Currently, the detail data is mostly surveyed by theodolites and satellite positioning instruments; however, it is time-consuming and labor-intensive. Additionally, the surveying result is usually local data. Recently, UAVs are increasingly being used as a low-cost, efficient system which can support in acquiring high-resolution data; therefore, this study attempts to use ResU-net to assist in extracting global terrain detail information from high-resolution UAV orthoimages of farm land readjustment areas, and evaluate the feasibility of using the post-processing results in the detail survey. Except for the high-resolution orthoimages, the digital surface model (DSM) by dense matching was also used. The results showed that if the label data covered the elevation changes, adding DSM data by dense matching could promote the accuracy of detection. The F Score in Yilan testing data was 0.73; Taichung testing data was 0.86. In terms of the planar position difference, the result showed that about 80% data meet the accuracy requirement and it demonstrated the feasibility of using deep learning to assist in extracting global terrain detail information for farm land readjustment areas.

Phase filtering is one necessary step in the Interferometric Synthetic Aperture Radar (InSAR) procedure. In this study, the Goldstein Filtering function of the SNAP software was used. The research questions to be explored are two: first, the influences of adjustable parameters of the Goldstein Filtering on the interferogram; second, the cause of the “cross-like patterns” and the initial analysis of their impacts on the accuracy. The results has shown that: first, adaptive filter exponent and FFT size possess the most significant effect on the strength of filtering. Second, although enhancing the strength of filtering could reduce speckle noises, it will also increase the cross-like patterns. If there are too many cross-like patterns in the interferogram, it will undermine image interpretation and the accuracy of the InSAR final products. Therefore, users should adjust the filtering parameters appropriately according to the conditions of the experimental area, so as to achieve a balance between the noise reduction and the existence of cross-like patterns.

Water body segmentation from remote sensing imagery is essential for monitoring and protecting water resources, as well as for assessing the risks of disasters such as flooding. However, traditional index-based approaches to water body identification have significant limitations. In this study, we applied and trained a convolutional neural network (CNN) called U-Net on FORMOSAT-5 imagery of the greater Tainan City area to identify water bodies. The experimental results of the U-Net model were compared with the Normalized Difference Water Index (NDWI) and convincingly showed that the U-Net model had achieved significantly better water body detection performance.

There are about 29,000 bridges in Taiwan and 31% of the bridges are over 30 years old. Traditional bridge inspection methods are too subjective, time-consuming, costly and expose inspectors to danger. Therefore, deep learning was used to replace the traditional inspection method and the Faster-RCNN model with ResNet 50 as the backbone of the convolutional neural network was chosen as the crack identification method using public crack dataset and self-photographed bridge crack data. The results of the study confirmed that the detection efficiency, safety and flexibility were relatively improved compared with the traditional bridge detection method. The average accuracy of the research results for crack identification reached 80.7% and the recall rate reached 77% to successfully detect the damaged area of the bridge. In addition, 87.76% of the test images were able to predict the crack location completely, and only 12.24% of the crack images with interference recognition were misidentified.