The inspection and maintenance of urban sewer infrastructure remain critical challenges for megacities, where conventional manual inspection approaches are labor-intensive, time-consuming, and prone to human error. Although deep learning has been increasingly applied to sewer inspection, the field lacks both a publicly available large-scale dataset and a systematic evaluation of CNN and transformer-based models on real sewer footage. The primary aim of this study is to systematically evaluate and compare state-of-the-art deep learning architectures for automated sewer defect detection using a newly introduced dataset. We present the Istanbul Sewer Defect Dataset (ISWDS), comprising 13,491 expert-annotated images collected from Istanbul’s wastewater network and covering eight defect categories that account for approximately 90% of reported failures. The scientific novelty of this work lies in both the introduction of the ISWDS and the first systematic benchmarking of YOLO (v8/11/12) and RT-DETR (v1/v2) architectures under identical protocols on real sewer inspection footage. Experimental results demonstrate that RT-DETR v2 achieves the best performance (F1: 79.03%, Recall: 81.10%), significantly outperforming the best YOLO variant. While transformer-based architectures excel in detecting partially occluded defects and complex operational conditions, YOLO models provide computational efficiency advantages for resource-constrained deployments. Furthermore, a QGIS-based inspection tool integrating the best-performing models was developed to enable real-time video analysis and automated reporting. Overall, this study highlights the trade-offs between accuracy and efficiency, demonstrating that RT-DETR v2 is most suitable for server-based processing. In contrast, compact YOLO variants are more appropriate for edge deployment.

Image colorization is the process of obtaining colored images by assigning RGB color values to a grayscale or panchromatic image. This technique has an important place in the field of computer vision because colored images provide a better visual experience and are widely used in areas such as image recognition and object detection. It also has many practical applications such as coloring historical photographs, adding colors to be used in the analysis of medical images, and improving the analysis of satellite images. Colorization methods are divided into two main categories: brush coloring and sample-based coloring. Both methods have certain limitations. The performance of these methods depends on the selected reference images and may sometimes contain false colors or significant errors. While these methods require operator intervention or pre-defined rules, deep learning based methods are largely automated and uses neural networks to understand the global and local context of an image, leading to more realistic and contextually accurate colorizations. The presented study uses the Denoising Diffusion Null-Space Model (DDNM) architecture. DDNM is a method that aims to obtain more efficient and high-quality results compared to the coloring approaches available in literature. In the study, the weight data of the DDNM architecture was used to predict colored images from panchromatic images using the SpaceNet 6 open access dataset. The SpaceNet 6 dataset includes a combination of Capella Space 0.5m Synthetic Aperture Radar (SAR) imagery and Maxar's 0.5m electro-optical (EO) imagery. In order to assess the results, Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) accuracy metrics are calculated.

Difference Normalised Burn Ratio (dNBR) is used as a reliable reference index since it has accepted thresholds for determining fire severity. However, the use of the dNBR index is limited if pre-fire data cannot be obtained due to weather and other conditions. This situation necessitates the development of alternative indices that are calculated only with post-fire satellite imagery. This study was conducted on forest fires in Mersin, Izmir and Mugla provinces of Turkey while aiming to evaluate the effectiveness of alternative indices to the dNBR index, which is widely used in determining ecosystem damage and fire severity after forest fires. Using Sentinel-2 satellite data operated by the European Space Agency (ESA), the study analysed the performance of the NDVI, NDMI, NBR, MSAVI, EVI and BAIS2 indices calculated using only post-fire data against the dNBR index calculated from pre- and post-fire imagery. The main methods used in this study include data processing and analyses performed on the Google Earth Engine (GEE) platform and comparisons made on the QGIS platform. In this study, the extent to which these alternative indices can be effective in accurately and reliably assessing post-fire ecosystem damage was investigated. The results of the analyses showed that the NBR and BAIS2 indices have the highest accuracy in detecting post-fire ecosystem damage. While both indices produced results close to the dNBR index, MSAVI and EVI were found to be effective in monitoring vegetation changes but insufficient in determining fire severity. In conclusion, BAIS2 and NBR provide strong alternatives to dNBR in analyses based on post-fire data, while the other indices used in the study are considered as complementary tools.