Nature's palette: from databases to networks for flowers, foliage, and gemstones classification

Abstract

The thesis entitled “Nature's Palette: From Databases to Networks for Flowers, Foliage, and Gemstones Classification” primarily focuses on real-life leveraging of smart agriculture and artificial intelligence. The work conducted in this thesis is mostly focused on curation of several databases which is essential for authenticated training of deep neural networks in order to prepare the networks to perform automated categorization tasks as well as design of light-weight dual attention network to achieve a perfect desired balance between computational and performance efficiencies. This thesis has mostly concentrated on curation of new databases for identification of distinct floral species growing in my native place Ranaghat which plays a significant role in enhancing the annual revenue growth of the floriculture industry of West Bengal as well as India. The thesis work is also extended to designing new databases focused on categorization of healthy and damaged insulin leaves which have inherent hyperglycemic properties and serve as natural diabetes healer, classification of five different foliage species which have immense medicinal properties and curation of a novel vegetable database comprising of fresh, damaged and severely damaged vegetables which is prepared for performing automated freshness evaluation of edibility of vegetables. The curation of databases have been one of the primary focus of this thesis owing to their notable significance in computer vision domain. Apart from curating new databases, the focus of thesis also lies in designing new deep neural based architecture for performing categorization task. This thesis also mentions a novel database of the precious gemstones which is created for performing automated classification task. This mentioned gemstone database fills the research gaps for designing an identification system in this domain. The novelties of this thesis work is highlighted as follows: • Newly curated databases 1. BongFloralpedia: A comprehensive collection of diverse types of flowers growing in fields of Ranaghat, Bengal 2. Nature's Diabetes Helpers: Insulin Leaves Repository 3. Herbal Leaf Repository: Unveiling Nature's Pharmacy 4. The Gem Atlas: A Pictorial Database of Precious Stones 5. Nature's Best vs. Bruised: A Veggie Evaluation • Deep neural architecture 6. GreenMedNet which is designed to perform categorization of damaged and healthy insulin leaves.