Crack segmentation and tracking method based on lightweight model

The invention discloses a crack segmentation and tracking method based on a lightweight model, which belongs to the technical field of road engineering. The method comprises the following steps that an original image is processed through employing a feature extraction network, and a feature map is o...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: LIU KUN, WANG LIANGCAI, ZHANG HANCHENG, MO HONGLIU, LIU YANFENG, HU JING, XIONG FENG, WU JIAN
Format: Patent
Sprache:chi ; eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The invention discloses a crack segmentation and tracking method based on a lightweight model, which belongs to the technical field of road engineering. The method comprises the following steps that an original image is processed through employing a feature extraction network, and a feature map is obtained; the target recognition module performs target detection on the pavement disease based on the feature map to obtain a surrounding frame of the disease; the instance segmentation module calculates the position of a feature map corresponding to a disease target based on the bounding box, and performs instance segmentation on the target through up-sampling, deconvolution and other calculations; and the visual tracking module calculates the position of the feature map corresponding to the disease target based on the bounding box and calculates the id feature of the feature map so as to carry out visual tracking in the next step. The method has important significance on model lightweight and deployment in embedd