Purpose: Gastrointestinal (GI) illness demands precise and efficient diagnostics, yet conventional approaches (e.g., endoscopy and histopathology) are time-consuming and prone to reader variability. This work presents GID-Xpert, a deep learning framework designed to improve feature learning, accuracy, and interpretability for GI disease classification. Methods: GID-Xpert integrates a hierarchical, multi-stage attention-driven mixture of experts with dynamic routing. The architecture couples spatial–channel attention mechanisms with specialized expert blocks; a routing module adaptively selects expert paths to enhance representation quality and reduce redundancy. The model is trained and evaluated on three benchmark datasets—WCEBleedGen, GastroEndoNet, and the King Abdulaziz University Hospital Capsule (KAUHC) dataset. Comparative experiments against state-of-the-art baselines and ablation studies (removing attention, expert blocks, and routing) are conducted to quantify the contribution of each component. Results: GID-Xpert achieves superior performance with 100% accuracy on WCEBleedGen, 99.98% on KAUHC, and 75.32% on GastroEndoNet. Comparative evaluations show consistent improvements over contemporary models, while ablations confirm the additive benefits of spatial–channel attention, expert specialization, and dynamic routing. The design also yields reduced computational cost and improved explanation quality via attention-driven reasoning. Conclusion: By unifying attention, expert specialization, and dynamic routing, GID-Xpert delivers accurate, computationally efficient, and more interpretable GI disease classification. These findings support GID-Xpert as a credible diagnostic aid and a strong foundation for future extensions toward broader GI pathologies and clinical integration.

Wildfires are among the most devastating natural disasters, causing significant damage to the environment, the economy, wildlife, and human lives, and requiring an early detection system. This paper presents a novel LightFire-Net deep cross model fusion architecture for Wildfire classification and interpretation from remote sensing images. The proposed model is computationally efficient and employs Efficient Flame Pattern blocks with depthwise separable convolutional layers, a Lightweight spectral spatial module, and Efficient Feature Fusion for real-time wildfire classification and detection. The model's multi-stage architecture enables hierarchical feature extraction, resulting in more accurate and efficient classification. Bayesian Optimization (BO) is employed to select hyperparameters, thereby improving the training of the proposed model on selected remote sensing datasets. A Softmax classification layer is employed for the final prediction. The proposed model achieves exceptional performance with above 98% prediction accuracy while maintaining only 1.37 million parameters and an inference speed of 7.74 ms. A comprehensive evaluation across multiple dataset splits (90:05:05-40:30:30) demonstrates robust generalization, with accuracy ranging from 98.02% to 99.25%. The five-fold mean accuracy further confirms the model's stability, at 98.62%±0.15%. The proposed model outperforms existing CNN and transformer-based architectures while being significantly more efficient, making it highly suitable and reliable for real-time monitoring in resource-constrained environments.

Land Use and Land Cover (LULC) classification plays a key role in the last decade for managing the decay of resources and mitigating the impact of population growth. It is used in several places, such as rapid urbanization, agriculture, climate change, coastal areas, and disaster recovery. The traditional remote sensing (RS) techniques encounter limitations in accurately classifying dynamic and complex Ariel Scenes such as coastal areas and LULC. This paper proposed a novel Squeeze-Excitation Multi-Scale Fusion Network (SEMSF-Net) to classify LULC and the coastal regions using remote sensing images. The proposed model is based on the squeeze-excitation block initially embedded with inception and dense blocks separately. These blocks are designed based on the multi-scale to generate more important features information that can later perform accurate classification. In the next phase, these blocks are fused at the network level, where bottleneck and inverted residual blocks are connected to reduce the learnable parameters and improve feature strength. The hyperparameters of this network are selected based on the several experiments utilized in the training of the proposed model. The trained SEMSF-Net architecture is employed further in the testing phase, and classification is performed. The GradCAM is also used to interpret the trained model's visual prediction. Three datasets are utilized for the experimental process: the Coastal dataset, MLRSNet, and NWPU. We obtained an improved accuracy of 94.94, 93.7, and 95.70% on these datasets, respectively. In addition, the macro recall rate is 79.0, 93.0, and 96%, respectively. Comparison with several recent techniques shows the proposed model outperforms the selected datasets.

When the number of white blood cells (WBCs) in the human body becomes imbalanced, leukemia (blood cancer) can develop, affecting individuals of all ages. Early and accurate detection is crucial for improving patient outcomes, yet manual diagnosis is time-consuming and prone to subjectivity. This study proposes an efficient hybrid convolutional neural network (CNN)-transformer model for automated blood cancer detection, integrating convolutional layers for localized feature extraction with transformer-based attention mechanisms for capturing global dependencies. The architecture employs depthwise separable convolutions, efficient multi-head self-attention (EMHSA), and an efficient multilayer perceptron (EMLP) block, optimized via Bayesian hyperparameter tuning. The proposed model was evaluated on two publicly available datasets: the Blood Cancer dataset (binary classification) and the Blood Cells Cancer (ALL) dataset (four-class classification). Using a 50:50 training-testing split, the model achieved 100% accuracy, 100% precision, 100% recall, and 100% F1-score on the Blood Cancer dataset, and 99% accuracy, 99% precision, 100% recall, and 99% F1-score on the ALL dataset. Additional experiments with 80:20, 70:30, and 60:40 splits confirmed consistent performance above 98% across all metrics, demonstrating strong robustness. The model contains only 2.04 million trainable parameters, significantly fewer than standard CNN or transformer-based architectures, making it computationally lightweight and suitable for deployment in resource-constrained clinical environments. These results highlight the potential of the proposed hybrid framework to provide accurate and efficient blood cancer classification, advancing the applicability of deep learning in hematological diagnostics.