Dual Discriminator and Adaptive Multisource Feature Fusion WGAN-GP for Coal-Rock Properties Recognition Under Limited Infrared Thermal Images

Accurate recognition of coal-rock properties (CRPs) is an important prerequisite for ensuring the intelligent control and stable operation of shearer. The main challenges in industrial applications are potential safety hazards and single cutting mode in coal mining face, resulting in samples with insufficient number, unbalanced distribution, and background interference. However, most current CRPs recognition methods rely on sufficient and balanced samples, limiting their application efficiency and scope. To this end, this article proposes a dual discriminator and adaptive multisource feature fusion Wasserstein generative adversarial network with gradient penalty for generating high-quality infrared thermal (IRT) images to assist in CRPs recognition under limited samples. First, dual discriminator network with U-Net model is designed to obtain the global and local feature and optimize the training strategy of the generator. Then, adaptive multisource feature fusion mechanism is constructed by combining random noise and intermediate feature to guide the generator for adversarial learning and avoid violent oscillations at the initial stage of model training. Finally, conditional random field is employed to perform postprocessing operations to improve the accuracy of the sample generation boundaries. The experimental results show that the proposed method outperforms other state-of-the-art models and effectively assists in the recognition of CRPs under limited samples.