【摘要】 为准确提取烟株茎秆特征,基于计算机视觉技术提出了一种动态识别烟株茎秆的方法。在图像预处理阶段,构造图像清晰度评价函数,采用迭代法提取清晰图像,利用加权平均法对其进行灰度化预处理;为消减光源强度、重影、地形等因素的影响,对灰度化后的图像进行降噪处理;为突出烟株茎秆特征,对图像进行灰度增强;运用自适应算法对图像进行阈值智能分割,最终提取出烟株茎秆特征。在光源充足、阴影、车辆颠簸3种条件下,分别对100株烟株进行茎秆识别模拟试验。结果表明,光源充足条件下识别率最高,达76%;颠簸状态下识别率最低,为69%;3种试验条件下平均识别率为72.67%。试验验证了在动态图像中获取烟株茎秆特征的可行性,配合仿生采摘执行机构可以完成烟叶的自动化采摘,并可有效控制烟叶的破损率,提高烟叶采摘效率。更多还原

【Abstract】 For accurate extraction of tobacco stalk features, a method for dynamic recognition of tobacco stalks was proposed based on computer vision technology. A function of image clarity evaluation was constructed at image preprocessing stage, an iterative method was used to extract distinct images and the images were grayed by weighted average method. For diminishing the influences from light source intensity, ghost, terrain and other factors, the grayed images were denoised, and the gray degree of images was intensified to make the features of tobacco stalks more prominent. Finally, self-adaptive algorithm was used to segment image threshold intelligently, the features of tobacco stalks were extracted. The simulation tests of tobacco stalk recognition were conducted with 100 tobacco plants separately under adequate light irradiation, shade, and bumping vehicle conditions. The results showed that: with average recognition rate of 72.67% under the three conditions, the recognition rate under adequate light irradiation was the highest(76%), while that in a bumping vehicle was the lowest(69%). The experiment verified the feasibility of obtaining the features of tobacco stalks in dynamic images, automatic tobacco leaf harvesting could be implemented by combining with a bionic actuator, it is helpful to lowering leaf damage and improving the efficiency of leaf harvesting.更多还原