As an all-weather ocean monitoring technology, Synthetic Aperture Radar (SAR) plays a crucial strategic role in ship classification. However, traditional classification methods still encounter significant challenges in feature representation and model generalization performance due to the widespread coherent speckle noise in SAR images, the multi-scale nature of ship targets, and the interference from complex sea conditions. To address these issues, this study proposes an improved MobileNetV2 approach incorporating an effective channel attention mechanism. The channel attention mechanism is introduced into the bottleneck residual module of MobileNetV2, enabling the model to adaptively adjust the weights of feature channels, dynamically emphasizing those that contain essential information, and improving the model’s feature perception capability. Furthermore, to tackle the problem of insufficient adaptability to multi-scale features, a receptive field module based on multi-scale feature fusion is designed. This module employs parallel convolution operations with different receptive fields to perceive and integrate multi-scale features, which in turn strengthens the model’s performance in feature extraction and representation for maritime targets. The proposed method achieves higher classification accuracy and demonstrates superior robustness and adaptability compared to mainstream models, as evidenced by the experimental results.