Authors: Himanshu Shahoo, GautamYadav, ChinmayeeTripathy, Padmaja Panda
Abstract: Flower identification is a crucial aspect of plant classification and ecological research, playing a significant role in understanding biodiversity and ecosystem dynamics. This research paper presents a new approach to flower identification using advanced deep learning techniques. The proposed system used folding networks (CNNs) to automatically extract hierarchical features from high-resolution images of flowers, allowing for more accurate and efficient classification. The procedure is implemented as a multi-stage process, beginning with data preprocessing to enhance image quality and remove noise. Using another data record, educated CNN models such as modified reset 50, VGG16, or Google are then fine-tuned with commented flower images. Furthermore, transfer learning is used to properly use knowledge from large data records and improve the ability of models to generalize different types of flowers.. In the end, our approach achieved an accuracy of 82.04% using VGG16, the highest compared to other algorithms.
DOI: http://doi.org/10.5281/zenodo.16610684
