Brain tumor segmentation from MRI images is a critical task for accurate diagnosis and treatment planning. While traditional segmentation approaches leverage deep learning models, such as U-Net and its variants, they often struggle with balancing precision and interpretability, especially in handling noisy and complex MRI data. Earlier works have demonstrated remarkable performance improvements through hybrid models, yet limitations persist in effectively capturing both global and local contexts, as well as ensuring model transparency for clinical application. These shortcomings highlight the need for frameworks that not only excel in segmentation accuracy but also provide explainability for reliable medical decision-making. To address these challenges, we propose a novel Explainable U-Net Transformer (ExU-Trans) framework for brain tumor segmentation in MRI images. ExU-Trans integrates the interpretability strengths of Vision Transformers with the spatial precision of U-Net in a dual-encoder structure. The Vision Transformer encoder employs Self-Explanatory Multi-Head Attention (SE-MHA) to generate noise-resistant attention maps, while a modified U-Net encoder focuses on extracting multiscale spatial features critical for tumor delineation. To enhance interpretability, a Discriminative Attribute Explainer (DAE) is embedded within the transformer encoder, pinpointing tumor-specific attributes at a pixel level. Furthermore, a Contextual Self-Attention mechanism enriches feature representations by capturing both global and local contexts, refining tumor detection. The outputs from both encoders are combined using a bivariate fusion module that incorporates spatial and channel attention, producing a unified feature representation. Finally, an attribute-guided loss function enables self-supervised learning, resulting in highly accurate and inherently interpretable segmentation outcomes. Extensive evaluations demonstrate the superiority of ExU-Trans in terms of segmentation accuracy and explainability, offering a reliable tool for clinical brain tumor diagnosis and treatment planning.

Gesture recognition remains a challenging task in human-computer interaction due to complex temporal dynamics and high intra-class variability. Traditional approaches often suffer from limited feature representation capabilities and lack of model interpretability, particularly when dealing with large gesture vocabularies in noisy environments. To address these limitations, this study presents an explainable multimodal gesture recognition framework that integrates time-series image fusion with deep residual learning to enhance both accuracy and interpretability. Three complementary image transformation techniques, recurrence plots (RP), Markov transition fields (MTF), and Gramian angular fields (GAF), are employed to convert one-dimensional temporal gesture sequences into two-dimensional representations, capturing diverse spatiotemporal characteristics. These image modalities are fused with the original time-series data to form enriched multimodal feature sets, which are processed by a customized deep residual network for end-to-end classification. Comprehensive experiments conducted on a self-constructed dataset collected under controlled indoor conditions containing 62 gesture classes (10 digits and 52 English letters) demonstrate recognition accuracies of 100%, 100%, and 99% for Arabic numerals, uppercase letters, and lowercase letters, respectively, achieving an overall accuracy of 99%, which represents a 5.5% improvement over conventional unimodal approaches. Furthermore, SHapley Additive exPlanations (SHAP) analysis provides quantitative interpretability by elucidating individual feature contributions, enhancing transparency and model trustworthiness. The proposed framework thus delivers a significant advancement in gesture recognition by simultaneously achieving high precision, multimodal robustness, and explainable decision-making.

Image dehazing plays a critical role in understanding complex real-world systems such as transportation, environmental monitoring, and healthcare imaging, where haze significantly degrades visual information and hinders decision-making. Traditional deep learning models emphasize accuracy but often neglect interpretability and resource efficiency, limiting their applicability in sustainable multi-modal fusion frameworks. We introduce SFX-GAN, a Sustainable and eXplainable Frequency-domain GAN that employs a novel channel-wise spectral fusion strategy, treating low, mid, and high-frequency components across RGB channels as complementary modalities. This multi-modal spectral integration enables the model to capture both global context and fine-grained structural cues, facilitating more reliable restoration of haze-free images. For interpretability, SFX-GAN incorporates a Concept Bottleneck Layer and Grad-CAM supervision, aligning network attention with haze-affected regions and enhancing trust in automated dehazing for safety-critical applications. Designed with sustainability in mind, SFX-GAN leverages frequency-domain processing to deliver efficient inference with reduced computational overhead. Experiments on RESIDE-SOTS, NH-HAZE, I-HAZE, O-HAZE, and D-HAZY datasets demonstrate state-of-the-art performance under both full-reference (PSNR, SSIM) and no-reference (NIQE, BRISQUE, FADE, SSEQ) metrics. Ablation studies confirm the effectiveness of each module in the fusion pipeline. Overall, SFX-GAN contributes to the intelligent fusion of spectral-spatial modalities for complex imaging systems, advancing responsible and explainable AI aligned with SDG 9 (Industry, Innovation, and Infrastructure) and SDG 11 (Sustainable Cities and Communities).

Abstract This paper presents LISA as a Language Independent Sentiment Analysis tool that exploits Graph Neural Networks, GNN, for sentiment analysis, SA, applications. To build up that analyzer two types of GNN are examined. These types are: (1) Graph convolutional Networks, GCN, and (2) Graph Attention Networks, GAT. The input text is transformed to a corresponding graph to bypass the multilingual differential representations between Natural Languages, NL. Accordingly, LISA includes input steam of text (e.g. tweets, reviews…), followed by a text to graph module which is connected to a tokenizer to determine the correct tokens and pass them to Word2vec embedding. The corresponding graph of the underlying text enters the GNN which has been implemented by either GCN or GAT to generate the output sentiment polarity (+ ve/−ve). The two GNN models are compared to explore the effect of attention on a SA. In addition, LISA performance is evaluated (using two multilingual datasets: amazon reviews, tweets). It has been found that its accuracy, F1 score and confusion matrices are in the same range of the state of the art (SOTA) framework.

Image captioning endeavors to generate text descriptions from images using the learned cross-modal generator. Automatically generating descriptions for life ecological experiment images of China’s space station can significantly expedite the comprehension of the intricate semantic details within the experimental data, facilitating the efficient management and accurate retrieval of massive experimental data. However, current image caption approaches trained on general image caption datasets cannot be directly applied to life ecological experiment scenes. In this work, we construct a Chinese caption dataset of life ecological experiment images by collecting and annotating the life ecological experiment image products of China’s space station (LEEoCSSCc). Furthermore, we propose an innovative image captioning approach by exploiting cross-modal prediction and data augmentation (IACPDA), which is designed to harness the raw image to guide the semantic constraints of the generated captions from different perspectives. IACPDA facilitates the transformation of both the original image and the resulting caption into a unified semantic space. It then leverages the predictions derived from the original image to offer constructive guidance for the generated captions. Moreover, we use a semantic-guided image augmentation method based on a diffusion model to generate augmented images that are distinct from the original training images while maintaining the essential semantics of the original images. Consequently, these augmented images can be effectively utilized as soft labels, enabling the learning of a more effective generator. The experiments indicate that our method achieves superior performance compared to other state-of-the-art baselines on the LEEoCSSCc. This paper is the first application of image captioning to the analysis of space science experiment images from China’s space station and further offers vital technical support for the understanding of space science experiments.

Abstract Pedestrian trajectory prediction from egocentric monocular video is hindered by camera motion, intermittent occlusions, and complex social interactions. We present NIM-STGCN, a unified framework whose core contribution is a differentiable view normalization(GVN) that couples an enhanced differentiable PnP layer (ED-PnP) with an $$\textrm{SE}(3)$$ SE ( 3 ) warp to align past observations into a single virtual static camera frame. Because GVN is trained end-to-end, forecasting losses back-propagate to pose estimation, yielding geometrically cleaner inputs. On the normalized histories, a lightweight Gated Convolutional Imputation Module (GCIM) recovers missing bounding-box measurements while preserving observed entries, and an efficient spatio-temporal GCN encodes agent dynamics and interactions (optionally augmented by a physics-guided kinematics–interaction prior, PKIM). A Gaussian-mixture predictor produces multi-modal futures and is optimized with a sequence-level negative log-likelihood together with a time-weighted position loss. Extensive experiments on the JAAD and PIE benchmarks show that NIM-STGCN reduces Average Displacement Error (ADE) and Final Displacement Error (FDE) by 12–18 % compared to state-of-the-art methods. Code is available at https://github.com/fantot/NIM-STGCN . Graphical abstract