改写：改进的通道注意力机制由全局池化层、卷积层、sigmoid激活函数和scale操作组成将需要进行融合的特征称为 和 其中 为任一阶段第一个SFRM模块的输出特征 为第二个SFRM模块的输出特征。

The enhanced channel attention mechanism is composed of a global pooling layer, a convolutional layer, a sigmoid activation function, and a scaling operation. The features that need to be fused are referred to as F1 and F2, where F1 denotes the output feature of the first SFRM module at any stage, and F2 denotes the output feature of the second SFRM module.