Trio-CNN: Detection of Railway Track Objects and Wheel Cracks Fastener Faults via GAN-based Super-Resolution Filters

Railways play a vital role in economic development and are an essential component of the country’s transportation system. It takes a significant amount of time to manually notice an object or flaw on the railroad track. To overcome this, a novel Trio-CNN has been proposed for detecting fastener faults, wheel cracks, and objects in railway tracks. Initially, video sequences are converted into images and pre-processed using GAN-based super-resolution filters to eliminate noise and enhance overall image quality. These enhanced images are then employed for feature extraction using deep learning models like Capsule Net, SegNet, and GoogleNet to efficiently identify objects in the railroad tracks, wheel cracks, and fastener flaws. In the proposed method, a Gannet Optimization Algorithm ensures that only the most relevant features are selected for accurate object detection. This method detects “objects on track”, “cracks on wheels”, and “faults on tracks” using Faster RCNN. The Grad-CAM approach identifies problematic regions on the railroad tracks and produces heat maps for improved interpretability to show the detected objects. Finally, a decision-making module evaluates detection results and alerts the local pilot and nearby control room, ensuring timely intervention and safety. The proposed method achieves an accuracy rate of 99.61% by using both the Rail-5k and RailSem19 datasets. The proposed Trio-CNN method achieves an overall accuracy of 1.01%, 1.69%, and 0.94% compared to the existing methods DFF-Net, FE-SSD, and ERTNet, respectively.