基于改进YOLOv5的轻量化航空目标检测方法

为解决硬件平台资源受限条件下的实时航空目标检测需求，在基于改进YOLOv5的基础上，提出了一种针对移动端设备/边缘计算的轻量化航空目标检测方法。首先以MobileNetv3为基础搭建特征提取网络，设计通道注意力增强结构MNtECA (MobileNetv3 with Efficient Channel Attention)提高特征提取能力；其次在深度可分离卷积层增加1×1的卷积，在减少卷积结构参数的同时提高网络的拟合能力；最后对检测网络进行迭代通道剪枝实现模型压缩和加速。实验选取DIOR (Object Detection in Optical Remote Sensing Images)数据集进行训练和测试，并在嵌入式平台(NVIDIA Jetson Xavier NX)对轻量级模型进行推理验证。结果表明，所提出的轻量级模型大幅降低了参数和计算量，同时具有较高精度，实现了移动端设备/边缘计算的实时航空目标检测。

Lightweight Aerial Object Detection Method Based on Improved YOLOv5

In order to solve the real-time aerial object detection requirements under the condition of limited hardware platform resources， we propose a lightweight aerial object detection method for mobile devices/edge computing based on the improved YOLOv5. Firstly， we build a feature extraction network based on MobileNetv3， and design a channel attention enhancement structure MNtECA to improve the feature extraction ability. Then， we add 1×1 convolution to the depthwise separable convolutions layer to reduce the parameters of convolution structure while improving the fitting ability of the network. Finally， we force iterative channel-level pruning to the detection network to achieve model compression and acceleration. We conduct the training and testing experiments on the DIOR dataset， and perform inference verification on the embedded platform (NVIDIA Jetson Xavier NX). The results prove that the proposed lightweight model reduces the number of parameters and the amount of calculation greatly with high accuracy， realizing the real-time aerial object detection of mobile devices/edge computing.