Lei Shi

Accurate grain quality classification at scale requires models that are simultaneously lightweight and highly discriminative, yet existing approaches struggle to satisfy both constraints. This paper proposes a lightweight complementary network integrating a convolutional neural network (CNN) and Transformer, termed the Lite Grain Classification Network (LGC-Net), for semantic alignment of heterogeneous features in multi-grain defect recognition. To overcome the quadratic complexity of conventional self-attention, we introduce Fast Channel Additive Attention, which applies additive attention to a partial channel subset, reducing complexity from quadratic to linear while preserving global semantic modeling capacity. A Dual Attention Fusion Module is further proposed to align local spatial features from the CNN branch with global representations from the Transformer branch, effectively leveraging their complementary strengths. Experiments on a dataset comprising five grain types (maize, wheat, rice, barley, and sorghum) with over ten defect categories show that LGC-Net-XXS achieves 89.59% average accuracy with only 0.93 M parameters and 0.43 G FLOPs, demonstrating its potential for real-time deployment in large-scale agricultural data analysis on resource-constrained edge devices.

Public Opinion Reports are essential tools for crisis management, yet their evaluation remains a critical bottleneck that often delays response actions. Recently, dominant Large Language Model (LLM)-based evaluators often overlook a critical challenge: highly open-ended dimensions such as “innovation” and “feasibility” require synthesizing diverse stakeholder perspectives, as different groups judge these qualities from fundamentally different perspectives. Motivated by this, we propose the Role-based Adaptive Evaluation (RAE) framework. This framework employs an adaptive mechanism leveraging multi-perspective evaluation insights through role-based analysis, and further introduces dynamically generated roles tailored to specific contexts for these dimensions. RAE further incorporates multi-role reasoning aggregation to minimize individual biases and enhance evaluation robustness. Extensive experiments demonstrate that RAE significantly improves alignment with human expert judgments, especially on challenging highly open-ended dimensions.

According to forecast data from the State Grid Corporation of China, the number of terminal devices connected to the power grid is expected to reach the scale of 2 billion within the next five years. With the continuous growth in the number and variety of terminal devices in the smart grid, traditional cloud-edge-end architecture will face the increasing issue of response latency. In this context, in situ computing, as a paradigm for local or near-source data processing within cloud-edge-end architecture, is gradually becoming a key technological pathway in industrial systems. The in situ server system, by deploying servers near terminals, enables the near-data-source processing of terminal-generated in situ data, representing an important implementation of in situ computing. To enhance the processing efficiency and response capability of in situ data in smart grid scenarios, this study designs an in situ data processing mechanism and an access demand management framework. Due to the heterogeneity of in situ server performance, there are variations in response capabilities for access demands across rounds. This study introduces a double deep Q-network with prioritized experience replay to assist in response decision-making. Simulation experiments show that the proposed method reduces waiting latency and response latency by an average of 67.69% and 68.77%, respectively, compared to traditional algorithms and other reinforcement learning algorithms, verifying its effectiveness in in situ data management. This scheme can also be widely applied to in situ computing scenarios with low-latency data management, such as smart cities and industrial IoT.

With the rapid expansion of urban populations and the accelerated pace of urbanization, the concept of urban air mobility (UAM) has emerged. During flights, UAM aircraft need to transmit real-time sensing information to base stations for further processing and analysis. Large-scale real-time data require leveraging the computing capabilities of edge servers at the network edge to reduce transmission delay and energy consumption of UAM aircraft. In cases where edge servers are unable to process information, an unmanned aerial vehicle (UAV) equipped with computing capabilities and operating in low-altitude airspace can serve as a relay to assist in communication and computation. Due to the limited payloads and flight times of UAVs and UAM aircraft, delay and energy consumption within the system pose significant challenges. To tackle these challenges, two fundamental objectives have been proposed: minimizing delay and minimizing energy consumption. Furthermore, an optimization problem has been proposed to minimize the weighted sum of delay and energy consumption. Then, a UAM federated twin delayed deep deterministic policy gradient (UF-TD3) algorithm has been proposed to solve the original problems characterized by complex, non-convex, and inseparable variables. Simulation results show that the proposed UF-TD3 algorithm converges quickly and significantly outperforms four other baseline algorithms in optimizing delay and energy consumption performance. Moreover, compared to the conventional delay minimization strategy and energy minimization strategy, the proposed strategy of minimizing the weighted sum of delay and energy consumption can reduce the delay by 63.8% and reduce energy by 73.96%.

To address the issues of insufficient samples, limited scene diversity, missing perspectives, and low resolution in existing UAV-based pedestrian detection datasets, this paper proposes a novel UAV-based pedestrian detection benchmark dataset named the Novel Surveillance View (NSV). This dataset encompasses diverse scenes and pedestrian information captured from multiple perspectives, and introduces an innovative data mining approach that leverages tracking and optical flow information. This approach significantly improves data acquisition efficiency while ensuring annotation quality. Furthermore, an improved pedestrian detection method is proposed to overcome the performance degradation caused by significant perspective changes in top-down UAV views. Firstly, the View-Agnostic Decomposition (VAD) module decouples features into perspective-dependent and perspective-independent branches to enhance the model’s generalization ability to perspective variations. Secondly, the Deformable Conv-BN-SiLU (DCBS) module dynamically adjusts the receptive field shape to better adapt to the geometric deformations of pedestrians. Finally, the Context-Aware Pyramid Spatial Attention (CPSA) module integrates multi-scale features with attention mechanisms to address the challenge of drastic target scale variations. The experimental results demonstrate that the proposed method improves the mean Average Precision (mAP) by 9% on the NSV dataset, thereby validating that the approach effectively enhances pedestrian detection accuracy from UAV perspectives by optimizing perspective features.

Micro-expressions often reveal more genuine emotions but are challenging to recognize due to their brief duration and subtle amplitudes. To address these challenges, this paper introduces a micro-expression recognition method leveraging regions of interest (ROIs). Firstly, four specific ROIs are selected based on an analysis of the optical flow and relevant action units activated during micro-expressions. Secondly, effective feature extraction is achieved using the optical flow method. Thirdly, a block partition module is integrated into a convolutional neural network to reduce computational complexity, thereby enhancing model accuracy and generalization. The proposed model achieves notable performance, with accuracies of 93.96%, 86.15%, and 81.17% for three-class recognition on the CASME II, SAMM, and SMIC datasets, respectively. For five-class recognition, the model achieves accuracies of 81.63% on the CASME II dataset and 84.31% on the SMIC dataset. Experimental results validate the effectiveness of using ROIs in improving micro-expression recognition accuracy.