基于优化损失函数的YOLOv2目标检测器

Object detector YOLOv2 based on optimized loss function

针对基于深度学习的单阶段目标检测器在训练过程中的样本不均衡问题,设计了一种新型的基于Softmax分类的动态调制交叉熵损失函数。此损失函数能够高效地降低训练过程中的易分负样本的损失权重,相应地提高困难样本的损失权重,从而可以使模型的整个训练过程变得高效。将基于Softmax分类的动态调制交叉熵损失函数代替标准的交叉熵损失函数用于YOLOv2训练中关于类别预测的损失计算,能够一定程度上提升YOLOv2的检测准确率。

To solve the sample imbalance problem of single-stage object detector based on deep learning in the training process,a new dynamic modulation cross entropy loss function based on Softmax classification is designed.This loss function can effectively reduce the loss weight of easily classified negative samples in the training process,and correspondingly improve the loss weight of difficult samples.Consequently,the whole model training process can become more efficient.By replacing the standard cross entropy loss function,the dynamic modulation cross entropy loss function based on Softmax classification is used for the loss calculation of category prediction in the YOLOv2 training.The YOLOv2 detection accuracy can be improved by a certain extent.