基于多轮廓距离形状描述子的植物叶片图像识别  

Leaf Image Recognition Based on Multi-profile Distance Shape Descriptor

作　　者：叶梦婕 吴旭姿 李海波 孙蕊 程勇 YE Mengjie;WU Xuzi;LI Haibo;SUN Rui;CHENG Yong(School of Information and Engineering,Jiangsu Open University,Nanjing 210003,China)

机构地区：[1]江苏开放大学信息工程学院,江苏南京210003

出　　处：《江苏海洋大学学报（自然科学版）》2023年第1期106-113,共8页Journal of Jiangsu Ocean University:Natural Science Edition

基　　金：国家自然科学基金资助项目(61972187);江苏省高等学校基础科学(自然科学)研究面上项目(21KJB510036)。

摘　　要：植物叶片形状一般具有小的类间差异和大的类内变化,加上叶片图像仿射变换,给叶片形状识别带来很大挑战。首先从叶片轮廓曲线偏移角度提出了最小投影距离特征和相对投影距离特征,并结合表征轮廓凸凹特性的拱高距离特征,提出一种多尺度多轮廓距离形状描述子。所提算法不仅能从两个正交维度较全面地描述不同尺度轮廓曲线特征,而且具有平移、缩放和旋转不变性。在Swedish和MEW2012叶片数据库上的实验结果表明,所提算法明显优于现有有效轮廓描述算法。此外,提出一种人工特征和深度特征归一化策略,探索低层次轮廓特征和高层次语义特征融合机理,在Swedish和MEW2012叶片数据库上的实验结果表明,融合特征能够显著提高叶片图像识别性能。The generally small inter-class difference and large intra-class variation of leaf shapes,and affine transformation of leaf images bring great challenge to leaf shape recognition.In this paper,the minimum projection distance feature and relative projection distance features are firstly proposed based on the offset of blade profile curve with combining the arch height distance features representing the convexity and concavity of the contour.The proposed can not only describe the features of contour curves of different scales comprehensively from two orthogonal dimensions,but also has translation,scaling and rotation invariance.Experimental results on Swedish and MEW2012 blade databases show that the proposed algorithm is superior to existing effective contour description algorithms.In addition,a normalization strategy of hand-crafted features and deep features is raised to explore the fusion mechanism of low-level contour features and high-level semantic features.Experimental results on Swedish and MEW2012 leaf databases show that the fusion can significantly improve the performance of leaf image recognition.

关 键 词：距离特征 形状描述子 叶片图像识别 特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]