| 摘要: |
| 目的 针对目前工业自动化生产中基于人工特征提取的包装缺陷检测方法复杂、专业知识要求高、通用性差、在多目标和复杂背景下难以应用等问题，研究基于深度学习的实时包装缺陷检测方法。方法 在样本数据较少的情况下，提出一种基于深度学习的Inception-V3图像分类算法和YOLO-V3目标检测算法相结合的缺陷检测方法，并设计完整的基于计算机视觉的在线包装缺陷检测系统。结果 实验结果显示，该方法的识别准确率为99.49%，方差为0.000 050 6，只使用Inception-V3算法的准确率为97.70%，方差为0.000 251。结论 相比一般基于人工特征提取的包装缺陷检测方法，避免了复杂的特征提取过程。相比只应用图像分类算法进行包装缺陷检测，该方法在包装缺陷区域占比较小的情况下能较明显地提高包装缺陷检测精度和稳定性，在复杂检测背景和多目标场景中体现优势。该缺陷检测系统和检测方法可以很容易地迁移到其他类似在线检测问题上。 |
| 关键词: 缺陷检测 Inception-v3 YOLO-V3 TensorFlow Serving MQTT 迁移学习 |
| DOI：10.19554/j.cnki.1001-3563.2020.07.025 |
| 分类号:TB487 |
| 基金项目: |
|
| An Industrial Automation Packaging Defect Detection Method Based on Deep Learning |
|
LI Jian-ming1, YANG Ting1, WANG Hui-dong2
|
|
1.Tianjin University, Tianjin 300072, China;2.Beijing University of Technology, Beijing 100124, China
|
| Abstract: |
| The work aims to study a real-time packaging defect detection method based on deep learning, in view of the problems such as complexity, considerable professional knowledge, poor generality, and difficulty in application under multi-objective and complex background of the current packaging defect detection methods based on artificial feature ex-traction in industrial automation production. In the case of small sample set, a defect detection method combining the In-ception-V3 image classification algorithm and YOLO-V3 target detection algorithm based on deep learning was proposed, and a complete online packaging defect detection system based on computer vision was designed. Experimental results showed that the recognition accuracy rate and variance of the proposed method were 99.49% and 0.000 050 6 respectively. The accuracy rate of using only Inception-V3 algorithm was 97.70% and its variance was 0.000 251. Compared with the general packaging defect detection method based on artificial feature extraction, the proposed method avoids the complex feature extraction process. Compared with the packaging defect detection only with image classification algorithm, the proposed method can obviously improve the accuracy and stability of packaging defect detection especially when the defect occupies a relatively small proportion, and performs well in complex detection background and multi-objective situation. At the same time, the defect detection system and detection method designed herein can be easily migrated to other similar online detection problems. |
| Key words: defect detection Inception-v3 YOLO-V3 TensorFlow Serving MQTT transfer learning |
渝公网安备 50010702501716号