Data Source

def get_array (path) :

    

    X = []

    y = []

    

    for label, disease in enumerate(os.listdir(path)) :

        

        

        

        

        # resize and append

        for count, image in enumerate(os.listdir(os.path.join(path, disease))) :

            X.append(img_to_array(

                fromarray(

                    cv2.resize(

                        cv2.imread(

                            os.path.join(os.path.join(path, disease),image)),(128,128)))))

            

            # only 5k imgs

            y.append(label)

            if count > 5e3 :

                break

    

    X = np.array(X)

    y = np.array(y)

    

    return X/255.0, to_categorical(y)







X_train, y_train = get_array(paths[0])

X_test , y_test  = get_array(paths[1])

X_valid, y_valid = get_array(paths[2])

plt.figure(figsize = (20,20))



i = 0



while i < 16 :

    

    x = np.random.randint(0, 20000)

    plt.subplot(4, 4, i+1)

    plt.imshow(X_train[x])

    plt.axis('off')

    plt.title(info[np.argmax(y_train[x])])

    

    i += 1

plt.show()

Model

Two methods were defined to construct the convolution block and fully connected block. The convolution blocks consists of Conv2D, MaxPooling2D, BatchNormalization, Dropout and LeakyReLU (with a slope of 0.2) layers. And the fully connected block consists of Dense with l2 regularization, BatchNormalization, Dropout and LeakyReLU layers.

def conv_layer (filterx) :

    

    model = Sequential()

    

    model.add(Conv2D(filterx, (4,4), padding = 'same', use_bias = False))

    model.add(BatchNormalization())

    model.add(Dropout(.2))

    model.add(LeakyReLU())

    model.add(MaxPooling2D((2, 2)))

    

    return model



def dens_layer (hiddenx) :

    

    model = Sequential()

    

    model.add(Dense(hiddenx, kernel_regularizer = 'l2'))

    model.add(BatchNormalization())

    model.add(Dropout(.2))

    model.add(LeakyReLU())

    

    return model

Conv2D -

Convolution layers are the feature detectors for images. The filters move across the width and height of the image performing convolution operations and storing the values.

MaxPooling2D -

Pooling layer with mode = ‘max’ is used to reduce the image and retain the dominant features in the region of the filter applied.

BatchNormalization -

Normalize the hidden features to have zero mean and unit standard deviation. It helps in faster convergence and prevents internal covariate shift.

def cnn (filter1, filter2, filter3, hidden1) :

    

    model = Sequential([

        

        Input((128,128,3,)),

        conv_layer(filter1),

        conv_layer(filter2),

        conv_layer(filter3),

        Flatten(),

        dens_layer(hidden1),

        Dense(4, activation = 'softmax')

    ])

    

    model.compile(loss = 'categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])

    return model

model summary,



Accuracy



Confusion Matrix