无线电工程

水下建筑物结构老化等缺陷使水电站运行存在巨大的安全风险。水下建筑物图像作为运维人员了解水下情况的主要来源,在水下缺陷检测中发挥着重要作用。针对水电站水下建筑物图像模糊、颜色失真导致的缺陷检测困难等问题,提出一种基于改进直方图校正和模糊估计(Improve Histogram Correction and Fuzzy Estimation, IHCFE)的多层次水下图像增强算法。该算法利用粒子群算法(Particle Swarm Optimization, PSO)自适应确定裁剪因子,并得到自适应对比度增强图像;进而利用IBLA算法解决图像颜色失真问题,得到颜色还原后图像;再以拉普拉斯金字塔对上述图像逐层分解融合,得到多尺度融合图像。将原图、限制对比度自适应直方图均衡化(Contrast Limit Adaptive Histogram Equalization, CLAHE)、综合色彩模型(Integrated Color Model, ICM)、水下图像增强卷积神经网络模型(Convolutional Neural Network Model for Underwater Image Enhancement, UWCNN)和所提算法增强后图像输入Fast RCNN中验证目标检测性能,所提算法拥有最高的目标检测准确率,说明了增强算法的有效性和优越性。

Defects such as structural deterioration of underwater buildings pose significant safety risks to the operation of hydropower plants. As the main source for operation and maintenance personnel to understand the underwater situation, underwater building images play an important role in the detection of underwater defects. To tackle the problems of defect detection difficulties caused by blurring and colour distortion of underwater building images of hydropower stations, a multilevel underwater image enhancement algorithm based on Improve Histogram Correction and Fuzzy Estimation( IHCFE) is proposed. The Particle Swarm Optimization( PSO) algorithm is used to adaptively determine the cropping factor to obtain an adaptive contrast-enhanced image.Furthermore, the IBLA algorithm is employed to address the problem of color distortion in the image, subsequently yielding an image with corrected colors; Then, the Laplacian pyramid is used to decompose and fuse the above images layer by layer to obtain a multiscale fused image; The original image, Contrast Limit Adaptive Histogram Equalization( CLAHE), Integrated Color Model( ICM),Convolutional Neural Network Model for Underwater Image Enhancement( UWCNN) and the enhanced image of the proposed algorithm are input into Fast RCNN to verify the object detection performance. The proposed algorithm has achieved the highest object detection accuracy, demonstrating the effectiveness and superiority of the proposed enhancement algorithm.