Plant species identification based on modified local discriminant projection
- PDF / 1,157,343 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 35 Downloads / 204 Views
(0123456789().,-volV)(0123456789().,-volV)
IAPR-MEDPRAI
Plant species identification based on modified local discriminant projection Shanwen Zhang1
•
Wenzhun Huang1 • Zhen Wang1
Received: 18 May 2018 / Accepted: 20 September 2018 Ó The Natural Computing Applications Forum 2018
Abstract Plant species identification based on plant leaves is important for biological science, ecological science, and agricultural digitization. Because of the complexity and variation of the plant leaves, many classical plant species identification algorithms using plant leaf images are not enough for practical application. A modified local discriminant projection (MLDP) algorithm is proposed for plant species identification. MLDP aims to extract discriminant features for plant species identification by taking class label information into account based on the property of locality preserving. The MLDP can preserve the local geometrical structure of leaves and extract the strong discriminative ability. The experimental results on the public ICL leaf image database show the effectiveness and feasibleness of the proposed method. Keywords Plant species identification Maximum margin criterion (MMC) Local discriminant projection (LDP) Modified LDP (MLDP)
1 Introduction Plant species identification using leaf images is a topic of research [1, 2]. Feature extraction and selection from leaf images is an important step in plant classification [3–5]. Jyotismita et al. [6] proposed a plant species recognition method by using texture and shape features of leaves with neural classifier. Trishen et al. [7] proposed a plant species identification method by using shape and color histogram of leaves. A lot of feature extraction methods and dimensionality reduction methods have been applied to plant species recognition [8–10], but many classical leaf-based plant classification and recognition methods cannot be effective and enough for plant classification system, due to the diversity of plant leaf shape, texture, and the large difference between within-class leaf leaves. In recent years, a lot of manifold learning algorithms have been proposed for pattern recognition by discovering the low-dimensional intrinsic embedding feature from the original image data [11, 12]. Yan et al. [13, 14] proposed a marginal Fisher & Shanwen Zhang [email protected] 1
Department of Information Engineering, Xijing University, Xi’an 710123, China
analysis (MFA) algorithm for face identification. Chen et al. [15] proposed a local discriminant embedding (LDE) algorithm for dimensional reduction and face recognition. Li et al. [16] proposed a manifold learning algorithm, namely maximum margin criterion (MMC), by maximizing the difference between-class scatter and meanwhile minimizing the within-class scatter by a discriminant criterion to preserve the global structures of the samples. Yu et al. [17] proposed a local discriminant projection (LDP) method for pattern recognition. In the method, the weight matrices of within-class and between-class in objective f
Data Loading...
