水下
亚像素渲染
人工智能
灰度
计算机科学
稳健性(进化)
计算机视觉
质心
结构光
分割
光学
遥感
地质学
像素
物理
生物化学
海洋学
化学
基因
作者
Shuaishuai Li,Xiang Gao,Zhengchao Xie
出处
期刊:Sensors
[MDPI AG]
日期:2023-12-15
卷期号:23 (24): 9839-9839
摘要
The non-uniform reflectance characteristics of object surfaces and underwater environment disturbances during underwater laser measurements can have a great impact on laser stripe center extraction. Therefore, we propose a normalized grayscale gravity method to address this problem. First, we build an underwater structured light dataset for different illuminations, turbidity levels, and reflective surfaces of the underwater object and compare several state-of-the-art semantic segmentation models, including Deeplabv3, Deeplabv3plus, MobilenetV3, Pspnet, and FCNnet. Based on our comparison, we recommend PSPnet for the specific task of underwater structured light stripe segmentation. Second, in order to accurately extract the centerline of the extracted light stripe, the gray level values are normalized to eliminate the influence of noise and light stripe edge information on the centroids, and the weights of the cross-sectional extremes are increased to increase the function convergence for better robustness. Finally, the subpixel-structured light center points of the image are obtained by bilinear interpolation to improve the image resolution and extraction accuracy. The experimental results show that the proposed method can effectively eliminate noise interference while exhibiting good robustness and self-adaptability.
科研通智能强力驱动
Strongly Powered by AbleSci AI