Zepa Yang

Zepa Yang


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
6 May 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
Muhammad Zaheer SajidMuhammad Zaheer Sajid
Muhammad Zaheer Sajid
MHMuhammad Fareed Hamid
Muhammad Fareed Hamid
KTKamran Bashir KAMRAN BASHIR Taas
Kamran Bashir KAMRAN BASHIR Taas
MKMuhammad Attique Khan
Muhammad Attique Khan
LAlatifah almuqren
latifah almuqren
ZYZepa Yang
Zepa Yang
Aerial scene recognition has progressed substantially with deep learning methods for RGB and hyperspectral imagery; however, existing approaches typically operate on single modalities or rely on explicit multimodal fusion, limiting scalability, flexibility, and deployment in heterogeneous sensing environments. To address this limitation, we propose a sensor-agnostic semantic representation learning framework that formulates multimodal learning as the unification of semantic representations rather than feature-level fusion. The proposed architecture employs modality-specific encoders and projection heads to map spatial and spectral–spatial features into a shared semantic embedding space, enabling modality-invariant representation learning while preserving discriminative characteristics of each sensing modality. A composite objective integrating cross-spectral alignment, intra-class compactness regularization, and prototype-based semantic anchoring is introduced to enforce consistent embedding geometry and improve class separability across modalities. A unified classifier operating within this shared space enables reliable inference from a single modality input without requiring paired data or explicit fusion. Extensive evaluations on multiple benchmark datasets, including Houston 2013 for cross-modality RGB–hyperspectral analysis, UC Merced for independent RGB aerial scene classification, and Indian Pines for hyperspectral land-cover recognition, demonstrate the robustness and generalization capability of the proposed framework. In Houston 2013, the method achieves 96.4% (RGB) and 97.3% (hyperspectral) overall accuracy, with cross-modality transfer performance of 87.2% (RGB → HSI) and 88.7% (HSI → RGB), further improving to 97.0% and 97.8% under joint training. On UC Merced and Indian Pines, the model attains 98.7% and 97.6% overall accuracy, respectively. These results establish semantic representation unification as a scalable and effective alternative to conventional multimodal fusion for heterogeneous remote sensing environments.

7 April 2026 in Frontiers in Oncology
Verified
Journal Article
Open access
PDF file
Muhammad Zaheer SajidMuhammad Zaheer Sajid
Muhammad Zaheer Sajid
MHMuhammad Fareed Hamid
Muhammad Fareed Hamid
ZYZepa Yang
Zepa Yang
MAMohammad Alhefdi
Mohammad Alhefdi
SAShrooq Alsenan
Shrooq Alsenan
YNYunyoung Nam
Yunyoung Nam
Accurate and reliable skin cancer detection from dermoscopic images remains challenging due to large visual variability, overlapping lesion appearances, and inherent clinical uncertainty. To address these issues, this work proposes a morphology-guided attention framework for explainable and uncertainty-aware skin lesion classification. The system integrates lesion segmentation to preserve clinically meaningful morphological structures, followed by an attention-based classification network that emphasizes diagnostically relevant regions while suppressing background artifacts. Visual attention and attribution maps are generated to provide transparent explanations aligned with established dermoscopic criteria. In addition, an uncertainty estimation module is incorporated to quantify prediction confidence and identify ambiguous or out-of-distribution cases for safe clinical triage. The proposed approach is evaluated on publicly available dermoscopic datasets and achieves classification accuracy 99.12% with a recall rate above 99% for malignant lesions, demonstrating strong sensitivity for early cancer detection. Experimental results show that morphology-guided attention improves both classification performance and interpretability compared to conventional deep learning models. Furthermore, uncertainty-aware predictions enhance model reliability by reducing overconfident errors in challenging cases. These findings indicate that the proposed framework offers a robust, explainable, and clinically relevant solution for automated skin cancer screening under real-world conditions.

3 June 2026 in Bioengineering
Verified
Journal Article
Open access
PDF file
MBMaria Altaib Badawi
Maria Altaib Badawi
EAEhtisham Arshad
Ehtisham Arshad
AAArmughan Ali
Armughan Ali
OSOumaima Saidani
Oumaima Saidani
Taoufik SaidaniTaoufik Saidani
Taoufik Saidani
Yunyoung NamYunyoung Nam
Yunyoung Nam
Breast cancer remains the leading cause of cancer-related deaths among women globally. Early detection through mammography is vital for improving survival rates; however, the large volume of medical images and subtle variations in lesion characteristics pose significant challenges to manual interpretation. Recent automated diagnostic models based on deep learning have shown strong potential for breast cancer classification, but challenges such as overfitting, high computational complexity, limited generalization, and insufficient interpretability remain unresolved. This paper proposes a computationally efficient and context-aware deep learning framework for breast cancer classification using transformer-based multi-scale attention mechanisms and explainable artificial intelligence (XAI). The proposed architecture integrates the Hierarchical Multi-Scale Transformer (HMT) and Edge-Aware Local Transformer (ELT) modules to jointly capture global contextual dependencies and boundary-sensitive local representations from mammographic images. ELT improves feature refinement in high-entropy regions, while HMT models global semantic interactions across multiple feature scales. In addition, an Adaptive Contextual Refinement (ACR) module is introduced to preserve semantically consistent feature representations across spatial resolutions. A Meta-Ensemble Classification (MEC) framework combining weighted SVM and K-Nearest Neighbors (KNN) classifiers is further employed using validation-guided class-adaptive weighting. The proposed framework is evaluated on four benchmark mammography datasets, namely CBIS-DDSM, DDSM, INBreast, and MIAS. The proposed model has demonstrated superior accuracy of over 99% across all breast cancer datasets. The model surpassed transformer-based baselines including Swin-T and ViT while maintaining lower parameter complexity and achieving approximately 7% higher accuracy on the CBIS-DDSM dataset. The proposed framework also demonstrated strong cross-dataset generalization and consistently achieved high precision, recall, and F1-score values across all benchmark datasets. To improve model interpretability, Grad-CAM, SHAP, Occlusion Sensitivity Analysis (OSA), and the proposed TIxAI consistency analysis framework are incorporated to provide preliminary explainability assessment for mammographic classification. The explainability analysis demonstrated spatially consistent saliency behavior across benchmark datasets; however, the current evaluation is based on internal saliency consistency rather than external clinical validation using expert lesion annotations. Overall, the proposed framework provides an effective and computationally efficient approach for automated breast cancer classification while improving model explainability and interpretability.

1 January 2026 in Computers, Materials & Continua
Verified
Journal Article
Open access
PDF file
MZMuhammad Naeem Zafar
Muhammad Naeem Zafar
YYYunfei Yin
Yunfei Yin
JAJunaid Abbas
Junaid Abbas
BABayan Alabdullah
Bayan Alabdullah
KAKhaled Alnowaiser
Khaled Alnowaiser
ZYZepa Yang
Zepa Yang
Accurate medical image classification increasingly relies on the joint modeling of global contextual semantics and fine-grained local structural cues, since many lesions are only reliably recognized when subtle local details are interpreted within their broader anatomical context. However, most recent hybrid CNN–Transformer and global–local frameworks still extract these features in separate streams and merge them only through late-stage static fusion, without explicit bidirectional interaction during representation learning. As a result, global context cannot effectively guide the refinement of subtle local structures, and local discriminative cues cannot recalibrate higher-level semantic reasoning before classification, which limits reciprocal feature refinement and undermines robust integration of contextual and structural information in challenging medical images. To address this limitation, we propose BIAC-Net, a bidirectional global–local feature communication network for medical image classification. BIAC-Net employs a Vision Transformer (ViT) backbone to capture contextual representations and introduces two complementary refinement branches that enhance global saliency and preserve local structural details. Unlike conventional hybrid designs that merge features directly, the proposed framework introduces a bidirectional cross-stream residual refinement mechanism between global and local representations, allowing contextual features to guide local refinement while local structural cues recalibrate global reasoning before fusion. The reciprocally refined representations are then integrated through an adaptive gated fusion mechanism that learns input-dependent feature weighting. Experimental results on the Kvasir and ISIC 2018 datasets demonstrate that BIAC-Net achieves 97.84% and 90.91% accuracy, respectively, outperforming several representative baseline models. Ablation studies further support the contribution of bidirectional communication, while Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations provide preliminary qualitative evidence that the model attends to diagnostically relevant image regions.

23 June 2026 in Journal of International Medical Research
Verified
Journal Article
Open access
BFBushra Fiaz
Bushra Fiaz
MKMuhammad Attique Khan
Muhammad Attique Khan
AZAfia Zafar
Afia Zafar
SAShrooq Alsenan
Shrooq Alsenan
YAYazan Alnsour
Yazan Alnsour
YNYunyoung Nam
Yunyoung Nam
Objective: Skin cancer diagnosis via automated image analysis remains a challenging task due to poor image contrast, visual similarity among lesion classes, and class imbalance in available datasets. To address these issues, this study proposes a novel Tri Model Dermatology Cancer Neural Network (TriDermCancerNet) for classifying skin cancer from dermoscopic images. Methods: Two publicly available datasets are used in this work: the International Skin Imaging Collaboration 2018 and 2019 datasets, which contain multiple classes of cancer types. In the proposed model framework, a contrast enhancement technique was applied to improve image quality, followed by data augmentation to balance the datasets. The proposed TriDermCancerNet architecture is designed based on the key challenges of this work, including dataset variability, interclass similarity, and model explainability. The proposed architecture integrates three modules: a 105-layer Inception module, a 186-layer Inverted Bottleneck Residual module, and the Dense-177 module. Each module was integrated into the proposed network in parallel rather than in series. Thereafter, each module was trained, and features were extracted and fused using a depth concatenation approach. During training, several important hyperparameters were selected using Bayesian optimization, and the final model was used for classification in the testing phase. Results: The fused TriDermCancerNet achieved 98.6% accuracy, 98.6% sensitivity, 98.6% F1-score, and an area under the curve of 1.0 on the International Skin Imaging Collaboration 2018 dataset and 99.7% accuracy, 99.6% sensitivity, 99.6% F1-score, and an area under the curve of 1.0 on the International Skin Imaging Collaboration 2019 dataset. Statistical significance testing confirmed that the fusion model outperforms each branch (p < 0.05). Conclusion: The proposed hybrid TriDermCancerNet approach enhances the precision and robustness of skin cancer classification frameworks, providing clinicians with a valuable diagnostic aid for early detection.

25 September 2026 in Journal of King Saud University Computer and Information Sciences
Verified
Journal Article
Open access
PDF file
JLJiawen Li
Jiawen Li
LCLing Chen
Ling Chen
XCXuanzhong Chen
Xuanzhong Chen
LAlatifah almuqren
latifah almuqren
ZYZepa Yang
Zepa Yang
XHXianglei Hu
Xianglei Hu
The escalating global burden of neuropsychological disorders demands objective and automated assessment tools, as traditional diagnostic methods often rely on subjective expert interpretation. Electroencephalography (EEG) offers a promising solution due to its non-invasiveness, cost-effectiveness, and high temporal resolution for capturing neural dynamics. Given the importance of long-term monitoring in managing these conditions, there is a critical need to develop portable and user-friendly EEG-based devices. In this context, channel selection is essential for minimizing hardware complexity and enhancing portability. To address this, we propose a novel framework utilizing an Adam-optimized adaptive feature fusion mechanism. Key multi-domain features, including Wavelet Convolution Standard Deviation (WCSD), Wavelet Time–Frequency Convolution Standard Deviation (WTF-CSD), spectral entropy, and waveform factor, are extracted to characterize complex neural dynamics. An adaptive weighting mechanism, optimized via the Adam algorithm, integrates these features into a discriminative representation, which is then evaluated using five machine learning classifiers: Support Vector Machine (SVM), Naive Bayes (NB), Generalized Additive Model (GAM), K-Nearest Neighbors (KNN), and Decision Tree (DT). Leave-One-Out Cross-Validation (LOOCV) is performed on public datasets for Schizophrenia (SZ) and Alzheimer's Disease (AD). Ablation experiments confirm that the adaptive fusion mechanism contributes significantly to performance gains. Furthermore, cross-dataset generalization is validated by training on one AD dataset (AHEPA) and testing directly on another (Florida-Based) without retraining. These results demonstrate that this approach achieves impressive performance using single-channel EEG data with the DT classifier. Specifically, the P4 channel achieves 81.90% ± 2.29% accuracy for the SZ dataset, while the T5 channel achieves 84.62% ± 1.09% accuracy for the AD dataset (AHEPA). By integrating Adam optimization into feature fusion, the proposed method eliminates multi-channel dependencies and minimizes electrode requirements for data-limited scenarios. Consequently, it offers a practical solution for developing a portable EEG-based diagnostic system.