import os

import cv2

import numpy as np

# Initialize data dictionaries for splits

data_dict = {"Train": [], "Valid": [], "Test": []}

labels_dict = {"Train": [], "Valid": [], "Test": []}

# Loop over folds

for fold in os.listdir(dataset_path):

fold_path = os.path.join(dataset_path, fold)

if not os.path.isdir(fold_path):

continue

for split in ["Train", "Valid", "Test"]:

split_path = os.path.join(fold_path, split)

if not os.path.exists(split_path):

print(f" Missing split folder: {split_path}")

continue

for cls in classes:

class_path = os.path.join(split_path, cls)

if not os.path.exists(class_path):

print(f" Missing class folder: {class_path}")

continue

for img_name in os.listdir(class_path):

img_path = os.path.join(class_path, img_name)

img = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)

if img is not None:

img = cv2.resize(img, (img_size, img_size))

data_dict[split].append(img.flatten()/255.0)

labels_dict[split].append(cls)

# Convert to numpy arrays

X_train = np.array(data_dict["Train"])

y_train = np.array(labels_dict["Train"])

X_val = np.array(data_dict["Valid"])

y_val = np.array(labels_dict["Valid"])

X_test = np.array(data_dict["Test"])

y_test = np.array(labels_dict["Test"])

print("Train:", len(X_train))

print("Val:", len(X_val))

print("Test:", len(X_test))

from sklearn.preprocessing import LabelEncoder

from tensorflow.keras.utils import to_categorical

le = LabelEncoder()

y_train = to_categorical(le.fit_transform(y_train), num_classes=len(classes))

y_val = to_categorical(le.transform(y_val), num_classes=len(classes))

y_test = to_categorical(le.transform(y_test), num_classes=len(classes))

from sklearn.utils import class_weight

import numpy as np

y_int = np.argmax(y_train, axis=1)

weights = class_weight.compute_class_weight(

class_weight="balanced",

classes=np.unique(y_int),

y=y_int

)

class_weights = dict(enumerate(weights))

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense

model = Sequential([

Dense(512, activation="relu", input_shape=(img_size*img_size,)),

Dense(256, activation="relu"),

Dense(len(classes), activation="softmax")

])

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

model.summary()

history = model.fit(

X_train, y_train,

validation_data=(X_val, y_val),

epochs=80,

batch_size=100,

class_weight=class_weights

)

loss, acc = model.evaluate(X_test, y_test)

print(f"Test Accuracy: {acc*100:.2f}%")