Flower Spices Detection and Classification Using Transfer Learning With Deep Convolutional Neural Network

Abstract

Flower species identification refers to a process of comparing defined characteristics of a given flower to allocate a particular species to a known taxonomic group. Biodiversity is declining rapidly, and the plant species in the world are disappearing faster than ever before. To categorize and classify a flower correctly, it is required to promptly identify the flower features and elements of a flower. Sometimes flowers might look identical, but they may differ with each other, which might create difficulty for a non-expert person to identify the flower correctly. Less significant work has been done to detect flower species diversity when more than one flower species is presented in the image or an overlap between flower species presented in one image. Flower objects sometimes appear in tiny or small shapes in a flower image. To locate and classify these small objects from the dataset is complex tasks, and for that, efficient techniques are needed to be used. Also, real-time flower images always own a complex background scene, and with a large number of species, it makes the task of flower object detection complex, time-consuming, and less accurate. Therefore, flower species detection comprises two tasks; image localization and image classification, as it is required to detect a flower in the image as an object and then recognize which species it belongs to. newlineRecently, Convolutional Neural Networks (CNN), an emerging field of Deep Learning (DL), has grown rapidly and widely applied to many domains with promising results, especially in object detection from visual images. Object detection and classification emerge as a vital research area used to solve the problems across many disciplines. Availabilities of large datasets, support of powerful processors, i.e., graphical processing unit (GPU) and advanced learning methods, make integration of CNN with computer vision accessible. newlinevi newlineThe major objective of this research is to bring out an optimized deep convolutional neural network model for classification and locali