草原典型牧草的特征提取与识别研究

The Study on Identification and Feature Extraction for the Typical Grassland Pasture

选择四种内蒙古乌兰察布市荒漠化草原的典型牧草,对牧草图像进行颜色、形状特征提取,以实现牧草的分类识别,为提高牧草管理信息化水平,实现多牧草种类分割识别及牧草盖度、频度等检测研究奠定基础。在自然光照情况下应用高清照相机采集尺寸、形状、大小各不相同的草原原始牧草图像,对羊草(Leymus chinensis)、鹅绒委陵菜(Potentilla anserine)、阿尔泰狗娃花(Heteropappus altaicus),冷蒿(Artemisia frigida)四种牧草图像提取9种RGB颜色一、二、三阶矩特征与6种宽长比等形状特征,统计获取相应的数据库,试验表明颜色矩、形状特征参数具有显著的差异。构建一种3层BP(Back propagation,BP)神经网络模型,应用主成分分析法(principal component analysis,PCA)优选特征明显的维数,降低计算成本、提升识别效率,四种牧草平均识别率为82.5%,上述试验证明该方法能够有效地对内蒙古草原典型牧草图像进行分类识别研究。牧草自动识别技术的发展为草业管理信息化提供数据支撑,是实现现代生态草原的重要途径。

Four typical pastures were selected in Inner Mongolia Ulanqab desert steppe to extract the color and shape characteristics and carry out the classification of pasture,which can improve informationalized level of pasture management and build a foundation for detecting forage coverage and frequency.Under natural light conditions,high-definition cameras were used to collect original grassland plant images with different sizes and shapes,in which 9 kinds of RGB color moment features and six kinds of shape features were extracted from the four plants(Leymus chinensis,Potentilla anserine,Heteropappus altaicus and Artemisia frigida)images.The corresponding databases were established by data statistics obtained from above four plants.The results showed that color moments and shape features had significant differences among four plants.The overall recognition rate of the four plants can reach to 82.5%.To reduce the computational cost and improve the recognition efficiency,a three-layer back propagation(BP)neural network model integrated with a method of principal component analysis was used to optimize parameters.In conclusion,the present experiment proved that above methods can efficiently identify typical plant images in Inner Mongolia grassland.The development of pasture automatic identification technology provided sufficient data for informationalized grassland management,which is an important approach to achieve a goal of digital grassland ecosystem.