Transfer Learning Implementation with MobileNetV2 for Cassava Leaf Disease Detection

Cassava (Manihot esculenta) is one of Indonesia’s key agricultural commodities but is vulnerable to various leaf diseases, such as Cassava Bacterial Blight (CBB) and Cassava Mosaic Disease (CMD). These diseases often exhibit similar visual symptoms, making it challenging for farmers to accurately identify them through manual observation. This study aims to develop an automatic cassava leaf disease detection system based on transfer learning, utilizing the MobileNetV2 architecture. The dataset used consists of 1,500 images, evenly distributed across three categories: CBB, CMD, and healthy leaves. The data underwent preprocessing, augmentation, and model training, including fine-tuning of the last 20 layers of the MobileNetV2 model. Evaluation results indicated that the model achieved an accuracy of 67% on the test set, with the highest performance in detecting Cassava Mosaic Disease, reflected by an F1-score of 0.75. These results demonstrate the potential of MobileNetV2 as a lightweight and efficient solution for detecting cassava leaf diseases, particularly when supported by a larger and more diverse dataset. This research serves as a foundation for developing mobile-based diagnostic tools to help farmers make faster and more accurate decisions in the field.