A Novel Moisture Measurement Method for the Sintering Mixture Based on Multivariate Feature Fusion

Real-time and accurate measurement of the sintering mixture moisture is crucial for improving product quality and the efficiency of the sintering process. However, this task remains challenging due to the difficulty in quantifying the expertise in moisture observation and insufficient use of on-site multisource information. Therefore, a novel method for moisture measurement based on multivariate feature fusion is proposed in this article. First, an on-site moisture measurement system is designed to collect mixture images. Then, shallow visual features, reflecting observational expertise in moisture estimation, are designed and extracted to quantify expert knowledge. Deep abstract features of mixture images are obtained using deep convolutional neural networks. In addition, sintering process variables that exhibit a high correlation with moisture are further extracted to fully leverage the sintering process information. To effectively integrate these multivariate features, a feature fusion model based on channel interaction and attention mechanism is proposed, which explicitly separates and recombines shared and private representations. Finally, a measurement model for the mixture moisture is developed using multimodal data as input. Industrial experiments showed that the proposed method could achieve accurate moisture predictions, providing references for practical moisture measurement tasks in the sintering process.