Slender Flexible Object Segmentation Based on Object Correlation Module and Loss Function Optimization

Slender flexible objects are ubiquitous in real-world circumstances. The existing object detection and segmentation algorithms have achieved high accuracy speed rigid objects, but the effect of slender is not ideal. Extreme aspect ratios dynamically changeable geometric appearance characterize to extent that it difficult locate them instance segmentation. In this paper, a new method based on correlation module loss function optimization proposed for objects. order achieve more accurate anchor box positioning, GIoU bounding-box regression selected overcome problem inconsistency between training objectives assessment indicators. Furthermore, due Mask Scoring R-CNN network ignoring relationships end-to-end learning modeling features all image improve accuracy. results experiments self-built dataset power grid operation sites demonstrate presented research can efficiently recognize segment with 44.8%. ablation experiment also shows addition revised both effective enhance by 1.2% 0.5%, respectively. considers characteristics improves increase