How to use Tensorflow and Qiskit to make a quantum neural network and recognize handwriting？
Instead of combining Tensorflow and Qiskit, you can use the Tensorflow Quantum package. It provides seamless integration of quantum simulation in the tensorflow framework. You can build layered approaches as you would with classical neural networks, and use the standard loss functions and optimizers that you might be used to.
A tutorial on how to build a QNN using Tensorflow Quantum for MNIST (recognize hand-written digits) is available at https://www.tensorflow.org/quantum/tutorials/mnist.