Deep Learning Network Algorithm Based on X-transfer Learning for Micro Object Detection

Abstract

In this paper, a low-resolution object detection algorithm was proposed based on X-transfer learning on GAN model. The proposed method is effective in improving detection of micro objects by optimizing with GAN network for super-resolution and an object recognition network. In addition, the proposed X-transfer learning technique alternately uses transfer learning and curriculum learning to overcome the lack of training data. This method can improve the accuracy, robustness, and localization performance of object recognition based on rich visual information on entire network. The proposed model was evaluated with remote sensing data sets. It was confirmed that the proposed method is more accurate than existing methods in terms of mAP@0.5 and F1 scores.