Electricity theft is a major source of non-technical losses in distribution networks, threatening both economic revenues and power supply reliability. This study addresses the identification of nodes exhibiting anomalous load behavior (anomalous nodes) in 10 kV distribution feeders. Based on the IEEE-33 bus benchmark system, the disturbance patterns induced by abnormal consumption are analyzed. The results show that voltage and current fluctuations intensify with increasing electrical distance from the power source, while branch loss peaks localize at the affected terminals and propagate unidirectionally along the power flow path. Building on these findings, an improved density-based spatial clustering of applications with noise (DBSCAN) method is proposed, integrating five spatial network features and sixteen temporal electrical features extracted from voltage, current, and power series. Prior to clustering, the features are standardized and reduced via principal component analysis (PCA), retaining over 90% of the cumulative variance. Validation on a hybrid dataset demonstrates that the proposed method achieves 90.7% accuracy, 87.5% recall, and an F1-score of 0.895, outperforming traditional K-means and approaching supervised CNN models without requiring labeled data. These results confirm the method’s robustness and suitability for practical deployment in distribution networks.

Unmanned surface vehicles (USVs) must detect small obstacles under glare, wave motion, poor visibility, and limited onboard computation. This review connects these sensing conditions to dataset selection, deep learning methods, multimodal fusion, and edge deployment. We distinguish direct USV evidence from transferable results and compare public datasets by acquisition viewpoint, target scale, sensor coverage, and annotation task. An accompanying benchmark evaluates YOLOv8s, YOLO11s, YOLO26s, and RT-DETR-L on FloW, SeaDronesSee v2, and WSODD using three training seeds, separate checkpoint-selection and assessment subsets for the new SeaDronesSee runs, and recall at matched false-positive rates. RT-DETR-L achieves the highest mean mAP50–95 under native preprocessing, with desktop-GPU latency 3.2–3.9 times that of the YOLO models. On FloW, stretching raises YOLOv8s and YOLO11s to 46.0% and 45.5% mAP50–95, respectively, compared with 45.5% for RT-DETR-L, showing the importance of preprocessing in model comparisons. Size-based recall and per-object diagnostics reveal different patterns of missed targets, localization error, and class confusion. The synthesis identifies priorities for preserving weak target cues, using temporal and complementary sensor observations, and evaluating detection quality together with device-specific latency and power.