A Composite Deep Learning Access for Leaf Species Classification
Yashvardhan Singh Rautela1, Kanu Garg2, Harshit Singh Chhabra3, Rahul Nijhawan4

1Yashvardhan Singh Rautela, Graphic Era University, Dehradun (Uttarakhand), India.

2Kanu Garg, Graphic Era University, Dehradun (Uttarakhand), India.

3Harshit Singh Chhabra, Graphic Era University, Dehradun (Uttarakhand), India.

4Rahul Nijhawan, Graphic Era University, Dehradun (Uttarakhand), India.

Manuscript received on 08 September 2019 | Revised Manuscript received on 17 September 2019 | Manuscript Published on 11 October 2019 | PP: 42-45 | Volume-8 Issue-11S September 2019 | Retrieval Number: K100909811S19/2019©BEIESP | DOI: 10.35940/ijitee.K1009.09811S19

Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open-access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: Plants are an integral part of the human life one way or the other. They have multi-dimensional use as food, medicine, clothing, art, industrial raw material and are vital for sustaining the ecological balance of our planet. All these real life applications make the identification of plants intensely important and useful. This dictates to design an accurate recognition system of plants. It will be useful to facilitate faster classification, management and apprehension. Almost all the plants are accompanied by unique leaves. In this paper, we have used this property of leaf identification for the identification of plants. In this study, we have applied a composite deep learning model, where Inception-v3 model is used for feature engineering and Stacking Ensemble model is used for the detection and classification of leaves from images. We have used a modified Flavia dataset of 1287 leaf images divided amongst 21 distinct plant species to test the proposed approach. on comparing our proposed work with other pre-existing algorithms (RF, SVM, kNN and Tree), it is found that it surpassed them, obtaining an accuracy of 99.5%.

Keywords: Deep Learning, Leaf Classification, Hybrid, Feature Extraction.
Scope of the Article: Deep Learning