UAS Inspection Image Enhancement Coupled with Denoise Algorithm Based on Deep Neural Network

Abstract

Unmanned Aerial System (UAS) technologies integrated with image processing algorithms are considered timely and useful for bridge inspections because of improved accessibility, recording ability, and cost-efficiency compared to the conventional inspection approach. The image processing algorithms can improve the ability of the UAS-aided bridge inspections in efficiently identifying and quantifying deterioration. This study was aimed to inspect an in-service single-span precast concrete bridge on a rural roadway in South Dakota using UAS technologies coupled with a Deep Neural Network (DNN) denoise algorithm. During the inspections, Phantom 4 and DJI Matrice 210 UASs recorded several videos for different bridge elements (e.g., girders and decking), and a total of 21,784 inspection images were extracted from the videos with a duration of more than 14 minutes. Deteriorations specific to the bridge elements such as spalling and rust were characterized by performing the DNN-aided image processing algorithm with the extracted inspection images. The DNN allowed for computation and analysis between input and output image data to reduce the noises on the images. Besides, a grayscale image enhancement algorithm was considered to improve the visibility of images by optimizing image contrast settings. With the visibility-improved images, detailed quantification on the detected deterioration per bridge element was carried out using a pixel-based measurement tool. Based upon the studys results, it was revealed that the UAS technologies with the DNN denoise algorithm were able to successfully characterize and quantify visible deteriorations to the certain bridge elements using pixel-based tools.