基于姿态描述的果园靶标害虫自动识别方法  

Orchard Pest Automated Identification Method Based on Posture Description

作　　者：李文勇[1,2] 陈梅香[2] 李明[2] 孙传恒[2] 杜尚丰[1] 

机构地区：[1]中国农业大学信息与电气工程学院,北京100083 [2]国家农业信息化工程技术研究中心,北京100097

出　　处：《农业机械学报》2014年第11期54-59,共6页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金青年基金资助项目(31301238);北京市自然科学基金资助项目(4132027);北京市农林科学院青年科研基金资助项目(QN201102);北京市农林科学院国际合作基金资助项目(GJHZ2013-4)

摘　　要：针对自动诱捕果园靶标害虫的姿态形体存在不确定性,增加果园害虫图像自动识别与计数的难度等问题,提出一种基于姿态描述的算法用于果园靶标害虫姿态表征与识别。首先分析了方法对靶标害虫在8个旋转角度、6种常见姿态形状的描述能力及稳定性,通过计算靶标害虫不同姿态的平均归一化傅里叶描述子和离散度阈值,确定了基准姿态特征向量和相似度差异判据值。对200幅包含3种果园害虫的样本图像进行了测试,当离散度阈值为0.021 26时,靶标害虫桃蛀螟识别的正确率为86.7%,误判率为2.6%。试验结果表明该方法具有稳定的姿态形状描述能力和良好的识别性能。Orchard pests monitoring is very important in precise pest management. The pests trapped by high-vohage current in orchard show different postures, which increase the difficulty of image automatic identification. A posture description-based method has been proposed to automatically represent image boundary and identify certain common pests in orchards. The performance Of proposed method in posture description with rotation, translation and wing scale of image edge was tested. Posture feature vector and similarity difference threshold were determined by the calculation of average normalized Fourier deseriptors（FDs） and discrete degree. Three types of pests with 200 sample images were tested, and it found that correct rate of target pest Dichocrocis punctiferalis （Guenee） was 86.7% and error rate was 2.6% when discrete degree threshold was 0. 021 26. The experiment results indicate that this approach has a stable description ability for posture-shape and good recognition performance.

关 键 词：果园精准管理 害虫识别 图像处理 姿态识别 傅里叶描述子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] S431.9[自动化与计算机技术—计算机科学与技术]