<![CDATA[The clove plantation sector (Syzygium aromaticum) in Indonesia plays an important role in supporting national economic stability. However, its productivity faces threats from Plant Pest Organisms (PPOs), such as Clove Leaf Cacar Disease (CDC), Sooty Mold, and Dieback, which can cause losses of 10%-40% of total harvest yields. Conventional visual diagnosis is prone to subjectivity and human error, while the implementation of large-scale deep learning models is constrained by the computational capacity of end-user devices. This study aims to develop an objective and lightweight diagnostic system through the CloveVision web application based on the MobileNetV2 architecture. The study employed a quantitative experimental approach involving data collection, image preprocessing, model development using transfer learning and fine-tuning, system conversion, and evaluation. The dataset consisted of 1,000 clove leaf images collected from Tajun Village, Buleleng, Bali, with 250 images per class. The model was converted to TensorFlow.js to facilitate client-side execution directly in a web browser. The test results showed an accuracy of 78.75% on an independent test dataset, a model size of 9.24 MB, and an inference time of less than 100 milliseconds without an internet connection. Black-box testing achieved a 100% success rate, while usability testing using the System Usability Scale (SUS) with 15 farmers produced an average score of 86.67 (Best Imaginable). The study concludes that MobileNetV2 effectively balances accuracy and computational efficiency to support smart farming.]]>
