2.构建基于GAN的战场目标图像数据生成模型为解决战场目标图像识别研究

2. Construction of GAN battlefield target image data generation model based on <BR>data generated model to solve the problem of insufficient training data model image recognition based on battlefield targets exist in paper-based GAN basic model derived, the characteristics of the Army Air Assault Brigade battlefield target image , battlefield targets established model enables generating image data among three different image conversion between domains. Model training method by studying the migration of part of the real battlefield using the target image data collected for a large number of virtual battlefield target image data and experimental results verify the feasibility of the diversity of the model and generate data.

2. Build a GAN-based battlefield target image data generation model In order to solve the problem of insufficient model training data in the field target image recognition research, the data generation model derived from the basic model of the paper is combined with the characteristics of the battlefield target image image of the Army Air Assault Brigade, and the battlefield target image data generation model can be used to convert between three different domain images. The model is trained by migration learning method, using some of the collected real battlefield target image data to generate a large amount of virtual battlefield target image data, and verifying the feasibility of the model and the diversity of the generated data.

2. Build the data generation model of battlefield target image based on GaN<BR>In order to solve the problem of insufficient model training data in the research of battlefield target image recognition, the data generation model derived from the basic Gan model in this paper, combined with the characteristics of battlefield target image of Army Air Assault brigade, established the battlefield target image data generation model which can realize the mutual conversion of three different domain images. The model is trained by the method of transfer learning, and a large number of virtual battlefield target image data are generated by using part of the real battlefield target image data collected. The feasibility of the model and the diversity of the generated data are verified by experiments.<BR>