While many aspects of the image recognition problem have been largely solved by presenting large datasets toconvolutional neural networks, there is still much work to do when data is sparse. For synthetic aperture radar(SAR), there is a lack of data that stems both from the cost of collecting data as well as the small size of thecommunity that collects and uses such data. In this case, electromagnetic simulation is an effective stopgapmeasure, but its effectiveness at mirroring reality is upper bounded both by the quality of the electromagneticprediction code as well as the fidelity of the target's digital model. In practice, we find that classificationmodels trained on synthetic data generalize poorly to measured data. In this work, we investigate three machinelearning networks, with the goal of using the network to bridge the gap between measured and synthetic data.We experiment with two types of generative adversarial networks as well as a modification of a convolutionalautoencoder. Each network tackles a different aspect in the problem of the disparity between measured andsynthetic data, namely: generating new, realistic, labeled data; translating data between the measured andsynthetic domain; and joining the manifold of the two domains into an intermediate representation. Classificationresults using widely-employed neural network classifiers are presented for each experiment; these results suggestthat such data manipulation improves classification generalization for measured data.
展开▼
