针对无人机拍摄的巡检低质量图像存在的光照异常、杆塔主体特征不明显等问题,提出一种基于小波变换和改进Retinex-Net相结合的图像增强方法。首先,利用小波变换将低质量图像分解为低高频图像,利用改进Retinex-Net网络处理低频图像,引入ASPP模块和SE模块强化网络特征提取能力,利用跳跃链接结构和最近邻插值法进行特征图缩放以减少背景噪声干扰；使用限制对比度自适应直方图均衡算法(CLAHE)增强高频图像,降低高频噪声干扰。最后,利用小波重构得到增强后的低质量图像。在自建数据集上的实验结果表明,相较于HE、MSRCR等算法,本文算法可以增强图像细节边缘特征、还原图像颜色、提高分辨率,帮助输电线路运维人员监测,提高分析工作的准确性。

Aiming at the problems of abnormal illumination and unobvious main features of the tower in the low-quality inspection image taken by UAV, an image enhancement method based on wavelet transform and improved Retinex-Net is proposed. Firstly, the low-quality image is decomposed into low-frequency images by wavelet transform, and the low-frequency images are processed by the improved Retinex-Net network. ASPP module and SE module are introduced to enhance the network feature extraction ability, and the feature map is scaled by using jump link structure and nearest neighbor interpolation method to reduce background noise interference. The contrast-limited adaptive histogram equalization algorithm (CLAHE) is used to enhance high-frequency images and reduce high-frequency noise interference. Finally, the enhanced low-quality image is obtained by wavelet reconstruction. Experimental results on self-built data sets show that, compared with HE and MSRCR algorithms, the proposed algorithm can enhance the edge features of image details, restore the image color, improve the resolution, help transmission line operators to monitor and improve the accuracy of analysis.

国家重点研发计划子课题，项目编号：2021YFF0603702；金华八达集团有限公司科技项目，项目编号：BD2022JH-KXXM007