摘要
针对红外场景中行人、车辆等目标识别率低且存在复杂背景干扰的问题,提出一种基于Effi-YOLOv3模型的红外目标检测方法。将轻量高效的EfficientNet骨干网络与YOLOv3网络相结合,提升网络模型的运行速度。通过模拟人类视觉的感受野机制,引入改进的感受野模块,在几乎不增加计算量的情况下大幅增强网络有效感受野。基于可变形卷积和动态激活函数构建DBD和CBD结构,提升模型特征编码的灵活性,扩大模型容量。选择兼顾预测框与真值框中心点距离、重叠率和长宽比偏差的CIoU作为损失函数,更好地反映预测框与真值框的重叠程度,加快预测框回归速度。实验结果表明,该方法在FLIR数据集上的平均精度均值达到70.8%,Effi-YOLOv3模型参数量仅为YOLOv3模型的33.3%,对于红外场景中的目标检测效果更优。
To improve the low recognition rate of persons and cars in infrared scenes and solve the problem of complex background interference,an infrared target detection method based on Effi-YOLOv3 is proposed.This method combines the lightweight and efficient EfficientNet backbone network and the YOLOv3 network to improve the training speed of the model. By simulating the receptive field mechanism of human vision,an improved Receptive Field Block(RFB)is introduced to significantly increase the effective receptive field of the network while increasing computation by a small amount. Then,based on deformable convolution and dynamic activation functions,DBD and CBD structures are constructed to improve the flexibility of model feature coding and increase network model capacity. Finally,the CIoU,which takes into account the distance between the center points of the prediction box and the ground truth box and the overlap ratio and deviation of aspect ratio between them,is selected as the loss function.This better reflects the degree of overlap between the prediction box and the ground truth box and accelerates the regression speed of the prediction box.The experimental results show that the mean Average Precision(mAP) of the proposed method on the FLIR dataset reaches 70.8%,the parameter quantity of the Effi-YOLOv3 model is only 33.3% of the YOLOv3 model,and the detection effect of infrared targets is significantly improved.
作者
秦鹏
唐川明
刘云峰
张建林
徐智勇
QIN Peng;TANG Chuanming;LIU Yunfeng;ZHANG Jianlin;XU Zhiyong(Key Laboratory of Beam Control,Chinese Academy of Sciences,Chengdu 610209,China;Institute of Optics and Electronics,Chinese Academy of Sciences,Chengdu 610209,China;School of Electronic,Electrical and Communication Engineering,University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《计算机工程》
CAS
CSCD
北大核心
2022年第3期211-219,共9页
Computer Engineering
基金
国家科技委创新项目(G158207)。
关键词
YOLOv3模型
红外目标检测
复杂背景
可变形卷积
动态激活函数
YOLOv3 model
infrared target detection
complex background
deformable convolution
dynamic activation function
