基于CENTRIST的植物叶片识别研究与实现  

Research and implementation of plant leaf recognition based on CENTRIST

作　　者：王艳菲[1] 朱俊平[1] 蔡骋[1] 

机构地区：[1]西北农林科技大学信息工程学院,陕西杨凌712100

出　　处：《计算机工程与设计》2012年第11期4343-4346,4397,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(60975007);陕西省自然科学基金项目(2010JQ8019)

摘　　要：针对传统植物识别方法工作任务量大,效率低下,难以保证数据客观性的问题,将CENTRIST视觉特征描述符应用于植物叶片的自动识别,设计出一个基于植物叶片图像的在线自动识别系统,对叶片图像进行预处理并提取其CEN-TRIST特征,使用最近邻分类器进行分类,计算并查询网络数据库中与之最匹配的图像,能够快速识别常见的32种植物叶片,平均正确识别率达到了90%以上。系统的后台使用PHP语言实现,所有叶片图像的信息存储于网络服务器的MySQL数据库中。实验结果表明,该系统对叶片识别具有非常好的效果,而且易于实现,也方便数据扩充和用户使用。In view of the problem of traditional plant identification method for large workload, low work efficiency, and difficult to guarantee the objectivity of the data, CENsus Transform hiSTogram （CENTRIST） is used as our visual descriptor for plant leaf recognition. An online automatic leaf recognition system is designed. First, the leaf images are preprocessed, and then CEN- TRIST feature of leaves is extracted, at last, nearest neighbor classifier is employed for classify, the most matching images are computed and queried from the online database. Our system can fast recognize common 32 species of leaves now, and the average correct recognition rate is greater than 90%. The background implementation of this system is accomplished by using PHP language; information of all the leaf images is stored in a MySQL database of the webserver. Experimental results show that this system facilitates leaf recognition, and has better performance. The most important thing, it is very easy to implement and it also facilitates new leaf data expansion.

关 键 词：叶片识别 视觉特征描述符 特征提取 最近邻分类器 统计变换直方图 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]