Automatic modulation classification (AMC) is a key technology of cognitive radio used in non-cooperative communication. Recently, deep learning has been applied to AMC tasks. In this paper, an AMC scheme based on deep learning is proposed, which combines random erasing and attention mechanism to achieve high classification accuracy. Firstly, we propose two data augmentation methods, random erasing at sample level and random erasing at amplitude/phase (AP) channel level. The former replaces training samples with noise information, while the latter replaces AP channel information of training samples with noise information. Erased data segments are randomly stitched to enable training data expansion. Training data of different qualities enables deep learning model to have stronger generalization capability and higher robustness. Then, we propose a single-layer Long Short-Term Memory (LSTM) model based on attention mechanism. In the first part of this model, we propose the signal embedding, which enables the input to contain modulation information more comprehensively and accurately. Then hidden state output by LSTM is input into the attention module, and weighting is applied to the hidden state to help the LSTM model capture the temporal features of modulated signals. Compared to a model without attention mechanism, this model has faster convergence speed and better classification performance. Lastly, we propose a random erasing-based test time augmentation (RE-TTA) method. Test data is randomly erased for multiple times and classification results are comprehensively evaluated, in order to further improve classification accuracy. Experimental results on dataset RML2016.10a show that classification accuracy of the proposed scheme is competitive compared with the state-of-the-art methods.

The early detection of ocular disorders, including Graves’ disease, myasthenia gravis, conjunctival hyperemia, conjunctivitis, and keratitis, which critically impair the vision of millions worldwide, necessitates large-scale screening predicated on ocular appearance measurements as a crucial diagnostic component. The emerging Internet of Medical Things (IoMT) introduces new avenues for local clinics to embrace portable and extensive diagnostics. However, the inherent heterogeneity and blurriness of ocular images, compounded by environmental noise, and the computational resource constraint hinder the high-precision diagnostics on IoMT devices. In response to these challenges, a Linguistic-Guided Generative Large Vision Model (LiGu-LVM) has been formulated to assist and enhance the diagnostic capability of IoMT-enabled ocular scanners, integrating a Dynamically Allocated High-Speed Quantization System (DAHSQS), a Linguistic-Guided Generative Local-isolation module (LiGu), an Oculo Visio Transformatrix Segmentum-Analytica Modulorum (OVT-SAM), and a Multi-Scale Recursive Attention Segmentation Engine (MuRASE). DAHSQS enables the flexible aggregation and transmission of patient imagery to shift heavy diagnostic tasks from IoMT-enabled mobile ocular scanners to computational clusters, facilitating rapid facial measurements and preliminary screening via dynamic task allocation and scalable server clusters. The LiGu module employs natural language guidance to generate key image locations, using extensive prior knowledge embedded within linguistic models for precise semantic isolation. OVT-SAM synthesizes multi-level features from large vision model, extracting intermediate characteristic information and addressing global features alongside deep semantic understanding in natural images collected from IoMT-enabled ocular scanners. MuRASE achieves high-fidelity segmentation of ocular images by incorporating contextual recursive attention mechanisms and skip connections with layer-wise reverse connectivity. Extensive experiments shown proposed method surpassing 80% IoU in ocular semantic segmentation on the CelebA-HQ dataset, achieving an IoU of 82.9%, thus exceeding the performance of existing models by 4.9%.

Ship detection based on Synthetic Aperture Radar (SAR) is crucial for maritime surveillance, yet existing methods struggle with two critical challenges: (1) missed detections in nearshore scenarios due to highly similar radar reflections between ships and shore objects (e.g., buildings, docks), and (2) feature vanishing of small ships caused by SAR resolution limits and downsampling operations. To address these issues, we propose Kolmogorov-Arnold-Enhanced YOLO (KaneYOLO), a novel SAR ship detector designed for robust discrimination in complex nearshore scenes and enhanced small-target detection. Firstly, we introduce KAN Block, a feature extraction module leveraging the Kolmogorov-Arnold theorem to model complex nonlinear relationships in SAR images. By integrating Gram polynomials and B-spline-based activation functions, KAN Block dynamically calibrates feature responses to suppress background clutter (e.g., shore structures) while amplifying subtle intensity patterns and edge transitions unique to ships. Secondly, we design a Feature Fusion Allocation Structure (FFAS) with parallelized deep convolutions to aggregate multi-scale features. This structure preserves small-target details by propagating enriched semantic information across detection layers, mitigating feature loss during downsampling. Finally, a Detail Enhancement Detection Head (DEDH) is proposed to reduce computational overhead through shared convolutional layers, while enhancing local feature utilization via direction-sensitive convolutions and group normalization. KaneYOLO achieved 93.9% AP on HRSID and 98.3% AP on SSDD. These results validate its robustness in handling SAR-specific challenges, offering significant potential for real-world maritime security applications.

Information retrieved from three dimensions is treated uniformly in CNN-based volumetric segmentation methods. However, such neglect of axial disparities fails to capture true spatio-temporal variations. This paper introduces the volumetric axial disentanglement to address the disparities in spatial information along different axial dimensions. Building on this concept, we propose the Post-Axial Refiner (PaR) module to refine segmentation masks by implementing axial disentanglement on the specific axis of the volumetric medical sequences. As a plug-and-play enhancement to existing volumetric segmentation architecture, PaR further utilizes specialized attention approaches to learn disentangled post-decoding features, enhancing spatial representation and structural detail. Validation on various datasets demonstrates PaR's consistent elevation of segmentation precision and boundary clarity across 11 baselines and different imaging modalities, achieving state-of-the-art performance on multiple datasets. Experimental tests demonstrate the ability of volumetric axial disentanglement to refine the segmentation of volumetric medical images. Code is released at https://github.com/IMOP-lab/PaR-Pytorch.