The rapid proliferation of social media has underscored the critical need for effective detection of hate content disseminated online, particularly as such content often comprises both textual and visual modalities. Existing multimodal detection methods frequently suffer from inadequate complementarity between text and image modalities, leading to an imbalance wherein one modality disproportionately influences the classification outcome. To address these limitations, we propose the Dual-Stream Attention Fusion Model (DAHM), specifically designed for hate content detection. DAHM employs a dual-stream architecture, wherein textual and visual features are extracted independently before undergoing an attention-based fusion process. This mechanism enhances the interaction between the two modalities, ensuring that both contribute meaningfully to the final classification. The model further integrates the outputs from each stream through a weighted aggregation strategy, thereby mitigating the risk of over-reliance on a single modality and enhancing overall detection performance. Experimental results show our model surpasses baselines on critical metrics, achieving a higher F1-score and a lower MMAE on challenging datasets. This underscores its effectiveness in delivering fair, robust, and reliable multimodal hate content detection.