Hey @Ahmad Alhashil
There are a lot of different ways you can try to increase accuracy of your model using transfer learning.
The first thing you can try is to unfreeze weights and biases of more layers in the pretrained network, and train it using your dataset. This will increase the time to train the network as there are more learnable parameters but will also make your network suitable for your dataset.
Other than that you can use techniques common to all the neural networks to increase the performance, which are augmenting your dataset, and finetuning hyperparameters like learning rate, mini batch size, adding layers, adding neurons, applying regularization techniques like L1, L2, dropout.
