LSKNet: A Lightweight Backbone for Remote Sensing LSKNet is a lightweight backbone network designed for remote sensing tasks, including scene classification, object detection, and semantic segmentation. It dynamically adjusts its large spatial receptive field to better model the contextual information needed for various remote sensing objects. Unlike previous methods, LSKNet introduces large and selective kernel mechanisms, which are novel in remote sensing. The network uses a spatial selective mechanism to efficiently weight features processed by large depth-wise kernels and then spatially merge them. This allows the model to adaptively use different large kernels and adjust the receptive field for each object in space. The paper presents an extended version of LSKNet, evaluating its performance on multiple remote sensing benchmarks, including scene classification, object detection, semantic segmentation, and change detection. LSKNet achieves state-of-the-art results on these tasks across 14 public datasets. The model's effectiveness is validated through comprehensive analysis, highlighting the importance of identified priors in remote sensing data. LSKNet's architecture is based on repeated LSK blocks, which consist of two residual sub-blocks: the Large Kernel Selection (LK Selection) sub-block and the Feed-forward Network (FFN) sub-block. The LK Selection sub-block dynamically adjusts the network's receptive field, while the FFN sub-block is used for channel mixing and feature refinement. The LSKNet backbone is efficient and lightweight, achieving high performance on various remote sensing tasks. The paper also compares LSKNet with other models, such as SKNet, and demonstrates its advantages in terms of performance and efficiency. The results show that LSKNet outperforms other models in remote sensing tasks, particularly in object detection, semantic segmentation, and change detection. The model's ability to adaptively select kernels based on the input data makes it effective in capturing and processing semantic features in remote sensing images. The proposed LSKNet backbone is a significant contribution to the field of remote sensing, offering a lightweight and efficient solution for various downstream tasks.LSKNet: A Lightweight Backbone for Remote Sensing LSKNet is a lightweight backbone network designed for remote sensing tasks, including scene classification, object detection, and semantic segmentation. It dynamically adjusts its large spatial receptive field to better model the contextual information needed for various remote sensing objects. Unlike previous methods, LSKNet introduces large and selective kernel mechanisms, which are novel in remote sensing. The network uses a spatial selective mechanism to efficiently weight features processed by large depth-wise kernels and then spatially merge them. This allows the model to adaptively use different large kernels and adjust the receptive field for each object in space. The paper presents an extended version of LSKNet, evaluating its performance on multiple remote sensing benchmarks, including scene classification, object detection, semantic segmentation, and change detection. LSKNet achieves state-of-the-art results on these tasks across 14 public datasets. The model's effectiveness is validated through comprehensive analysis, highlighting the importance of identified priors in remote sensing data. LSKNet's architecture is based on repeated LSK blocks, which consist of two residual sub-blocks: the Large Kernel Selection (LK Selection) sub-block and the Feed-forward Network (FFN) sub-block. The LK Selection sub-block dynamically adjusts the network's receptive field, while the FFN sub-block is used for channel mixing and feature refinement. The LSKNet backbone is efficient and lightweight, achieving high performance on various remote sensing tasks. The paper also compares LSKNet with other models, such as SKNet, and demonstrates its advantages in terms of performance and efficiency. The results show that LSKNet outperforms other models in remote sensing tasks, particularly in object detection, semantic segmentation, and change detection. The model's ability to adaptively select kernels based on the input data makes it effective in capturing and processing semantic features in remote sensing images. The proposed LSKNet backbone is a significant contribution to the field of remote sensing, offering a lightweight and efficient solution for various downstream tasks.