Foliage-Pattern-Identifier: A Hybridized Grey Wolf & Adam optimized Deep Learning architecture designed for identification of similarly patterned leaves - Université d'Évry
Conference Papers Year : 2023

Foliage-Pattern-Identifier: A Hybridized Grey Wolf & Adam optimized Deep Learning architecture designed for identification of similarly patterned leaves

Abstract

Automated leaf pattern identification has emerged as a prominent research area in the field of computer vision, given the crucial role of leaves in monitoring plant health. Recognizing the significance of this research topic in preventing crop yield loss, we have developed a novel and computationally efficient deep learning architecture called Foliage-Pattern-Identifier for the classification of five distinct types of leaves: Wax gourd, Cucumber, Onion, Garlic and Tuberose. These leaves exhibit high inter-class similarities in their patterns. The designed architecture incorporates depth-wise convolution, inception blocks, and multiple skip connections, resulting in improved multi-scale contextual feature extraction and reduced computational cost. Additionally, our work introduces a hybridized optimization technique, diverging from the conventional Adam optimization method with default momentum values. In our approach, we leverage Grey-Wolf Optimization (GWO) to select the momentum values of the Adam optimizer. This hybridized optimization technique not only enhances the efficiency of our proposed network but also demonstrates benefits for other well-known networks. To facilitate our research, we have curated a unique database, FoliageDB, comprising a comprehensive collection of 699 leaf images belonging to the aforementioned five species. To the best of our knowledge, no existing databases encompass such samples. The designed network has given over-whelming performance in classifying these five similarly patterned leaves, achieving an overall test accuracy as high as 99.72%. The superior performance of our network, compared to popular existing networks, is demonstrated in the Results section.
No file

Dates and versions

hal-04355054 , version 1 (20-12-2023)

Identifiers

Cite

Sriparna Banerjee, S.K. Daud Hassan, Suranjana Mukherjee, Sheli Sinha Chaudhuri, Khalifa Djemal, et al.. Foliage-Pattern-Identifier: A Hybridized Grey Wolf & Adam optimized Deep Learning architecture designed for identification of similarly patterned leaves. Twelfth International Conference on Image Processing Theory, Tools and Applications (IPTA 2023), Oct 2023, Paris, France. pp.1-6, ⟨10.1109/IPTA59101.2023.10320017⟩. ⟨hal-04355054⟩
30 View
0 Download

Altmetric

Share

More