To address the challenges in fine-grained image classification stemming from significant intra-class variations and subtle inter-class differences, as well as the performance instability caused by simplistic fusion of deep and shallow features in conventional methods, this paper proposes a long–short range features network (LSNet), a novel gated convolutional model based on long/short-range feature integration. The model innovatively reformulates the convolution process as a sequential task, leveraging the specific order of each convolutional module in the network architecture to implement long–short-term memory network (LSTM)-based gating mechanisms for layer-wise feature selection and adaptive fusion between modules. LSNet achieves phased integration of spatial-static features from convolutional models and temporal-dynamic characteristics from recurrent models, effectively resolving the interference between deep and shallow features. Experimental results demonstrate that under equivalent parameter counts, LSNet achieves classification accuracy improvements of 2.58% and 4.80% over the baseline model on the FGVC-Aircraft and Stanford Cars datasets, with final accuracies reaching 90.15% and 91.97%, respectively, while its architecture allows seamless integration into any deep convolutional model. Gradient-based class activation mapping (Grad-CAM) visualizations further validate the model’s capability to capture discriminative subtle features. Additionally, we have successfully applied LSNet in intelligent livestock farming, achieving precise facial recognition of pigs in high-density breeding environments. The model achieves a recognition accuracy of 92.82% on the pig face recognition dataset. This research not only provides an effective solution for fine-grained classification but also demonstrates extensibility to other vision tasks requiring refined feature extraction.

Electromagnetic structure (EMS) design plays a critical role in developing advanced antennas and materials, but remains challenging due to high-dimensional design spaces and expensive evaluations. While existing methods commonly employ high-quality predictors or generators to alleviate evaluations, they are often data-intensive and struggle with real-world scale and budget constraints. To address this, we propose a novel method called Progressive Quadtree-based Search (PQS). Rather than exhaustively exploring the high-dimensional space, PQS converts the conventional image-like layout into a quadtree-based hierarchical representation, enabling a progressive search from global patterns to local details. Furthermore, to lessen reliance on highly accurate predictors, we introduce a consistency-driven sample selection mechanism. This mechanism quantifies the reliability of predictions, balancing exploitation and exploration when selecting candidate designs. We evaluate PQS on two real-world engineering tasks, i.e., Dual-layer Frequency Selective Surface and High-gain Antenna. Experimental results show that our method can achieve satisfactory designs under limited computational budgets, outperforming baseline methods. In particular, compared to generative approaches, it cuts evaluation costs by 75$\tilde$85%, effectively saving 20.27$\tilde$38.80 days of product designing cycle.

Collaborative perception allows connected vehicles to share sensor data through Vehicle-to-Vehicle (V2V), helping each vehicle to see beyond its own sensing range and make safer driving decisions. However, the large volume of perception data brings heavy computational pressure, and centralized offloading methods struggle to respond quickly in dynamic and decentralized vehicular networks. To address this challenge, we propose an intelligent distributed task offloading framework built upon non-cooperative game theory. In this framework, each vehicle acts as an autonomous decision-maker that selects its offloading strategy using only local observations, without relying on any central controller. A lightweight game-theoretic optimization algorithm is designed to iteratively adjust strategies and reach a Nash equilibrium, achieving real-time task allocation with minimal communication overhead. Simulation results show that our method significantly reduces perception latency and achieves better load balancing compared with fixed and random allocation schemes. These findings demonstrate that the proposed approach provides a practical solution for real-time collaborative perception in highly dynamic vehicular environments.

Test-time adaptation provides an effective means of tackling the potential distribution shift between model training and inference, by dynamically updating the model at test time. This area has seen fast progress recently, at the effectiveness of handling test shifts. Nonetheless, prior methods still suffer two key limitations: 1) these methods rely on performing backward computation for each test sample, which takes a considerable amount of time; and 2) these methods focus on improving the performance on out-of-distribution test samples and ignore that the adaptation on test data may result in a catastrophic forgetting issue, \ie, the performance on in-distribution test samples may degrade. To address these issues, we propose an efficient anti-forgetting test-time adaptation (EATA) method. Specifically, we devise a sample-efficient entropy minimization loss to exclude uninformative samples out of backward computation, which improves the overall efficiency and meanwhile boosts the out-of-distribution accuracy. Afterward, we introduce a regularization loss to ensure that critical model weights tend to be preserved during adaptation, thereby alleviating the forgetting issue. Extensive experiments on CIFAR-10-C, ImageNet-C, and ImageNet-R verify the effectiveness and superiority of our EATA.