Page 88 - Read Online

Page 394 Wang et al. Intell Robot 2022;2(4):391-406 https://dx.doi.org/10.20517/ir.2022.25

Figure 1. Research framework.

The use of intelligent information processing methods to remove background noise, capture multiscale and
diverse representation information, and extract its related sensitive features is the basis for detecting
concrete bridge cracks.

In the aspect of feature extraction for concrete bridge cracks, the traditional classical methods include
principal component analysis (PCA), sparse decomposition, wavelet transform (WT), and adaptive neural
fuzzy inference system (ANFIS).

PCA obtains different principal components via the matrix transformation of signals, arranges the sizes of
the principal components according to the variance, and compares the contribution rate of each component
with the threshold value, thus realizing effective feature extraction [29,30] . In the research on concrete structure
crack recognition based on multiple image features, Yu et al. collected 1200 bridge crack images and
employed integral projection and PCA to extract effective features sensitive to the crack for crack edge
[31]
detection . Mei et al. collected acceleration data from all the vehicles within a certain period and extracted
the transformed features that are related to bridge damage with Mel-frequency cepstral coefficients and
PCA to identify the damage by comparing the distributions of these transformed features . PCA is a widely
[32]
utilized method to remove noise and extract effective features, but it cannot accurately analyze the real
subspace structure of data.