A Typical Remote Sensing Object Detection Method Based on YOLOv3

considering the problem of typical remote sensing object detection, this paper proposes an improved remote sensing object detection method based on YOLOv3 algorithm. A new activation function is designed automatically by genetic algorithm. The classification task training result of the convolutional neural network on remote sensing image is used as the evaluation function of genetic algorithm. The function enhances its nonlinear transformation ability to remote sensing features; then the activation function is embedded in the YOLOv3 algorithm, and the object detection task is trained on the self-made typical remote sensing target dataset. The experimental results show that the method can significantly improve the detection accuracy of various typical remote sensing targets.