<p>Accurate identification of Mesoscale convective systems (MCSs) is important for disaster prevention and management. Traditional MCSs identification methods are based on temperature and area thresholds. These methods are highly accurate for MCSs identification, but consume large computational resources. Deep learning models have made significant progress in the field of object identification. However, existing deep learning models do not take into account the characteristics of excessive differences in MCSs area coverage and uneven spatial distribution. Direct application of these models to this task may lead to the problems of missed identification of small MCSs and ambiguous identification of MCSS boundaries. In this paper, we construct a MCSs recognition dataset and propose a Multiscale Feature Enhancement Network (MFE-Net) for identifying MCSs. This model consists of two main modules: Feature Alignment Distribution Module (FADM) and Asymmetric Feature Recovery Module (AFRM). The FADM, which is used to aggregate multiscale information during the downsampling phase, aims to improve the perception of small-scale MCSs and prevent them from being lost during the downsampling process of the model. AFRM is designed to extract and preserve spatial edge features of MCSs from different spatial dimensions and accurately identify the edges of MCSs. The experimental results show that our MFE-Net achieves excellent performance both quantitatively and qualitatively.</p>