In order to improve the automation and intelligence level of aerial reconnaissance and forensics, this paper proposes an automatic target classification for aerial reconnaissance and forensics based on deep learning and FPGA software and hardware co-design. Multi-channel collection and collation of military aircraft data sets, using deep neural network model to carry out automatic recognition, verify the feasibility of automatic target classification, and deploy to the FPGA development evaluation board for accuracy, throughput and other performance verification. The experimental results show that various network models can effectively realize the automatic classification of aerial reconnaissance and forensics targets, and have high throughput and accuracy, and the timeliness can meet the actual needs of the battlefield.