Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Kavimbi Chipusu has a total of 23 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    UAV-based tree species classification using DenseNet121 with transfer learning on visible light images

    by Wei Zeng, Xianxian Luo, Jianbing Xiahou, Jianlong Huang, Kavimbi Chipusu, Muhammad Awais Ashraf, Songya Xu

    This study introduces a cutting-edge approach to tree species classification in environmental monitoring by leveraging UAV-based visible light imagery. Focusing exclusively on visible light images captured by unmanned aerial vehicles (UAVs), we address the limitations of traditional classification methods with the powerful DenseNet121 deep learning architecture. By harnessing DenseNet121's dense connectivity, the model significantly enhances feature extraction and information flow, resulting in superior performance for complex image classification tasks. UAV-captured visible light images of diverse tree species were processed and used to create a robust dataset for model training. To further optimize performance, transfer learning was employed to fine-tune the pre-trained DenseNet121 model for the specific tree species dataset. Experimental results reveal that DenseNet121, combined with transfer learning, outperforms a range of well-established models such as ResNet18, ResNet50, VGG16, VGG19, MobileNet, EfficientNet-B0, MobileNetV3, ConvNeXt-Tiny, EdgeNeXt, and Swin-Tiny, delivering exceptional classification accuracy as measured by precision, recall, and F1-score metrics. This work underscores the transformative potential of combining pre-trained architectures like DenseNet121 with transfer learning for processing UAV-based visible light imagery, offering a highly effective and scalable solution for tree species classification. It establishes a reliable, cost-effective baseline for forestry monitoring and provides an invaluable tool for environmental and ecological studies, especially in data-scarce scenarios.

  • Open Access

    Physiological classification of Parkinson’s disease severity using multimodal speech biomarkers with a hybrid CNN-Mamba framework

    by Taisheng Zeng, Yuguang Ye, Yunyi Zeng, Jianshe Shi, Yifeng Huang, Bijiao Ding, Kavimbi Chipusu, Jianlong Huang

    Introduction: Hypokinetic dysarthria in Parkinson’s disease provides an accessible non-invasive biomarker, but multi-class severity grading remains difficult because of overlapping acoustic patterns and limited long-range temporal modeling in existing approaches. Methods: We developed a hybrid CNN-Mamba framework using multimodal speech features transformed into 2D representations. The model was trained and validated on speaker-disjoint PC-GITA Spanish data and tested on an independent Mandarin clinical cohort, with additional external evaluation on a public Parkinsonian speech corpus. Speaker-level results were obtained by aggregating segment predictions within each subject. Results: Segment-level accuracy reached 97.8% on PC-GITA and 95.4% on the Mandarin cohort. Speaker-level accuracy reached 94.0% and 91.2% using majority voting, improving to 94.8% and 91.9% with mean-probability aggregation. SHAP analysis supported physiological interpretability, and ablation studies showed advantages over CNN-BiLSTM, Transformer, and SVM baselines. Discussion: The proposed CNN-Mamba framework provides an interpretable, computationally efficient, and non-invasive approach for Parkinson’s disease severity assessment and remote monitoring, with promising cross-lingual transfer under structured clinical speech tasks.


  • Open Access

    Robust ventricular segmentation in cardiac MRI via fused attention and capsule networks

    by Yuguang Ye, Taisheng Zeng, Kavimbi Chipusu, Bijiao Ding, Yifeng Huang, Jingwei Guo, Jianlong Huang

    Objectives: An effective automatic system for ventricular segmentation from MRI is vital for diagnosing cardiovascular diseases, yet challenges persist due to anatomical variations and artifacts. Methods: We preprocess cardiac MRI with min-max normalization, then propose a hybrid segmentation network (ResFAU-net) integrating residual blocks, attention gates, and a Fused Accumulation Bridge module to delineate ventricle boundaries. The segmented regions are classified by the HAMC3 model, which combines cascaded capsule networks, CNNs, and hierarchical attention, with parameters optimized via the Coati Optimization Algorithm (COA). Results: Rigorous assessment on our CMRI dataset using metrics (Dice, IoU, accuracy, precision, etc.) demonstrates the model’s high performance in segmenting and classifying the left and right ventricles achieving an IoU of 96.29 % and accuracy of 99.03 %. Conclusions: The proposed ResFAU-net and HAMC3 framework offers a robust, end-to-end solution for precise ventricular cardiac analysis, demonstrating strong potential to automate and enhance the efficiency of cardiovascular diagnosis in clinical MRI workflows.


  • Top Cited Papers
    Open Access

    Large Language Models Evaluation of Medical Licensing Examination Using GPT-4.0, ERNIE Bot 4.0, and GPT-4o

    by Luoyu Lian, Xin Luo, Kavimbi Chipusu, Muhammad Awais Ashraf, Kelvin K. L. Wong, Wenjun Zhang

    This study systematically evaluated the performance of three advanced large language models (LLMs)—GPT-4.0, ERNIE Bot 4.0, and GPT-4o—in the 2023 Chinese Medical Licensing Examination. Employing a dataset of 600 standardized questions, we analyzed the accuracy of each model in answering questions from three comprehensive sections: Basic Medical Comprehensive, Clinical Medical Comprehensive, and Humanities and Preventive Medicine Comprehensive. Our results demonstrate that both ERNIE Bot 4.0 and GPT-4o significantly outperformed GPT-4.0, achieving accuracies above the national pass mark. The study further examined the strengths and limitations of each model, providing insights into their applicability in medical education and potential areas for future improvement. These findings underscore the promise and challenges of deploying LLMs in multilingual medical education, suggesting a pathway towards integrating AI into medical training and assessment practices.


  • Open Access

    A Hybrid AI Framework for Integrated Predictive Maintenance and Mineral Quality Assessment in Mining

    by Wanji Mwale, Zhixiang Liu, Kavimbi Chipusu

    In the mining industry, operational efficiency, equipment reliability, and mineral quality assessment are paramount for cost-effective and sustainable production. Traditional approaches often address equipment maintenance and quality control as separate challenges, leading to suboptimal operational synergy. This paper proposes a novel artificial intelligence (AI) framework that integrates predictive maintenance with real-time mineral quality assessment through advanced sensor fusion and deep learning. Our model leverages a hybrid architecture, combining Convolutional Neural Networks (CNNs) for analyzing visual and spectral data of iron ore with Long Short-Term Memory (LSTM) networks for processing temporal sensor data (vibration, thermal, acoustic) from critical equipment like crushers and conveyors. A dedicated fusion layer synthesizes these spatial and temporal features to simultaneously predict equipment failure probability and classify mineral quality. Validated on a real-world dataset from active iron ore mines, the system demonstrates a significant 20–30% reduction in projected maintenance downtime and a 15% improvement in mineral classification accuracy compared to baseline models while achieving real-time inference speeds of less than 10 milliseconds. This work underscores the transformative potential of unified AI-driven systems in enhancing the intelligence, resilience, and productivity of modern mining operations.


  • Top Cited Papers
    Open Access

    Hybrid CNN-BLSTM architecture for classification and detection of arrhythmia in ECG signals

    by Yuguang Ye, Kavimbi Chipusu, Muhammad Awais Ashraf, Bijiao Ding, Yifeng Huang, Jianlong Huang

    This study introduces a robust and efficient hybrid deep learning framework that integrates Convolutional Neural Networks (CNN) with Bidirectional Long Short-Term Memory (BLSTM) networks for the automated detection and classification of cardiac arrhythmias from electrocardiogram (ECG) signals. The proposed architecture leverages the complementary strengths of both components: the CNN layers autonomously learn and extract salient morphological features from raw ECG waveforms, while the BLSTM layers effectively model the sequential and temporal dependencies inherent in ECG signals, thereby improving diagnostic accuracy. To further enhance training stability and non-linear representation capability, the Mish activation function is incorporated throughout the network. The model was trained and evaluated using a combination of the widely recognized MIT-BIH Arrhythmia Database and de-identified clinical ECG recordings sourced from collaborating healthcare institutions, ensuring both diversity and clinical relevance of the dataset. Notably, the framework operates with minimal preprocessing, underscoring its practical viability for real-time implementation. Experimental results demonstrate the model’s exceptional performance, achieving an overall classification accuracy of 99.52%, sensitivity of 99.48%, and specificity of 99.85%. These outcomes highlight the model’s robustness, generalizability, and strong potential for integration into clinical decision support systems, particularly in high-throughput or resource-constrained healthcare environments.