Coastline detection is an important research topic in the field of remote sensing, particularly in the context of climate change, sea-level rise, and environmental change. Rapid coastline detection plays a crucial role in environmental monitoring and management. This study combines water indices and deep learning for coastline detection. This study employs the Normalized Difference Water Index (NDWI) and the Automated Water Extraction Index (AWEI), among other related indices, to conduct coastline detection research. By applying five relevant water indices within the Google Earth Engine (GEE) platform, water and land areas in the Waisanding Sandbar region are distinguished to generate binary maps. Subsequently, Otsu's method is used to automatically determine the optimal threshold and refinement for land and sea classification. The results are compared and analyzed to evaluate the automation process of the coastline detection workflow. Finally, a deep learning approach is applied to produce the coastline detection results. A U-Net deep learning model is trained to automatically detect coastlines directly from satellite imagery. The model achieved an accuracy of 99.6% and an F1 score of 0.72, indicating its effectiveness in detecting the coastline of the Waisanding Sandbar.

The operation of rain gauges is influenced by environmental shielding, where excessive obstruction may lead to precipitation measurement biases. To ensure the representativeness of measurement results, rain gauges should be installed in open areas, and their obstruction conditions should be continuously monitored. To address this need, this study proposes an image-based algorithm to estimate the rain gauge exposure. Through capturing sky images above the rain gauge using a fisheye lens and applying the fisheye object-image correspondence, and then defines parameters based on exposure geometric conditions and devises a tilt error correction method. Experimental results demonstrate that, in a correction mode that is convenient for field application procedures and takes into account various errors, the proposed method for exposure measurement can achieve an elevation angle accuracy within 1 degree, meeting the quality requirements of exposure measurement.

This study applies the RGB-D sensor, which is one of low-cost devices, to capture point clouds of the indoor environment for aiding 3D reconstruction manually. Based on the goal of reducing the cost, the proposed solution is compared with that of measurements and 3D reconstructions by manually. For measuring the indoor environment, the manual approach is faster than the RGB-D scanning, saving 1 hour and 25 minutes. The RGB-D sensor captures point clouds with multiple stations and data registrations causing time consuming operations. For reconstruction, the RGB-D based method improves than the manual approach to save 4 hours. In summary, this study demonstrates the feasibility of the low-cost strategy. The significant improvements of measurement will be expected by upgrading the specification of the used sensor or modeling the complex and irregular targets.

With the advancement of UAV and image analysis technologies, precision agriculture has become a key strategy for enhancing productivity. This study uses high-resolution UAV imagery to evaluate three machine learning and three deep learning models for detecting rice paddy plots in Yilan County, Taiwan. Among the machine learning models (BPNN, logistic regression, and the C5.0 decision tree), the BPNN performed best, achieving 95.62% accuracy and a Kappa coefficient of 0.912. In the deep learning category (AlexNet, VGG16, and VGG19), VGG16 yielded the highest performance with 93.83% accuracy and a Kappa of 0.894 using RGB bands and basic enhancements bands. Although certain machine learning models demonstrated slightly higher accuracy, they required complex and time-consuming feature engineering. In contrast, deep learning models produced competitive interpretation results using only RGB bands and simple enhancement techniques. These findings demonstrate the superior capability of image-based (2D) convolutional neural networks (CNNs) in land cover interpretation and highlight their strong potential for application in agricultural environmental surveys.