Deep Lichen-Net: Deep learning based automatic classification and segmentation of Lichens in Western Ghats, India

Abstract

Lichens are a symbiotic association between fungal and photoautotrophic algal partners that exhibit vast diversity in India with around 2,300 species recorded. In this research, a novel deep learning method known as LichenNet is proposed for the classification of lichens gathered from Western Ghats, India. Initially, the gathered images are denoised with Bright contrast dynamic histogram equalization (BCDC) filter to enhance the image quality and these are augmented to increase the images in the dataset. The Region of Interest (ROI) method is applied for generating the image patches by dividing the non-overlapping segments. The Dilated LinkNet is integrated with local and global sampling to extract the fine features while the Pelican Optimization (PeO) algorithm selects the best features for classification. The proposed LichenNet achieves the classification accuracy of 99.26%. The proposed LichenNet progresses the overall accuracy of 2.19%, 4.29%, and 14.36% for XGBoost, SIFT and CNN respectively.