Abstract: Dredging project plays a critical role in the constructions of port channels and marine exploitation. As a main working equipment, the cutter suction dredgers (CSDs) are commonly used for conveying the excavated sediments to the designated discharge sites by pipelines. Therefore, to optimize the sediment-conveying efficiency of the CSDs, it is critical to monitor key factors such as flow velocity in the pipeline in real time. However, the challenging working environment imposes great constraints to the wild applications of the traditional physical sensors due to their high cost and complex maintenance requirements. For measuring the flow velocity in the pipeline, therefor, a soft sensoring method ICBFormer is proposed. This ICBFormer integrates the Interactive Convolutional Block (ICB) and the Transformer Model in order to replace the traditional flowmeters, in which the ICB is used to capture the complex inter-variable relationships and extract multiscale temporal features. Subsequently, the dynamic relationships among variable data sequences are high-effectively handled with the help of the advantages of the Transformer Model in long-sequence feature extraction, thus achieving the accurate prediction of flow velocity in the pipeline. This soft sensoring method has been evaludated through data collecting by building a simulation experiment platform for conveying by dredging mud pump. The results show that the ICBFormer proposed in the study has a significant advantage in the flow velocity prediction and provides a new solution for reducing the sensor cost and maintenance expense of the dredgers.