Description

ABSTRACT

we summarize the history and state of the art in Convolutional Neural Networks (CNNs), which constitute a significant advancement in pattern recognition. As a demonstration of capability, we address the problem of automatic aircraft identification during refueling approach. In this paper we describe the history of CNN development and provide a high level overview of the state of the art and a summary of leading CNN libraries with CUDA support. Finally, we demonstrate an application of CNN technology to autonomous aerial refueling and identify areas of follow on research.

INTRODUCTION

Recent emergence of high performance deep learning toolkits have spurred advances in deep learning research as summarized. In particular, the ImageNet competition of the last 5 years has produced a number of meaningful advances to convolutional neural network architectures and techniques that make this a feasible technology for a variety of object recognition applications. To give the reader a sense for the considerable impact of this burgeoning field, this paper traces its roots to biological origins and early development prior to the ImageNet competition, describing major issues that hindered earlier success and how they have been overcome. We then describe the general structure of a CNN, illustrated with a tour of the LeNet-5 framework, as a pedagogical example of modern implementations. The most recent advances that constitute the current state of the art are manifested in AlexNet which contextualizes developments in deep learning theory and practice As a case study that shows the potential for CNNs in a specific setting, we show an illustrative example of aircraft identification for aerial refueling. We describe the issues in training and classification using CNNs from a 76,000 image dataset. We implement a deep learning classifier

PROPOSED METHOD

we address the problem of automatic aircraft identification during refueling approach. In this paper we describe the history of CNN development and provide a high level overview of the state of the art and a summary of leading CNN libraries with CUDA support. Finally, we demonstrate an application of CNN technology to autonomous aerial refueling and identify areas of follow on research.

BLOCK DIAGRAM

CONCLUSION

The dataset is successfully trained to perform fine grain aircraft recognition of ten models of United States Air Force (USAF) aircraft, largely independent of perspective. Optimization of classifier test performance was accomplished by analysing the bias-variance trade-off relationship between test set performance and classifier size