Cover Source Mismatch (CSM) poses a significant challenge to image steganalysis. The varying distributions of images across different stego carriers reduce the effectiveness of steganalysis networks trained in the source domain when detecting images in the target domain. Existing methods lack consideration for multi-source scenarios and often fail to fully exploit the prior knowledge from the source domain as well as the distribution information from the target domain, resulting in limited detection accuracy. To address these challenges, this paper, for the first time, proposes a novel multi-source domain adaptation method for image steganalysis. First, our method design a dual-stage alignment strategy for image steganalysis. The first stage aligns the distribution differences among multiple source domains and the target domain. The second stage aligns the prediction outputs of different classifiers to ensure consistency in the predictions of the subnetworks. This strategy enables the network to learn more effective domain-invariant features. Second, combined with the subsequent embedding strategy, we designed a progressive feature fusion module. Through this module, the network effectively fuses steganographic noise information from both the source and target domains. This enables the smooth transfer of the knowledge obtained through pre-training in the source domain to the target domain. Extensive experiments demonstrate that the proposed network achieves significant improvements in detection accuracy across various CSM scenarios.