Detection of mining land is of great importance for sustainable development of mining land geologic environment. The current mining land remote sensing dataset is relatively less, which is not conducive to the construction of deep learning model. The background composition of mining land targets is complex and the scale difference is large, which is easy to cause difficulties in recognizing mining land features and miss detection of weak targets. In this study, a target detection dataset of mining land based on multispectral imagery in a large area was constructed. Then, a feature exchange and distribution-based mining land detection method (FED-MD) was proposed and tested. (1) Channel switching and attention module: Channel attention is used to perform channel weighting and exchange of multi-level RGB and NIR features to achieve information complementation. Intra-view and cross-view attention were used to fuse the former exchanged features guiding the model to focus on mining land features, further enhancing the model's ability to identify mining lands in complex backgrounds. (2) Feature aggregation and distribution module: the features obtained in (1) were further weighted and fused to realize the aggregation of multi-level spatial and semantic information. The aggregated features were distributed by down sampling to align and fuse with the multi-scale features in the feature pyramid. This module can enhance the feature representation capability. Experiments demonstrate that the proposed FED-MD model can effectively improve the accuracy of mining land detection.