To address the current challenge in accurately estimating the state of health (SOH) of lithium-ion battery packs, a high-precision SOH estimation method integrating multi-scale features of the overall degradation and individual cell inconsistencies of the battery pack is designed. In this method, a deep learning model convolutional neural network Kolmogorov-Arnold network-Bahdanau attention (CNN-KAN-BA) combining a convolutional neural network (CNN), a Kolmogorov-Arnold network (KAN), and a Bahdanau attention (BA) mechanism is proposed. In the proposed SOH estimation process, systematic aging experiments are first conducted on a six-cell series-connected 18650 battery pack to obtain full life-cycle data. Then, the incremental energy analysis (IEA) method is adopted to extract the incremental energy curve length feature that characterizes the overall degradation of the battery pack. Simultaneously, the median absolute deviation of individual cell voltages within the pack and the temperature kurtosis are calculated as key individual features reflecting the evolution of inconsistency. Thereby, a multi-scale feature set that comprehensively describes the coordinated “overall-individual” degradation of the battery pack is constructed. The CNN-KAN-BA estimation model is trained using the features from the training data and is validated with the test data. The results show that this method can achieve high-precision SOH estimation, with a mean absolute error of 0.587 4%, a root mean square error of 0.699 0%, and an average coefficient of determination higher than 98%, all of which are superior to other common SOH estimation methods. The proposed method can effectively solve the problem of precise SOH estimation for lithium-ion battery packs.