A COGNITIVE OPTICAL ANOMALY RECOGNITION SYSTEM FOR RAILWAY INFRASTRUCTURE INTEGRITY MONITORING

Abstract

Technological evolution is observed in railway track inspection methods due to the passage from traditional image processing methods to more innovative data-driven models, owing to the inflexibility of more traditional methods upon being exposed to real-world scenarios. Though its capability of handling uncertain inputs has been improved with the inclusion of fuzzy-logic principles in it, owing to its reliance on manually coded knowledge, it has become less relevant. The current paper discusses a new complete process for railway track inspection that combines the autonomous identification power of convolutional neural networks and the logical deduction power of fuzzy logic inference. This new process combines the autonomous identification power of CNNs with the logical deduction power of fuzzy logic inference.