The invention provides an
abutment tooth selection multi-
element analysis method based on improved YOLOv11. Comprising the following steps: preprocessing an original desensitized tooth panoramic x-
ray film, constructing and marking a self-made
data set, and carrying out target detection and tooth key point detection; in the target detection, a dynamic
hybrid convolutional network (Dynamic Inception Mixer) and a small object enhanced
pyramid (SOEP) are introduced into an original model, and the
feature extraction fusion capability is enhanced; a
frequency domain-space attention mechanism (FSA) and a ShapeIoU
loss function are integrated, the
small target recognition and bounding box regression precision is improved, and a DSF-YOLOv11s-detect model is generated. In key point detection, a high-frequency enhanced residual block (HFERB) and a multi-scale attention mechanism (MSGA) are introduced, so that the positioning accuracy is improved, an SIoU
loss function is adopted to enhance the sensitivity to tooth direction features, and an HMS-YOLOv11s-
pose model is generated. Experiments show that the improved model is better in recognition and positioning effect. The
abutment tooth selection multi-
element analysis method integrates two models, solves the problems of low minimal
lesion recognition rate and insufficient multi-element comprehensive analysis ability of an original model, and promotes precise and intelligent auxiliary
decision making of the dental department.