<![CDATA[Operating artificial intelligence on smartphones attracted interest in various applications, but in practice, device capacity limited AI capabilities. Limited processing power, restricted memory capacity, and unstable network connectivity could make AI models difficult to use outside lab environments. In this work, we describe Leafy AI, a mobile application that identifies ornamental plants designed to work fully on the device. The classifier is based on MobileNetV2 and trained with transfer learning using 67,200 images from 112 plant categories. Images were resized to 224 × 224 pixels and normalized before training. After training, the model was converted into TensorFlow Lite format and integrated within a Flutter application. A lightweight service layer manages preprocessing and inference so that the interface remains simple for the user. Evaluation using 13,440 test images achieved a top-one accuracy of 0.89. A smaller field experiment involving 226 photos captured under real-world conditions resulted in lower accuracy, primarily due to variations in lighting and background. Nevertheless, the system remained reliable in offline mode. The findings show that recognition of ornamental plants can be carried out on ordinary smartphones and that further improvements are possible through augmentation, domain adaptation, quantization, and hardware acceleration.]]>
Copyrights © 2026
