基于图像增强和改进YOLOv8的煤矿低光照目标检测

打开文本图片集

中图分类号：TD67 文献标志码：A

Abstract： At present， the existing image enhancement techniques for underground coal mines suffer from insufficient stabilityand large fluctuations in the quality of generated images，which afect theaccuracyof subsequent target detection.Meanwhile，target detection methods based on YOLOv8 also face certain limitations in low-light environments due to weakened image features and information loss.To address these problems，a low-light target detection algorithm for coal mines based on image enhancement and improved YOLOv8 was proposed. The Denoising Diffusion Probabilistic Model （DDPM） was used to denoise and enhance the original images，restoring ilumination and detail information. Based on YOLOv8n， improvements were made by introducing a Low-Frequency Filter Enhancement Module （LEF） and a Feature Enhancement Module （FEM） to enhance feature extraction performance for low-light images. The original CIoU regression loss function in YOLOv8n was replaced with MPDIoU，yielding the YOLOv8-DLFM model.The YOLOv8-DLFM was then used for target detection to improve accuracy and robustness. Experimental results showed that： ① compared with mainstream image enhancement methods， DDPM achieved a peak signal-to-noise ratio of 28.379dB ， a structural similarity index of 0.886，and a perceptual similarity of O.104，demonstrating superior image reconstruction quality and structural similarity. ② YOLOv8-DLFM exhibited excellent overall performance， with precision， recall， and mAP @0.5 reaching 0.878， 0.791，and 0.896， respectively， and a frame rate of 88.6 frames/s. Compared with the original YOLOv8n model， the precision， recall， and mAP@0.5 ofYOLOv8-DLFM increased by 8.13% 0 6.6% ，and 8.74% ，respectively. ③ Compared with mainstream target detection models， YOLOv8-DLFM demonstrated stronger robustness and higher detection accuracy under complex low-light environments.It also exhibited high robustnessand adaptability under typical conditions such as target occusion， lighting interference， sparse targets， and dense targets.

Key words： underground target detection; low light; image enhancement; YOLOv8n; Denoising Diffusion Probabilistic Model; low-frequency filtering; feature enhancement

0引言

作为智能化建设的关键组成部分，煤矿井下智能监控系统能够实时监测环境变化，迅速识别潜在的安全隐患，从而有效降低事故发生的风险[1-2]。（剩余18156字）