Medical Bottles
收藏资源简介：
针对生产线药液瓶在线自动检测问题，设计了一种植入式视觉检测系统，提出了基于多特征融合的集成学习检测算法。为视觉检测系统设计了隧道结构，可以在不改变原有工艺和设备的情况下实现瓶子检测的自动化。提出了一种高精度的药液瓶视觉检测方法。利用局部背景差分法作为软开关来捕捉瓶子图像。并采用图像灰度均衡预处理技术消除光照影响。设计了三个特征，包括梯度块状直方图、灰度块状直方图和原始像素。在理论上分析了精度提升对集成学习的前提条件，提出了一种基于独立性检验和多特征融合的集成学习算法。一些分析和比较结果证明，与基线方法相比，该方法具有先进性。特别是，当样本标签中存在一些噪声时，我们的方法具有显着的优势。然后，我们在一条实际生产线上进行了72小时连续测试，检测错误率小于1‰。在时间和精度上都优于传统的人工检测。试验结果证明所设计的视觉检测系统和算法具有先进性和实用性。
Aiming at the online automatic detection problem of liquid medicine bottles on production lines, an implantable visual inspection system was designed, and an integrated learning detection algorithm based on multi-feature fusion was proposed. A tunnel structure was designed for the visual inspection system, enabling automated bottle detection without modifying the original production process and equipment. A high-precision visual detection method for liquid medicine bottles was proposed. The local background difference method was used as a soft switch to capture bottle images, and image grayscale equalization preprocessing technology was adopted to eliminate the impact of illumination. Three features were designed, including gradient block histogram, grayscale block histogram, and raw pixel values. The prerequisite for accuracy improvement in integrated learning was theoretically analyzed, and an integrated learning algorithm based on independence test and multi-feature fusion was proposed. Multiple analysis and comparison results demonstrate that the proposed method outperforms baseline methods, especially showing significant advantages when there is noise in sample labels. Subsequently, a 72-hour continuous test was conducted on an actual production line, with the detection error rate lower than 1‰. The proposed method outperforms traditional manual detection in terms of both time efficiency and accuracy. The test results prove that the designed visual inspection system and algorithm have advanced performance and practical applicability.
