摘要
露天矿低质图像增强是无人驾驶宽体车感知系统的重要环节,单目相机获取的图像易受到矿区粉尘、雨雪雾、剧烈震动等多种因素影响。针对传统图像增强算法在处理露天矿图像时存在噪声大、图像颜色失真等问题,提出了改进Retinex-Net算法对露天煤矿图像进行增强。使用循环对抗生成网络和双通道残差网络来改进增强和去噪部分。循环对抗生成网络通过学习低光照图像和正常光照图像之间的映射关系,生成更自然和真实的增强结果。双通道残差网络通过同时处理亮度和色度信息,有效去除低光照图像中的噪声和伪影。试验结果表明:该方法在客观和主观评价指标上均优于现有方法。所提改进Retinex-Net算法为解决露天矿图像质量问题提供了一种有效方案。
Low-quality image enhancement is an important part of the perception system of unmanned wide-body vehicle in open-pit mine.The image acquired by monocular camera is susceptible to many factors such as dust,rain,snow and fog in mining area and violent vibration.Aiming at the problems of high noise and color distortion in the traditional image enhance-ment algorithm,an improved Retinex-Net algorithm is proposed to enhance the image of open-pit mine.Cyclic adversarial gen-eration network and two-channel residual network are used to improve the enhancement and denoising parts.The cyclic adver-sarial generation network generates more natural and realistic enhanced results by learning the mapping relationship between low-light and normal-light images.The dual-channel residual network can effectively remove noise and artifacts in low-light im-ages by processing brightness and chrominance information simultaneously.The experimental results show that the proposed method is superior to the existing methods in both objective and subjective evaluation indexes.The proposed Retinex-Net algo-rithm provides an effective scheme to solve the image quality problem in open-pit mine.
作者
孟保威
陈曦
MENG Baowei;CHEN Xi(Dananhu No.2 Mine,State Grid Energy Hami Coal and Electricity Co.,Ltd.,Hami 839000,China)
出处
《金属矿山》
CAS
北大核心
2024年第3期221-228,共8页
Metal Mine
基金
国家自然科学基金项目(编号:51804249)。
