孙林,陈圣,姚旭龙,等. 矿井智能监控目标识别的图像增强方法与应用[J]. 煤炭学报,2024,49(S1):495−504. DOI: 10.13225/j.cnki.jccs.2023.0489
引用本文: 孙林,陈圣,姚旭龙,等. 矿井智能监控目标识别的图像增强方法与应用[J]. 煤炭学报,2024,49(S1):495−504. DOI: 10.13225/j.cnki.jccs.2023.0489
SUN Lin,CHEN Sheng,YAO Xulong,et al. Image enhancement methods and applications for target recognition in intelligent mine monitoring[J]. Journal of China Coal Society,2024,49(S1):495−504. DOI: 10.13225/j.cnki.jccs.2023.0489
Citation: SUN Lin,CHEN Sheng,YAO Xulong,et al. Image enhancement methods and applications for target recognition in intelligent mine monitoring[J]. Journal of China Coal Society,2024,49(S1):495−504. DOI: 10.13225/j.cnki.jccs.2023.0489

矿井智能监控目标识别的图像增强方法与应用

Image enhancement methods and applications for target recognition in intelligent mine monitoring

  • 摘要: 煤矿井下安全违章智能识别技术成为安全管理与信息获取的主要手段,但由于井下空间环境受低照度、点光源、高粉尘等因素的影响,极大地降低了智能识别的准确率。基于此,提出了一种多权重融合的图像增强方法,实现了图像亮度加强和照度均衡的融合增强。利用Gamma算法实现井下监控图像的亮度增强,在亮度增强的基础上,进行HSV空间变换,保持色调分量和饱和度分量不变,利用多尺度高斯函数提取光照分量,再利用改进的二维伽马函数对光照分量过强和过弱的区域进行调整,实现照度均衡。结合拉普拉斯对比度、亮度、饱和度3个权重并通过高斯和拉普拉斯金字塔融合亮度增强与照度均衡的图像,实现矿井智能监控图像增强。通过矿井监控中的安全帽智能识别实验,对提出的图像增强方法和多尺度视网膜算法(MSR)、具色彩保护的多尺度Retinex算法(MSRCP)、带色彩修复的多尺度Retinex算法(MSRCR)、色彩增益加权的AutoMSRCR算法(AMSRCR)利用标准差、峰值信噪比、信息熵等图像评价指标进行评判。峰值信噪比相较于上述算法平均提高了32.44%;标准差相较于原图平均提高了115.38%,相较于上述算法平均提高了47.30%。安全帽识别准确率达到了86.7%,相较于上述算法平均提高了47.52%。实验验证结果表明,所构建的图像增强方法可有效提高矿井图像的亮度、清晰度、对比度,减少色彩失真、光晕等现象,并明显提高了矿井智能监控中目标识别的准确率,可为煤矿井下安全违章智能识别奠定坚实基础。

     

    Abstract: Intelligent recognition technology for underground safety violations in coal mines has become the primary means of safety management and information acquisition. However, the accuracy of intelligent recognition is reduced by factors such as low illumination, point light sources, and high dust in the underground spatial environment. To overcome those problems, this paper proposes a multi-weight fusion image enhancement method, which achieves the fusion enhancement of image brightness and illumination balance. The monitoring image brightness is increased by using the Gamma algorithm. Based on the brightness enhancement, the HSV spatial transformation is carried out to keep the hue component and saturation component unchanged, the illumination component is extracted by using a multi-scale Gaussian function, and then the regions where the illumination component is too strong and too weak are adjusted by using an improved two-dimensional gamma function to achieve an illumination equalization. Combining the three weights of Laplacian contrast, luminance, and saturation and fusing the luminance-enhanced and illuminance-balanced images by Gaussian and Laplacian pyramids, the intelligent surveillance image enhancement is achieved. Through the experimental verification of the intelligent recognition of helmets in mine monitoring, the image enhancement method proposed in this paper and the MSR, MSRCP, MSRCR, and AMSRCR algorithms are evaluated in terms of image evaluation indexes such as standard deviation, peak signal-to-noise ratio, and information entropy, and the peak signal-to-noise ratio is improved by 32.44% on average compared with other algorithms, the standard deviation is improved by 115.38% on average compared with the original image, and the average improvement in standard deviation is 115.38% compared with the original image and 47.30% compared with other algorithms, and the accuracy of helmet recognition reaches 86.7%, an average improvement of 47.52% compared with other algorithms. The results show that the algorithm in this paper can effectively improve the contrast and clarity of mine images, reduce the halo phenomenon, and significantly improve the accuracy of target recognition in intelligent mine monitoring. It can lay a solid foundation for the intelligent identification of safety violations in underground coal mines.

     

/

返回文章
返回
Baidu
map