| 姜明.基于图像处理的轮胎X光图像缺陷检测[J].轮胎工业,2024,44(4):0250-0255 |
二维码（扫一下试试看！） |
| 基于图像处理的轮胎X光图像缺陷检测 |
| tire X-ray image；image processing；deep learning；defect detection |
| 投稿时间：2022-11-16 修订日期：2022-11-16 |
| DOI： |
| 中文关键词: 轮胎X光图像 图像处理 深度学习 缺陷检测 |
| 英文关键词: Deep Learning Defect Detection Tire X-ray Images Faster R-CNN Image Cutting |
| 基金项目:泰山学者基金(tshw201502042); 山东省重点研发计划(2017CXGC0607和2017GGX30145); 中国国家自然科学基金(61702295) |
| 中图分类号: |
|
| 摘要点击次数: 130 |
| 全文下载次数: 121 |
| 中文摘要: |
| 在图像处理的基础上使用深度学习方法对轮胎X光图像缺陷进行自动检测。轮胎X光图像具有高分辨率、形状狭长、缺陷目标较小的特点，通过将每张X光图像按照640×640像素进行切割，对切割出的每个区域进行标注，把存在缺陷的区域划分到训练集，对训练集进行直方图均衡化以增强图像前景与背景的对比度，进而继续对训练集进行数据增强以提高模型的泛化能力，最后在Faster R-CNN深度学习缺陷检测模型上训练出最优权重。在模型推理阶段，完整的X光图像会被送入模型，缺陷范围被框出，重组为原始X光图像；若某个缺陷具有多个选框，则将所有邻近的选框合成为一个选框。该方法能够有效降低小缺陷目标的漏检率，提高检测的准确率，间接解决了原始X光图像特征丢失的问题。 |
| 英文摘要: |
| The defects in tire X-ray images were detected automaticly by using deep learning methods based on image processing.Tire X-ray images had the characteristics of high resolution，narrow shape and small defect targets.By dividing each X-ray image into 640×640 pixels cutted，annotating each cutted region，the defective regions were divided into a training set，and the training set was histogram balanced to enhance the contrast between the foreground and background of the image.Further data augmentation was performed on the training set to improve the model’s generalization ability.Finally，the optimal weights were trained on the Faster R-CNN deep learning defect detection model.In the model inference stage，the complete X-ray image would be fed into the model，the defect range would be framed，and reassembled into the original X-ray image，and if a defect had multiple boxes，all adjacent boxes were combined into one box.The method could effectively reduce the missed detection rate of small defect targets，improve the accuracy of detection，and indirectly solve the problem of feature loss in the original X-ray image. |
|
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
