训练神经网络时振荡火车/阀门精度图

Question

我正在做手势识别，我将视频分为几帧。 我已经使用CNN3D

如下训练了我的模型

Layer (type)                 Output Shape              Param #   
=================================================================
conv3d_1 (Conv3D)            (None, 10, 116, 116, 32)  12032     
_________________________________________________________________
max_pooling3d_1 (MaxPooling3 (None, 5, 58, 58, 32)     0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 5, 58, 58, 32)     0         
_________________________________________________________________
conv3d_2 (Conv3D)            (None, 3, 56, 56, 64)     55360     
_________________________________________________________________
max_pooling3d_2 (MaxPooling3 (None, 1, 28, 28, 64)     0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 1, 28, 28, 64)     0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 50176)             0         
_________________________________________________________________
dense_1 (Dense)              (None, 256)               12845312  
_________________________________________________________________
dropout_3 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 128)               32896     
_________________________________________________________________
dropout_4 (Dropout)          (None, 128)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 5)                 645       

categorical_accuracy 与 val_categorical_accuracy 的关系图看起来像

这可能是什么原因？我想念什么或做错什么了？