Accurate estimation of the state of charge (SOC) of batteries is one of the key problems in a battery management system. This paper proposes an adaptive SOC estimation method based on unscented Kalman filter algorithms for lithium (Li)-ion batteries. First, an enhanced battery model is proposed to include the impacts due to different discharge rates and temperatures. An adaptive joint estimation of the battery SOC and battery internal resistance is then presented to enhance system robustness with battery aging. The SOC estimation algorithm has been developed and verified through experiments on different types of Li-ion batteries. The results indicate that the proposed method provides an accurate SOC estimation and is computationally efficient, making it suitable for embedded system implementation.

To enable smart home applications, embedding home monitoring systems with different sensors can be a feasible remedy to capture the multimodal sensory information from daily life. However, the previously developed monitoring systems have high implementation cost, large power consumption, and complicated device configuration, which are not conducive to achieving carbon neutrality and carbon peak goals. Here, we demonstrate a multimodal neuromorphic sensory-processing system with memristor circuits for smart homes, offering a more environmentally friendly approach with low cost and easily deployable hardware. We used three components to facilitate the multimodal neuromorphic sensory-processing system design. First, a memristor crossbar array is fabricated using low-cost, reliable and eco-friendly two-dimensional (2D) materials; enabling a parallel connection to reduce the system complexity and the deep-learning computational cost. Second, a neuromorphic multimodal sensory-processing module is designed for collecting and processing multiple sensory cues efficiently; to establish accurate depiction of the environment and realize timely response. Third, the biomimetic hierarchical learning module is realized for smart home monitoring and relative applications (e.g., indoor human behavior recognition). The developed multimodal neuromorphic sensory-processing system effectively combines the nanotechnology, intelligent sensing, and deep learning; indicating an advancement in smart home applications.

Accurate estimating the state-of-charge (SOC) of Li-ion battery contributes significantly to electric vehicle safety. Existing methods typically focus on the traditional recurrent neural networks to encode time series features for SOC estimation. However, these methods rely solely on their own structure to extract time series correlated features, ignoring a significant amount of information on temporal dimension. To address this issue, this paper proposes a temporal transformer-based sequence network (TTSNet) that can make full use of temporal dimensional information to model the relationship between the input and SOC. Specifically, the proposed network splits the raw data into three branches including voltage, current, and temperature, as well as extracts the corresponding primary semantic features. It then uses a temporal transformer to effectively encode the features of temporal dimensional information. The resulting features are further fed into an attention-guided feature fusion module to interact information among voltage, current, and temperature branches for subsequent SOC estimation. To enhance the network's resilience for long time sequences, a sliding time window technique is introduced to pre-process the raw data. Besides, a Kalman filter is incorporated as post-processing to smooth the output to guide a more accurate SOC estimation. Comprehensive experiments are conducted on battery open datasets and vehicle operation datasets to verify the proposed method. The results demonstrate that the proposed method achieves high accuracy and strong robustness in both datasets, with average MAE, RMSE, and $\mathrm{R^{2}}$ values of 0.506%, 0.694%, and 99.791%, respectively. The code is available at https://github.com/haooozi/TTSNet .

Traffic sign detection is of great significance to the development of the Intelligent Transportation System (ITS) as a database for environmental awareness. The main challenges of existing traffic sign detection method are inaccurate small object detection, difficult mobile deployment, and complex working environment. Based on these, a vehicle-mounted adaptive traffic sign detector (VATSD) for small-sized signs in multiple working conditions is proposed in this paper. First, the Backbone of the detector is optimized. A feature tight fusion structure is designed to constitute a new feature extraction module, DCSP, which improves the feature extraction capability and the detection accuracy of small objects with negligible additional parameters. Second, an image enhancement network IENet with an adaptive joint filtering strategy is proposed. The IENet enables the dynamic selection of filters and thus adaptively optimizes low-quality images under multiple conditions to improve the accuracy of subsequent detection tasks. The proposed method has experimented on three traffic sign datasets and the detection accuracy increased by up to 7.6% compared to the original. The proposed detector demonstrates superiority over other state-of-the-art (SOTA) methods in terms of small object detection accuracy, detection speed, and environmental adaptability. Further, we deployed VATSD to Jetson Xavier NX and achieved a detection speed of 21.6 FPS, meeting real-time requirements.

With the widespread use of deep learning in single object tracking task, mainstream tracking algorithms treat tracking as a combined classification and regression problem. Classification aims at locating an arbitrary target, and regression aims at estimating the corresponding bounding box. In this paper, we focus on regression and propose a novel box estimation network, which consists of a transformer encoder target pyramid guide (TPG) and transformer decoder target pyramid spread (TPS). Specifically, the transformer encoder TPG is designed to generate fine-grained prior knowledge with explicit representation for template targets. In contrast to the raw transformer encoder, we capture the visual dependence through local-global self-attention and deem the multi-scale target regions as the “local” region. Using this fine-grained prior knowledge, we design the transformer decoder TPS to spread it to the subsequent search regions with high affinity to accurately estimate the bounding boxes. Considering that self-attention fails to model information interaction across channels between the template target and search regions, we develop a channel-wise cross-attention block within the TPS as compensation. Extensive experiments on the OTB100, UAV123, NFS, VOT2020, VOT2021, LaSOT, LaSOT_ext, TrackingNet and GOT-10k benchmarks show that the proposed box estimation network outperforms most existing box estimation methods. Furthermore, our trackers based on this estimation network exhibit a competitive performance against state-of-the-art trackers.