Automatic target detection plays a major role in automated war operations. The key concept behind automated target detection is military objects recognition from the captured images. For object recognition in the given image, Convolutional Neural Network (CNN) is a powerful classification network. But in general CNNs are trained for general object recognition. But, the performance of CNN depends mainly on the size of the training set. The size of the training data is generally available in less proportion for military objects due to its operational and security issues. Hence the performance of CNN may degrade sharply. To address the issue of military objects, a relatively new neural network architecture called Capsule Network (CapsNet) is introduced. Hence, in this article, a variant of CapsNet called Multi-level CapsNet framework is projected for military object recognition under the case of small training set. The introduced framework of this paper is validated on a dataset of military objects which are collected from the internet. The dataset contains particularly five military objects and the similar civil ones. The proposed framework demonstrates a large improvement of 96.54% of accuracy for military object recognition. Experiments demonstrate that the proposed framework can accomplish a high recognition precision, superior to many other algorithms such as conventional Support Vector Machines and transfer learning based CNNs.