Question

我想对振动测量结果进行二进制分类。在一篇论文中，我找到了以下网络，我想对其进行简化复制。（CNN-LSTM structure，LINK to the paper - download）

我现在对TimeDistributed（Conv1D（））的输入有问题，找不到错误。错误消息将复制到代码下方。它说我缺少一个输入变维。

我的数据具有以下结构：

2951次测量（标签=良好）+ 3043次测量（标签=不良）= 5994个样本。

每次测量16000个测量点

一维测量

希望您能帮助我，非常感谢您的宝贵时间！

我的代码：

xGood = dfGood.values
yGood = pd.DataFrame(index=range(len(dfGood)), columns=range(1))
yGood.iloc[:len(yGood)] = 0
yGood = yGood.values

xBad = dfBad.values
yBad = pd.DataFrame(index=range(len(dfBad)), columns=range(1))
yBad.iloc[:len(yBad)] = 1
yBad = yBad.values

x = np.concatenate((xBad, xGood), axis=0)
y = np.concatenate((yBad, yGood), axis=0)



trainX, testX, trainy, testy = train_test_split(x, y, test_size=0.33, random_state=42)
trainy = to_categorical(trainy)
testy = to_categorical(testy)
n_timesteps, n_outputs = trainX.shape[1], trainy.shape[1]
trainX = np.expand_dims(trainX, axis=2)
testX = np.expand_dims(testX, axis=2)

#Model
def evaluate_model(trainX, trainy, testX, testy):
    verbose, epochs, batch_size = 1, 50, 400
    n_timesteps, n_outputs = trainX.shape[1], trainy.shape[1]

    model = Sequential()
    #CNN Model
    model.add(TimeDistributed(Conv1D(filters=16, kernel_size=12001, activation='relu'), input_shape=(5994,n_timesteps, 1)))
    model.add(TimeDistributed(Conv1D(filters=250, kernel_size=3985, activation='relu')))
    model.add(TimeDistributed(MaxPooling1D()))
    model.add(TimeDistributed(Flatten()))
    #LSTM Model
    model.add(LSTM(16))
    model.add(Dense(n_outputs, activation='softmax'))

    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
    # fit network
    model.fit(trainX, trainy, epochs=epochs, batch_size=batch_size, verbose=verbose)
    # evaluate model
    _, accuracy = model.evaluate(testX, testy, batch_size=batch_size, verbose=0)
    print(model.summary())
    print(accuracy)

    model.save(pathModel)
    return accuracy

evaluate_model(trainX, trainy, testX, testy)

我得到的错误：

ValueError：检查输入时出错：预期time_distributed_28_input具有4维，但数组的形状为（4017，16000，1）

通过CNN-LSTM网络进行振动测量（1D）的二进制分类。输入形状问题

0 个答案: