实时目标检测算法YOLO的批再规范化处理

针对实时目标检测YOLO(You Look Only Once)算法中存在的检测精度低和网络模型训练速度慢等问题，结合批再规范化算法处理小批样本以及非独立同分布数据的优势，提出了在YOLO网络结构加入批再规范化处理的改进算法。该YOLO改进算法把卷积层经过卷积运算产生的特征图看作一个个神经元，然后对这些神经元进行规范化处理。同时，在网络结构中移除了Dropout，并增大了网络训练的学习率。实验结果表明，该改进算法相对于原YOLO算法具有更高的检测精度、更快的实时检测速度以及通过适当设置批样本大小可使网络模型在训练时间和硬件设备方面成本有一定的降低。

Batch renormalize the realtime object detection algorithm YOLO

Abstract: In order to overcome the shortcomings that low detection accuracy and slow network training of the realtime object detection algorithm YOLO (You Look Only Once), combining the characteristic of Batch Reormalization, which has the strength to deal with small or non-i.i.d. minibatches, this paper proposed an algorithm that add the Batch Renormalization to the YOLO network structure. The improved algorithm view the feature maps, which generated from the convolutional layers, as activations, and then batch renormalize the activations, meanwhile it remove the Dropout from the original network structure and increase learning rate. The experimental results shows that the proposal algorithm own better detection accuracy and faster than before, and furthermore, it can decrease the model training time and the requirement of hardware equipment.