Accurately estimating the state-of-health of batteries is critical for effective battery monitoring and management. However, the estimation remains challenging due to dynamic operation environments and complex battery degradation patterns. In this study, a feature fusion-based convolutional neural network is proposed for battery state-of-health estimation based on voltage measurements obtained during a partial cycle. Instead of directly feeding the voltage data as input, three feature sequences are first extracted, including the capacity versus voltage curve and its differentiation with respect to voltage and life cycle. The objective is to exploit more effective information from intra-cycle and inter-cycle perspectives. Then, an element-wise addition is embedded as a feature fusion operation in the proposed convolutional neural network to generate more efficient feature maps when dealing with multiple model inputs. To validate the performance of the proposed methodology, eighteen batteries from three battery datasets are utilized for comparative studies. Experimental results demonstrate that the data preprocessing from both intra-cycle and inter-cycle perspectives, along with the adoption of the feature fusion operation, significantly improve the accuracy of battery state-of-health estimation, with an average mean absolute error and mean absolute percentage error being no more than 0.0028 and 0.32 %, respectively. • A feature fusion-based convolutional neural network is proposed. • Three feature sequences are extracted from partial voltage curves. • Feature information is considered from both intra-cycle and inter-cycle perspectives. • The method is verified using batteries with different materials and test conditions. • Estimation on different model architectures, feature inputs, and voltage ranges is discussed. [ABSTRACT FROM AUTHOR]