结合MobileNetV2与YOLOv5s的轻量级遥感目标检测方法研究

Research on lightweight remote sensing target detection method combining MobileNetV2 and YOLOv5s

针对目前遥感影像目标检测方法模型结构复杂、计算量大,且难以在边缘计算硬件上部署并流畅运算的问题,提出一种基于YOLOv5s的轻量化遥感影像目标检测模型。将使用轻型通道注意力机制的MobileNetV2网络放入YOLOv5s内代替CSPDarknet53进行特征提取,然后使用多路径融合网络PANet实现多尺度特征图融合增强,最后输出3个尺度的特征图参与检测。在公开遥感目标检测数据集上对模型训练和测试。实验结果表明,改进后模型相比原模型YOLOv5s,在检测速度与模型体量方面均有明显提升,同时mAP相比对照组内的MobileNet-SSD、YOLOv4-tiny模型,分别提高了23.1%和21.4%。以上结果充分说明本文提出模型能够在边缘计算硬件上实现对遥感影像目标的精准、实时检测,可以在城市规划、应急救灾等领域发挥出重要的应用价值。

In view of the problem that the current remote sensing image target detection method has a complex model structure and a large amount of calculation,which makes it difficult to deploy and operate smoothly on edge computing hardware,a lightweight remote sensing image target detection model based on YOLOv5s is proposed.The MobileNetV2 network used by the light channel attention mechanism is put into YOLOv5s to replace CSPDarknet53 for feature extraction,and then the multi-path fusion network PANet is used to achieve multi-scale feature map fusion enhancement,and finally three-scale feature maps are output to participate in detection.The model was trained and tested on the public remote sensing target detection data set.The experimental results show that the improved model has significantly improved detection speed and model size compared with the original model,and the detection accuracy is compared with other groups of lightweight models.The comparison models have improved to varying degrees 23.1%and 21.4%.The above results fully demonstrate that the model proposed in this article can achieve accurate and real-time detection of remote sensing image targets on edge computing hardware,and may play an important application value in urban planning,emergency disaster relief and other fields.