Romário Da Costa Silva
Research Subjects
Romário Da Costa Silva has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Carlos Renato Lisboa Francês
Federal University of Pará (UFPA)
2 shared publications - Jasmine Araújo
Universidade federal do Pará
2 shared publications Carlos André de Mattos Teixeira
Graduate Program of Electrical Engineering, Universidade Federal do Pará, Belém, Pará, Brazil
1 shared publicationCarlos Renato Lisboa Francês
Graduate Program of Electrical Engineering, Universidade Federal do Pará, Belém, Pará, Brazil
1 shared publication- Diego Lisboa Cardoso
Federal University of Para
1 shared publication - Elainne Ladislau Ferreira Pereira
Department of Electrical Engineering, Federal University of Par (UFPA), Belm, Brazil
1 shared publication
Latest Publications
- Open Access
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 83991-84015
Landfill Site Allocation Multi-Objective Optimization With NSGA-II: A Case Study on Fogo Island, Cape Verde
Effective landfill site selection is essential for minimizing environmental impact and construction costs while mitigating public health risks, particularly in Small Island Developing States (SIDS) where geographical and financial constraints are severe. Traditional spatial analysis methods aggregate conflicting objectives into single suitability scores, masking crucial trade-offs and neglecting operational resilience. This paper introduces a rigorously validated multi-phase framework integrating Geographic Information Systems (GIS) with the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to address the Facility Location Problem on Fogo Island, Cape Verde. The model balances three conflicting objectives, namely implementation cost, environmental impact, and Social Safety, across 473 candidate sites. A minimum pairwise distance constraint (10 km) enforces geographic dispersion, reducing the feasible solution space from 17.5 million to 2.90 million valid configurations while supporting geographic resilience against localized natural hazard scenarios. Validated against a complete deterministic Exhaustive Search, NSGA-II captured 99.50% of the true Pareto front hypervolume with an optimality gap of 0.50% in the best run (mean gap: 2.45% across 30 independent runs) and a 4.0-fold reduction in computational time. The Best Compromise Solution forms a triangular network with an average inter-facility spread of 13.41 km, covering two of the island’s three municipalities and achieving a 50.3% improvement in geographic distribution at a trade-off ratio of 8.52:1 between spread gain and hypervolume reduction. This work transitions waste infrastructure planning from static suitability mapping to dynamic, validated, and resilient systems engineering.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 83991-84015
Open AccessLandfill Site Allocation Multi-Objective Optimization With NSGA-II: A Case Study on Fogo Island, Cape Verde
Effective landfill site selection is essential for minimizing environmental impact and construction costs while mitigating public health risks, particularly in Small Island Developing States (SIDS) where geographical and financial constraints are severe. Traditional spatial analysis methods aggregate conflicting objectives into single suitability scores, masking crucial trade-offs and neglecting operational resilience. This paper introduces a rigorously validated multi-phase framework integrating Geographic Information Systems (GIS) with the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to address the Facility Location Problem on Fogo Island, Cape Verde. The model balances three conflicting objectives, namely implementation cost, environmental impact, and Social Safety, across 473 candidate sites. A minimum pairwise distance constraint (10 km) enforces geographic dispersion, reducing the feasible solution space from 17.5 million to 2.90 million valid configurations while supporting geographic resilience against localized natural hazard scenarios. Validated against a complete deterministic Exhaustive Search, NSGA-II captured 99.50% of the true Pareto front hypervolume with an optimality gap of 0.50% in the best run (mean gap: 2.45% across 30 independent runs) and a 4.0-fold reduction in computational time. The Best Compromise Solution forms a triangular network with an average inter-facility spread of 13.41 km, covering two of the island’s three municipalities and achieving a 50.3% improvement in geographic distribution at a trade-off ratio of 8.52:1 between spread gain and hypervolume reduction. This work transitions waste infrastructure planning from static suitability mapping to dynamic, validated, and resilient systems engineering.
- Open Access
- 13 March 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3708
Assessing the efficacy of convolutional neural networks for Pap smear classification: a real world analysis
Background: Undetected cervical lesions can progress to cancer, a leading cause of mortality among women worldwide. While automated analysis of Papanicolaou (Pap) smear images using convolutional neural networks (CNNs) has demonstrated significant potential for screening, most existing studies rely on single curated datasets. This aspect limits the understanding of model generalization to the noise and variability inherent in real-world clinical cytology. Methods: We evaluated three CNN architectures (VGG16, ResNet50, and InceptionV3) across four curated Pap smear datasets using stratified 5-fold cross-validation. For each dataset, the model achieving the highest mean Macro-F1 score was selected for further analysis. To assess robustness against domain shift, we performed an external evaluation using a non-curated, Real-World dataset comprising routine clinical images. Results: All architectures achieved robust performance on the curated benchmarks, with mean Macro-F1 scores ranging from 73.58% to 99.28%. However, performance dropped significantly when models were evaluated on the Real-World dataset (Macro-F1: 33.25–55.91%), highlighting the severity of the domain gap. Notably, the model trained on a combined heterogeneous dataset achieved the highest inter-domain performance, suggesting that data diversity improves robustness. Class-wise analysis revealed that high-grade lesions were most sensitive to real-world variability. Conclusions: Although CNNs achieve state-of-the-art results on curated benchmarks, their direct applicability to routine cytology workflows is hindered by domain shift. Our findings emphasize that evaluating models across heterogeneous, multi-source datasets is a prerequisite for reliable clinical deployment.
- 13 March 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3708
Open AccessAssessing the efficacy of convolutional neural networks for Pap smear classification: a real world analysis
Background: Undetected cervical lesions can progress to cancer, a leading cause of mortality among women worldwide. While automated analysis of Papanicolaou (Pap) smear images using convolutional neural networks (CNNs) has demonstrated significant potential for screening, most existing studies rely on single curated datasets. This aspect limits the understanding of model generalization to the noise and variability inherent in real-world clinical cytology. Methods: We evaluated three CNN architectures (VGG16, ResNet50, and InceptionV3) across four curated Pap smear datasets using stratified 5-fold cross-validation. For each dataset, the model achieving the highest mean Macro-F1 score was selected for further analysis. To assess robustness against domain shift, we performed an external evaluation using a non-curated, Real-World dataset comprising routine clinical images. Results: All architectures achieved robust performance on the curated benchmarks, with mean Macro-F1 scores ranging from 73.58% to 99.28%. However, performance dropped significantly when models were evaluated on the Real-World dataset (Macro-F1: 33.25–55.91%), highlighting the severity of the domain gap. Notably, the model trained on a combined heterogeneous dataset achieved the highest inter-domain performance, suggesting that data diversity improves robustness. Class-wise analysis revealed that high-grade lesions were most sensitive to real-world variability. Conclusions: Although CNNs achieve state-of-the-art results on curated benchmarks, their direct applicability to routine cytology workflows is hindered by domain shift. Our findings emphasize that evaluating models across heterogeneous, multi-source datasets is a prerequisite for reliable clinical deployment.
- Open Access
- 3 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 19302-19319
Pineapples Health Detection Using Deep Learning Models
Early detection of diseases in agricultural crops remains a significant challenge, especially for small producers who still rely on manual visual inspections—time-consuming, subjective processes with low reproducibility. In pineapple (Ananas comosus) cultivation, phytosanitary monitoring is essential to prevent losses associated with diseases such as fusarium wilt, wet rot, and wilt, as well as infestations by pertussis and nutritional deficiencies. This study presents an innovative automated monitoring approach for detecting visual anomalies in pineapple fruits, based on computer vision techniques applied primarily to images obtained under real field conditions, considering variations in lighting, occlusions, leaf overlap, different fruit orientations, and background interference. Furthermore, a comparative analysis was conducted between different versions of the You Only Look Once model (YOLOv5, YOLOv8, and YOLOv9) to identify the most suitable solution for different production profiles—from small family farms with limited computational resources to large farms with advanced technological infrastructure. The evaluation considered the metrics of Precision, Recall, Average Precision (mAP), and inference time, as well as operational feasibility in low-infrastructure contexts. The results demonstrate consistent performance among the tested models, with YOLOv8 standing out as the most suitable alternative for practical applications in the field and in embedded devices, establishing itself as a robust, accessible, and efficient solution for automated phytosanitary monitoring of pineapple crops.
- 3 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 19302-19319
Open AccessPineapples Health Detection Using Deep Learning Models
Early detection of diseases in agricultural crops remains a significant challenge, especially for small producers who still rely on manual visual inspections—time-consuming, subjective processes with low reproducibility. In pineapple (Ananas comosus) cultivation, phytosanitary monitoring is essential to prevent losses associated with diseases such as fusarium wilt, wet rot, and wilt, as well as infestations by pertussis and nutritional deficiencies. This study presents an innovative automated monitoring approach for detecting visual anomalies in pineapple fruits, based on computer vision techniques applied primarily to images obtained under real field conditions, considering variations in lighting, occlusions, leaf overlap, different fruit orientations, and background interference. Furthermore, a comparative analysis was conducted between different versions of the You Only Look Once model (YOLOv5, YOLOv8, and YOLOv9) to identify the most suitable solution for different production profiles—from small family farms with limited computational resources to large farms with advanced technological infrastructure. The evaluation considered the metrics of Precision, Recall, Average Precision (mAP), and inference time, as well as operational feasibility in low-infrastructure contexts. The results demonstrate consistent performance among the tested models, with YOLOv8 standing out as the most suitable alternative for practical applications in the field and in embedded devices, establishing itself as a robust, accessible, and efficient solution for automated phytosanitary monitoring of pineapple crops.