一般情况，抽样层跟随在卷积层后面，对前一层特征映射图进行二度特征提取。

In general, the sampling layer follows the convolutional layer and performs two-degree feature extraction on the feature map of the previous layer. The convolutional layer benefits from its structural characteristics. Although it can significantly reduce the connection paths between neurons and reduce network parameters, the total number of neurons in each output feature map obtained by the convolutional layer has not been significantly reduced. If a fully connected classifier is directly added after the convolutional layer, the input dimension of the classifier needs to be very high, which is likely to cause over-fitting problems

In general, the sampling layer follows the reel layer and extracts the feature map of the previous layer for two degrees. The reel layer benefits from its structural characteristics, although it can significantly reduce the connection path between neurons and reduce network parameters, but the total number of neurons for each output feature map obtained by the reel layer has not decreased significantly. If a fully connected classifier is added directly behind the reel layer, the classifier's input dimensions need to be high and prone to overfitted problems

In general, the sampling layer follows the convolution layer, and the second degree feature extraction is carried out on the feature map of the previous layer. The convolution layer benefits from its structural characteristics. Although it can significantly reduce the connection path between neurons and reduce the network parameters, the total number of neurons in each output feature map obtained by convolution layer does not significantly decrease. If a fully connected classifier is added directly after the convolution layer, the input dimension of the classifier needs to be very high, which is easy to cause the problem of over fitting<br>