无人机图像小目标检测实验的设计与实现

Design and Implementation of Small Object Detection Experiment in UAV Image

针对无人机图像小目标检测存在目标密集和遮挡等问题,设计了卷积注意力、深度特征融合和多尺度特征融合模块以增强图像小目标的局部信息和语义信息,从而提高对无人机图像小目标特征的提取能力,最后结合改进的损失函数加快了模型的收敛。基于VisDrone2019数据集的实验结果表明,所提模型提高了无人机图像小目标检测精度,减小了目标误检和漏检概率,为复杂场景下无人机小目标检测提供了实验基础。

In order to solve the problems such as target density,and occlusion in detecting the small objects of unmanned aerial vehicle(UAV)images,the convolutional attention,deep feature fusion,and multi-scale feature fusion modules are designed to enhance the local and semantic information,which can improve the capability of feature extraction in the small objects of UAV images.Meanwhile,the loss function is improved to accelerate the convergence of proposed model.The experimental results based on the VisDrone2019 dataset show that the proposed model improves the detection accuracy of small objects,reduces the probability of false and missed detections,and provides an experimental basis for small target detection in more complex scenes.