Medical imaging is used to assist doctors in clinical diagnosis and treatment. Automated generation of accurate and reliable medical imaging reports can reduce the burden on physicians. The lesion region in a chest radiograph is only a small part of the whole image, and many studies use only the overall features to generate reports. In this paper, this study proposes a new framesswork for SAENet. Firstly, multi-layer feature separation is performed on radiographs. Then, the multi-layer features are fused. These fused features are fed into the developed feature enhancement fusion module, which outputs the enhanced features. Finally, the caption generator uses these features to generate a report. We performed an experimental evaluation on the publicly available dataset IU X-Ray and the improvement of results show that: BLEU-1:6.4%, BLEU-2:11.2%BLEU-3:9.6%, BLEU-4:4.2%, METEOR:19.8%.