摘要
合理的烧结混合料粒度分布是影响烧结矿产量和质量的重要因素。对具有代表性的多批次烧结混合料颗粒图像,采用加权平均灰度处理、滤波降噪和直方图均衡化增强等技术进行了图像预处理,获得了构建混合料粒度识别模型所需的基础数据集。选用5种评价指标,对比分析了Pspnet、Mask-Rcnn、Deeplabv3+、Unet等语义分割网络模型的准确性,确定使用Re-Unet网络构建了烧结混合料粒度识别模型。通过跳连接方式优化网络结构降低特征损失,采用复合损失函数与自定义指标验证了模型的性能,应用可视化技术实现了烧结混合料粒度分布计算。测试结果表明,模型的输出结果与实际值误差小于2%,可用于对烧结混合料粒度的精准识别。
Reasonable particle size distribution of sintered mixture is an important factor affecting the yield and quality of sinter.Representative particle images of multiple batches of sintered mixture were preprocessed using techniques such as weighted average grayscale processing,filtering noise reduction,and histogram equalization enhancement to obtain the basic dataset required to construct the particle size recognition model.Five evaluation indicators were used to compare and analyze the accuracy of semantic segmentation network models such as Pspnet,Mask-Rcnn,Deeplabv3+,and Unet,and the Re-Unet network was selected to construct the sintered mixture particle size recognition model.The network structure was optimized using skip connections to reduce feature loss,and the performance of the model was verified using composite loss functions and custom metrics.Visual techniques were applied to calculate the particle size distribution of sintered mixture.The test results show that the error between the output of the model and the actual value is less than 2%,indicating that the model can be used for accurate identification of sintered mixture particle size.
作者
李福民
侯炬才
刘颂
刘然
刘连继
吕庆
LI Fumin;HOU Jucai;LIU Song;LIU Ran;LIU Lianji;LüQing(College of Metallurgy and Energy,North China University of Science and Technology,Tangshan 063210,Hebei,China;College of Artificial Intelligence,Tangshan College,Tangshan 063000,Hebei,China;Chief Engineering Office,HBIS Group Tangsteel Company,Tangshan 063000,Hebei,China)
出处
《钢铁研究学报》
CAS
CSCD
北大核心
2023年第11期1347-1357,共11页
Journal of Iron and Steel Research
基金
河北自然科学基金资助项目(E2020209208)
唐山市应用基础研究科技计划资助项目(21130233C)。
