Application and research of intelligent inspection robots in hydropower stations

摘要

A machine vision based inspection robot system is designed in this paper to address the issues of high cost and difficulty in ensuring detection accuracy in traditional manual pumped storage station crack and seepage detection. Construct a convolutional neural network that integrates cross entropy and Dice cost function, and establish an evaluation function based on total pixel accuracy, intersection to union ratio, and F1 score to ensure accurate detection of common cracks. In order to verify the effectiveness of the robot inspection system, the convolutional neural network was tested and compared with common computer vision and manual detection methods in terms of performance. The comparison results show that the neural network constructed in the article has made significant progress in both detection accuracy and detection efficiency.