Chapter 9: SaltSpot: A convolutional neural network approach for classifying salt contamination damage on civil infrastructure

Civil infrastructure, such as bridges, houses and buildings, is vital to societal functionality. The focus of civil engineering aims to preserve and maintain existing infrastructure to ensure optimal conditions, thereby upholding security, durability and efficient operationality. Salt contamination represents pervasive damage to civil infrastructure, directly influencing its longevity and safety while incurring substantial reparative costs. Consequently, the refinement of detection methodologies for such damage assumes paramount importance.

At present, visual inspection is the predominant form of monitoring salt damage. However, this approach demands considerable time, human and financial resources, rendering the task subjective and inconsistent. This chapter delves into the potential optimisation of salt damage detection on civil infrastructure by applying innovative methods supported by machine learning. Moreover, this research introduces the ‘SaltSpot dataset’, a cutting-edge dataset that can be used for benchmarking.

The numerical outcomes depict the efficacy of employing a sophisticated convolutional neural network for salt damage classification. This approach yields an accuracy rate of approximately 95%, overcoming the performance of traditional methods while substantially reducing the time investment. The findings underscore the transformative potential of ResNet50 architecture into salt damage detection processes, contributing to advancements in the efficiency and reliability of civil infrastructure inspection protocols.