Shengxiang Zhang

Currently, unmanned aerial vehicles (UAV) are applied to routine inspection tasks of power transmission devices. Deep-learning algorithm and machine vision have attracted much attention in the field of the UAV’s autonomous control as it’s an effective way to improve the efficiency of inspection. Considering the differences between the distant and close view, this paper adopts Mask R-CNN to detect various components of power transmission devices but use the methods such as processing of the edge, hole filling and Hough Transform identify the wires in distant. Some major components, such as pole, truss, cross arm, insulator string and so on, can be 100% recognized. This proposed model shows the characteristics of high recognition speed and high accuracy, which can assist UAV to inspect well.

This paper demonstrate a novel methodology for safety verification of hybrid systems. The methodology uses functions of state termed barrier certificates that guarantee the safety requirement of hybrid systems, instead of computing reachable sets. The barrier certificates can be computed using the sum of squares decomposition and semidefinite programming. The unmanned aerial vehicles which used to perform a given mission and execute the flight plan are modeled by hybrid automata. Barrier certificates are constructed using the software SOSTOOLS for safety verification of flight plans of unmanned aircraft.