关键词: 虚拟现实　, 图像增强　, ZeroDCE　, 轻量化

Abstract: The virtual reality video images of mines often encounter problems such as low brightness,noise,and blurriness,
which seriously affect the effective operation of the subsequent intelligent recognition and safety monitoring systems in the
mines. To address the issue of image quality degradation caused by weak illumination in mines,an image enhancement model
based on the improved Zero-reference Deep Curve Estimation (ZeroDCE) is proposed. Based on the traditional ZeroDCE model,
a Depthwise Separable Convolution (DSC) method is adopted to replace the convolution layer in the traditional model. At
the same time,the reuse of curve estimation parameters,downsampling,and pruning operations are adopted to achieve lightweight
processing of the model,generating a lightweight ZeroDCE model. The results show that after the image is processed by
the lightweight ZeroDCE model,not only the brightness is improved,but more details are retained,and the original tone is not
changed;the Peak Signal to Noise Ratio (PSNR) and Visual Information Fidelity (VIF) of this model reach 26. 84 and 3. 85
respectively,which are superior to the DeepUPE,EnlightenGAN,and URetinexNet models,and also have advantages in running
time,which can meet the dual requirements of image quality and real-time processing of the mine virtual reality system to a certain
extent.