Marko Petrov

Mr. Marko Petrov


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 February 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
Marko PetrovMarko Petrov
Marko Petrov
Ema PandilovaEma Pandilova
Ema Pandilova
IDIvica Dimitrovski
Ivica Dimitrovski
Dimitar TrajanovDimitar Trajanov
Dimitar Trajanov
VSVlatko Spasev
Vlatko Spasev
IKIvan Kitanovski
Ivan Kitanovski
Semantic segmentation in remote sensing images, which is the task of classifying each pixel of the image in a specific category, is widely used in areas such as disaster management, environmental monitoring, precision agriculture, and many others. However, traditional semantic segmentation methods face a major challenge: they require large amounts of annotated data to train effectively. To tackle this challenge, few-shot semantic segmentation has been introduced, where the models can learn and adapt quickly to new classes from just a few annotated samples. This paper presents a comprehensive review of recent advances in few-shot semantic segmentation (FSSS) for remote sensing, covering datasets, methods, and emerging research directions. We first outline the fundamental principles of few-shot learning and summarize commonly used remote-sensing benchmarks, emphasizing their scale, geographic diversity, and relevance to episodic evaluation. Next, we categorize FSSS methods into major families (meta-learning, conditioning-based, and foundation-assisted approaches) and analyze how architectural choices, pretraining strategies, and inference protocols influence performance. The discussion highlights empirical trends across datasets, the behavior of different conditioning mechanisms, the impact of self-supervised and multimodal pretraining, and the role of reproducibility and evaluation design. Finally, we identify key challenges and future trends, including benchmark standardization, integration with foundation and multimodal models, efficiency at scale, and uncertainty-aware adaptation. Collectively, they signal a shift toward unified, adaptive models capable of segmenting novel classes across sensors, regions, and temporal domains with minimal supervision.

22 August 2026 in Sensors
Verified
Journal Article
Open access
PDF file
Ema PandilovaEma Pandilova
Ema Pandilova
ASAleksandar Stojmenski
Aleksandar Stojmenski
ICIvan Chorbev
Ivan Chorbev
Marko PetrovMarko Petrov
Marko Petrov
IKIvan Kitanovski
Ivan Kitanovski
Dimitar TrajanovDimitar Trajanov
Dimitar Trajanov
Quantitative EEG features such as frontal alpha asymmetry, spectral ratios and signal-complexity measures are often presented as interpretable biomarkers of emotion. Such claims require the markers to generalize across individuals, yet common evaluation protocols allow overlapping epochs and recordings from the same participants to appear in both training and test sets. We re-evaluated qEEG-based valence and arousal recognition on DEAP and DREAMER under trial-grouped, participant-independent, within-participant and cross-dataset protocols. Epoch-pooled evaluation on DEAP gave ROC-AUC values of 0.689 for valence and 0.711 for arousal, whereas participant-independent evaluation of the same features and model returned 0.493 and 0.447. Grouping epochs by trial accounted for about 0.06 of that difference and separating participants for a further 0.13 to 0.17. The same features identified participants with accuracy of 0.998 on DEAP and 0.891 on DREAMER, and a predictor that used no EEG, assigning each trial its participant’s training-set positive rate, accounted for 42 to 84 percent of the above-chance discrimination of the epoch-pooled model. Emotion-related effects were reproducible within participants on DEAP but close to zero on DREAMER, and their direction reversed for about 40 percent of features across participants. In a matched participant-level comparison using a single fixed estimator in both arms, training on a participant’s own data improved DEAP valence by 0.092 AUC (95% CI 0.029 to 0.157, Holm-adjusted p=0.042) and gave no reliable benefit for DEAP arousal or for either DREAMER target. Across the channels shared by the two datasets, per-feature arousal effect sizes correlated moderately, although no individual feature reached false-discovery-rate significance in both datasets. Pooled qEEG emotion-recognition scores can therefore reflect participant-specific recording structure rather than transferable affective information. Population-level claims require participant-independent evaluation, while personalization should be considered only where stable within-person effects are demonstrated.

3 November 2026 in Diagnostics
Verified
Journal Article
Open access
PDF file
Ema PandilovaEma Pandilova
Ema Pandilova
Marko PetrovMarko Petrov
Marko Petrov
ICIvan Chorbev
Ivan Chorbev
IKIvan Kitanovski
Ivan Kitanovski
ASAleksandar Stojmenski
Aleksandar Stojmenski
Dimitar TrajanovDimitar Trajanov
Dimitar Trajanov
Background/Objectives: Deep learning segmentation of nasopharyngeal carcinoma is usually reported as one Dice similarity coefficient (DSC) against one reference mask, although multi-sequence MRI is often delineated separately on each sequence. We aimed to quantify how far the choice of reference changes reported performance, and to compare this with the disagreement between radiologists. Methods: We analysed a public dataset of 277 patient records with separately delineated T1-weighted, T2-weighted and contrast-enhanced T1-weighted MRI, 30 of whom were annotated by two radiologists. Tumour volumes were compared without registration. Spatial agreement was measured after rigid registration, with the loss from transferring masks between acquisition grids estimated separately. nnU-Net models were scored against every reference, including the second radiologist’s, in 102 held-out patient records. Results: Without registration, the sequence effect on tumour volume was 10.9%, against a reader effect of 5.7% (ratio 1.9, 95% CI 1.40 to 2.49). After rigid registration, per-sequence references agreed at a median DSC of 0.783, against 0.898 between radiologists; in the double-annotated patient records, the paired gap narrowed only from 0.129 to 0.083 (95% CI 0.063 to 0.111) when grid transfer was matched. One fixed set of predictions scored 0.726, 0.685 and 0.673, depending only on the reference (median per-patient reference spread 0.075), and switching the reference sequence moved scores 2.7 to 3.6 times as much as switching the radiologist. Conclusions: Part of the cross-sequence difference is a difference in what each sequence makes visible rather than annotation error, so a single-reference DSC on multi-sequence data depends on a rarely stated choice. Performance should be reported against every available reference and read against measured inter-reader agreement.

17 August 2026 in Information
Verified
Journal Article
Open access
PDF file
Ema PandilovaEma Pandilova
Ema Pandilova
Marko PetrovMarko Petrov
Marko Petrov
ICIvan Chorbev
Ivan Chorbev
Dejan GjorgjevikjDejan Gjorgjevikj
Dejan Gjorgjevikj
Dimitar TrajanovDimitar Trajanov
Dimitar Trajanov
Retrieval-augmented generation (RAG) is increasingly deployed in safety-critical cybersecurity question answering, where an answer must be both factually correct and faithful to its sources. Evaluation in this space routinely conflates the two, treating grounding as a proxy for correctness. We show that this conflation is unsafe. We introduce SENTINEL-QA, a bilingual English and Macedonian cybersecurity QA benchmark of 70 hand-authored items over a curated corpus of authoritative source documents, and use it to compare four answer generation pipelines (zero-shot, Classic RAG, DSPy-structured RAG, and GraphRAG) across three generators, judged by a four-model LLM panel on both factual correctness and grounding. The two metrics decouple under graph-based retrieval augmentation: on all three generators our chunk-similarity GraphRAG makes answers look better sourced without making them more correct, and as the retrieval budget grows it can even degrade factual correctness while grounding holds steady. Structured, minimal prompting improves both qualities, at lower latency and cost; a four-cell ablation attributes the factual gain to pruning the instruction, while the declarative schema contributes on grounding. The judge panel mirrors the split, agreeing more on what is well grounded than on what is true, under absolute as well as rank-based agreement criteria. We conclude that grounding alone is an insufficient target for RAG evaluation and that pruning the instruction, rather than expanding retrieval, is what improves factual correctness, without eliminating the grounded-but-wrong regime.