model.predict() - Model with accuracy near to 1 predicts wrong classes

I created a model to recognize license plates. It is this one:

def create_model(input_shape = (224, 224, 3)):
    input_img = Input(shape=input_shape)
    model = efnB0_model (input_img)
    model = GlobalAveragePooling2D(name='avg_pool')(model)
    model = Dropout(0.2)(model)
    backbone = model

    branches = []
    for i in range(7):
            branches.append(backbone)
            branches[i] = Dense(360, name="branch_"+str(i)+"_Dense_360")(branches[i])
            branches[i] = BatchNormalization()(branches[i])
            branches[i] = Activation("relu") (branches[i])
            branches[i] = Dropout(0.2)(branches[i])
                       
            branches[i] = Dense(35, activation = "softmax", name="branch_"+str(i)+"_output")(branches[i])
        
    output = Concatenate(axis=1)(branches)
    output = Reshape((7, 35))(output)
    model = Model(input_img, output)

    return model

I used this DataGenerator:

import tensorflow.keras as keras
from skimage.io import imread
from skimage.transform import resize
import numpy as np
import math

class DataGenerator(Sequence):

    def __init__(self, x_set, y_set, batch_size):
        self.x, self.y = x_set, y_set
        self.batch_size = batch_size

    def __len__(self):
        return math.ceil(len(self.x) / self.batch_size)

    def __getitem__(self, idx):
        batch_x = self.x[idx*self.batch_size : (idx + 1)*self.batch_size]
        batch_x = np.array([resize(imread(file_name), (224, 224)) for file_name in batch_x])
        batch_x = batch_x * 1./255
        batch_y = self.y[idx*self.batch_size : (idx + 1)*self.batch_size]
        batch_y = np.array(batch_y)

        return batch_x, batch_y

Therefore, I onehot-encoded every license plate (length 7 characters and 35 possible character for every position using this code:

#One Hot Encoding der Labels, Zielarray hat eine Shape von (7,35)
from numpy import argmax
# define input string

def my_onehot_encoded(label):
    # define universe of possible input values
    characters = '0123456789ABCDEFGHIJKLMNPQRSTUVWXYZ'
    # define a mapping of chars to integers
    char_to_int = dict((c, i) for i, c in enumerate(characters))
    int_to_char = dict((i, c) for i, c in enumerate(characters))
    # integer encode input data
    integer_encoded = [char_to_int[char] for char in label]
    # one hot encode
    onehot_encoded = list()
    for value in integer_encoded:
        character = [0 for _ in range(len(characters))]
        character[value] = 1
        onehot_encoded.append(character)

    return onehot_encoded

For the license with label "7CT2498" I get the following onehot-encoded output:

[[0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0], [0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]

Now, when running the model I created above for 10 epochs on 10.000 training data and 3.000 validation data I get a training accuracy of 0.9969 and a validation accuracy of 0.9798, so not too bad.

But now I tried to predict a license plate with this model (the image is from the same dataset as my training and validation data is).

I used this code:

model = keras.models.load_model(
    "/path/to/model.h5", compile=True)
opt = keras.optimizers.Adam(learning_rate=0.0001)
model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=["accuracy"])
img = cv2.imread('/path/to/image.png')
img = cv2.resize(img,(224,224))
img = np.reshape(img,[1,224,224,3])

classes = model.predict(img)

print(classes)

And I do only get one correctly predicted class. Is there something wrong with my code?