Vehicle Target Detection Algorithm Based on Improved Faster R-CNN for Remote Sensing Images

Aiming at the problems that remote sensing image vehicle targets are susceptible to complex background interference, multi-scale differences, and difficulties in detecting small targets, this paper proposes a remote sensing image vehicle target detection algorithm based on improved Faster R-CNN. In this paper, based on the framework of Faster R-CNN, firstly, a multi-scale feature extraction network (EM-FPN) is designed by using the FPN structure and ResNet50 network, so that the network extracts rich target features; secondly, the ECA attention mechanism is introduced, so that the feature extraction network focuses on the target features, suppresses the interference of irrelevant background information, and constructs the multirate dilated convolution module (MDCM) to enhance the network's ability to perceive the contextual information of the target; finally, ROI Align is used instead of ROI Pooling to reduce the feature quantization error. The experimental results prove that the accuracy of the proposed algorithm reaches 88.6%, which can effectively detect vehicle targets in remote sensing images.