A Comprehensive Biometric Approach for Taxonomic Identification of Plant Species of the Western Ghats

Abstract

Diversity is a remarkable feature of organic species. In spite their differences the organisms are newlinegrouped into taxa. The assignment of an unknown living thing to a taxon is called identification. newlineKnowledge of the species is crucial for conservation of the biodiversity of a region. The world newlineinherits a very large number of plant species. Current estimates of flowering plant species newline(angiosperms) range between 220,000. Plant identification is essential for ecological newlinemonitoring and thereby especially for biodiversity conservation. The conventional manual newlinetechniques used to identify plants is a cumbersome, complex and time-consuming process and newlineit requires the skills of an experienced botanist to identify rare and endangered species of plants. newlineIt can be a huge hurdle for novices or beginners who would want to acquire species knowledge, newlinewhich is hard to overcome. The declining and partly non-existent taxonomic knowledge within newlinethe general public has been termed taxonomic crisis . newlineThe primary goal of this research is to automate plant identification and classification newlineby utilising the leaf as a unit of organisation. Additionally, the research focused on developing newlinea novel dataset, the western ghats leaf dataset. Plant leaves come in a variety of forms, both newlinesimilar and dissimilar between and within species. As a result, developing a common strategy newlinecapable of classifying a wide variety of leaf forms without prior classification is rather a newlinedifficult undertaking, and the inadequacy of the methods provided in the literature is reflected newlinein their low recall rates. newlineWe collected the requisite photos of Western Ghats leaves for the construction of a newlinestandard dataset during our research from the Pilikula biodiversity reserve, the data has newlineundergone rigorous pre-processing for deem it suitable for classification. Using various data newlineaugmentation techniques, increased the dataset from 950 to 121000 images. Extracted features newlinefrom a particular leaf sample using a variety of feature extraction approaches, which newlinecontributed significantly in selecting the appropriate features for classification. After denoising newlinethe images, they were subjected to a variety of machine learning techniques to get varying newlinedegrees of accuracy. We have constructed and developed innovative deep learning models newlineLeafNet-1, LeafNet-2 and LeafNet-3 for identification purposes. The models have achieved newlineaccuracies ranging from for these models. in the next stage we have also done a comparative newlinestudy to assess the performance of our novel models to check their performance analysis. newline