Research on robustness technology of power equipment image recognition algorithm model in confrontation scenario

Abstract

In the process of controlling the traditional identification means, it is difficult to meet the current three-dimensional transmission inspection requirements. Therefore, transmission line inspection combined with artificial intelligence has gradually come into people’s vision. At present, there are some explorations in the performance, stability, robustness and use value of the model. However, the exploration of the security performance of the model in the current relevant literature often occurs in the defense and structure of different samples, and the robustness evaluation content is not reasonably standardized. Aiming at the power inspection scenario of power grid application, this paper integrates the artificial intelligence model evaluation indicators in academia and the model evaluation system in the specific scenario of industry, designs customized evaluation indicators in combination with the power scenario, focuses on the robustness indicators of power grid power inspection, and studies the countermeasure sample generation technology of power image. The generated countermeasure samples will effectively simulate the attacker’s intention, and propose effective countermeasure algorithm measurement indicators; For the evaluation of the robustness against various hacker and organizational attacks, an artificial intelligence model robustness evaluation system is formed, which can effectively support the high-quality development of big data artificial intelligence platform and promote the safe, effective, reliable and stable operation of artificial intelligence models in power scenarios.