Object detection in satellite images by faster R-CNN incorporated with enhanced ROI pooling (FrRNet-ERoI) framework

In this research work, the author proposes a new model of FrRNet-ERoI approach merely utilized to detect object within the remote sensing image. Here, we model a Faster R-CNN procedure comprise of network layer such as backbone ResNet-101 CNN network, HoG Feature Pyramid, Multi-scale rotated RPN and Enhanced RoI pooling network. To implement the proposed technique, the deep network containing respective layers which are trained through MATLAB software. In backbone layer, ResNet-101 network is preferred which effectively trained with the help of imagenet dataset. Then HoG feature pyramid inputted the final residual feature map to extract local informative features by image gradient approach. RPN poses to build an anchor box which is used to detect several numbers of objects. Finally, the enhanced RoI developed to optimize the model using bat algorithm. It performs several strategies to fine-tune the network. Because of this design approach, the proposed approach sustains to show high efficiency with reduced training time. The result concludes that proposed work achieves better detection accuracy than the several state-of-art techniques.