Bridging a Gap in SAR-ATR: Training on Fully Synthetic and Testing on Measured Data
نویسندگان
چکیده
Obtaining measured synthetic aperture radar (SAR) data for training automatic target recognition (ATR) models can be too expensive (in terms of time and money) complex a process in many situations. In response, researchers have developed methods creating SAR targets using electro-magnetic prediction software, which is then used to enrich an existing dataset. However, this approach relies on the availability some amount data. work, we focus case having 100% data, while testing only We use SAMPLE dataset public released by AFRL, find significant challenges learning generalizable representations from due distributional differences between two modalities extremely limited sample quantities. Using deep learning-based ATR models, propose augmentation, model construction, loss function choices, ensembling techniques enhance representation learned ultimately achieved over 95% accuracy analyze functionality our saliency feature-space investigations them learn more cohesive Finally, evaluate out-of-library detection performance synthetic-only that they are nearly 10% effective than baseline at identifying test samples do not belong class set. Overall, their compositions significantly feasibility trained exclusively
منابع مشابه
a study on rate making and required reserves determination in reinsurance market: a simulation
reinsurance is widely recognized as an important instrument in the capital management of an insurance company as well as its risk management tool. this thesis is intended to determine premium rates for different types of reinsurance policies. also, given the fact that the reinsurance coverage of every company depends upon its reserves, so different types of reserves and the method of their calc...
Bridging the Gap Between Synthetic and Real Data
There is a long tradition of using generative models in combination with discriminative classifiers [5, 6, 7]. Equally the recently successful deep learning technique [3] use jittering techniques [1, 2] that imply sampling from an underlying distribution. Although in both cases the the model is postulated and all parameters are in our control, we rarely achieve an accurate representation of the...
متن کاملin translation: translators on their work and what it means
کتاب در باب ترجمه، اثر استر آلن و سوزان برنوفسکی منتشر شده در ماه می 2013 توسط نشریه کلمبیا است. نویسندگان در این کتاب به بررسی 18 مترجم با در نظر گرفتن نقش آثاری که این مترجمان ترجمه کرده اند میپردازند. کتاب به دو بخش تقسیم میشود: " مترجم در جهان" و " کار مترجم" این دو بخش مقالات همیشگی ترجمه و موقعیت خاص ادبیات بیگانه در جهان وسیع امروزی را مورد خطاب قرار میدهد. در این کتاب مقالات متعددی از ن...
ذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
سال: 2021
ISSN: ['2151-1535', '1939-1404']
DOI: https://doi.org/10.1109/jstars.2021.3059991