Intelligent Coal–Gangue Sorting Method Based on CCD and DE-XRT Image Fusion

In response to the issues of missed detection, false detection, and low accuracy often encountered in single-modal ore sorting under complex environments, this paper proposes an intelligent coal-gangue sorting method based on the fusion of industrial CCD (Charge Coupled Device) images and DE-XRT (Dual Energy X-ray Transmission) images. First, a data acquisition device was set up in the laboratory to collect CCD and DE-XRT images of coal and gangue. Adaptive weighted fusion was applied to preprocess the high- and low-energy X-ray images. Subsequently, we constructed a dual-branch backbone network capable of simultaneously processing CCD and XRT images. To achieve a balance between efficiency and performance, this network is built upon a lightweight MobileNetV4 architecture, where the width multiplier is adjusted to significantly reduce the parameters and computational complexity. Furthermore, the SimAM attention mechanism is incorporated to enhance feature representation and improve recognition accuracy. The proposed method effectively combines the advantages of CCD images (providing clear surface features) and XRT images (enabling internal component identification) and achieves real-time accurate ore sorting. Experimental results demonstrate that the image fusion approach maintains excellent performance while significantly reducing model complexity: compared to the single-modal CCD and XRT models, our fusion model reduces FLOPs by 64.9%, decreases parameters by 61.7%, and improves recognition accuracy by 7.6% and 28.3%, respectively. Finally, the model achieved 100% recognition accuracy, with no missed or false detections.