The growing demand for large-scale image retrieval calls for methods that balance efficiency and accuracy. Deep hashing accelerates image retrieval using compact binary codes, but most existing approaches rely on a single type of backbone, such as Convolutional Neural Networks or Vision Transformers. Consequently, they face difficulties in capturing both global context and detailed local features. To tackle this, we present FusedHash, a novel model that combines the strengths of multiple feature extractors in a unified framework. Specifically, FusedHash adopts a dual-branch architecture with heterogeneous backbones to learn more discriminative features. We further introduce an efficient fusion strategy that combines global semantics with local details, generating multi-scale representations. In addition, we optimize hash target construction to better distinguish visually similar categories. On three standard benchmarks, FusedHash delivers improved performance compared to most existing methods. In low-bit scenarios, it achieves up to a 6.28% improvement in retrieval performance, validating the effectiveness and potential of FusedHash in image retrieval applications.