摘要
准确测试出车牌位置是车牌识别系统(LPRS)的关键步骤,为了提高车牌检测的性能,本文提出多级融合的倒置金字塔注意力U-net (AUMFIF)。先将原始图像送入改进网络,其次通过原始图像的卷积运算生成多个级别的特征图,通过多层拼接和卷积来检测更多含有丰富的空间信息的特征图。然后以分层的形式,将这些特征图与倒置特征金字塔网络逐层进行融合连接,所得到的特征图不仅有丰富的空间信息还具有饱满的语义信息。最后,将注意力机制用于保留重要区域的信息,并且抑制无关背景区域,便可得出车牌位置的分割图像。为了验证所提出方法的有效性,本文在AOLP车牌数据集上进行了一系列实验。实验结果表明,该方法能够有效地提高车牌位置检测的性能。
Accurate detection of license plate position is a key step of License Plate Recognition System (LPRS). In order to improve the performance of license plate detection, this paper proposed multi-fusion inverted pyramid feature attention U-net (AUMFIF). Firstly, the original image is sent to the im-proved network, and then the convolution of the original image is used to generate feature maps at multiple levels. More feature maps with rich spatial information can be detected through mul-ti-layer splicing and convolution. Then, in a hierarchical form, these feature maps are fused and connected with the inverted feature pyramid network layer by layer. The resulting feature maps not only have rich spatial information, but also have full semantic information. Finally, the attention mechanism is used to retain the information of important regions, and suppress irrelevant back-ground regions, then the segmented image of license plate position can be obtained. In order to ver-ify the effectiveness of the proposed method, a series of experiments are carried out on AOLP da-taset. The experimental results show that this method can effectively improve the performance of license plate position detection.
出处
《应用数学进展》
2023年第2期494-504,共11页
Advances in Applied Mathematics
