Comparison of Convolutional Neural Network Architectures for Classification of Tomato Plant Diseases

Tomato plants are highly affected by diverse diseases. A timely and accurate diagnosis plays an important role to prevent the quality of crops. Recently, deep learning (DL), specifically convolutional neural networks (CNNs), have achieved extraordinary results in many applications, including the classification of plant diseases. This work focused on fine-tuning based on the comparison of the state-of-the-art architectures: AlexNet, GoogleNet, Inception V3, Residual Network (ResNet) 18, and ResNet 50. An evaluation of the comparison was finally performed. The dataset used for the experiments is contained by nine different classes of tomato diseases and a healthy class from PlantVillage. The models were evaluated through a multiclass statistical analysis based on accuracy, precision, sensitivity, specificity, F-Score, area under the curve (AUC), and receiving operating characteristic (ROC) curve. The results present significant values obtained by the GoogleNet technique, with 99.72% of AUC and 99.12% of sensitivity. It is possible to conclude that this significantly success rate makes the GoogleNet model a useful tool for farmers in helping to identify and protect tomatoes from the diseases mentioned.