A pole tilt detection and risk assessment method based on semantic segmentation and evidence theory

Abstract

Now that UAVs （unmanned aerial vehicles） have been widely used for power inspection， the UAV image analysis method based on computer vision technology can provide auxiliary means for pole tilt detection and risk assessment. This paper proposes a method of pole tilt detection and risk assessment based on semantic segmentation and evidence theory. The pole image segmentation method based on U-Net and connected component labeling algorithm is used to classify the images pixel by pixel according to semantic categories， and non-pole regions are eliminated by using the connected component labeling algorithm. By the construction of pole tilt criterion， a method for pole tilt detection based on evidence theory is proposed to fuse the conclusions of multiple criteria by using synthetic rules. The tilt risk assessment method is given， and the tilt level is taken as the assessment result. The experimental results show that the proposed method can improve detection accuracy.

Keywords

• semantic segmentation
• evidence theory
• morphological characteristics
• pole tilt
• risk assessment