Non-contact Identification Method for Carbon Steel Corrosion Grade of Transmission Tower Based on Hyperspectral Technology

Abstract

The corrosion status of transmission line towers is difficult to detect. Once corrosion damage occurs, it will not only cause equipment and facilities to fail prematurely, be scrapped, and shorten their lifespan, but also cause significant economic losses, and even cause serious personal injuries. Traditional detection methods such as weightlessness method need to destroy the material structure, which is cumbersome to operate on site. This paper proposes a non-contact identification method based on hyperspectral technology for carbon steel corrosion grade of transmission towers. Collect hyperspectral images of carbon steel samples of different corrosion grades, use the pre-processed full-band spectral data to establish the K-nearest neighbor algorithm (KNN) model and the partial least squares discriminant analysis (PLS-DA) model. It is found through comparison that PLS-DA model classification effect is better. Through competitive adaptive reweighted sampling algorithm (CARS) and principal component analysis (PCA), the full-band spectral data of different corrosion grade carbon steel samples were extracted, and a PLS-DA model based on the optimal band was established. The results show that the use of characteristic waveband modeling greatly reduces the interference of redundant information, and the classification accuracy is better than that of the full waveband. The PLS-DA model based on the characteristic waveband has an accuracy of 95% for the classification of different corrosion levels in the verification set. Therefore, this method can be applied to the non-destructive and rapid detection of the corrosion level of carbon steel, and provides a new idea for the identification of the corrosion level of carbon steel in transmission towers.