Estimation of Deterioration Levels of Transmission Towers via Deep Learning Maximizing Canonical Correlation Between Heterogeneous Features

Abstract

This paper presents estimation of deterioration levels of transmission towers via deep learning maximizing the canonical correlation between heterogeneous features. In the proposed method, we newly construct a correlation-maximizing deep extreme learning machine based on a local receptive field (CMDELM-LRF). For accurate deterioration level estimation, it is necessary to obtain semantic information that effectively represents deterioration levels. However, since the amount of training data for transmission towers is small, it is difficult to perform feature transformation by using many hidden layers such as general deep learning methods. In CMDELM-LRF, one hidden layer, which maximizes the canonical correlation between visual features and text features obtained from inspection text data, is newly inserted. Specifically, by using projections obtained by maximizing the canonical correlation as weight parameters of the hidden layer, feature transformation for extracting semantic information is realized without designing many hidden layers. This is the main contribution of this paper. Consequently, CMDELM-LRF realizes accurate deterioration level estimation from a small amount of training data. Index Terms-Deterioration level estimation, deep extreme learning machine, canonical correlation analysis. I. Introduction A LL countries have critical infrastructures such as power grids, railways, tunnels, bridges and transmission towers. In order to maintain these infrastructures, visual inspection has usually been performed by inspectors. Visual inspection is a labor-intensive and time-consuming process [1]. In order to reduce costs, new techniques for supporting maintenance inspection are required [2]-[6]. Some methods have been proposed for supporting various inspection tasks for infrastructures including distress classification [7], [8], detection of specific distresses [9]-[17], analysis of surface status [18]-[20] and deterioration level estimation [21], [22]. Support for maintenance inspection of transmission towers is important since inspection is performed by inspectors ascending towers, which is a dangerous inspection task [23].

Distress Image Retrieval for Infrastructure Maintenance via Self-Trained Deep Metric Learning Using Experts’ Knowledge

“…Consequently, records often include images that are unrelated to their deterioration levels, and the annotations of deterioration levels for records are coarse and inexact [8], [9]. However, conventional methods for distress type classification [10], [11], deterioration level estimation [12] and similar image retrieval [6], [7] assume a one-to-one correspondence between images and their deterioration levels. That is, these methods cannot consider whether an input image is suitable for deterioration level determination.…”

“…In order to determine whether repairs should be conducted in the actual inspection of distresses, not only detection and classification of the types of distress but also determination of the deterioration level is necessary, and several methods for estimating the deterioration level have been proposed. For example, the literature [12] proposed deterioration level estimation using canonical correlation between heterogeneous features, and the literature [31] proposed automatic evaluation of degrees of corrosion deterioration of a transmission tower.…”

Music Video Recommendation Based on Link Prediction Considering Local and Global Structures of a Network

“…To verify whether link prediction scores can represent the degrees to which users desire music videos, we calculated Pearson's correlation coefficients between link prediction scores and the ground truth. The evaluation scheme is motivated by the papers [87], [88], in which the discriminant power of features was evaluated by using Pearson's correlation coefficients between the features and labels. Specifically, we calculated the correlation coefficients between link prediction scores of all music videos and labels that indicate 1 if each music video is the ground truth and 0 otherwise.…”

Correlation-Aware Attention Branch Network Using Multi-Modal Data For Deterioration Level Estimation Of Infrastructures

“…It should be noted that engineers often record text data that indicate useful information such as the distress regions and their materials during the actual inspection [13]. It has been reported that canonical features obtained by calculating canonical correlations between heterogeneous features from distress images and their text-data have better discriminative ability than the original features [14,15]. That is, these heterogeneous features are semantically related.…”