Detect of cassava diseases by computer vision methods

Abstract

Background. Development of a convolutional neural network model for detect-ing cassava diseases from a mobile phone photo. Materials and methods. The material for the research was taken images with various types of cassava diseases, published in open access of the Kaggle platform. Research methods: theory of design and development of information systems, pro-gramming, methods of augmentation and extension of datasets for computer vision problems, methods of tuning hyperparameters for training neural network models. Results. Cassava is one of the key crops for agriculture in many regions of the world. One of the main reasons for poor yields is a different type of disease. For the prevention and early warning of the spread of plant diseases, a tool is needed in the form of a neural network model that allows to determine the presence of the disease from a photo from a smartphone. We used the methods of deep learning of convolutional neural networks, as well as the concept of “transfer learning”. On the basis of the ResNet 50 network, the neural network model was trained that allows determining the presence of disease in the cassava plant from the image with accuracy 0,93 according to the F1-score metric. Conclusion. Has been prepared the dataset of cassava images, included five classes, for efficient classification by the neural network. Four classes with signs of certain cassava leafs diseases and one class for healthy plants. Has been built and trained model for the task of classification to detect cassava leafs disease by images from a smartphone.