합성곱 신경망 기반의 헬기 기종 분류 모형 연구

Abstract

In various surveillance systems, it is difficult to identify objects by relying on the naked eye. As the number of troops and the length of military service has been reduced in the Republic of Korea Army, there is a growing need for autonomous surveillance systems that could replace soldiers in the field of military surveillance operations. Thus, in this paper, we used binary classification based on a CNN (Convolutional Neural Network), which is a deep learning algorithm, to develop an autonomous model that classifies types of helicopters among the military weapon systems. We then evaluated the model’s performance. Consequently, the identification model demonstrates 98.9%, 99.4%, 98.4%, and 98.9% accuracy, precision, recall, and F-measure, respectively. In addition, we analyzed the feature map on the convolution layers of the identification model to check which area of imagery is highly weighted. In general, the rotary shaft of the rotating wing, wheels, and air-intake in both mobile and attack helicopters play a major role in the performance of the identification model.