Unmanned Aerial Vehicle (UAV)/drone vision and navigation require low-power 3D depth-sensing with robustness against strong/weak light and various weather conditions. CMOS image sensor (CIS) and light detection and ranging (LiDAR) can provide high-fidelity imaging. However, CIS lacks depth sensing and has difficulty in low light conditions. LiDAR suffers from heavy fog and is costly. Radar is resilient in poor weather but tends to be bulky and expensive. Ultrasound imaging system (UIS), on the other hand, is robust in various weather and light conditions and is cost-effective; however, it usually has low imaging resolution and low frame rate. Using a higher frequency ultrasound theoretically enables higher resolution, yet this comes at the cost of short range due to greater attenuation and severe grating lobe phenomenon. Compared to prior air-channeled ultrasound imaging systems, this work is first to integrate low-frequency (LF) and high-frequency (HF) dual band ultrasound imaging into SoC, achieving highest angular resolution so far with near-zero frequency switching latency and typical 11.04M focal point/s throughput to enable real-time high resolution 3D imaging. System level validation including signal-to-noise ratio (SNR) VS range characterization, hybrid LF–HF imaging, two-point separation demonstrating 0.5° angular resolution, and in-air measurements on a flying drone confirms reliable operation at a 7 m range and 21 frames per second (fps) for both LF and HF modes. These results establish the hybrid dual-band SoC as an effective and practical 3D depth-sensing solution for UAV navigation under challenging environmental conditions. The SoC is implemented in 180 nm 1P6M Standard CMOS, occupies 30mm${}^{\mathbf {2}}$and consumes 116.1 mW.

Emotion recognition has a wide range of applications in human-computer interaction, marketing, healthcare, and other fields. In recent years, the development of deep learning technology has provided new methods for emotion recognition. Prior to this, many emotion recognition methods have been proposed, including multimodal emotion recognition methods, but these methods ignore the mutual interference between different input modalities and pay little attention to the directional dialogue between speakers. Therefore, this article proposes a new multimodal emotion recognition method, including a cross modal context fusion module, an adaptive graph convolutional encoding module, and an emotion classification module. The cross modal context module includes a cross modal alignment module and a context fusion module, which are used to reduce the noise introduced by mutual interference between different input modalities. The adaptive graph convolution module constructs a dialogue relationship graph for extracting dependencies and self dependencies between speakers. Our model has surpassed some state-of-the-art methods on publicly available benchmark datasets and achieved high recognition accuracy.

Ensuring construction site safety requires accurate and real-time detection of workers’ safety helmet use, despite challenges posed by cluttered environments, densely populated work areas, and hard-to-detect small or overlapping objects caused by building obstructions. This paper proposes a novel algorithm for safety helmet wearing detection, incorporating a dynamic attention within the detection head to enhance multi-scale perception. The mechanism combines feature-level attention for scale adaptation, spatial attention for spatial localization, and channel attention for task-specific insights, improving small object detection without additional computational overhead. Furthermore, a two-way fusion strategy enables bidirectional information flow, refining feature fusion through adaptive multi-scale weighting, and enhancing recognition of occluded targets. Experimental results demonstrate a 1.7% improvement in mAP@[.5:.95] compared to the best baseline while reducing GFLOPs by 11.9% on larger sizes. The proposed method surpasses existing models, providing an efficient and practical solution for real-world construction safety monitoring.

Over the past decades, first-principles real-time time dependent density functional theory(rt-TDDFT) simulations have been limited to systems with only thousands of atoms. We propose a novel method based on the discontinuous Galerkin adaptive local basis, significantly reducing global communication in rt-TDDFT. We further introduce a tensor compression technique that leverages basis locality to avoid repeated evaluation of multi-center integrals in hybrid functionals, greatly reducing computational cost. To overcome the projection bottleneck in our basis sets, we design a fused Gemm-Reduce operation that achieves several times higher floating-point efficiency than standard BLAS combination. Our implementation reaches 34.8% of theoretical peak performance on 524,288 CGs of the New Sunway supercomputer and simulates electronic dynamics of systems with over one million atoms for both local-semi-local and hybrid functionals. This work improves computational scale by two orders of magnitude, opening new possibilities for exploring ultrafast dynamics in large-scale materials and nanophotonic devices.

GW approximation is a powerful approach to accurately describe the excited-state of semiconductors. However, GW incurs high computational cost $\mathcal{O}\left(N^{4}\right)$ and large memory usage $\mathcal{O}\left(N^{3}\right)$, limiting its applications to thousands of (2,742) atoms even on leadership supercomputers. Herein we present a massively parallel implementation of accurate and efficient cubic-scaling plane-wave GW calculations by using low-rank approximations and high-performance computing on leadership supercomputers. By using a series of low rank approximations, we can reduce the expensive GW calculations to the cubic-scaling computational cost $\mathcal{O}\left(N^{3}\right)$ and quadratic memory usage $\mathcal{O}\left(N^{2}\right)$. With the help of parallel and communication optimization, the plane-wave GW calculations gain an overall speedup of over 70x and efficiently scale up to 13,824 atoms within a few minutes using 449,280 cores on new Sunway supercomputer. This accomplishment paves the way for excited-state quantum mechanical material simulations at mesoscopic scale (10K atoms) and for the design of next-generation semiconductor devices.

Aiming at the problem of low-quality electronic medical record(EMRs), the medical knowledge graph is used to enhance the quality and trustworthiness of EMRs. We propose a novel EMR quality control method that integrates a knowledge graph with deep learning techniques and complement it with a system implementation. The entities and relationships are extracted by the End to End Medical Information Extraction based on Knowledge Base (MIE-KB), which leverages an encoder-decoder framework and self-supervised pre-training model based on BERT. We introduce a bidirectional QC system for assessing the content normality and diagnostic rationality of medical records to classify and detect contradictions. By using more than 500,000 desensitized medical records covering 129 diseases in 7 categories, deep learning methods are used to learn the relationship between the desensitized medical records and the diagnosis features, and to determine whether the diagnosis is reasonable through the inference of medical knowledge mapping. We use the UMLS and other standards to evaluate the extraction accuracy and the diagnostic accuracy of the QC system. The system evaluation results show that the system can effectively boost the quality control effect of EMRs. The end-to-end pre-training model and bidirectional QC system improve the quality of EMRs in primary hospitals.