Real-Time Classification of Hydroponic Vegetable Types on Mobile Devices Using Lightweight Deep Learning Models

Abstrak

Hydroponic cultivation requires precise monitoring to ensure crop quality and productivity, yet manual identification of vegetable varieties and their growth status remains labor-intensive and prone to error. This study aims to develop a real-time, mobile-based classification system for hydroponic vegetables using lightweight Deep Learning models optimized for edge computing. The proposed method evaluates two distinct architectures, MobileNetV3 and YOLO-Nano, trained via transfer learning on a dataset comprising major hydroponic crops such as Lettuce, Pak Choy, Mustard Greens, and Cherry Tomatoes. Experimental results demonstrate that while YOLO-Nano offers superior inference speed (~55 FPS), MobileNetV3 achieves a significantly higher classification accuracy of 96.4% while maintaining a real-time performance of ~35 FPS on standard mobile hardware. The study concludes that MobileNetV3 provides the optimal balance between accuracy and computational efficiency for handheld agricultural applications. This research contributes a scalable, low-cost solution for smart farming, enabling producers to perform rapid, on-site digital inventory and quality assessment without reliance on internet connectivity.