Applying Convolutional Neural Network for Military Object Detection on Embedded Platform

Object detection has always been an important part in the field of image processing. The traditional object detection algorithm has complex structure and operations. With the continuous development of deep learning technology, Convolutional Neural Network (CNN) has become an advanced object detection method. Because of its high accuracy, stability, and speed of operation, this method is widely used in many fields. In this work, we use CNN to achieve the detection of military objects. It uses the idea of regression to build a model, which is fast and accurate and can achieve detection in real-time. Unlike image classification, image detection requires more parameters and calculations, and therefore it is difficult to be placed on a small embedded platform. We analyzed some of state-of-the-art object detection network, replace the traditional fully connected layer with global average pool layer, generate region proposals using the anchor boxes, and apply it to military object detection. Finally, we deployed it successfully on TMS320C6678, which is a low-cost, low-power embedded platform. A well-performing and easy-to-deploy military object detection system is realized, which helps to improve the accuracy and efficiency of military operations.