Early detection of ToBRFV disease in tomato plants using deep learning methods

In this thesis, early detection of tomato brown rugose fruit virüs (ToBRFV) seen in tomato plants was studied with artificial intelligence. In the experiment conducted in fully controlled greenhouses of Adana Biological Control Research Institute within the scope of the priority area project (ÖNAP) carried out by Eskişehir Osmangazi University, images were collected daily from plant leaves for 29 days for the classification of healthy and ToBRFV of commercial tomato varietiy. Four original datasets were created by taking images at 800 nm and 1000 nm wavelengths, 400-1100 nm wide spectral range and 400-700 nm visible light spectrum. Densenet169, Mobilenet V2, Resnet50 V2 and Xception deep learning models and a convolutional neural network (CNN) based model were proposed and used to determine in which wavelength or spectral range the early detection of ToBRFV is classified. In the proposed model, a more efficient and computationally cost-effective network model was created compared to other models by using the bottleneck residual block and channel attention module. Test analyzes of deep learning models were performed with two approaches. In the first approach, in order to understand how the symptoms of ToBRFV on plant leaves change over time, day-based cumulative and day-based classification was performed with three time series models. In this approach, both healthy and infected plants were detected with 100% success at 1000 nm wavelength with Xception and the proposed model. In the second approach, non-time-based classification was performed using images from all days and 98% success was achieved in the prediction of infected plants at 1000 nm spectral band with the proposed model. The results show that the proposed CNN model performs early detection of ToBRFV with high accuracy at 1000 nm spectral band in tomato variety.