提取深度网络某一层特征

1.利用框架：

    TensorFlow+keras

2.提取深度网络某一层特征的核心函数：

    import keras.backend as K

    def get_activation(model, layer, X_batch):
        get_activations= K.function([model.layers[0].input, K.learning_phase()],[model.layers[layer].output])
        activations=get_activations([X_batch, 0])
        return activations

    

    使用示例：

    # generate extract-feature data
    X_extractfeature = HDF5Matrix(hdf5path, 'X', 0, dataSetLength, normalizer = scaleData)
    y_extractfeature = HDF5Matrix(hdf5path, 'y', 0, dataSetLength)

   count=range(0, len(X_extractfeature), 100)
    for i in range(len(count)):
    if i<len(count)-1:
        myfeature=get_activation(model, 11, X_extractfeature[count[i]:count[i+1]])
    if i==len(count)-1:
        myfeature=get_activation(model, 11, X_extractfeature[count[i]:len(X_extractfeature)])

3.提取深度网络各层信息：

    for layer in model.layers:
       print("{} output shape: {}".format(layer.name, layer.output_shape))
       print layer.output