面向遥感图像实时目标检测任务的轻量级神经网络设计与实现

Design and Implementation of Lightweight Neural Network for Real-time Target Detection Tasks in Remote-sensing Images

神经网络的表征能力给遥感图像目标检测任务提供了一个的便捷工具。然而,目前主流神经网络模型计算代价高限制了其在遥感图像实时目标检测任务的应用。提出一种轻量级神经网络模型用于遥感图像实时目标检测。实验结果表明,文中提出的方法在保持与Yolov3检测精度相当的情况下,模型大小约为Yolov3的十五分之一,网络模型在目标检测精度以及计算开销上可达到更好的均衡。

The characterization ability of neural networks provides a convenient tool for remote-sensing image object detection tasks.However, the current high computational cost of mainstream neural network models has limited their application to real-time object detection tasks in remote-sensing images. This paper presents a lightweight neural network model for real-time object detection in remote-sensing images. The experimental results show that the method proposed in this paper keeps the detection accuracy equivalent to Yolov3, the model size is about one fifteen of Yolov3, and the network model can achieve a better balance in object detection accuracy and computational overhead.