Incremental learning-based land mark recognition for mirco-UAV autonomous landing

In order to expand the application fields of micro-UAVs, the ability of land mark recognition and autonomous landing is one of the key technologies for UAVs flighting in complex environment. For achieving more robust and precise relative pose estimation, we propose to apply an ellipse feature-based pose estimation method instead of QR code features. Considering the poor calculating ability on-board, the land mark recognition algorithms based on deep learning are difficult to be used in micro-UAVs. Hence, we put forward a new strategy for target recognition by taking advantage of incremental learning. Concretely, we select to use broad learning system (BLS) to replace the classification layer of MobileNetV3, and design a new target recognition network that may be named as MobileNetV3-BLS. To verify the effectiveness of proposed MobileNetV3-BLS, we use PASCAL VOC2007 and data set collected in our university, and carry out a series of comparative experiments on Nvidia TX2. Results of experiments show that MobileNetV3-BLS can progressively increase the accuracy of landmark recognition online. In addition, the proposed MobileNetV3-BLS does meet the need of deployment on Nvidia TX2 and the real-time requirement of on-board calculation in mirco-UAV avionics systems.