Deep Learning for Grapevine Leaf Identification: A MobileNetV3Large-based Approach

In viticulture, accurate identification of grapevine leaf kinds is vital since it helps to better manage the vineyard, avoid diseases, and increase grape harvest. This work aims to categorize five different grapevine leaf variants: Ak, Ala_Idris, Buzgulu, Dimnit, and Nazli by use of a refined MobileNetV3Large model. Using 100 high-quality photographs taken from the Kaggle platform, the dataset used in this study reflects the five variants with a rather equal distribution of images. While extra fully linked dense layers are fine-tuned for the purpose of leaf classification, the model architecture includes pre-trained MobileNetV3 Large layers, which remain static. Generalization of the model was enhanced by means of image pretreatment and augmentation methods. With precision and recall values ranging from 0.95 to 1.00 over the five grapevine varietals, the refined model attained an overall accuracy of 96%. The model also showed outstanding F1-scores; Buzgulu and Ak leaves respectively scored 0.98 and 0.97. Early grapevine leaf identification and sustainable vineyard management would benefit much from the model's accuracy and efficiency shown by these results. This work shows the possibilities of using deep learning models such as MobileNetV3Large for highly resource-efficient, high-performance agricultural uses.