Understanding the impact mechanisms of lighting conditions on robust coal and gangue image recognition

Lighting conditions, as a critical measurement factor affecting image quality and feature clarity, play a decisive role in coal and gangue image recognition. However, in real industrial scenarios, complex and dynamic lighting variations pose challenges to reliable recognition. To reveal the underlying mechanism by which lighting conditions affect recognition, this paper develops an experimental image acquisition platform and constructed datasets under seven controlled lighting conditions. Grayscale and texture features are extracted and analyzed. Tree models and neural network models are trained to assess how lighting conditions influence image processing performance. Experimental results show that a lighting intensity of 500 Lux provides the highest feature separability and optimizes classification performance. The comparative evaluation between convolutional neural networks-based frameworks and tree models under varying lighting conditions shows that the former maintain consistently higher accuracy and robustness, indicating their stronger capacity to extract illumination-invariant features. This study provides a valuable reference for designing instrumentation and measurement that enhance the accuracy of intelligent sensing and monitoring systems.