This paper proposes a novel Frequency Adaptive Mixture of Experts (FAME) framework for pan-sharpening, which aims to enhance the performance of pan-sharpening methods by effectively recovering high-frequency information. The framework consists of three key components: the Adaptive Frequency Separation Prediction Module, the Sub-Frequency Learning Expert Module, and the Expert Mixture Module. The first module uses the discrete cosine transform (DCT) to perform frequency separation by predicting a frequency mask. The second module employs low-frequency and high-frequency MOE to enable effective reconstruction of low- and high-frequency information. The final module dynamically weights high- and low-frequency MOE knowledge to adapt to remote sensing images with significant content variations. Quantitative and qualitative experiments on multiple datasets demonstrate that the proposed method outperforms existing state-of-the-art methods and exhibits strong generalization ability. The method leverages the frequency domain information and dynamic network structure to adapt to different remote sensing images, enabling the network to learn and process high-frequency information effectively. The framework is evaluated on three typical datasets, including WorldView-II, Gaofen2, and WorldView-III, and shows significant improvements in PSNR and other metrics. The method also performs well on full-resolution scenes, demonstrating its adaptability to remote sensing images. The proposed FAME framework is an innovative approach that combines MOE with frequency domain information, enabling the network to learn image features at different frequencies, particularly high-frequency information. The framework is designed to adapt to remote sensing images with significant content variance, thereby enhancing its generalization ability. The FAME framework comprises three key modules: Frequency Mask predictor, Sub-frequency learning experts module, and Experts Mixture module. The Mask predictor generates frequency masks that segregate the image into high-frequency and low-frequency parts, enabling effective processing of the image content. The Frequency experts consist of two MOE components, namely low-frequency MOE and high-frequency MOE, which are used for processing low-frequency and high-frequency information of the image. The final experts mixture part dynamically fuses high- and low-frequency features, as well as PAN and MS features, to adapt to remote sensing images with significant content variations. The final output is obtained by dynamically adding multiple different frequency experts. The method is evaluated on multiple datasets and shows significant improvements in performance compared to existing methods. The proposed method is effective in generating clear images with fine textures and strong generalization ability.This paper proposes a novel Frequency Adaptive Mixture of Experts (FAME) framework for pan-sharpening, which aims to enhance the performance of pan-sharpening methods by effectively recovering high-frequency information. The framework consists of three key components: the Adaptive Frequency Separation Prediction Module, the Sub-Frequency Learning Expert Module, and the Expert Mixture Module. The first module uses the discrete cosine transform (DCT) to perform frequency separation by predicting a frequency mask. The second module employs low-frequency and high-frequency MOE to enable effective reconstruction of low- and high-frequency information. The final module dynamically weights high- and low-frequency MOE knowledge to adapt to remote sensing images with significant content variations. Quantitative and qualitative experiments on multiple datasets demonstrate that the proposed method outperforms existing state-of-the-art methods and exhibits strong generalization ability. The method leverages the frequency domain information and dynamic network structure to adapt to different remote sensing images, enabling the network to learn and process high-frequency information effectively. The framework is evaluated on three typical datasets, including WorldView-II, Gaofen2, and WorldView-III, and shows significant improvements in PSNR and other metrics. The method also performs well on full-resolution scenes, demonstrating its adaptability to remote sensing images. The proposed FAME framework is an innovative approach that combines MOE with frequency domain information, enabling the network to learn image features at different frequencies, particularly high-frequency information. The framework is designed to adapt to remote sensing images with significant content variance, thereby enhancing its generalization ability. The FAME framework comprises three key modules: Frequency Mask predictor, Sub-frequency learning experts module, and Experts Mixture module. The Mask predictor generates frequency masks that segregate the image into high-frequency and low-frequency parts, enabling effective processing of the image content. The Frequency experts consist of two MOE components, namely low-frequency MOE and high-frequency MOE, which are used for processing low-frequency and high-frequency information of the image. The final experts mixture part dynamically fuses high- and low-frequency features, as well as PAN and MS features, to adapt to remote sensing images with significant content variations. The final output is obtained by dynamically adding multiple different frequency experts. The method is evaluated on multiple datasets and shows significant improvements in performance compared to existing methods. The proposed method is effective in generating clear images with fine textures and strong generalization ability.