【作者】 叶飞强； 蒋朝辉； 周昊； 桂卫华；

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

【机构】 中南大学自动化学院；

【摘要】 烧结矿、球团矿等是高炉炼铁生产过程中的重要原料,其粒径大小将直接影响着高炉内部的煤气流分布,进而影响高炉内部的铁的还原。在实际炼铁生产过程中,高炉炉料矿石是通过离线筛分的方式进行粒径检测,检测方式效率低,精度差。使用机器视觉的方式可以有效提高检测效率,但由于矿石形状不规则,且矿石表面粗糙,在使用传统的图像处理方式会使得图像出现严重的过分割。深度学习方法可以基于深度特征对图像进行按像素的分割,可以有效地对高炉原料矿石图像进行分割。本文提出一种基于多特征融合的Unet矿石图像分割方法。首先对采集的矿石图像进行预处理,进行数据集制作;其次对Unet网络每层的卷积层输出进行密集叠加,进行多特征融合,利用制作的数据集在改进的Unet网络上进行训练;最终使用训练好的模型进行预测,对预测结果进行后处理,得到矿石图像分割结果。实验结果表明,本文方法在IoU和F1Scorce两项指标上均好优于经典Unet网络和传统的图像分割方法,能够有效的实现炉料矿石图像分割。更多还原

【Abstract】 Sinter, pellets, etc. are important raw materials in the process of blast furnace ironmaking, and their particle size will directly affect the gas flow distribution inside the blast furnace, and then affect the reduction of iron inside the blast furnace. In the actual ironmaking production process, the blast furnace charge ore is tested for particle size by offline screening, which has low efficiency and poor accuracy. The use of machine vision can effectively improve the detection efficiency, but due to the irregular shape of the ore and the rough surface of the ore, the use of traditional image processing methods will cause serious over-segmentation of the image. The deep learning method can segment the image by pixel based on the depth feature, and can effectively segment the blast furnace raw material ore image. This paper proposes a Unet ore image segmentation method based on multi-feature fusion. First collect ore images and make data sets; Second, improve the Unet by connecting the output of the convolutional layer of each layer, and use the prepared data set to train the improved Unet network; Finally, use the trained model to make predictions, and post-process the prediction results to obtain the ore image segmentation results. The experimental results show that the method in this paper is better than the classic Unet network and the traditional image segmentation method in IoU and F1 Scorce, and can effectively achieve the image segmentation of the charge ore.更多还原