Abstract
This research introduces a novel approach utilizing the Frilled Lizard Optimization (FLO) algorithm to enhance the hyperparameters of the CatBoost model for rail transit safety. By integrating the Figma platform for collaborative design and the FLO-CatBoost model for high-precision detection, the study addresses the critical need for automated foreign object detection on rail tracks. Using a dataset of 6,000 images, the FLO-CatBoost model achieved a Root Mean Square Error (RMSE) of 0.274, significantly outperforming traditional optimization algorithms and providing a robust framework for intelligent rail transit system design.
Introduction
As rail transit speeds continue to increase globally, the reliance on human operators for track monitoring has become a significant safety bottleneck. Detecting foreign objects—such as pedestrians, dropped tools, or debris—requires real-time, high-precision automated systems. However, developing these systems involves two primary challenges: the lack of efficient, collaborative design environments for safety interfaces and the difficulty of tuning complex machine learning models like CatBoost to avoid local optima in high-stakes environments.
Collaborative Design with Figma
To address design bottlenecks, the researchers utilized the Figma platform to facilitate a modern UI/UX workflow for the autonomous detection system. Figma’s cloud-based infrastructure allows for real-time collaboration between engineers and designers, enabling rapid prototyping and versioning of the system's logic and interface.
By moving from low-fidelity sketches to high-fidelity interactive prototypes within Figma, the team could simulate the system's feedback loops before deployment. This integration ensures that the "design effect" is quantitatively scored and optimized alongside the underlying algorithmic performance.
System Architecture and Intelligent Processing
The proposed detection system operates through a multi-layered architecture designed to move from raw data acquisition to actionable safety alerts. This structure ensures that the high-speed data from rail sensors is processed efficiently.
The architecture is divided into three primary layers:
- Image Acquisition Layer: Handles data receiving and preprocessing of visual inputs from the track.
- Intelligent Processing Layer: Executes track direction detection and orbital foreign object detection using the optimized machine learning model.
- Decision-Making Control Layer: Triggers necessary safety protocols, such as voice and light alarms, based on the detection results.
The FLO-CatBoost Optimization Model
The core of the intelligent processing layer is the CatBoost model, chosen for its superior handling of categorical data. To overcome the sensitivity of CatBoost to hyperparameter settings, the researchers implemented the Frilled Lizard Optimization (FLO) algorithm. FLO mimics the hunting and tree-climbing behaviors of the frilled lizard through two distinct phases:
- Exploration (Hunting): A global search phase where data points explore the mathematical space to identify promising parameter regions.
- Exploitation (Tree-Climbing): A local search phase that fine-tunes parameters for maximum precision.
This hybrid FLO-CatBoost model specifically optimizes the number of iterations, learning rate, tree depth, and L2 regularization, ensuring the model reaches a global optimum rather than getting stuck in local minima.
Experimental Results and Performance Analysis
The model was evaluated using a comprehensive dataset of 6,000 images categorized into seven classes, including straight tracks, pedestrians, and various hazards like wrenches. The FLO-CatBoost model demonstrated exceptional predictive consistency and accuracy:
- Accuracy: The model achieved an RMSE of 0.274, a drastic improvement over the baseline CatBoost (1.452) and other meta-heuristic optimizers like the Marine Predators Algorithm (MPA).
- Reliability: The Mean Absolute Percentage Error (MAPE) was recorded at 0.1218, with confusion matrices showing over 90% detection correctness across all hazard categories.
- Efficiency: The evaluation time was optimized to 0.091 seconds, meeting the requirements for real-time rail transit applications.
Conclusion and Future Directions
The integration of FLO-CatBoost within a Figma-based design framework represents a significant advancement in rail transit safety. While the current model excels with static images, future research will focus on extending the algorithm to handle high-speed video data and diverse weather conditions. By closing the loop between collaborative design and bio-inspired algorithmic optimization, this approach provides a scalable path for the next generation of intelligent transit systems.
Summary updated