Detection and Counting of Grape Leaves Using YOLOv8 via TFLite on Mobile Applications

By creating a mobile app framework that uses the YOLOv8 model to identify and count grape leaves. The goal is to improve grape (Vitis vinifera L.) growth monitoring by accurately counting leaves, which is critical for evaluating plant development. YOLOv8 is optimized for mobile applications using TensorFlow Lite, ensuring efficient processing. The methodology of this research includes dataset acquisition, annotation, and training of the YOLOv8 model with 100, 200, and 300 images. The model was then implemented on Android devices using TFLite to customize the performance of YOLOv8. The findings show that the model achieves up to 93% detection accuracy with the largest dataset (300 images) and mAP50-90 of 60%. Detection speed and accuracy are affected by dataset size, with larger datasets improving generalization but slightly slowing down inference time. The integration of YOLOv8 and TFLite into a mobile app provides an accessible and efficient tool for farmers to monitor crop growth by detecting and counting from leaves on grapes. This innovation has a significant impact on precision agriculture by enabling more accurate and timely decisionmaking, ultimately improving agricultural productivity and sustainability. The novelty of this research lies in the successful application of advanced deep learning models in a mobile framework, which offers practical solutions to real-world agricultural challenges.