提出一种基于整体植物外观特征提取的植物自动识别方案.首先,用普残差法对植物图像进行显著性区域检测,较粗略地得到植物对象,再结合色调信息进行细分割.接着提取该对象区域的SIFT特征作为底层局部特征,建立视觉词包模型,最后设计分类器进行分类.选取了9种常见的室内盆栽,每种植物各28个样本.在实验中,分别对比当前流行的BP神经网络、SVM和ELM三种分类器的分类性能.实验结果发现,支持向量机和极限学习机有较好的分类效果,识别率可以达到90%左右.这对植物识别的研究及应用推广都具有一定的积极作用.%In this paper,we propose an algorithm for plant species recognition based on whole appearance features.First,the Spectral Residual method was adopted in salient region detection to segment the plant object roughly.And then,the hue information was used to obtain the precise object.Second,SIFT in the object region was extracted to build the BOV model.Finally,three classifiers were designed and implemented to classify the plant species.In our experiments,there were nine different plant species,and 28 examples of each species.BP neural network,SVM and ELM,these three different classifiers were implemented and compared.The experimental results show that the SVM and ELM classifiers were better than BP neural network,and are able to identify about 90% of these plants correctly.It is important for the research and application of plant species recognition.
展开▼