3.构建基于GAN的战场目标图像识别模型针对陆军空中突击旅战场目标易出

3. Build battlefield target image recognition model based GAN <BR>smaller target partial occlusion and long-range aerial reconnaissance due to the characteristics of a plan for the Army Air Assault Brigade prone battlefield targets, GAN-based model to feature the battlefield target image recognition model extraction network training, and strengthen the network of strong noise image to extract and partially occluded object profiling capabilities. By changing the target recognition classification and regression model input network conditions, without increasing the amount of calculation premise, the method of FIG resolution input features to achieve improved by lifting the small target object recognition model to identify the effect, and so constructed battlefield target image recognition model. Model with real battlefield target image data and generates a virtual battlefield target image data composed of data sets were trained to test, through performance comparison with the current mainstream model of object recognition, object recognition verify the validity of the model built by the paper. <BR>Through this study, in theory, to explore the use of the Army Air Assault Brigade in combat in the future to achieve the goal of intelligent information processing new style; on the method provides a new way to solve the problem of inadequate military training data model; in the application, to achieve an effective recognition of the complexity of battlefield target depth study of the use of technology.

3. Build a GAN-based battlefield target image recognition model In view of the characteristics of the low target caused by the easy partial masking of the battlefield target of the Army Air Assault Brigade and the low target caused by the long-range view reconnaissance in the air, the characteristic extraction network of the battlefield target image recognition model is trained according to the GAN model, and the characteristic extraction ability of the network to the strong noise image and partial masking target is strengthened. By changing the input conditions of classification regression network in the target recognition model, the target recognition model is improved by improving the resolution of the input feature map by means of improving the resolution of the input feature map, and the battlefield target image recognition model is constructed accordingly.<BR>The model uses the data set of real battlefield target image data and the generated virtual battlefield target image data to carry out training tests, and verifies the validity of the target recognition model built by the paper by comparing with the performance of the current mainstream target recognition model. Through the research of this paper, in theory, the new style of combat application under the future realization of intelligent target intelligence processing of the Army Air Assault Brigade is explored, the method provides a new way to solve the problem of insufficient military model training data, and in the application, the effective identification of complex battlefield targets by using deep learning technology is realized.

3. Build the model of battlefield target image recognition based on GaN<BR>In view of the characteristics that part of the battlefield targets of the Army Air Assault brigade are easy to be occluded and the targets are small due to aerial long-distance overhead reconnaissance, the feature extraction network of the battlefield target image recognition model is trained based on the Gan model to enhance the feature extraction ability of the network for strong noise images and partially occluded targets. By changing the input conditions of the classification regression network in the target recognition model, and on the premise of not increasing the amount of calculation, the method of improving the resolution of the input characteristic map is used to improve the recognition effect of the target recognition model on the small target, and based on this, the battlefield target image recognition model is constructed. The model uses the data set composed of the real battlefield target image data and the generated virtual battlefield target image data to train and test. Through the performance comparison with the current mainstream target recognition model, the validity of the target recognition model established in this paper is verified.<BR>Through the research of this paper, in theory, it explores the new operational mode of Army Air Assault brigade under the future intelligent target information processing; in method, it provides a new idea to solve the problem of insufficient military model training data; in application, it realizes the effective recognition of complex battlefield targets by using deep learning technology.<BR>