Plant Leaf Recognition for Ficus Deltoidea Jack (Moraceae) Varieties: A Hyperparameter Tuning Classifier on Feature-Based Transfer Learning

The identification of plant species through leaf characteristics is a significant area of study within computer vision applications. However, research on modeling for the recognition of Ficus deltoidea varieties is currently limited, with only a few existing models available, mostly relying on traditional methods. The application of Transfer Learning (TL) has been demonstrated as a powerful method for extracting essential features, leading to a substantial reduction in training time. Furthermore, the feature extraction model has exhibited outstanding performance within the TL approach across a wide range of applications. The aim of this research is to establish a suitable pipeline that combines TL with traditional classifiers to proficiently distinguish between different varieties of F. deltoidea. In the process of feature extractionVGG16, VGG19, InceptionV3 models were used and then integrated with either k-Nearest Neighbor (k-NN) or Support Vector Machine (SVM) classifiers. A dataset comprising 420 F. deltoidea leaf images from six different varieties was collected, and it was subsequently split into an 80:20 ratio. The optimization of hyperparameters for both k-NN and SVM classifiers was conducted using the Grid Search method. The findings demonstrated that the most effective pipelines were attained by pairing VGG16 and VGG19 with the k-NN classifier, although the SVM classifier also yielded favorable results. The top-performing pipeline achieved a macro-average classification accuracy of 0.98 on both the training and test sets. To summarize, these results affirm that TL models have effectively showcased their ability to distinguish between F. deltoidea images for variety classification.