基于计算机视觉的鞘翅目储粮害虫检测与分类识别研究

项目名称： 基于计算机视觉的鞘翅目储粮害虫检测与分类识别研究

项目编号： No.30871449

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 毛罕平

作者单位： 江苏大学

项目金额： 31万元

中文摘要： 以米象、谷蠹等 15 类小麦粮仓中危害严重的鞘翅目粮虫的成虫为研究对象，研究了死活虫体的代谢差别对近红外光谱反射特征的影响，建立了粮虫生命体征与其近红外反射光谱之间的相关关系，确定了能有效区分活虫和死虫的最优光谱波长及最短可区分时间，探讨了基于近红外计算机视觉的粮虫生命体征检测机理。提出基于可见光-近红外双目计算机视觉的粮虫活虫检测方法，运用近红外相机和可见光相机的融合信息精确定位活虫和死虫，活虫的判别准确率达到100%，解决了计算机视觉检测系统无法判别活虫的难题。针对细小形态差异的粮虫分类，提出基于兴趣区间对偶点分析的粮虫局部特征的提取方法，提取了尾部弧度等7个局部形态学特征，构建了优化的特征空间，研制了可见光-近红外双目计算机视觉粮虫自动检测系统，实现了小麦粮仓中比较常见的15类粮虫的有效检测，活虫的分类正确率达到94.8%，解决了基于计算机视觉的活虫准确识别及识别种类、识别率增加的难题。

中文关键词： 储粮害虫；计算机视觉；高光谱成像；特征提取；识别

英文摘要： The fifteen species of the adult stored-grain insects of Coleoptera are used as the research object. They are seriously rampant in wheat grain-depot, such as Sitophilus oryzae(L.)and Rhyzopertha dominica(F.). The impact of the metabolic difference between the live insects and the death insects to the characteristics of near-infrared spectral reflectance was studied. The correlation between the insect vital signs and the near-infrared reflectance spectroscopy was set up. The optimal wavelength and the differentiable shortest time between the live and the dead were determined. The detection mechanism of the insect vital signs was discussed based on the NIR computer vision. The novel detection method of the live stored-grain insects was proposed based on the visible and near-infrared binocular computer vision. The live and the dead were located accurately by the information fusion of the near-infrared camera and the visible camera. And the identification accuracy of the live insects was 100%. The problem that the computer vision detection system could not identify the live insects was solved. For the classification of the insects being of small morphological differences, an effective feature extraction method based on the dual points in the interest region was proposed to extract the local morphological features of the insects. The seven local morphological features like the tail radian were extracted. The feature space of the stored-grain insects was constructed and optimized. The detection system for the live insects was developed based on the visible and near-infrared binocular computer vision. The fifteen species of the common stored-grain insects in wheat granary could be detected effectively. And the recognition accuracy of the live insects was 94.8%. The problems involving the accurate identification of the live insects based on computer vision, and the increasement of the species and the accuracy of recognition were solved.

英文关键词： stored-grain insects;computer vision;hyperspectral imaging;feature extraction;recognition