基于烧结机尾红外热图像特征的烧结完成度智能识别

Intelligent recognition of sintering completeness based on infrared thermal image features of sintering machine tail

烧结完成度是烧结热状态的直观体现,其精准识别对提升烧结过程稳定性和生产效率、降低烧结生产能耗具有重要意义。由于烧结现场生产环境恶劣,燃烧反应复杂,烟气和物料对流换热频繁,烧结完成度实时精准识别困难,目前仍依赖于“人工看火”。针对存在的问题,提出一种基于烧结机尾红外热图像特征的烧结完成度智能识别方法。该方法首先对红外热图像进行预处理,提出一种双阶段关键帧提取方法,获取受扬尘干扰小,红层完整清晰的关键帧图像;然后基于热图像RGB属性中R通道值变化趋势精准提取烧结红层ROI区域;并基于图像结构纹理分解挖掘隐含的料层信息,获取分区料层高度;设计并提取燃烧层高度占比分布等特征表征烧结完成度;最后构建DS-CEFNet智能识别模型,融合图像深层特征和手工设计的浅层特征,实现烧结完成度精准识别。结果表明,所提模型对烧结完成度的识别准确率达96%,具有良好的应用价值。

The sintering completeness is the intuitive embodiment of the sintering thermal state,and its accurate identification is of great significance to improve the stability and production efficiency of the sintering process and reduce the energy consumption of sintering production.Due to the harsh production environment at the sintering site,the complex combustion reaction,and the frequent convection&thermal exchange between flue gas and materials,it is difficult to accurately identify the sintering completeness in real time.At present,it still relies on“artificial fire watching”.In order to solve the existing problems,an intelligent recognition method of sintering completeness based on infrared thermal image features of sintering machine tail is proposed.In this method,firstly,the infrared thermal image is preprocessed,and a two-stage keyframe extraction method is proposed to obtain the keyframe image with little dust interference and complete and clear red layer;and then the ROI area of the sintered red layer is accurately extracted based on the change trend of the R channel value in the RGB attribute of the thermal image,and the hidden layer information is mined based on the image structure texture decomposition to obtain the height of the partitioned layer,and the characteristics such as the proportion distribution of the burning layer height are designed and extracted to characterize the sintering completeness;finally,the DS-CEFNet intelligent recognition model is constructed,which integrates the deep features of the image and the hand-engineered shallow features to achieve accurate recognition of sintering completeness.The results show that the model proposed has an accuracy of 96%for the recognition of sintering completeness,which has good application value.