Comparison of Modern Description Methods for the Recognition of 32 Plant Species

Abstract

Plants are one kingdom of living things. They are essential to the balance of nature and people's lives. Plants are not just important to human environment, they form the basis for the sustainability and long-term health of environmental systems. Beside these important facts, they have many useful applications such as medical application and agricultural application. Also plants are the origin of coal and petroleum. In order to plant recognition, one part of it has unique characteristic for recognition process. This desired part is leaf. The present paper introduces bag of words (BoW) and support vector machine (SVM) procedure to recognize and identify plants through leaves. Visual contents of images are applied and three usual phases in computer vision are done: (i) feature detection, (ii) feature description, (iii) image description. Three different methods are used on Flavia dataset. The proposed approach is done by scale invariant feature transform (SIFT) method and two combined method, HARRIS-SIFT and features from accelerated segment test-SIFT (FAST-SIFT). The accuracy of SIFT method is higher than other methods which is 89.3519 %. Vision comparison is investigated for four different species. Some quantitative results are measured and compared.