Abstract: To address the shortcomings of existing estimation methods in feature extraction and time series modeling, a lithium battery state-of-health estimation method that integrates multi-scale convolutional neural networks with gated dual attention units is proposed. Firstly, health indicators reflecting the aging process are extracted from the battery charging curve. Subsequently, local fluctuations and long-term trend features in the battery degradation sequence are extracted in parallel using convolutional kernels of different sizes, and the gated dual attention unit is utilized to enhance the model's ability to model long-term dependencies and improve estimation accuracy. Finally, verification and analysis are conducted on the aging dataset. The results demonstrate that the proposed method significantly outperforms traditional models such as SVR, BiLSTM, and GRU.