Application of Minnan Folk Light and Shadow Animation in Built Environment in Object Detection Algorithm

To resolve the problems of deep convolutional neural network models with many parameters and high memory resource consumption, a lightweight network-based algorithm for building detection Minnan folk light synthetic aperture radar (SAR) images is proposed. Firstly, based on rotating target R-centernet, Ghost ResNet constructed to reduce number model by replacing traditional convolution in backbone convolution. Secondly, channel attention module integrating width height information proposed enhance network’s ability accurately locate salient regions images. Content-aware reassembly features (CARAFE) up-sampling used replace deconvolution fully incorporate feature map during improve detection. Finally, dataset rotated annotated shadow SAR trained tested using improved R-centernet algorithm. The experimental results show that improves accuracy 3.8%, recall 1.2% speed 12 frames/second compared original