Classification of rice leaf diseases using transfer learning based on MobileNetV2 architecture

Rice leaf disease is a serious problem in the agricultural sector that can significantly reduce productivity. Early detection is necessary to minimize losses, but manual identification by agricultural experts is time-consuming and prone to error. This study aims to develop an automatic classification system for rice leaf disease using a Convolutional Neural Network (CNN) with a transfer learning approach on the MobileNetV2 architecture. The training process was carried out using several scenarios, namely Scratch, Fixed Feature, Fine Tuning First, Fine Tuning Middle, and Fine Tuning Last, with training parameters of 50 epochs, batch size 32, Adam optimizer, and learning rate 0.0001. The results indicated that the Fine Tuning First scenario worked best, with an accuracy of 78%, a Macro Average F1-score of 76%, and a Weighted Average F1-score of 78%. These findings indicate that the application of transfer learning on the MobileNetV2 architecture is capable of automatically detecting rice leaf diseases with fairly satisfactory results. In addition, this method has the potential to be further developed to support artificial intelligence- based smart farming systems.