Power outages impede economic operations, because numerous processes rely on electricity. Doing regular maintenance and inspection is one solution. However, traditional foot patrol inspection methods can be risky, slow, and tedious. An innovative contemporary alternative involves inspecting distribution networks by unmanned aerial vehicles (UAVs). Thus, this paper introduces a three-tier inspection procedure: track power lines, detect utility assets (transformer bank, low-voltage bushing, high-voltage bushing, arrester, radiator fin, and cutoff fuse), and spot foreign object obstructions (snake, gecko, and tree branch). This work exhibits two contributions. First, it introduces new learning representations for detecting power lines, utility assets, and foreign obstructions. The image processing-based techniques are built by modifying pre-trained neural networks – Point Instance Network (PINet) for power lines and You Only Look Once (YOLO) for utility assets and obstructions. Second, it presents trained detection models with image datasets tailored to each problem. The datasets consist of images from both obstruction-free and shelled UAVs for the three detection tasks. Experimental results indicate no significant disparity in utilizing two distinct UAV variants, with or without a shell. The proposed method, leveraging Transfer Learning with PINet, yields around 80% accuracy and 72% f1-score across obstruction-free and obstructed image datasets. Moreover, the retrained YOLO checkpoints with yolov5m and yolov5l weights showcase suitability for deployment in detecting utility assets and foreign object obstructions that achieved mAPs around 0.96 and 0.85, respectively. Despite a slight decline in performance metrics with obstructed images, the models signify a balance between detection accuracy and the safety conferred by the insulated shell.

Hazardous and remote environments, such as volcanic regions and disaster zones, require regular monitoring due to the significant risks they pose. Innovative solutions like aerial and mobile robots offer safer monitoring options, though drones are limited by short battery life, and mobile robots face mobility challenges. Collaborative aerial-ground robots have been developed to address these issues, with drones transporting and recovering ground robots. Deploying the ground robot is easy, but its recovery is challenging due to drone Global Positioning System (GPS) inaccuracies. It highlights the need for a precise visual alignment technique. This study utilized YOLOv8n segmentation model to develop a ground robot detection model. The development involved gathering images of ground robots in various environmental settings such as rocks, seashores, sand, grasslands, limestone, and soil. Three models were tested in this study: plate detection, ground robot detection, and a combined plate and ground robot detection model. The ground robot detection model emerged as the most robust, achieving an F1 score of 99.7% and an mAP of 95.4% at 0.5 to 0.96 IoU thresholds. The model was integrated into a small computer on a visual servoing drone and tested outdoors. The alignment threshold was set at 65 cm, based on the size of the retrieval mechanism. The experiments indicate that the drone achieved an average alignment error of 41.985 cm within 7.82 seconds. The experimental results exhibit the effectiveness of the UAV-UGV alignment for the retrieval task in hazardous and remote areas.

Retrieving ground robots from dangerous environments after their operation is a challenging task that poses risks for the personnel. Some researchers often employ drones for retrieval, which makes operations safer. However, this set-up requires an accurate method that guarantees drone and ground robot alignment due to inaccuracies in standard GPS devices, drone drifts, and wind gusts. Hence, this research article introduces simultaneous object detection and tilt correction as part of visual servoing to achieve precise drone-rover alignment. Drone detection using YOLOv8 and a tilt correction algorithm was integrated for the proposed visual servo of the ground robot. The study collected 3024 images as a data set for drone detection. The experimental results show that the trained instance segmentation model detected and captured drone objects. The study conducted an initial test for visual servo control of the ground robot in various surface terrains, resulting in a maximum alignment error on rough surfaces. Furthermore, the study conducted drone-ground robot alignment real test in an outdoor field setting. The alignment between the drone and the ground robot produced a maximum alignment error of 20.3 cm, below the threshold error. The open field experiments verified the effectiveness of the ground robot's visual servo control with an actual drone operation.

Structural crack detection remains a key component of structural health monitoring for large-scale civil infrastructure, where manual inspection is costly and hazardous. While aerial imaging enables rapid data acquisition, reliable crack analysis is limited by the difficulty of obtaining high-resolution images at close range without risking UAV damage. To address this, we introduce a UAV equipped with a protective spherical shell that allows safe surface contact. Although CNN-based architectures such as FPN and U-Net are established for crack segmentation, their encoders struggle to model long-range dependencies and fine edge structures, which result in poor delineation of thin, low-contrast, and discontinuous cracks. This limitation remains a persistent research gap in aerial crack analysis. To address this gap, this study investigates two complementary strategies: (i) replacing conventional CNN encoders with hierarchical Transformer backbones to capture global contextual relationships, and (ii) introducing edge-aware input augmentation using luminosity-based grayscale and Sobel gradient maps to optimize boundary sensitivity. We conducted experiments on our CrackUAS dataset, which consists of aerial images collected from infrastructure inspection scenarios. Integrating a hierarchical Transformer encoder with Sobel edge augmentation yields consistent performance gains across architectures. The ablation study concludes that F1-scores increase by approximately 33% for U-Net and 13% for FPN relative to their baseline configurations. These results show that integrating Transformer-based feature encoding with edge-aware input augmentation improves aerial crack segmentation.