2024102301

from keras import Input
from keras.layers import Conv2D, MaxPooling2D, Dropout, BatchNormalization
from keras.layers import concatenate, Conv2DTranspose, Activation
from keras.models import Model
from tensorflow.keras.optimizers import Adam


def EnhancedNetOptimized(pretrained_weights=None):
    input_shape = (None, None, 1)
    inputs = Input(shape=input_shape, name='input_img')

    conv1 = Conv2D(16, 3, activation='relu', padding='same')(inputs)
    bn1 = BatchNormalization()(conv1)
    drop1 = Dropout(0.3)(bn1)
    pool1 = MaxPooling2D(pool_size=(2, 2))(drop1)

    conv2 = Conv2D(24, 3, activation='relu', padding='same')(pool1)
    bn2 = BatchNormalization()(conv2)
    drop2 = Dropout(0.3)(bn2)
    pool2 = MaxPooling2D(pool_size=(2, 2))(drop2)

    conv3 = Conv2D(32, 3, activation='relu', padding='same')(pool2)
    bn3 = BatchNormalization()(conv3)
    drop3 = Dropout(0.3)(bn3)
    pool3 = MaxPooling2D(pool_size=(2, 2))(drop3)

    conv4 = Conv2D(40, 3, activation='relu', padding='same')(pool3)
    bn4 = BatchNormalization()(conv4)
    drop4 = Dropout(0.3)(bn4)

    up5 = Conv2DTranspose(32, 2, strides=(2, 2), padding='same')(drop4)
    concat5 = concatenate([drop3, up5], axis=3)
    conv5 = Conv2D(32, 3, activation='relu', padding='same')(concat5)
    bn5 = BatchNormalization()(conv5)
    drop5 = Dropout(0.3)(bn5)

    up6 = Conv2DTranspose(24, 2, strides=(2, 2), padding='same')(drop5)
    concat6 = concatenate([drop2, up6], axis=3)
    conv6 = Conv2D(24, 3, activation='relu', padding='same')(concat6)
    bn6 = BatchNormalization()(conv6)
    drop6 = Dropout(0.3)(bn6)

    up7 = Conv2DTranspose(16, 2, strides=(2, 2), padding='same')(drop6)
    concat7 = concatenate([drop1, up7], axis=3)
    conv7 = Conv2D(16, 3, activation='relu', padding='same')(concat7)
    bn7 = BatchNormalization()(conv7)
    drop7 = Dropout(0.3)(bn7)

    conv8 = Conv2D(1, 1, padding='same')(drop7)
    output = Activation('sigmoid')(conv8)

    model = Model(inputs=inputs, outputs=output)

    opt = Adam(learning_rate=1e-4)
    model.compile(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])

    if pretrained_weights:
        model.load_weights(pretrained_weights)

    return model