Visual Question Answering Model Using Transformer-Based Multi-Head Cross Attention Network with Optimized Gold Rush Octave Convolution Network