竹纤维改性合成混凝材料表面缺陷检测技术优化

Optimization of detection method on surface defects of jute fiber concrete

以黄麻纤维混凝土为研究对象,提出用卷积神经网络识别混凝土材料中的竹纤维,进而提高混凝土材料表面裂缝的准确率。首先,将采集到的黄麻纤维混凝土材料扫描图片进行数据增强,然后搭建DeepLabV3+的竹纤维快速识别网络,并对网络进行训练,最后输入采集的原始图片,对网络模型进行识别测试。结果表明,通过训练后得到的网络对竹纤维混凝土识别的ACC=81.7%,F1=78.3%,证明DeepLabV3+深度学习网络模型对竹纤维混凝土材料具有较高的识别精准度,可用于实际的化工工业和工程缺陷检测中。

Taking jute fiber concrete as the research object, it is proposed to use DeepLabV3+network to identify chopped jute fibers in concrete materials, so as to improve the accuracy of surface cracks in jute concrete materials.First of all, the collected nano CT scanning pictures of jute fiber concrete material are enhanced, and then a rapid identification network of chopped jute fiber of DeepLabV3+is built, and the network is trained.Finally, the collected original pictures are input, and the network model is identified and tested.The results show that ACC=99.3%,F1=80.4%,for the recognition of chopped jute fiber concrete by the network obtained after training, which proves that DeepLabV3+deep learning network model has high recognition accuracy for chopped jute fiber concrete materials, and can be used in the actual chemical industry and engineering defect detection.