A hybrid approach combining CUR matrix decomposition and weighted kernel sparse representation for plant leaf recognition. Issue 8 (14th September 2021)

• Record Type: Journal Article
• Authors: Pankaja, K.; Suma, V.
• Is Part Of: International journal of computers and applications. Volume 43:Issue 8(2021)
• Page Start: 830
• Page End: 840
• Publication Date: 2021-09-14
• Subjects: Plant species recognition -- CUR decomposition -- dimensionality -- dictionary learning -- sparse modeling -- weighted kernel sparse representation; Computers -- Periodicals; Computer software -- Periodicals; Computer networks -- Periodicals; Multimedia systems -- Periodicals; Internet -- Periodicals; World Wide Web -- Periodicals; Minicomputers -- Periodicals; Microcomputers -- Periodicals; 004.05
• DOI: 10.1080/1206212X.2019.1632031
• Languages: English
• ISSNs: 1206-212X

Abstract

Abstract : Plant species recognition is one of the important and difficult research areas. Many existing plant recognition and classification methods cannot meet the requirements of the automatic plant recognition system, due to the irregular, complex, and diverse nature of plant leaves. In this paper, a plant recognition approach is proposed by combining CUR decomposition and weighted kernel sparse representation (WKSR). The proposed method is different from traditional plant leaf classification methods. Instead of establishing a classification model by extracting the color, shape, and texture classifying features, the proposed method directly reduces the dimensionality of image and recognizes the test samples based on WKSR coefficients. In order to reduce the recognition time the proposed method uses class specific dictionary learning for sparse modeling. By using the comparison analysis, the proposed method is verified on four plant leaf datasets namely, Flavia leaf dataset, Swedish leaf dataset, Original leaf images and Leafsnap dataset with existing plant classification methods such as Random Forest, Support Vector Machine and K-Nearest Neighbors. The accuracy of the proposed method is 98% which is the best classification rate.