A self-adaptive exhaustive search optimization-based method for restoration of bridge defects images

Existing bridges are aging and deteriorating. Furthermore, large number of bridges exist in transportation networks meanwhile maintenance budgets are being squeezed. This state of affairs necessities the development of automatic bridge defects evaluation model using computer vision technologies to overcome the limitations of visual inspection. The digital images are prone to degradation by noises during the image acquisition phase. The absence of efficient bridge defects image restoration method results in inaccurate condition assessment models and unreliable bridge management systems. The present study introduces a self-adaptive two-tier method for detection of noises and restoration of bridge defects images. The first model adopts Elman neural network coupled with invasive weed optimization algorithm to identify the type of noise that corrupts images. In the second model, moth-flame optimization algorithm is utilized to design a hybrid image filtering protocol that involves an integration of spatial domain and frequency domain filters. The proposed detection model was assessed through comparisons with other machine learning models as per split validation and tenfold cross validation. It attained the highest classification accuracies, whereas the accuracy, sensitivity, specificity, precision, F-measure and Kappa coefficient are 95.28%, 95.24%, 98.07%, 95.25%, 95.34%. 95.43% and 0.935, respectively in the separate noise recognition module. The capabilities of the proposed restoration model were evaluated against some well-known good-performing optimization algorithms in addition to some conventional restoration models. Moth-flame optimization algorithm outperformed other restoration models, whereas peak signal to noise ratio, mean-squared error, normalized absolute error and image enhancement factor are 25.359, 176.319, 0.0585 and 7.182, respectively.