Javokhir Musaev
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Javokhir Musaev has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Dosam Hwang
Yeungnam University
17 shared publications - Abdulaziz Anorboev
Yeungnam University
12 shared publications - Ngoc Thanh Nguyen
Wrocław University of Science and Technology
12 shared publications - Yeong-Seok Seo
Yeungnam University
10 shared publications - Jeongkyu Hong
University of Seoul
7 shared publications - Sarvinoz Anorboeva
Yeungnam University
4 shared publications
Latest Publications
- 21 September 2026
- Published by Springer Nature
- p. 183-197
Learning to Fuse Multi-enhanced YOLO26n Predictions via MLP for Improved Landmark Recognition
- 21 September 2026
- Published by Springer Nature
- p. 183-197
Learning to Fuse Multi-enhanced YOLO26n Predictions via MLP for Improved Landmark Recognition
- Open Access
- 21 August 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (8) , 397
Landmark Recognition Beyond Curated Benchmarks: Cross-Domain Evaluation of a Multi-Threshold Selective YOLO11 Ensemble on User-Generated Imagery, with a Zero-Shot Multimodal LLM Baseline
Landmark recognition for smart tourism is usually validated on curated benchmark images. In deployment, however, the classifier must handle user-generated photographs whose viewpoint, lighting, resolution, occlusion, and compression differ sharply from curated data. This paper evaluates a previously published multi-threshold enhancement and selective YOLO11n-cls ensemble under this shift, and provides a preliminary zero-shot comparison of three general-purpose multimodal large language models (MLLMs) on the same task. To measure the shift, we build Samarkand v2-SNS, a 300-image out-of-distribution test set of social-media photographs of 12 Samarkand landmarks, disjoint from the training and validation data. Under the shift, four supervised baselines fall by 12.73–22.08 percentage points to 73–80% accuracy, and their in-distribution ranking does not hold. The selective ensemble degrades least (99.24% to 93.00%, −6.24 points) and outperforms the strongest baseline by 13 points. A capacity-matched ablation shows that most of this robustness comes from enhancement diversity, not from generic ensembling. In a preliminary comparison, zero-shot MLLMs (GPT-5, Claude Sonnet 4.5, Gemini 2.5) reach only 24.81–54.26%, far below deployment needs. The results argue for reporting out-of-distribution accuracy alongside curated benchmarks, and for hybrid systems that pair compact specialised recognisers with MLLM-based interpretation.
- 21 August 2026
- Published by MDPI AG in Journal of Imaging
- Vol. 12 (8) , 397
Open AccessLandmark Recognition Beyond Curated Benchmarks: Cross-Domain Evaluation of a Multi-Threshold Selective YOLO11 Ensemble on User-Generated Imagery, with a Zero-Shot Multimodal LLM Baseline
Landmark recognition for smart tourism is usually validated on curated benchmark images. In deployment, however, the classifier must handle user-generated photographs whose viewpoint, lighting, resolution, occlusion, and compression differ sharply from curated data. This paper evaluates a previously published multi-threshold enhancement and selective YOLO11n-cls ensemble under this shift, and provides a preliminary zero-shot comparison of three general-purpose multimodal large language models (MLLMs) on the same task. To measure the shift, we build Samarkand v2-SNS, a 300-image out-of-distribution test set of social-media photographs of 12 Samarkand landmarks, disjoint from the training and validation data. Under the shift, four supervised baselines fall by 12.73–22.08 percentage points to 73–80% accuracy, and their in-distribution ranking does not hold. The selective ensemble degrades least (99.24% to 93.00%, −6.24 points) and outperforms the strongest baseline by 13 points. A capacity-matched ablation shows that most of this robustness comes from enhancement diversity, not from generic ensembling. In a preliminary comparison, zero-shot MLLMs (GPT-5, Claude Sonnet 4.5, Gemini 2.5) reach only 24.81–54.26%, far below deployment needs. The results argue for reporting out-of-distribution accuracy alongside curated benchmarks, and for hybrid systems that pair compact specialised recognisers with MLLM-based interpretation.
- Open Access
- 15 August 2026
- Published by Elsevier in SSRN Electronic Journal
Why Lightweight Convolutional Neural Networks Struggle with Small Datasets: A Multi-Scale Analysis and a Dual-Attention Solution
Lightweight convolutional neural networks are commonly selected for image classification on edge devices based on their ImageNet rankings. However, no prior studies have systematically examined whether these rankings remain valid for small-scale, domain-specific tasks. We address this gap and demonstrate that rankings can completely reverse: ShuffleNetV2, the weakest model among four representative baselines on ImageNet, becomes the strongest on our benchmarks, highlighting feature diversity rather than model capacity as the critical factor when training data is limited. Our analysis further identifies two architectural limitations that become pronounced in small-data regimes: (i) Ghost-based architectures extract features at a single spatial scale per block, lacking the multi-granularity context necessary for fine-grained discrimination, and (ii) squeeze-and-excitation (SE) attention introduces 2𝐶2∕𝑟 parameters per block, imposing a disproportionate overhead when the total model budget is only a few million parameters. To jointly address these limitations, we propose the efficient channel attention multi-scale ghost network (ECA-MGNet). This 2.50 million-parameter architecture augments a pretrained GhostNet backbone with two novel modules: a multi-scale refinement block, which injects multi-granularity context into the ghost framework via parallel 1 × 1, 3 × 3, 5 × 5, and global pooling branches, and a dual attention module, which replaces SE with ECA paired with spatial attention, reducing attention parameters by approximately 92,000× to fewer than 200. On a 10-class Flowers102 subset and EuroSAT (European satellite image dataset), ECA-MGNet matches or exceeds the accuracy of EfficientNetV2-Small and ConvNeXt-Tiny, while outperforming all lightweight baselines by 7.55 percentage points (𝑝 < 0.01, McNemar’s test).
- 15 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessWhy Lightweight Convolutional Neural Networks Struggle with Small Datasets: A Multi-Scale Analysis and a Dual-Attention Solution
Lightweight convolutional neural networks are commonly selected for image classification on edge devices based on their ImageNet rankings. However, no prior studies have systematically examined whether these rankings remain valid for small-scale, domain-specific tasks. We address this gap and demonstrate that rankings can completely reverse: ShuffleNetV2, the weakest model among four representative baselines on ImageNet, becomes the strongest on our benchmarks, highlighting feature diversity rather than model capacity as the critical factor when training data is limited. Our analysis further identifies two architectural limitations that become pronounced in small-data regimes: (i) Ghost-based architectures extract features at a single spatial scale per block, lacking the multi-granularity context necessary for fine-grained discrimination, and (ii) squeeze-and-excitation (SE) attention introduces 2𝐶2∕𝑟 parameters per block, imposing a disproportionate overhead when the total model budget is only a few million parameters. To jointly address these limitations, we propose the efficient channel attention multi-scale ghost network (ECA-MGNet). This 2.50 million-parameter architecture augments a pretrained GhostNet backbone with two novel modules: a multi-scale refinement block, which injects multi-granularity context into the ghost framework via parallel 1 × 1, 3 × 3, 5 × 5, and global pooling branches, and a dual attention module, which replaces SE with ECA paired with spatial attention, reducing attention parameters by approximately 92,000× to fewer than 200. On a 10-class Flowers102 subset and EuroSAT (European satellite image dataset), ECA-MGNet matches or exceeds the accuracy of EfficientNetV2-Small and ConvNeXt-Tiny, while outperforming all lightweight baselines by 7.55 percentage points (𝑝 < 0.01, McNemar’s test).
- Open Access
- 28 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 138020-138036
IMed-CNN: Ensemble Learning Approach With Systematic Model Dropout for Enhanced Medical Image Classification Using Image Channels and Pixel Intervals
To obtain light ensemble model through clearly explained effective ensemble member selection and finding data representation in various valuable forms are major challenges in medical image classification tasks. Despite numerous deep learning (DL) models were developed to solve the generalization problem in DL through ensemble learning, most of the models lacks the evaluation of the effects of each ensemble member for the final ensemble model. Additionally, existing ensemble models tend to include huge number of parameters and use images only in RGB channels or greyscale forms missing the crucial representations of the dataset to reach robust classification outcomes, particularly in medical applications. In this study, novel ensemble model IMed-CNN proposed solution to above-mentioned gaps in the field by introducing the data in ten various forms to the model and applying systematic model dropout (SMDE) with unique true prediction (UTP) analysis which insures to choose only useful ensemble members. To verify the performance of the IMed-CNN, extensive experiments was designed for testing. Results illustrate that IMed-CNN outperformed baseline models.
- 28 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 138020-138036
Open AccessIMed-CNN: Ensemble Learning Approach With Systematic Model Dropout for Enhanced Medical Image Classification Using Image Channels and Pixel Intervals
To obtain light ensemble model through clearly explained effective ensemble member selection and finding data representation in various valuable forms are major challenges in medical image classification tasks. Despite numerous deep learning (DL) models were developed to solve the generalization problem in DL through ensemble learning, most of the models lacks the evaluation of the effects of each ensemble member for the final ensemble model. Additionally, existing ensemble models tend to include huge number of parameters and use images only in RGB channels or greyscale forms missing the crucial representations of the dataset to reach robust classification outcomes, particularly in medical applications. In this study, novel ensemble model IMed-CNN proposed solution to above-mentioned gaps in the field by introducing the data in ten various forms to the model and applying systematic model dropout (SMDE) with unique true prediction (UTP) analysis which insures to choose only useful ensemble members. To verify the performance of the IMed-CNN, extensive experiments was designed for testing. Results illustrate that IMed-CNN outperformed baseline models.
- Open Access
- 28 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 101609-101623
Selective Intensity Ensemble Classifier (SIEC): A Triple-Threshold Strategy for Microscopic Malaria Cell Image Classification
Accurate malaria detection in resource-limited settings requires robust solutions—here, we introduce a selective intensity ensemble classifier (SIEC) that applies a triple-threshold strategy for enhanced microscopic image classification. This involves training three separate convolutional neural network models on the same images processed with different pixel-intensity thresholds: original, pixels above 100, and pixels above 200. This approach enables the ensemble to capture complementary low-, mid-, and high-intensity features, enhancing feature diversity and classification accuracy. The experiments were conducted on the publicly available Malaria Cell Dataset, consisting of 27,558 images. The proposed SIEC achieved an accuracy of 95.09%, with a precision of 95.27%, and matching recall and F1 scores of 95.09%, consistently outperforming six standard CNN models, including ResNet50, VGG16, Inception, and MobileNetV2. Notably, the combination of the 100-pixel filtered and original images yielded the highest classification performance, demonstrating the ensemble’s ability to integrate detailed and abstracted features effectively. These findings highlight SIEC as a promising and scalable solution for automated malaria detection and broader diagnostic imaging tasks.
- 28 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 101609-101623
Open AccessSelective Intensity Ensemble Classifier (SIEC): A Triple-Threshold Strategy for Microscopic Malaria Cell Image Classification
Accurate malaria detection in resource-limited settings requires robust solutions—here, we introduce a selective intensity ensemble classifier (SIEC) that applies a triple-threshold strategy for enhanced microscopic image classification. This involves training three separate convolutional neural network models on the same images processed with different pixel-intensity thresholds: original, pixels above 100, and pixels above 200. This approach enables the ensemble to capture complementary low-, mid-, and high-intensity features, enhancing feature diversity and classification accuracy. The experiments were conducted on the publicly available Malaria Cell Dataset, consisting of 27,558 images. The proposed SIEC achieved an accuracy of 95.09%, with a precision of 95.27%, and matching recall and F1 scores of 95.09%, consistently outperforming six standard CNN models, including ResNet50, VGG16, Inception, and MobileNetV2. Notably, the combination of the 100-pixel filtered and original images yielded the highest classification performance, demonstrating the ensemble’s ability to integrate detailed and abstracted features effectively. These findings highlight SIEC as a promising and scalable solution for automated malaria detection and broader diagnostic imaging tasks.
- 17 April 2025
- Published by Springer Nature
- p. 247-260
Vision-Based Malware Detection: Integrating CNN and MobileNetV2 for Dynamic Cybersecurity Challenges
- 17 April 2025
- Published by Springer Nature
- p. 247-260
Vision-Based Malware Detection: Integrating CNN and MobileNetV2 for Dynamic Cybersecurity Challenges
- 16 July 2024
- Published by Springer Nature
- p. 132-143
Enhancing Classification of Parasite Microscopy Images Through Image Edge-Accentuating Preprocessing
- 16 July 2024
- Published by Springer Nature
- p. 132-143
Enhancing Classification of Parasite Microscopy Images Through Image Edge-Accentuating Preprocessing
- Open Access
- 17 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 117685-117697
MICL-UNet: Multi-Input Cross-Layer UNet Model for Classification of Diseases in Agriculture
Agricultural diseases severely impact productivity and result in significant economic losses in the agricultural sector. Current monitoring practices are predominantly individualized, making them inefficient and challenging for farmers to manage on a large scale. Hence, there’s an urgent need for an effective solution to identify and classify these diseases swiftly and accurately. Addressing this need, we introduce the multi-input cross-layer UNet model (MICL-UNet). Enhanced with numerous residual connections, our model ensures minimal information loss from the input. Given the role of nonlinear activation functions at every layer, we strategically incorporated input data at multiple stages in the model, reinforcing it with terminal-layer residual connections. This design aims to distribute critical image features evenly across all layers. The MICL-UNet displayed superior performance over baseline models, achieving accuracy scores of 99.63% for Vegetable datasets, 99.25% for PLD datasets, and 98.7% for Guava Disease datasets. Beyond its academic implications, the MICL-UNet offers significant practical benefits. By improving disease detection accuracy, it promises to elevate agricultural product quality, reducing production uncertainties. This approach has the potential to foster more sustainable and resilient agricultural practices, paving the way for robust economic growth in the sector.
- 17 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 117685-117697
Open AccessMICL-UNet: Multi-Input Cross-Layer UNet Model for Classification of Diseases in Agriculture
Agricultural diseases severely impact productivity and result in significant economic losses in the agricultural sector. Current monitoring practices are predominantly individualized, making them inefficient and challenging for farmers to manage on a large scale. Hence, there’s an urgent need for an effective solution to identify and classify these diseases swiftly and accurately. Addressing this need, we introduce the multi-input cross-layer UNet model (MICL-UNet). Enhanced with numerous residual connections, our model ensures minimal information loss from the input. Given the role of nonlinear activation functions at every layer, we strategically incorporated input data at multiple stages in the model, reinforcing it with terminal-layer residual connections. This design aims to distribute critical image features evenly across all layers. The MICL-UNet displayed superior performance over baseline models, achieving accuracy scores of 99.63% for Vegetable datasets, 99.25% for PLD datasets, and 98.7% for Guava Disease datasets. Beyond its academic implications, the MICL-UNet offers significant practical benefits. By improving disease detection accuracy, it promises to elevate agricultural product quality, reducing production uncertainties. This approach has the potential to foster more sustainable and resilient agricultural practices, paving the way for robust economic growth in the sector.
- 22 September 2023
- Published by Springer Nature
- p. 175-188
Classifying Chicken-Made Food Images Using Enhanced MobilNetV2
- 22 September 2023
- Published by Springer Nature
- p. 175-188
Classifying Chicken-Made Food Images Using Enhanced MobilNetV2