Enhancing Scientific Visual Question Answering via Vision-Caption aware Supervised Fine-Tuning