Abstract: In the textile manufacturing industry, the process of yarn dropping in spinning machine is critical as it directly impacts yarn quality, production output, and factory operations. Traditional automatic fine yarn dropping machines, while capable of performing operations automatically, often suffer from imprecise of operations, resulting in missed of empty tube insertion or tube yarn extraction during the yarn dropping process. The current coping approach is to assign dedicated personnel to track, inspect, and address the issue, which increases labor costs and restricts the level of automation and intelligence in the workshop.
This paper aims to address the issues of yarn tube extraction missing and empty tube insertion missing in traditional automatic fine yarn dropping machines by introducing machine vision technology to accurately identify tube yarns and yarn spindles, thereby reducing the probability of missed extractions and insertions in subsequent operations. However, traditional image processing techniques face challenges in complex fine yarn workshop environments, such as complex target backgrounds, lighting variations, obstructions, and shooting angles. In addition, it is necessary to set parameters and thresholds based on experience during the identification process of yarn spindles and tube yarns, which increases the difficulty of recognition. Therefore, accurate identification of tube yarns and yarn spindles is crucial for automatic yarn dropping machines based on machine vision technology. To solve the interference problem of images gathered by visual sensors in workshop environments, this paper proposes a basic image processing method. By adjusting the contrast between the objects and the background, the characteristics of yarn spindles and tube yarns can be highlighted. This method also performs initial and secondary identification in the objects identification process to  obtain the object areas more accurately. In order to further improve the identification accuracy of yarn spindles and tube yarns, the basic image processing method is improved in image preprocessing phase and object identification phase respectively. In the preprocessing phase, image multiplication fusion is used to obtain high-contrast images to reduce the difficulty of subsequent image segmentation. For the interference introduced by the support rod of the spinning machine in the original image, find the nearest two horizontal contour bodies in the object region, and eliminate the smaller contour body in the object selection decision, until the remaining identification quantity is consistent with the set number.
The experimental results demonstrate that the identification accuracy of yarn spindles and tube yarns in different scenarios exceeded 98.40%, realizing the accurate identification of spindles and tube yarn in complex background. In future work, machine learning methods can be introduced to solve the problem that individual parameter setting cannot adapt to special scenes such as strong light and reflection, and further improve the recognition accuracy of spindles and pipe yarn.

Key words: machine vision, doffer, spindle identification, yarn identification

摘要： 为了解决传统自动细纱落纱机在拔管和插管操作中存在的漏拔、漏插问题，将机器视觉技术引入自动细纱落纱机中，提出了一种两次识别分割的方法，以准确识别管纱和纱锭，降低漏拔和漏插发生的概率。针对复杂环境对原始图像造成的严重干扰，首先在图像预处理时调整目标与背景之间的对比度，突出纱锭和管纱的图像特征；然后在目标识别过程中进行初始识别和二次识别。为进一步提高识别的准确率，分别在图像预处理和目标识别阶段进行改进，在预处理中通过图像乘法融合的方式进一步增强对比度，在目标筛选决策中引入目标数量特征以提高识别准确率。结果表明，该方法对不同场景下的纱锭和管纱的识别准确率超过98.40%，实现了在复杂背景下纱锭和管纱的精确识别。

关键词: 机器视觉, 落纱机, 纱锭识别, 管纱识别