 This paper proposes a transformer-based semantic change detection, SCD, model, pyramid SCD former, that accurately recognizes small changes and find details in remote sensing images with multiple changes from different scales. The SCD model selectively merges different semantic tokens in multi-head self-attention block to obtain multi-scale features, which is crucial for extraction information of RSI's. Experimental results demonstrate that pyramid SCD former outperforms existing state-of-the-art CD models and obtains an improvement in MIOU F1 of up to 19.53% on the Landsat SCD dataset for change classes proportion less than 1%. The recognition performance for small scale and fine edges of change types has greatly improved. This article was offered by Pan Li Yuan, Qing Zhang Zhao, Qing Biao Zhao, and others.