CNN for analyzing gas-chromatography mass-spectrometry (GC-MS) data

Beginners
1

The ultimate goal is to create a model that can classify two (for now) different species of wasps by analyzing their GC-MS data.
I’m trying to A) set up an new CNN model, and B) use a big pre trained model (InceptionV3) to achieve this goal.
Since my dataset has only 240 samples (each is a matrix 800x535) i have problems to train the models properly (they have problems to generalize → high train_acc low valid_acc), if you have an idea i would be really happy to hear your thoughts!
Also if you are interested in the background i’ll be happy to explain

19_02_25_AoI_diff
19_02_25_AoI_diff1920×640 127 KB

model = Sequential([
    Conv2D(32, (3, 3), activation='relu', padding='same', input_shape=(800, 535, 1)), 
    BatchNormalization(),  
    MaxPooling2D((2, 2)),
    Dropout(0.25), # prevents overfitting by deactivating randomly neurons

    Conv2D(64, (3, 3), activation='relu', padding='same'),
    BatchNormalization(),
    MaxPooling2D((2, 2)),
    Dropout(0.3),
    
    Conv2D(128, (3, 3), activation='relu', padding='same'),
    BatchNormalization(),
    MaxPooling2D((2, 2)),
    Dropout(0.2),
    
    GlobalAveragePooling2D(),
    Dense(128, activation='relu', kernel_regularizer=l2(0.001)), 
    Dropout(0.5),
    Dense(1, activation='sigmoid')
    # -> After Dense Layers: (1) (final output)
])
1 Like