摘要
目标检测作为计算机视觉的重要分支而备受人们关注。现有的检测方法普遍模型尺寸较大,设备要求高,难以用于移动端或嵌入式设备中实时处理。因此,本文利用MobileNetV3结合SSD的轻量级网络,再加入双向特征金字塔模型对特征进行融合,以此提高网络的准确率。实验结果表明,加入双向特征金字塔的轻量级目标检测网络在Pascal VOC数据集上mAP达到了73.65%,比单独的MobileNetV3-SSD目标检测网络提高了2.63%,模型所占内存大小比SSD降低了64.3%,检测速度比SSD提高了61%,使其更适应于移动端和嵌入式设备。
Object detection has received much attention as an important branch of computer vision.The existing detection methods generally have a large model size and high equipment requirements,which are difficult to be used for real-time processing in mobile terminals or embedded devices.Therefore,this paper uses the lightweight network of MobileNetV3 combined with SSD,and in order to improve the accuracy of the network,a bidirectional feature pyramid model is added to fuse the features.The experimental results show that the lightweight target detection network with bidirectional feature pyramid achieves mAP of 73.65%on the Pascal VOC dataset,which is 2.63%higher than the single MobileNetV3-SSD target detection network,and the memory size of the model is 64.3%lower than that of SSD and the detection speed is 61%higher than that of SSD.Make it more suitable for mobile and embedded devices.
作者
王贺
樊星
WANG He;FAN Xing(College of Physics and Electronic Engineering,Shanxi University,Taiyuan,030006,China)
出处
《测试技术学报》
2023年第2期152-157,共6页
Journal of Test and Measurement Technology
