【作者】 周昊； 蒋朝辉； 叶飞强； 桂卫华；

【Author】 Hao Zhou;Zhaohui Jiang;Feiqiang Ye;Weihua Gui;School of Automation,Central South University;

【机构】 中南大学自动化学院；

【摘要】 高炉上料过程物料料种是入炉原料重要的质量指标。针对现有高炉上料物料料种识别存在依赖人工经验、缺少实时检测手段以及现有矿石识别方法对象单一等问题,提出了一种基于机器视觉针对高炉上料过程堆叠物料的实时识别分类方法。以高炉上料过程传送带物料,包括焦炭、块矿、球团为研究对象,首先利用高速相机采集传送带各类物料图像并进行ROI区域提取与图像滤波,然后基于物料图像分析物料特性信息提取物料浅层特征,根据造料工艺物料破碎度的料间差异性,通过迁移学习方法,基于改进深度学习语义分割网络,利用多类物料图像数据对网络进行微调获得物料分割图像并提取物料分割特征,最后基于模糊支持向量机FSVM学习算法,融合物料图像浅层特征与深度学习分割特征完成完成高炉上料过程物料料种分类识别。实验结果表明,该方法能够做到有效降低料种识别错分率,提高上料物料识别的准确率,为高炉上料过程炉料料种检测提供新的方法。更多还原

【Abstract】 The material type in the feeding process of the blast furnace is an important quality indicator of the raw material entering the furnace. Aiming at the shortcomings of the existing blast furnace feeding materials and the identification of reliance on human experience, the lack of real-time detection methods, and the single object of existing ore identification methods, a real-time identification and classification method of stacked materials based on machine vision for the blast furnace feeding process is proposed. Taking the conveyor belt materials in the blast furnace feeding process, including coke, lump ore, and pellets as the research object, the prime minister used high-speed cameras to collect images of various materials on the conveyor belt and performed ROI area extraction and image filtering. Then analyze the material characteristics information based on the material image to extract the shallow features of the material, and according to the material difference of the material fragmentation in the manufacturing process, through the migration learning method based on the improved deep learning semantic segmentation network, the network is fine-tuned using the multi-type material image data.The material segmentation image is obtained and the material segmentation features are extracted. Finally, based on the fuzzy support vector machine FSVM learning algorithm, the shallow features of the material image and the deep learning segmentation features are combined to complete the classification and identification of the materials in the blast furnace feeding process. The experimental results show that this method can effectively reduce the misclassification rate of the material type identification, improve the accuracy of the charging material identification, and provide a new method for the detection of the charge type in the blast furnace charging process.更多还原