基于Tiny-YOLOv3的小目标检测仿真

Simulation of small object detection based on Tiny-YOLOv3

针对轻量级神经网络模型检测精度不高,容易对小目标物体产生漏检的问题,该文提出了基于Tiny-YOLOv3的目标检测改进算法。将Tiny-YOLOv3模型中的池化层用卷积核为3×3、步长为2×2的卷积层代替,对输入图像的尺寸进行调整,对特征提取网络最后4层的特征图尺寸与通道数进行修改,并在原有模型的基础上添加了一层特征融合层。在VOC2007数据集上进行仿真实验,改进后的模型mAP上升了3.79%,瓶子这类物体的AP值提高了14%,说明小目标物体的检测效果得到了提升,降低了中小目标检测过程中的漏检率。

In order to solve the problem that the lightweight neural network model has low precision and is easy to miss small objects,an improved object detection algorithm based on Tiny-YOLOv3 is proposed.The pooling layer in Tiny-YOLOv3 model is replaced by a convolution layer with convolution kernel of 3×3 and step size of 2×2,the size of the input image is adjusted,the size of the feature map and the number of channels in the last four layers of the feature extraction network are modified,and a feature fusion layer is added to the original model.The simulation experiment on VOC2007 dataset shows that the mAP(mean average precision)of the improved model has increased by 3.79%,and the AP value of objects like bottles has increased by 14%,which shows that the detection effect of small target is improved,and the miss detection rate is reduced.