Remote target detection in remote sensing image usually adopts single mode feature extraction.The feature information extraction of salient targets is not comprehensive,which will lead to poor detection of significant targets.Therefore,a method of significant target detection in UAV lidar remote sensing image is proposed.With the support of VGG16 network,the multimodal features of lidar remote sensing images are extracted by combining cascade operation and ReLu activation function to comprehensively describe the features of significant targets.Based on the results of multimodal feature extraction,the correlation between different levels is identified using multi-branch group fusion and single group fusion,and the feature fusion of each level is completed through Conv+ReLu layer.Specific weight values are assigned based on the importance of features,a spatial competition function is used to stack the weighted sum of each layer,and a significant target map is generated to achieve significant target detection in drone laser remote sensing images.The experimental results show that the proposed detection method has high accuracy in detecting significant targets and low detection time cost,with a maximum time cost of around 14 seconds.
NETL
NSTL
万方数据
