An advanced deep architecture for high accuracy Dianthus Species identification using Transfer Learning Methods

Flowers are botanically important structures and food resources for many insects and animals. Flowers also protect seeds to ensure the reproduction and survival of the plant species, which results existence of plants for generation after generation. Flower species detection is the process of identifying and classifying different species of flowers based on shape, size, and color. The identification of flower species helps to provide information about each species, and care instructions and also plays a vital role in flower cultivation, breed innovation, and in flower business. This species detection can be done both manually by botanists and by using machine-learning algorithms to analyze images of flowers. It helps to classify flowers into different species. Flower species detection can be useful in various fields including gardening, agriculture, and environmental monitoring. It can help researchers to study and better understand the different species of flowers and their characteristics. Dianthus is a versatile plant in flowerbeds or rock gardens. Dianthus offers an array of colors, easy cultivation, and the possibility of medicinal use. It is also used for culinary and in agriculture, it can be used for gardening and monitoring flowers’ growth and health. In this paper, we proposed an advanced deep architecture to identify species of Dianthus flower from images using transfer-learning techniques, which are DenseNet201, Inception v3, ResNet50, and ResNet101. We utilized end-to-end Transfer learning methods for our dataset by collecting 2716 Dianthus flower images of several different species. Among these methods, DenseNet201 achieves 100.00% accuracy which is the highest score in our dataset.