基于改进可变形部件模型与判别模型的葡萄叶片检测与跟踪

发布时间：2018-11-21 16:24

【摘要】：为解决酿酒葡萄生长状态自动监测问题,该文提出基于机器视觉和视频处理技术的自动监测系统,开发了多角度可变形部件模型的葡萄叶片检测算法和基于颜色特征的判别模型跟踪算法。在叶片检测方面,该算法对颜色特征图像采用可变形部件模型训练出多角度叶片检测器,通过多模型匹配后产生叶片检测候选集合,选择集合中得分最高的检测结果作为最后的定位信息;在跟踪方面,结合图像中目标的颜色直方图,建立具有区分背景和目标的组合判别模型,并将位置函数的最大值作为相邻帧的目标位置,从而实现对叶片目标的实时跟踪。试验结果表明,该文检测算法对自然条件下的葡萄叶片平均检测率为88.31%,误检率为8.73%;叶片跟踪的准确性也相对较高,其重叠率高达0.83,平均中心误差为17.33像素,其证明了该算法的有效性,研究结果为葡萄生长状态的自动分析提供参考。
[Abstract]:In order to solve the problem of automatic monitoring of grape growth status, this paper presents an automatic monitoring system based on machine vision and video processing technology. The detection algorithm of grape leaves based on multi-angle deformable component model and the tracking algorithm of discriminant model based on color feature are developed. In the aspect of blade detection, the algorithm uses deformable component model to train multi-angle blade detector for color feature image, and generates candidate set of blade detection after multi-model matching. Select the highest score in the set of detection results as the final location information; In the aspect of tracking, combining the color histogram of the target in the image, the combined discriminant model with distinguishing background and target is established, and the maximum value of the position function is taken as the target position of the adjacent frame to realize the real-time tracking of the leaf target. The results showed that the average detection rate of grape leaves under natural conditions was 88.31, and the false detection rate was 8.73. The accuracy of leaf tracking is also relatively high, with an overlap rate of 0.83 and an average center error of 17.33 pixels, which proves the validity of the algorithm and provides a reference for the automatic analysis of grape growth status.
【作者单位】：甘肃农业大学工学院;西北民族大学电气工程学院;
【基金】：国家自然基金(61461005) 甘肃省科技重大专项计划(1502NKDF023)
【分类号】：S663.1;TP391.41

【相似文献】