基于改进的YOLOv3⁃Tiny深度网络在架图书错序检测方法

On-shelf books out-of-sequence detection based on improved YOLOv3-Tiny deep network

基于索书号识别的在架图书错序检测方法对构建智能化图书馆非常重要。为实现快速而准确的在架图书错序检测,文中提出一种基于改进的YOLOv3-Tiny深度网络,实现图书索书号区域定位。该方法在原深度网络模型的基础上增加一个52×52的尺度输出和两个跳跃连接块,以增强网络对微小目标的检测能力并抑制由于网络深度的增加带来的梯度消失或爆炸问题,从而实现对图书索书号区域的准确定位。然后将索书号区域从原图像上分割,经预处理后使用光学字符识别技术完成索书号的识别。最后,根据识别出的索书号的排序关系判别书架上是否存在错序的图书。实验结果表明,所提方法相比于其他深度网络模型具有更好的检测精度,对实现图书馆智能化的图书管理具有一定的实际应用价值。

The on-shelf books out-of-sequence detection method based on the recognition of call numbers is important for the establishment of intelligent libraries.In order to realize fast and accurate out-of-sequence detection of books on shelves,an improved YOLOv3-Tiny deep network is proposed to realize the location of the book call number area.In this method,a 52×52 scale output and two jump connection blocks are added to enhance the detection capability of network for small targets and suppress the gradient disappearance or explosion problem caused by the increase of network depth,so as to achieve accurate positioning of the book call number area.The call number area is segmented from the original image,and optical character recognition technology is used to complete the identification of call number after preprocessing.The sorting relationship of the identified call number is used to determine whether there are out-of-sequence books on shelves.The experimental results show that,in comparison with other deep networks,the proposed method has better detection accuracy,which has a certain practical application value for realizing intelligent library management.