Evaluation of deep learning models for RGB image-based detection of potato virus y strain symptoms (O, NO, and NTN) in potato plants

Abstract

Potato virus Y (PVY) has been a long-standing problem for potato growers over the world, due to its ability to cause significant reductions in crop yields. The yield losses due to PVY may range from 10% to 80%, depending on the severity of the infection and the potato variety. The new necrotic strains of PVY cause mild symptoms in the foliage, making it challenging to detect infected plants. Consequently, identifying and disposing of infected plants (known as “roguing”) has become more difficult. There is a growing demand to create solutions that aid growers in identifying potato plants that have been infected with PVY. In past studies, deep learning-based convolutional neural networks (CNNs) have shown the ability to successfully make distinctions between various healthy plants, disease plants, and weeds. In this study, the use of these models for the detection of infected plants with different strains of PVY has been explored and extended. Different deep learning models, specifically EfficientNet, VGGNet-19, DenseNet-201 and ResNet-101 are trained on the imagery dataset of healthy and PVY-infected potato plants grown under greenhouse conditions. The evaluation metrics used were accuracy, precision, recall, and F1 Score. The trained models achieved classification accuracy scores of 85% while classifying the healthy and PVY-infected potato plants. The models were also able to accurately detect PVY-infected plants even when the symptoms were mild, which is essential for early detection and prevention of the spread of the virus. These models may assist roguers in the real-time identification of PVY-infected plants that may help in controlling the disease spread and improving the crop yield.

Keywords

• PVY
• Smart agriculture
• Machine learning
• Artificial intelligence
• CNN