Fast Sort YoloV11 Object Detection approach towards detection and classification of Foreign Objects in Railway lines and Tracks with Multiattention mechanism

Abstract

Due to rapid increase in usage of railway transformation across the world due to its multiple benefits such as cost effectiveness, reliability and environment friendliness for long distance transportation of passengers and goods. Thus it becomes extremely significant to detect the foreign objects intrusion on the high speed railway lines and its track beds to safeguarding train operations. Traditionally many Object detection methods have been employed to detect the foreign objects in railway lines to enhance safety, operational efficiency, and the overall reliability of rail transportation. Despite of several advantages, those architectures fails to address the following challenges such as feature extraction and aggregation of the object detection and recognition approaches increase the complexity of the model and it leads to poor efficiency in detecting the variation in the object appearance due to viewpoint, deformation , occlusion and lightening conditions. In this paper, new fast sort Yolov11 object detection architecture is designed to enhance the detection accuracy and reduce complexity of the model on incorporation of attention mechanism to detect the foreign objects in railway track and railway lines particularly small and occluded object effectively. In this work, RailFOD23 dataset is extracted from Figshare repository for training and testing the model which gathered data using LiDAR, radar, thermal imaging, and sensor networks. Proposed architecture is composed of multiple components to perform object detection. Initially extracted data is applied to backbone component which contains DenseNet121 architecture is to segment and extract the features of the image through convolution layer and fully connected layer of the model. Extracted features are employed to Neck component which performs feature aggregation for better feature presentations. Finally aggregated features is processed in the head component which detect and classify the foreign object in both railway line and railway track with high detection accurately and efficiency. Experimental analysis of the proposed model is performed and it is identified to efficient to process diverse objects such as trains, maintenance equipment, trespassers, or obstructions. Further performance analysis of the model is performed on basis of detection accuracy and detection efficiency, proposed model produces 97.2% accuracy and it is proved to be highly efficient while compared to existing object detection architectures.