基于BP神经网络的典型牧草特征提取与识别研究

Study on Identification and Feature Extraction for the Typical Pasture Based on BP Neural Network

针对内蒙古荒漠化草原牧草监测与数字化程度较低的问题,实现了三种典型牧草的特征提取与图像识别,为多牧草种类识别与草业管理提供依据.运用MATLAB图像处理技术,对糙苏、羊草及鹅绒萎菱菜进行图像预处理,提取了12种颜色矩特征和4种形状特征,利用BP神经网络法实现了三种牧草的图像识别,识别正确率为86.9%.试验结果表明,基于颜色矩特征和形状特征的BP(back propagation,BP)神经网络图像识别方法能够有效地实现典型牧草的图像分类研究.自动识别牧草是草业数字化的重要组成部分,可为监测植被物种多样性、草种退化及病虫草害的控制提供科学依据.

Aimed at the problems of pasture monitoring and low-level digitization in Inner Mongolia desert steppe, the feature extraction and image recognition for three typical pastures are conducted so as to provide a basis for grass species identification and grassland management. Images of three typical pasture＇s leaves （Phlomis umbrosa ,Leyrnus chinensis and Potentilla anserine ） were preprocessed by using the Matlab image processing technology, 12 kinds color moment features and 4 kinds shape features were extracted,and then a BP neural network was built for realizing the image recognition of these three pastures. An 86.9 % of final overall recognition rate and an effective classification result of three pasture images were received. The experimental results indicate that the image classification were perfectly realized by using the BP neural network model based on color moment features and shape features. Automatic identification of grass plays an important role in the grass digitizationlthe recognition for forage image provides a scientific basis for monitoring vegetation species diversity, grass degradation and pest control.