Autonomous Classification and Early Warning Framework for UAV Images Combining Improved ShuffleNet-V2 and Attention Mechanism
To realize Unmanned Aerial Vehicle(UAV)autonomous monitoring and early warning in disaster events,a novel architecture combining channel-by-channel attention mechanism and efficient convolutional neural network is proposed.Taking into account the resource constraints of embedded platforms,the lightweight ShuffleNet-V2 is used as the backbone network,by which more information is efficiently encoded and the network complexity is reduced as much as possible.In order to further improve the accuracy of disaster scene classification,a Squeeze-Excitation(SE)module is incorporated to implement a channel-wise attention mechanism,which significantly enhances the attention to important features.A UAV aerial disaster event dataset containing 12 876 images is obtained through data acquisition and enhancement technique.The performance of the proposed method is evaluated and compared with that of other advanced models.The results show that the proposed method achieves an average accuracy of 99.01％,the model size is only 5.6 MB,and the processing speed on UAV-onboard platform exceeds 10 FPS,which can meet the practical needs of UAV platform for autonomous disaster monitoring tasks.
万方数据
维普
