In order to solve the problems of low illumination and poor image quality in video monitoring systems caused by complex lighting conditions in underground mines, a low-light image enhancement method based on knowledge distillation is proposed. By establishing Net-S (student network) and Net-T (teacher network) for feature simulation, this method improves the image reconstruction ability of the model while transferring knowledge and improves the network efficiency and image enhancement effect at the same time. The improved RIR (residual in residual blocks) module is added to improve the feature extraction ability of the model. At the same time, the loss function for low illumination image characteristics is designed to improve the accuracy of the network model in image reconstruction. Compared with the traditional knowledge distillation network model, Net-T and Net-S of the model output different images respectively. Net-T extracts features from high-quality images. Net-S simulates features under the guidance of Net-T, and finally obtains enhanced images. The experimental results show that the model can effectively remove part of the noise, improve the brightness and quality of the image, and improve the visual effect of the image. Moreover, the model has good generalization ability. Compared with CLAHE and LIME, the absolute cumulative change of the normalized objective index of this method is increased by 42.21%, 7.47% and 30.18% respectively. In view of the low illumination image enhancement of the mine, this method has certain advantages, which can significantly improve the overall brightness of the image, reduce the noise, and is more in line with the human visual perception. To a certain extent, it meets the needs of the construction of intelligent mine video monitoring system.

Caoshuai Kang. Low-light images enhancement method in coal mine based on knowledge distillation. Academic Journal of Engineering and Technology Science (2021) Vol. 4, Issue 2: 52-56. https://doi.org/10.25236/AJETS.2021.040208.

[1] YAO H, WANG H, LI Y, et al. Three-dimensional spatial and temporal distributions of dust in roadway tunneling [J]. International Journal of Coal Science & Technology, 2020, 7(1): 88-96. 

[2] PIZER S M, AMBURN E P, AUSTIN J D, et al. Adaptive histogram equalization and its variations [J]. Computer Vision, Graphics, and Image Processing, 1987, 39(3): 355-68.

[3] Suo J, Zhu S C, Shan S, et al. A compositional and dynamic model for face aging[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2009, 32(3): 385-401.

[4] Zeng J, Ma X, Zhou K. CAAE++: Improved CAAE for age progression/regression[J]. IEEE Access, 2018, 6: 66715-66722.

[5] Lample G, Zeghidour N, Usunier N, et al. Fader networks: Manipulating images by sliding attributes[A]. Advances in Neural Information Processing Systems[C]. 2017: 5967-5976.

[6] XUAN D, GUAN W, YI P, et al. Fast efficient algorithm for enhancement of low lighting video; proceedings of the 2011 IEEE International Conference on Multimedia and Expo, F 11-15 July 2011, 2011 [C].

[7] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. The journal of machine learning research, 2014, 15(1): 1929-1958.

[8] BREIMAN LEO. Two-Eyed Algorithms and Problems [M]. 2003: 9.

[9] ISOLA PHILLIP, ZHU JUN-YAN, ZHOU TINGHUI, et al. Image-to-Image Translation with Conditional Adversarial Networks [M]. 2017: 5967-76.