Optimized Tank Detector Based on Modern Convolutional Neural Networks

This paper suggests a method to assist selecting the appropriate architecture that achieves right speed and accuracy balance for detecting military tanks. Nowadays, convolutional neural networks(CNN) have emerged as the powerful tool in the object detection area. In this paper, we have considered the faster region based convolutional neural networks(R-CNN) and single shot detectors(SSD) architectures. We show that, using different feature extractors and varying the critical parameters such as learning rate, momentum optimizer value, intersection over union(IOU) threshold, stride size and maximum region proposals we can obtain optimized tank detector.