Multi-Layer Vision Transformer Approach for Identification and Classification of Flowers

Abstract

Beauty is incomplete without flowers. Flowers have been employed for many human-beneficial purposes. Currently, roughly 400,000 different flower kinds are available worldwide. The similarity in shape and color amongst the blooms causes them to vary from one another. The classification of flowers is a challenging subject because of the variety of their shapes, color distribution, illumination, and exposure deformation. Vision Transformer model is applied to explore the flowers shape and color with inter and intra similarity parameter for classification. The two datasets are tested for performance. One is 5-category flowers dataset that is already labeled publicly, and second one is Oxford 102 Flowers. Oxford flower dataset is preprocessed using SIFT algorithm to gain Isomap of different flowers parameters like as color features and shapes. After getting the parameters, the Vision Transformer (ViT) model is applied on feature parameters and our model achieve the competitive and reliable results. In addition, the proposed model has capability to identify and classify the flowers category, also produced necessary details including as flower classification and name for identification. The highest accuracy achieved through proposed model is 97.5% on 5-category flowers dataset and 99.31% on Oxford 102 dataset.