In this research, 200 corrosion images of steel and 500 crack images of rubber bearing are collected and manually labeled to build the data set. Then the two data sets are respectively adopted to train VGG-Unet models in two methods, aiming to conduct Damage Segmentation by inputting different size of data set. One method is Squashing Segmentation to input squashed images from high resolution directly into VGG-Unet model while Cropping Segmentation uses cropped image with size 224 × 224 as input images. Because the proportion of damage pixels in the data set is different, the results produced by the two data sets are quite different. For large size damage (such as corrosion) segmentation, Cropping Segmentation has a better result while for minor damage (such as crack) segmentation, the result is opposite. The main reason is the gap in the concentration of valid data from the data set. To improve the capability of crack segmentation based on Cropping Segmentation, Background Data Drop Rate (BDDR) is adopted to reduce the quantity of background images to control the proportion of damage pixels from the data set in pixel-level. The ratio of damage pixels from the data set can be decided by different value of BDDR. By testing, the accuracy of Cropping Segmentation becomes relatively higher under BDDR being 0.8.

Recently, with the increase of aging deterioration bridges and the shortage of experienced bridge engineers, innovative methods are needed instead of convenience manual bridge inspections. New methods such as unmanned aerial vehicle (UAV)-based bridge inspections have been studied to fetch high-quality images and videos, and to help detecting damages on the surface of bridges. Even though the UAV cameras could provide high-resolution data, it still costs a greater manual processing time to find out and mark the damage on its location of bridge structure and to diagnose its status. This research aims to provide a deep learning-based method to find different categories of damage from a great number of images. In this study, a data set is generated by labeling damages pixel by pixel, manually on the images sliced from high-resolution videos installed on UAVs. The data set is used to train a Fully Convolutional Neural Network (FCNN) model for detecting damage image and location on small pieces of images. And the location of damage and its category can be shown in a stitched image visualizing the damage and its distribution.