为解决变量喷洒对杂草识别速度与正确率的要求，提出了一种基于多光谱图像和SVM 的杂草识别新方法。通过多光谱成像仪获得玉米与杂草图像，采用 IR-R 的多光谱融合并结合 Otsu 分割法完成背景分割；随后对植被图像进行目标分割与形态学处理，提取出所有植被叶片图像，在此基础上提取了叶片11个形状特征参数和纹理特征参数。为提高算法的实时性，对叶片的特征参数进行主成分分析，将前3个主成分作为支持向量机的输入建立模式识别模型。结果表明，降维后对于未知预测样本的识别正确率达到85％，用时0．001415 s。与直接利用支持向量机的90％的识别率和0．105165 s的用时相比，该算法在满足识别率的同时，用时更少，为田间杂草的快速识别提供了一种新方法。%To solve the requirement of variable spray for weed identification speed and accuracy rate , proposed a new method based on multi-spectral images and SVM weed identification .By the corn and weed images of multi-spectral im-age , IR-R multi-spectral fusion and combination of Otsu segmentation method was used to complete the background seg -mentation .Then vegetation image object segmentation and morphological processing was taken before extract all the vege -tation leaf images .Based on this , 11 leaf characteristic parameters of shape and texture was extracted .To improve the real-time , principal component analysis was taken for the characteristic parameters of leaves , and the first-three princi-pal components was taken as input of support vector machines to establish pattern recognition model .The results showed that 85%recognition accuracy for unknown prediction samples after dimensionality reduction , with a time of 0.001 415s, compared with recognition rate of 90%and 0 .105 165 s of the direct use the support vector machine .The algorithms cost a little more time to meet the require recognition accuracy , and provides a new method for the rapid identification of weed .
展开▼
