Current change detection methods in remote sensing mainly rely on CNNs and Transformers. CNNs often lack global context, while Transformers are computationally expensive due to their quadratic self-attention. The recent Mamba model offers linear complexity with global modeling but struggles with texture and color inconsistencies in remote sensing images. To overcome these limitations, we propose Omnidirectional Divide-and-Conquer Convolutional Mamba Unet (ODCCMamba-Unet), a UNet-based model featuring the Omnidirectional Divide-and-Conquer Convolutional Mamba (ODCCMamba) module. ODCCMamba introduces a novel Omnidirectional Divide-and-Conquer Scanning (ODCScanning) strategy, which first performs local then global omnidirectional scans to effectively capture both local and global features. We also integrate Hue information from HSV space and Gabor-based texture features to enhance change detection. Furthermore, we design a Triple-branch Cross-modal ODCCMamba (ODCC-TCMamba) module to fuse three types of features effectively. Experiments on CDD, LEVIR-CD, and WHU-CD datasets show that our method outperforms Transformer-based, Mamba-based, and hybrid Mamba-CNN methods. Our code is available at https://github.com/zjp-zjp-zjp/ODCCMamba-Unet .