In recent years, query-based object detection methods have made significant progress in several fields, but still face the challenges of high computational complexity and slow convergence, especially when dealing with complex scenes and high resolution images. In addition, the layer-by-layer dependency structure of the model may lead to cascading errors, affecting stability and detection performance. To address these issues, we design a new detection model, Query-based Dynamic Sampling and Cascade Fusion Object Detector (DSCF-DE). The model is improved in two main aspects. First, DSCF-DE employs a dynamic sampling strategy that reduces redundant attention computations in the decoder, reduces the reliance on complex attention mechanisms, significantly reduces the computational burden, and enhances the query representation through spatial and channel blending. Secondly, DSCF-DE introduces a cascade fusion module, which utilizes the history query cache to complement and correct the current query, effectively mitigating the cascade error during training and enhancing the convergence speed and stability of the model. The experimental results verify the effectiveness of DSCF-DE.