Detection of Tomato Brown Rugose Fruit Virus in Anadol Peppers Using Deep Neural Networks and Multispectral Imaging

Tomato brown rugose fruit virus (ToBRFV) is a virus that has recently begun affecting plants such as tomatoes and peppers in fields and greenhouses, spreading rapidly. Once ToBRFV is detected, the spread of the virus can be minimized by removing and destroying the infected plants. The aim of this study is to investigate ToBRFV detection accuracies of two dense deep neural architectures for Anadol pepper plant using visible and near infrared images. In the study, Anadol pepper plants were cultivated in a fully controlled greenhouse environment. Images were taken from the plant using DSLR (Digital Single Reflex) and VNIR (Visible Near Infrared) cameras on a daily basis and data sets were constructed. Then binary classification was performed using the ResNet50 V2 and Inception V3 deep neural network (DNN) models, achieving over 81% accuracy in the results. The experimental results show that Inception V3 model gives better performance than ResNet50 V2. It has also been observed that the RGB images obtained from DSLR camera for Anadol pepper plant are classified with higher accuracy compared to VNIR images.