The degradation of image quality caused by adverse weather conditions presents a significant challenge to existing object detection methods. To address this issue, we propose an end-to-end object detection network for hazy weather conditions, called Dual-Path Supervised Network (DPSNet). This network adopts a dual-path architecture that integrates image dehazing into the detection task, strengthens the network’s feature extraction capability through a dehazing-based supervision mechanism, and enhances feature representation stability using a contrastive learning-based regularization constraint. Firstly, we designed a dual-path module, which achieves feature-level complementarity and enhancement through multi-scale feature fusion between the dehazing feature extraction path and the detection feature extraction path. Secondly, we utilize an image dehazing-based supervision mechanism, combined with frequency domain loss, to optimize the dehazing effect. This indirectly guides the model to focus on richer details, thereby improving detection performance. In addition, we propose a contrastive learning-based semantic consistency regularization method, which maximizes the feature similarity between an image and its spatially transformed sample (e.g., rotated or flipped), ensuring that the model can extract highly representative semantic features under varying input conditions. Finally, experimental results show that DPSNet achieves the highest mAP on both synthetic and real hazy datasets, surpassing advanced object detection methods. In the synthetic hazy dataset VOC-FOG-test, the mAP reaches 79.73%; on the real hazy datasets RTTR and Foggy Driving, the mAPs are 62.13% and 36.09%, respectively. These results demonstrate that DPSNet can perform robustly on detection tasks involving hazy-degraded images.

Federated Learning (FL) has emerged as a promising collaborative learning paradigm that enables to train machine learning models across decentralized devices, while keeping the training data localized to preserve user privacy. However, the heterogeneity in both decentralized training data and distributed computing resources has posed significant challenges to the design of effective and efficient FL schemes. Most existing solutions either focus on tackling a single type of heterogeneity, or are unable to fully support model heterogeneity with low communication overhead, fast convergence, and good interpretability. In this paper, we present CloREF, a novel rule-based collaborative learning framework that allows devices in FL to use completely different local learning models to cater to both data and resource heterogeneity. In CloREF, each rule is represented as a linear model, which provides good interpretability. Each participating device chooses a local model and trains it using its local data. The decision boundary of each trained local model is then approximated using a set of rules, which effectively bridges the gap arising from model heterogeneity. All participating devices collaborate to select the optimal set of rules as the global model, employing evolutionary optimization to effectively fuse the knowledge acquired from all local models. Experimental results on both synthesized and real-world datasets demonstrate that the rules generated by our proposed method can mimic the behaviors of various learning models with high fidelity ($\gt $0.95 in most tests), and CloREF gives competitive performance in accuracy, AUC, and communication overhead, compared with both the best-performing model trained centrally and several state-of-the-art model-heterogeneous federated learning schemes.

In this paper, a new structure of a single-phase hybrid active power filter (HAPF) with adaptive passive part and active part is proposed. The HAPF as a state-of-the-art power quality compensator is combining the advantage of passive filter and active filter. However, it may still suffer the high initial and operational costs situation under the capacitive load case. With the proposed topology, the reactive power and harmonics of the loading can be compensated dynamically. The structure can be applied to general household appliances including inductive and capacitive loading. Compared with the traditional APF for inductive loading compensation and HAPF for capacitive loading compensation, the proposed HAPF requires a low dc-linked voltage under both inductive and capacitive loading compensation, which can reduce the initial and operational costs. Especially, the transient condition has been analyzed when the passive part is switched to verify the feasibility of this topology. Finally, representative simulation and experimental results are presented to prove the compensation effect of this single-phase HAPF with adaptive passive part.

The basic Zero-Voltage Switching (ZVS) three-level DC-DC converter has one clamping capacitor to realize the ZVS of the switches, and two clamping diodes to clamp the voltage of the clamping capacitor. In order to reduce the reverse recovery loss of the diode as well as its cost, this paper proposes to remove one of the clamping diodes in basic ZVS three-level DC-DC converter. With less components, the proposed converter can still have a stable clamping capacitor voltage, which is clamped at half of the dc link voltage. Moreover, the ZVS performance will be influenced by removing the clamping diode. But as long as the clamping capacitor is properly selected, the degradation of the ZVS performance can be neglected. The impact of the clamping capacitor on the ZVS performance is mathematically analyzed as well.