![]() To address this, we extend our well-established port monitoring ATR techniques, initially trained on generic ship classes, to two newly curated datasets: aircraft carriers and other military warships.Ī key challenge in this type of extension is that these more precise targets of interest are also rarer, making them harder to find and train against using traditional supervised deep learning approaches. For critical defense applications, we need to know that these techniques continue to perform on more granular and difficult class types, especially those that are highly related and challenging to disambiguate from one another. These preliminary techniques were developed against generic ship classes that are relatively prevalent, and they are largely distinct from one another. ![]() ![]() In the past, we have shown that u-net based ship detection and hybrid/external convolutional neural networks (CNNs) are effective for Port Monitoring ATR and generalize well to new scenes and applications, such as object level change detection. However, the generic and often disparate or unrelated class labels in these datasets are not as useful for benchmarking overhead imagery-based defense applications. As deep learning techniques have become the dominant approach to automatic target recognition (ATR), several datasets (e.g., Imagenet) have been established as the standards for performance assessment and optimization. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |