摘要
针对植物分类识别算法在应用中的局限性,提出了带预处理项的尺度不变特征变换(SIFT)算法,该算法可以针对暴露在摄像头之下的植物叶片进行植物自动分类。首先,采集了53种植物叶片作为样本并搭建样本数据库,方便样本管理;然后,基于类间方差方法自适应地提取任意输入图片的前景,有效的前景提取可以提高识别的准确率;最后,采用尺度不变特征变换算法对输入图片进行匹配和识别。通过理论描述和实验证明,分别论述了新提出算法的有效性和可行性。结果表明,该分类识别算法应用局限性小,能克服图像采集过程中受到的光线、旋转以及拉伸等影响,可以有效地实现植物的分类和识别。
To address the limitation of plant classification algorithm in practical application, a Scale Invariant Feature Transform （SIFI＇） algorithm with pre-processing was proposed. With this algorithm, the plant leaf images could be acquired by camera and classified automatically. Firstly, 53 plant leaf samples were collected to set up a database for ease of the sample management. Secondly, the foreground of the input image was extracted using the Otsu＇ s method. The effective extraction could raise the recognition rate. Finally, the SIFT algorithm was adopted to classify and identify the input images. The effectiveness and feasibility of the proposed algorithm were described by theoretical formulation and validated by experiments, respectively. The results demonstrate that the proposed algorithm is efficient in overcoming the adverse effects caused by light, spinning and stretching while taking the leaves＇ photos. Thus it can effectively classify and identify plant species.
出处
《计算机应用》
CSCD
北大核心
2016年第A02期203-205,共3页
journal of Computer Applications
基金
厦门大学本科生创新训练项目(2015Y0644)
关键词
植物识别
前景提取
尺度不变
plant identification
foreground extraction
scale invariance
