利用CBIA与WSN构建的植物叶片分类系统

Plant leaf classification system using CBIA and WSN

为了将计算机辅助植物叶片分类算法从理论研究向实际应用推进,利用基于内容的图像分析与无线传感器网络技术实现了移动设备终端的植物叶片分类功能。利用基于Sobel边缘检测子的全自动图像分割方法获取叶片的准确形状,利用基于霍特林变换的方法对叶片进行旋转预处理并提取傅里叶描述子等九种形状特征,然后使用多类支持向量机分类器对叶片进行分类,再进一步使用早期融合的方法对分类结果进行加强,随后利用以上叶片分类方法作为核心技术建立无线传感器网络,最后利用Java与安卓技术实现移动客户端的应用功能。实验结果显示,对于两个叶片数据库,分别达到了80%的分类准确率水平,与国际同类研究水平相当;对于无线传感器网络,移动终端用户可在9 s内从服务器得到叶片分类的反馈结果;移动客户端实现了安卓操作系统上的应用程序。综上所述,研究已经取得了显著的阶段性成果,并将在下一阶段的工作中引入更加新颖高效的方法来进一步提高叶片分类准确率。

In order to use computer aided plant leaf classification algorithms in a practical way, this paper introduced a plant leaf classification system using content-based image analysis and wireless sensor network （WSN） techniques. First, it used a Sobel edge detector based full-automatic image segmentation method to obtain the accurate shapes of leaves. Second, it applied a Hotelling transform based method to rotate the obtained shapes and extracts nine shape features, including fourier descriptor and so on. Thirdly, it indentified different classes of leaves by a multi-class support vector machine classifier and evaluates the classification result by the classification accuracy. Furthermore, it used an early fusion approach to enhance the classification result by combine different features. Fourthly, it used the above classification method as the core technique to establish a WSN. Finally, it applied Java and Android techniques to implement an internet application on the mobile client. In experiments, it obtained good classification accuracies of 80% on two datasets, which were similar to that in other previous researches. Furthermore,it designed a brief WSN framework and was able to finish a data transmission in 9 seconds. Lastly,it used Java technique to implement an application in Android system for image capturing and data transmission. In conclusion, this paper shows a remarkable result in the current phase, and it will be improved by more effective methods in the future work.