Local Flower Detection Through Deep Learning Approach

Abstract

Flowers are some of the most beautiful creations of the Almighty. There are numerous seasonal flowers in Bangladesh. Each flower has its unique color, shape, and name. In our daily lives, we can see a variety of flowers when walking along highways, rail lines, and even in our gardens. However, the majority of our population is unaware of the significance of this beautiful creation. The identification and classification of local flower species using deep learning techniques has important implications for botanical research, biodiversity protection, and ecological monitoring. In this paper, we present an in-depth analysis of multiple deep learning models for detecting and classifying flowers from a dataset of 2400 photos of various flower species the tested models consist of VGG-16, MobileNet, M-1, and ResNet-50. Each model was selected based on its distinct architecture and established efficacy in image recognition tasks. The results of our study indicate that although all models work well, there are significant variations in their abilities. Specifically, the VGG16 model demonstrates great potential for real-time applications and achieving high accuracy of this dataset, with a rate of 97.73%.Six different kinds of flowers have been collected for our research. Our dataset consists of 2450 photos overall and 410 images per flower. We allocated 60% of the images to the training set in order to train the dataset. The validation set is assigned 20% of the total data, while another 20% is allocated to the test set for the purpose of testing. To improve the models' robustness and generalizability, we apply preprocessing techniques like resizing, normalization, and data augmentation. Six different kinds of flowers are used in this system. However, we plan to add more flowers in the future to make our system better.