Selected work

Representative Papers

Multi-Class Brain Tumor Classification Using Advanced Deep Learning Models: A Comparative Study

Jun 17, 2026

Accurate classification of multiple brain tumor types—particularly meningiomas, which exhibit subtle imaging features—remains challenging in MRI. This study systematically evaluates the performance of five convolutional neural network architectures, including a custom model and established networks (VGG16, VGG19, DenseNet121, and EfficientNetB0), on approximately 10,000 clinical MRI images within a unified experimental framework. Employing consistent transfer learning and fine-tuning strategies, the findings indicate that architectural efficiency outweighs model depth in determining performance. EfficientNetB0 achieves a markedly superior overall accuracy of 95% and, notably, elevates the recall for meningiomas from approximately 20% to 89% on large-scale clinical data—a substantial improvement that underscores its significant clinical utility.

0 citationsRead paper

A Hybrid YOLOv5-SSD IoT-Based Animal Detection System for Durian Plantation Protection

Nov 01, 2025

To address recurrent crop damage and economic losses in durian plantations caused by elephant, wild boar, and monkey intrusions, this paper proposes an end-to-end intelligent IoT-based protection system. The method integrates a hybrid object detection model combining YOLOv5 and SSD to enhance robustness in multi-species wildlife recognition; augments it with edge-based video analytics, real-time Telegram alerts, and directional acoustic deterrents to realize a fully automated “detection–alerting–response”闭环. Experimental results demonstrate daytime detection accuracies of 90%, 85%, and 70% for elephants, wild boars, and monkeys, respectively—substantially outperforming single-model baselines. The system requires minimal human intervention, exhibits high deployability on resource-constrained edge devices, and shows strong cross-scenario adaptability. This work establishes a scalable, practical technical paradigm for intelligent wildlife management in tropical orchards.

0 citationsRead paper

Two-Factor Authentication Smart Entryway Using Modified LBPH Algorithm

Aug 19, 2025

To address the lack of robust two-factor authentication and IoT interoperability in smart access control systems under pandemic-related mask-wearing conditions, this work designs and implements a lightweight edge-based access control system built on Raspberry Pi. We propose an enhanced Local Binary Patterns Histograms (LBPH) algorithm to improve facial feature extraction under partial occlusion, and integrate face-plus-PIN two-factor authentication with Telegram Bot API for remote alerting, real-time surveillance coordination, and user registration management. All recognition and response operations are executed locally at the edge, achieving an average accuracy of 70%, precision of 80%, and recall of 83.26%. The system demonstrates high user acceptance and practical deployability. Key contributions include: (1) an occlusion-aware LBPH optimization tailored for masked-face recognition; and (2) a unified edge architecture that synergistically combines bi-modal authentication with IoT-enabled remote management.

0 citationsRead paper

Prognosis Of Lithium-Ion Battery Health with Hybrid EKF-CNN+LSTM Model Using Differential Capacity

Apr 16, 2025

Lithium-ion batteries exhibit unclear degradation mechanisms and low State of Health (SOH) estimation accuracy under multi-rate operating conditions. Method: This paper proposes a differential capacity analysis (DCA)-based degradation modeling and prediction framework. We innovatively design an EKF-CNN-LSTM hybrid model to enable DCA-feature-driven, end-to-end SOH estimation—the first such approach. A Peak Identification Method (PIM) is introduced to quantify multidimensional degradation indicators, and systematic DCA-based analysis reveals accelerated aging mechanisms under fast charging (0.2C–1.5C) and discharging (0.5C–1.6C). Results: Experiments demonstrate that the proposed method achieves MSE and RMSE both below 0.001%, significantly outperforming state-of-the-art approaches. Furthermore, comparative analysis confirms superior degradation robustness of LiFePO₄ over NCA across wide load ranges.

0 citationsRead paper

Recent publications

Latest Papers

Multi-Class Brain Tumor Classification Using Advanced Deep Learning Models: A Comparative Study

Jun 17, 2026

Accurate classification of multiple brain tumor types—particularly meningiomas, which exhibit subtle imaging features—remains challenging in MRI. This study systematically evaluates the performance of five convolutional neural network architectures, including a custom model and established networks (VGG16, VGG19, DenseNet121, and EfficientNetB0), on approximately 10,000 clinical MRI images within a unified experimental framework. Employing consistent transfer learning and fine-tuning strategies, the findings indicate that architectural efficiency outweighs model depth in determining performance. EfficientNetB0 achieves a markedly superior overall accuracy of 95% and, notably, elevates the recall for meningiomas from approximately 20% to 89% on large-scale clinical data—a substantial improvement that underscores its significant clinical utility.

0 citationsRead paper

A Hybrid YOLOv5-SSD IoT-Based Animal Detection System for Durian Plantation Protection

Nov 01, 2025

To address recurrent crop damage and economic losses in durian plantations caused by elephant, wild boar, and monkey intrusions, this paper proposes an end-to-end intelligent IoT-based protection system. The method integrates a hybrid object detection model combining YOLOv5 and SSD to enhance robustness in multi-species wildlife recognition; augments it with edge-based video analytics, real-time Telegram alerts, and directional acoustic deterrents to realize a fully automated “detection–alerting–response”闭环. Experimental results demonstrate daytime detection accuracies of 90%, 85%, and 70% for elephants, wild boars, and monkeys, respectively—substantially outperforming single-model baselines. The system requires minimal human intervention, exhibits high deployability on resource-constrained edge devices, and shows strong cross-scenario adaptability. This work establishes a scalable, practical technical paradigm for intelligent wildlife management in tropical orchards.

0 citationsRead paper

Two-Factor Authentication Smart Entryway Using Modified LBPH Algorithm

Aug 19, 2025

To address the lack of robust two-factor authentication and IoT interoperability in smart access control systems under pandemic-related mask-wearing conditions, this work designs and implements a lightweight edge-based access control system built on Raspberry Pi. We propose an enhanced Local Binary Patterns Histograms (LBPH) algorithm to improve facial feature extraction under partial occlusion, and integrate face-plus-PIN two-factor authentication with Telegram Bot API for remote alerting, real-time surveillance coordination, and user registration management. All recognition and response operations are executed locally at the edge, achieving an average accuracy of 70%, precision of 80%, and recall of 83.26%. The system demonstrates high user acceptance and practical deployability. Key contributions include: (1) an occlusion-aware LBPH optimization tailored for masked-face recognition; and (2) a unified edge architecture that synergistically combines bi-modal authentication with IoT-enabled remote management.

0 citationsRead paper

Prognosis Of Lithium-Ion Battery Health with Hybrid EKF-CNN+LSTM Model Using Differential Capacity

Apr 16, 2025

Lithium-ion batteries exhibit unclear degradation mechanisms and low State of Health (SOH) estimation accuracy under multi-rate operating conditions. Method: This paper proposes a differential capacity analysis (DCA)-based degradation modeling and prediction framework. We innovatively design an EKF-CNN-LSTM hybrid model to enable DCA-feature-driven, end-to-end SOH estimation—the first such approach. A Peak Identification Method (PIM) is introduced to quantify multidimensional degradation indicators, and systematic DCA-based analysis reveals accelerated aging mechanisms under fast charging (0.2C–1.5C) and discharging (0.5C–1.6C). Results: Experiments demonstrate that the proposed method achieves MSE and RMSE both below 0.001%, significantly outperforming state-of-the-art approaches. Furthermore, comparative analysis confirms superior degradation robustness of LiFePO₄ over NCA across wide load ranges.

0 citationsRead paper