The paper introduces the Poly Kernel Inception Network (PKINet) for remote sensing object detection, addressing the challenges of varying object scales and diverse contextual information. PKINet employs multi-scale depth-wise convolutions without dilation to extract dense texture features across different scales, while a Context Anchor Attention (CAA) module captures long-range contextual information. These components work together to improve performance on four challenging benchmarks: DOTA-v1.0, DOTA-v1.5, HRSC2016, and DIOR-R. The method outperforms previous state-of-the-art methods in terms of accuracy and computational efficiency, demonstrating its effectiveness and adaptability to various remote sensing datasets.The paper introduces the Poly Kernel Inception Network (PKINet) for remote sensing object detection, addressing the challenges of varying object scales and diverse contextual information. PKINet employs multi-scale depth-wise convolutions without dilation to extract dense texture features across different scales, while a Context Anchor Attention (CAA) module captures long-range contextual information. These components work together to improve performance on four challenging benchmarks: DOTA-v1.0, DOTA-v1.5, HRSC2016, and DIOR-R. The method outperforms previous state-of-the-art methods in terms of accuracy and computational efficiency, demonstrating its effectiveness and adaptability to various remote sensing datasets.