Fine-grained agricultural pest image-text retrieval demands robust cross-modal matching capable of handling subtle appearance differences and structured multi-sentence descriptions. Yet, small-scale domain data introduces two key challenges in supervision and representation. First, existing corpora typically pair each image with a single coarse paragraph, providing incomplete supervision that omits life-stage appearances and subtle morphological traits, while high-frequency redundant fields shared across pest categories further cause semantic confounding. Second, prevailing retrieval models rely on global pooled alignment, which averages out low-occupancy but discriminative local cues such as stripe layouts, body segmentation, and wing venation, entangling near-neighbor categories. To address these issues, we propose StruMorAlign, a parameter-efficient cross-attention CLIP framework for fine-grained pest image-text retrieval. We first construct GjPest4CMR, an English goji pest image-text retrieval dataset created via LLMs-assisted re-writing with expert-in-the-loop verification, providing multi-sentence structured descriptions. On the modeling side, we design StruPD (Structured Semantic Deconfounding) to provide multi-granularity structured supervision and strengthen coarse-to-fine consistency learning on morphology-relevant local regions, addressing the first challenge; and introduce GateA2 (Gated Adaptive Alignment) self-distillation with high-momentum EMA teacher for parameter-efficient, morphology-aware cross-modal alignment, addressing the second challenge. On the GjPest4CMR dataset, StruMorAlign achieves mR 78.72% for both image-to-text and text-to-image retrieval, with I2T R@1=53.11% and T2I R@1=51.53%, outperforming the strongest baseline HarMA by +9.72 mR. Ablation studies confirm the complementary benefits of StruPD and GateA2, and qualitative analyses including t-SNE visualizations and attention heatmaps further validate the effectiveness of our approach.

Multimodal Image Fusion (MMIF) aims to integrate complementary information from different imaging modalities to overcome the limitations of individual sensors. It enhances image quality and facilitates downstream applications such as remote sensing, medical diagnostics, and robotics. Despite significant advancements, current MMIF methods still face challenges such as modality misalignment, high-frequency detail destruction, and task-specific limitations. To address these challenges, we propose AdaSFFuse, a novel framework for task-generalized MMIF through adaptive cross-domain co-fusion learning. AdaSFFuse introduces two key innovations: the Adaptive Approximate Wavelet Transform (AdaWAT) for frequency decoupling, and the Spatial-Frequency Mamba Blocks for efficient multimodal fusion. AdaWAT adaptively separates the high- and low-frequency components of multimodal images from different scenes, enabling fine-grained extraction and alignment of distinct frequency characteristics for each modality. The Spatial-Frequency Mamba Blocks facilitate cross-domain fusion in both spatial and frequency domains, enhancing this process. These blocks dynamically adjust through learnable mappings to ensure robust fusion across diverse modalities. By combining these components, AdaSFFuse improves the alignment and integration of multimodal features, reduces frequency loss, and preserves critical details. Extensive experiments on four MMIF tasks-Infrared-Visible Image Fusion (IVF), Multi-Focus Image Fusion (MFF), Multi-Exposure Image Fusion (MEF), and Medical Image Fusion (MIF)-demonstrate AdaSFFuse's superior fusion performance, ensuring both low computational cost and a compact network, offering a strong balance between performance and efficiency. The code will be publicly available athttps://github.com/Zhen-yu-Liu/AdaSFFuse.

Remote sensing image-text retrieval (RSITR) has become a research hotspot in recent years for its wide application. Existing methods in this context, based either on local or global feature matching, overlook the sensing variation-leaded visual deviation and geographically nearby image-text mismatching problems of remote sensing (RS) images. This work notes that this would limit the retrieval accuracy for RSITR. To handle this, we present IERR, an implicit-explicit relation reasoning framework that learns relations between local visual-textual tokens and enhances global image-text matching without requiring additional prior supervision. Specifically, masked image modeling (MIM) and masked language modeling (MLM) are used for symmetric mask reasoning consistency alignment. Meanwhile, masked features (i.e., implicit relation) and unmasked features (i.e., explicit relation) are fed into a multimodal interaction encoder to enhance the representations of the textual-visual features. Extensive experimental results on the RSICD and RSITMD datasets demonstrate the superiority of IERR compared with 17 baselines.