摘要
实现矿区遥感图像增强处理,有助于提升后续图像判别以及相关监测分析效率。以往矿区遥感图像增强一般采用滤波、灰度变换等方法,往往会导致图像大量细节信息丢失,在很大程度上影响了后续判读分析。近年来,深度学习方法逐步应用于图像增强处理,但该方法很大程度上依赖于模型设计和参数合理取值,需要进行大量的试验和优化方可取得理想效果。将深度学习方法(Deep Learning,DL)与离散小波变换(Discrete Wavelet Transform,DWT)相结合,提出了一种基于离散小波域深度残差学习的矿区遥感图像增强算法。首先将图像进行单级二维离散小波变换,得到4个子带;然后将4个子带系数输入深度学习残差网络,预测相应的残差图像增加4个子带图像和残差图像作为二维小波变换的新子带;最后通过二维离散小波逆变换得到增强图像。试验结果表明:所提算法相对于直方图均衡化和超分辨率重建等方法而言,无论在图像视觉效果以及峰值信噪比、结构相似性、均方误差等评价指标上都具有较好优势,反映出将离散小波变换与深度学习方法相结合,有助于提升矿区遥感图像视觉效果,方便后续图像解译判读工作。
Enhancing remote sensing images of mining areas can significantly improve subsequent image interpretation and monitoring analysis efficiency.Traditional methods for enhancing remote sensing images in mining areas often involve filte-ring,grayscale transformations,etc.,which can lead to significant loss of detail in the image,greatly affecting image interpreta-tion.In recent years,deep learning methods have gradually been applied to image enhancement processing.However,this meth-od heavily relies on model design and parameter tuning,requiring a large number of experiments and optimizations to achieve desirable results.Combining deep learning(DL)with discrete wavelet transform(DWT),a mining area remote sensing image enhancement algorithm based on deep residual learning in the discrete wavelet domain is proposed.Firstly,the image is subjec-ted to single-level 2D discrete wavelet transform to obtain 4 subbands.Then,the coefficients of the 4 subbands are input into a deep residual network to predict corresponding residual images.These residual images are added to the original 4 subband ima-ges to create new subbands for the 2D wavelet transform.Finally,the enhanced image is obtained through 2D inverse discrete wavelet transform.The test results show that:compared with methods such as histogram equalization wavelet transform and su-per-resolution reconstruction convolutional neural network,the proposed algorithm has a good advantage in terms of image visu-al effect,peak signal-to-noise ratio,structural similarity,mean square error and other evaluation indicators,reflecting that the combination of discrete wavelet transform and deep learning is helpful to improve the visual effect of remote sensing images in mining areas and facilitate subsequent image interpretation and interpretation.
作者
李亦珂
王春梅
LI Yike;WANG Chunmei(Department of Information Engineering,Shanxi Institute of Mechanical and Electrical Engineering,Changzhi 046011,China;School of Information Engineering,Shandong Universtity of Aeronautics,Binzhou 256600,China)
出处
《金属矿山》
CAS
北大核心
2024年第4期215-220,共6页
Metal Mine
基金
山西省自然科学(青年科技研究)基金项目(编号:2021021022-5)
山西机电职业技术学院院级课题(编号:JWC-L20017)。
关键词
矿区遥感图像
离散小波变换
深度学习
图像增强
remote sensing image of mining area
discrete wavelet transform
deep learning
image enhancement
