Aerial image detection and recognition system based on deep neural network

Aerial image detection aims to find objects of interest and give their locations, and the goal of aerial image recognition is to classify the scenes or objects. Deep neural network is by far the best model for image detection and recognition. But still it fails to meet the requirements of high precision and low missing alarm. To solve this issue, we propose a two-step system including detection and fine-grained recognition modules. The first module is based on one-stage detection method YOLOv3. Furthermore, the metric-learning-based magnet loss is introduced to realize fine-grain recognition in the second step. The experiments prove the effectiveness of their combinations on improving precision and reducing missing alarm rate.