidant compounds to the human diet. Garden strawberries are the most grown and consumed berry crop in the world. The restraining factors for increasing the production of garden strawberries are the significant damage of cultivated varieties to a wide range of diseases, the main pathogens of which are fungi. The Russian Federation does not produce technical means for diagnosing diseases available to producers of this berry crop. The purpose of the research is to develop a rational method for assessing several biotic stresses of garden strawberries for a wide range of farmers. Among modern ground-based diagnostic methods, preference is given to the computer vision method, which is capable of detecting the presence of 3 pathogenic fungi: white spot pathogens (Ramularia Tulasnei Sacc), brown spot (Marssonina potentillae Desm) and angular leaf spot (Dendrophoma obscurans). Using deep learning with convolutional neural networks (CNN), this method can be implemented as an application for a smartphone or other gadgets popular among the majority of the population. The most common CNN models were selected for deep learning. A dataset of 2,671 images was generated from the Internet of Things (IoT) network and divided into 4 classes: white spot (544 pcs.); brown spot (1,109 pcs.); angular spot (392 pcs.); unaffected leaves (626 pcs.). The dataset was divided into a training set of 70%, a validation set of 10%, and a test set of 20%. When conducting deep learning and neural network model improvement operations, the MobileNetV2 neural network model showed the best metrics: classification accuracy – 0.99; F-measures from 0.94 to 1.00.