Comparison of Accuracy in Detecting Tomato Leaf Disease with GoogleNet VS EfficientNetB3

Abstract

Tomato diseases vary greatly, one of which is tomato leaf disease. Some variants of leaf diseases include late blight, septoria leaf, yellow leaf curl virus, bacteria, mosaic virus, leaf fungus, two-spotted spider mite, and powdery mildew. By knowing the disease on tomato leaves, you can find medicine for the disease. So that it can increase the production of tomatoes with good quality and a lot of quantity. The problem that often occurs is that farmers cannot determine the disease in plants, they try to find suitable herbal medicines for their plants. After being given the drug, many plants actually died due to the pesticides given to the tomato plants. This is detrimental to tomato farmers. This problem is caused by incorrect disease detection. Therefore, this study aims to solve the problem of disease detection in tomato plants, in a more specific case, namely tomato leaves. Detection in this study uses a deep learning algorithm that uses a Convolutional Neural Network, specifically GoogleNet and EfficientNetB3. The dataset used comes from kaggle and google image. Both data sets have been pre-processed to match the data set class. Image preprocessing is performed to produce appropriate image datasets and improve performance accuracy. The dataset is trained to get the model. The training using GoogleNet resulted in an accuracy of 98.10%, loss of 0.0602 and using EfficientNetB3 resulted in an accuracy of 99.94%, loss: 0.1966.