Vehicle re-identification method based on improved depth relative distance learning model
The invention discloses a vehicle re-identification method based on an improved deep relative distance learning model, and the method comprises the steps: extracting the color information and model information of a vehicle through employing the characteristics of RepNet and a coarse-grained learning...
Saved in:
Main Authors | , , , , , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
10.11.2020
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The invention discloses a vehicle re-identification method based on an improved deep relative distance learning model, and the method comprises the steps: extracting the color information and model information of a vehicle through employing the characteristics of RepNet and a coarse-grained learning channel responsible for the label attribute classification, and feeding back the extracted featuresto the subsequent fine-grained similarity learning through an inhibition layer. According to the method, coarse-grained attributive characters embedded into a fine-grained learning channel are eliminated, so that the training time is saved, more attention can be paid to extraction of complex characters during fine-grained learning, and the recognition precision is improved.
本发明公开一种基于改进深度相对距离学习模型的车辆再识别方法,利用RepNet的特性,负责标签属性分类的粗粒度学习通道将车辆的颜色信息和车辆型号信息提取出来,通过抑制层将提取的特征对后面的细粒度相似性学习进行反馈,消除掉那些嵌入到细粒度学习通道中的粗粒度属性特征,这样既节省了训练时间,又可以让细粒度学习时将更多地注意力关注在复杂特征的提取上,来提高识别精度。 |
---|---|
Bibliography: | Application Number: CN202010685999 |