摘要
为了实现橘小实蝇虫口密度的精准监测,该文将机器视觉技术作为田间橘小实蝇成虫入侵自动化监测的感知方法。通过对监测区域内运动目标和背景的颜色分析,提出了基于卡尔曼(Kalman)滤波的运动目标颜色均值漂移跟踪算法,优化了多目标运动轨迹跟踪效果。该算法通过图像处理和匹配技术提取了橘小实蝇成虫在虫口监测区域二维平面X轴和Y轴方向上的位置坐标和速度分量,推算了橘小实蝇成虫运动轨迹递推关系。基于动态系统的状态序列线性最小方差估计理论和成虫运动轨迹关系分析,构建了卡尔曼滤波状态估计模型,并建立其预测和修正方程实现了橘小实蝇成虫运动目标位置估计。通过在虫口监测区域开展单目标和多目标分散及粘连条件下的成虫跟踪试验,试验结果表明颜色均值漂移跟踪算法对橘小实蝇成虫单目标跟踪具有较好的适应性,成虫监测计数准确率达100%,对于多目标分散和粘连情况跟踪处理效果较差,计数准确率分别下降至86%和76%;通过在颜色空间均值漂移跟踪算法的基础上引入Kalman滤波器估算目标运动的近似位置,实现了对橘小实蝇成虫分散和粘连多目标运动的持续跟踪优化,监测计数准确率分别提升至96%和93%。机器视觉技术实时跟踪橘小实蝇成虫在虫口监测区域运动轨迹试验进一步验证了橘小实蝇成虫虫口密度监测的可行性,为田间橘小实蝇成虫发生自动化监测技术研究提供了参考。
Bactrocera Dorsalis (Hendel) are invasive pests which occur frequently and are seriously harmful to the growth of fruit trees, and they have been ranked an important quarantine object in many countries and regions. The regular manual survey used as the routine predicting method for Bactrocera Dorsalis (Hendel) has not accomplished the requirement of real-time and precise monitoring and warning by means of the adult trapping and monitoring device deployed in orchards. With the development of science and technologies, the method of the automatic machine monitoring for pests has been studied including detection of sound characteristics, radar monitoring and spectral monitoring. Considering the characteristic with randomness, migratory and hiding for Bactrocera Dorsalis (Hendel), there were some problems such as timing, processing and costs in monitoring pests with the aid of combining the above monitoring and the traditional method. In order to accomplish precise monitoring for Bactrocera Dorsalis (Hendel), machine vision technologies were used as an in-field automatic detecting method for the Hendel adults in this paper. Considering the problem with tracking Bactrocera Dorsalis (Hendel) object disappearance in multi-objects with more closer condition by means of the mean shift algorithm in color space according to previous machine vision technology research results, the fusion algorithm based on mean shift and Kalman filter theories for moving objects was proposed for optimizing multi-objects moving trace tracking by means of colorful analysis for moving objects and background in monitoring zones. The recurrence relation of the adults moving trace was obtained, and position coordinate, X-component and Y-component of speed in the 2D plane were extracted by image processing and matching technologies in this algorithm. By analyzing the state sequence linear minimum variance estimate theory of dynamic system and recurrence relation of the adults moving trace, the model of state estimate based on a Kalman filter was built to achieve the position estimation of the adults using the prediction and modified equation of the model. The experiment of the adults tracking under the condition of single object and the condition of multi-objects with scatter and gathering indicated that the mean shift algorithm was adaptive to track the adults in the condition of single object with monitoring precision of 100%, was not adaptive to the condition of multi-objects with scatter and gathering since corresponding monitoring precisions were 86%and 76%respectively. The cooperation of mean shift and Kalman filter algorithm estimating of moving objects’ approximate location could achieve the stable and continuous tracking in the condition of multi-objects with scatter and gathering with corresponding monitoring precision of 96%and 93%. The real-time tracking experiment of the adults moving trace in pest monitoring zones by the aid of machine vision further validated the practicability of the Hendel adults population density monitoring for providing a theoretical and practical basis for in-field Hendel adults automatic monitoring technology studies.
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2014年第15期197-205,共9页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家现代农业产业技术体系建设专项资金(CARS-27)
公益性行业(农业)科研专项经费项目(200903023,201203016)
国家自然科学基金(31101077)
湖南省自然科学基金(14JJ3115)
湖南省农业机械化管理局科技计划项目(KJ2013-18)
中南林业科技大学引进高层次人才科研启动基金(104-0259)
关键词
监测
害虫防治
机器视觉
橘小实蝇
团块跟踪
均值漂移
卡尔曼滤波
monitoring
pest control
computer vision
Bactrocera Dorsalis (Hendel)
blob tracking
mean shift
Kalman filter