AŞAbdullah Şener

Abdullah Şener

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Abdullah Şener has a total of 10 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A Hybrid Deep Learning Approach for Early Detection of Wildfires Using Transformer Models and Feature Fusion Techniques

    by Abdullah Şener, Burhan Ergen, Kubilay Demir, Serdar Ekinci, Vedat Tümen

    Natural disasters, particularly wildfires, pose a serious threat to ecosystems, human lives, and infrastructure, making rapid and accurate fire detection essential for effective disaster management. Despite recent advances in deep learning-based wildfire detection, conventional CNN-based approaches are inherently limited in modeling long-range spatial dependencies, while relying on a single deep architecture may provide an incomplete representation of complex and multi-scale fire patterns. Furthermore, high-dimensional deep features may contain redundant information, increasing computational complexity and limiting the efficiency of real-time monitoring systems. Therefore, there remains a need for an efficient framework that can exploit complementary global representations from multiple transformer architectures while retaining only the most discriminative features. To investigate this application-specific problem, this study evaluates a hybrid framework that integrates three heterogeneous Vision Transformer architectures—DeiT3, MaxViT, and Swin—with feature-level fusion, mRMR-based feature selection, and SVM classification for binary wildfire image classification. The publicly available FlameVision dataset, consisting of aerial images categorized into fire and non-fire classes, was used for evaluation. Deep feature representations extracted from the three transformer architectures were systematically fused to exploit their complementary characteristics. Minimum Redundancy Maximum Relevance (mRMR) feature selection was subsequently employed to reduce feature redundancy and retain the most discriminative information, followed by classification using Support Vector Machines (SVM). The experimental results demonstrated an overall classification accuracy of up to 100% on the internal hold-out test partition and in five-fold cross-validation within the FlameVision dataset. Because both evaluation procedures were conducted using images originating from the same source dataset, these results should be interpreted as within-dataset classification performance rather than evidence of external generalization. Validation on independently collected wildfire datasets is therefore required to determine the robustness of the proposed framework under different geographical, sensor, environmental, and acquisition conditions. The proposed framework demonstrates that complementary transformer-derived representations combined with feature selection can provide highly discriminative wildfire representations while substantially reducing the number of features required for classification. These findings demonstrate that the proposed approach can retain high classification performance using a substantially more compact feature representation. However, the resulting reduction in feature dimensionality should not be interpreted as evidence of improved runtime efficiency, since computational time before and after feature selection was not experimentally measured. However, real-time performance is not claimed in the present study because FPS, end-to-end latency, runtime memory consumption, and deployment on continuous UAV or satellite streams were not experimentally evaluated. Such operational evaluation is left for future work.

  • Quantifying Post-Earthquake Urban Green Space Loss Using an Optimized Deep Learning-Based Semantic Segmentation Framework

    by Abdullah Şener, Muhammed Aydın, Burhan Ergen

    Large-scale earthquakes and the rapid reconstruction processes that follow lead to significant spatial transformations in urban ecosystems, particularly exerting pronounced impacts on intra-urban green spaces. This study quantitatively evaluates changes in urban green areas that occurred after the 2023 Kahramanmaraş earthquakes in Hatay (Antakya) and Elazığ (Güneykent) by utilizing high-resolution satellite imagery and an optimized deep learning–based semantic segmentation approach. In the proposed framework, an enhanced U-Net architecture incorporating asymmetric and dilated convolutional layers was designed to improve parameter efficiency while preserving extensive contextual information. The proposed model, comprising approximately 1.7 million parameters, achieved 88% accuracy, 0.76 Intersection over Union (IoU), and a 0.86 Dice score, demonstrating competitive performance compared to contemporary segmentation models with substantially higher parameter counts. Pixel-level comparison of pre- and post-earthquake imagery revealed reductions in green space reaching up to 89% in the examined sub-regions, with even higher rates observed in certain localized areas. The findings indicate that intensive post-disaster construction and urban reorganization processes may impose substantial pressure on environmental sustainability. The proposed method provides a computationally efficient and scalable decision-support framework for monitoring post-disaster environmental changes and developing sustainable urban planning strategies.

  • Open Access

    CalFireSegNet: a lightweight hybrid attention–transformer network for post-wildfire building footprint change analysis using high-resolution satellite imagery

    by Abdullah Şener, Vedat Tümen, Burhan Ergen, Erdal Akin

    Natural disasters, particularly wildfires, cause severe human, environmental, and economic losses worldwide. Rapid and accurate identification of building footprints and potential potential structural changes is essential for effective emergency response, search-and-rescue operations, and post-disaster recovery planning. To address the challenges of identifying building loss from remote sensing imagery, this study proposes CalFireSegNet, a lightweight hybrid attention–transformer network for post-wildfire building footprint extraction and loss proxy detection from satellite imagery. The proposed architecture integrates depthwise convolutions, convolutional block attention modules (CBAM), atrous spatial pyramid pooling (ASPP), and Transformer blocks to effectively capture both local structural details and long-range contextual dependencies while maintaining low computational complexity. The model was trained and evaluated using benchmark building segmentation datasets (Inria and WHU) and subsequently applied to pre- and post-event satellite imagery from the recent California wildfires. Experimental results demonstrate that CalFireSegNet achieves superior performance compared with several state-of-the-art semantic segmentation models, including U-Net, PSPNet, DeepLabv3+, ENet, HRNet, and SegNet, obtaining 98.45% accuracy, 94.35% mIoU, and 95.03% Dice Similarity Score while requiring only 3.72 million parameters. Furthermore, a lightweight mask-difference framework was developed to generate a spatial proxy of potential building footprint loss using pre- and post-event satellite pairs. Since publicly available building-level damage annotations for recent California wildfire events remain limited, the real-world wildfire experiments are presented as a validation of cross-domain applicability rather than a fully supervised structural loss proxy estimation benchmark.

  • Brain stroke detection and classification using deep learning-based transCBAMSegNet and hybrid transformer models

    by Abdullah Şener

    Stroke is a life-threatening neurological condition that requires rapid assessment to reduce mortality and long-term disability. Accurate lesion segmentation and reliable stroke type classification from brain computed tomography (CT) images are essential to support timely clinical decision-making. In this study, a deep learning-based framework is proposed for automatic stroke lesion segmentation and CT-based stroke type classification. In the segmentation stage, the developed TransCBAMSegNet model achieved an accuracy of 99.60%, a mean intersection over union (mIoU) of 75.98%, and a Dice Similarity Score (DSS) of 75.50%. In the classification stage, a hybrid MaxxViT–ViT–SwinV2 feature fusion strategy achieved an accuracy of 97%, enabling reliable differentiation between acute/hyperacute ischaemic stroke, haemorrhagic stroke, and normal cases. Experimental results show that the proposed multi-transformer fusion approach provides improved performance compared to single-backbone configurations under identical evaluation settings. The framework is designed as a computer-aided CT-based stroke assessment tool and may support early-stage imaging-based decision processes. Overall, the proposed methodology offers a robust and reproducible approach for automated stroke analysis using CT imaging data.

  • Comparative Evaluation of Machine Learning Models for Gender Classification from Speech Signals

    by Abdullah Şener, Burhan Ergen

    In this study, a machine learning method was developed to determine the gender of people based on voice signals. For this purpose, spectral, temporal, pitch, MFCC (Mel-Frequency Cepstral Coefficients) and energy-based features of the voice data were extracted and analysed. Different machine learning algorithms were compared to determine the model with the highest hit rate. Experimental analyses showed that the ExtraTreesClassifier model performed the best with an accuracy of 98.81%. The accuracy, loss and confusion matrix analyses showed that the model had a high generalisation ability and did not suffer from overfitting. The error rate was calculated to be 1.56% for the female class and 0.79% for the male class. The results of this study suggest that speech-based gender classification systems could contribute to potential applications in areas such as customer service, medical diagnostics and human-computer interaction.


  • Open Access

    Enhancing Geometry Education through Deep Learning Models: Addressing Challenges in Three-Dimensional Shape Visualization

    by Abdullah Şener, Serdal Poçan, Burhan Ergen

    Integrating technology into mathematics education is crucial for enhancing the understanding of mathematical concepts and skills, as well as increasing motivation. This is particularly applicable in geometry classes, where technology can facilitate the detection of geometric shapes, impacting both the learning and teaching processes. In this context, emerging concepts of artificial intelligence and deep learning can be utilized as tools to overcome such limitations. This study addresses the challenges that teachers face when drawing three-dimensional geometric shapes in digital environments. Shapes drawn manually in digital environments can often be complex, making it difficult for teachers to create accurate and precise drawings. Deep learning models can assist teachers in correcting drawing errors, thereby providing students with clearer and more comprehensible visuals to facilitate the learning of geometric concepts. The study emphasizes the high accuracy rates achieved using various deep learning models, highlighting their impressive capabilities in accurately classifying geometric shapes.


  • Open Access

    Automatic detection of gastrointestinal system abnormalities using deep learning-based segmentation and classification methods

    by Abdullah Şener, Burhan Ergen

    Early diagnosis and precise treatment of gastrointestinal (GI) diseases are crucial for reducing mortality and improving quality of life. In this context, the detection and classification of abnormalities in endoscopic images is an important support for specialists during the diagnostic process. In this study, an innovative deep learning approach for the segmentation and classification of pathological regions in the GI system is presented. In the first phase of the study, a novel segmentation network called GISegNet was developed. GISegNet is a deep learning-based architecture tailored for accurate detection of anomalies in the GI system. Experiments conducted on the Kvasir dataset showed that GISegNet achieved excellent results on performance metrics such as Jaccard and Dice coefficients and outperformed other segmentation models with a higher accuracy rate (93.16%). In the second phase, a hybrid deep learning method was proposed for classifying anomalies in the GI system. The features extracted from the transformer-based models were fused and optimized using the Minimum Redundancy Maximum Relevance (mRMR) algorithm. The classification process was performed using Support Vector Machines (SVM). As a result of feature fusion and selection, the second model, which combined features from DeiT and ViT models, achieved the best performance with an accuracy rate of 95.2%. By selecting a subset of 300 features optimized by the mRMR algorithm, the accuracy (95.3%) was maintained while optimizing the computational cost. These results show that the proposed deep learning approaches can serve as reliable tools for the detection and classification of diseases of the GI system.