PENERAPAN MOBILENETV3 DALAM DETEKSI KEKURANGAN NUTRISI TANAMAN PADI MELALUI ANALISIS CITRA DAUN BERBASIS ANDROID

Oryza sativa (Rice) is one of the main commodities in Indonesia’s agricultural sector. However, rice productivity often declines due to deficiencies in essential macronutrients such as Nitrogen (N), Phosphorus (P), and Potassium (K). This study aims to develop a nutrient deficiency detection system for rice plants through leaf image analysis using the Convolutional Neural Network (CNN) model MobileNetV3, integrated into an Android application. The method used combines a Research and Development (R&D) approach with quantitative evaluation. The dataset consists of 1,318 rice leaf images categorized into four classes: Nitrogen deficiency, Phosphorus deficiency, Potassium deficiency, and healthy leaves. The MobileNetV3 model was trained using TensorFlow, employing image augmentation techniques to balance the data distribution. The model was then converted to TensorFlow Lite format for deployment in an Android application built with Flutter. Evaluation was conducted using accuracy, precision, recall, and F1-score metrics, yielding a test accuracy of 52.63% and a macro-average F1-score of 0.5243. In addition, the application was tested through black box testing and user satisfaction surveys involving 10 farmer respondents. The questionnaire results showed an average feasibility score of 79.6% (Agree, close to Strongly Agree), with the highest ratings in detection speed (90%) and ease of use (86%), while the lowest rating was in detection clarity (68%). These findings indicate that the image-based mobile detection system can assist farmers in quickly and practically identifying nutrient deficiency symptoms. Although the model’s accuracy can still be improved, the system demonstrates strong potential as an adaptive and accessible precision agriculture tool.
Keywords: Nutrient detection, MobileNetV3, TensorFlow Lite, Flutter, Android application, CNN.