Few-shot object detection, which focuses on detecting novel objects with few labels, is an emerging challenge in the community. Recent studies show that adapting a pre-trained model or modified loss function can improve performance. In this paper, we explore leveraging the power of Contrastive Language-Image Pre-training (CLIP) and hard negative classification loss in low data setting. Specifically, we propose Re-scoring using Image-language Similarity for Few-shot object detection (RISF) which extends Faster R-CNN by introducing Calibration Module using CLIP (CM-CLIP) and Background Negative Re-scale Loss (BNRL). The former adapts CLIP, which performs zero-shot classification, to re-score the classification scores of a detector using image-class similarities, the latter is modified classification loss considering the punishment for fake backgrounds as well as confusing categories on a generalized few-shot object detection dataset. Extensive experiments on MS-COCO and PASCAL VOC show that the proposed RISF substantially outperforms the state-of-the-art approaches. The code will be available.

This paper proposes the application of compressive sensing (CS) to reduce the measurement time in electrochemical impedance spectroscopy (EIS) for lithium-ion batteries. EIS is a non-destructive frequency response technique that provides valuable information on the state and degradation mechanisms occurring inside a battery. However, EIS measurement time is lengthy, making it impractical for evaluating the state of operating cells. CS is a signal-processing technique that enables the efficient acquisition and reconstruction of signals from a reduced number of measurements. The study aims to identify a suitable transform domain using dictionary learning that facilitates the adoption of CS techniques for the compression of the EIS data obtained from lithium-ion batteries. Thanks to the reduced number of EIS measurements, the proposed CS-based EIS achieves approximately 40% reduction in measurement time for open-source and in-house collected data, respectively, with minimal accuracy degradation.

Lithium plating poses a significant challenge to the performance and safety of lithium-ion batteries. As a non-destructive detection method, voltage relaxation profile (VRP) analysis shows great potential for effective lithium plating detection. However, the conventional VRP analysis suffers from a lengthy experimental time requirement, which fundamentally hinders the further development of the VRP-based detection. To overcome this limitation, this paper proposes a new lithium plating indicator by fully exploiting distinctive behaviors of differential voltage$(dV/dt)$profiles depending on the amount of lithium plating. The proposed indicator focuses on a local maximum value in the$dV/dt$profiles, which allows for achieving robust and fast prediction under cell-to-cell variation and aging while preserving the quantitative information obtained from the conventional detection. Based on experiments using battery cells, the adoption of the proposed indicator reduces the lithium plating detection time by 40% with a minimum error compared with the conventional method. Furthermore, applying the same approach to reference datasets further validates the efficacy of the proposed indicator across various conditions, including varying charging currents and temperatures, which confirms the reliability and accuracy of the proposed fast lithium plating detection.

Unmanned Aerial Vehicles (UAVs) are a promising application to deal with diverse industrial and social problems. In order to increase the utilization and reliability of UAVs, batteries play a big role. The flight feasibility evaluation of UAVs should be done in terms of battery before starting the flight missions. The properties of batteries are sensitive to temperature since they generate power through an electrochemical reaction. In this paper, a battery temperature estimator is presented for a system-level UAV flight evaluation considering the usability of end-users. The proposed estimator consists of flight history-aware data preprocessing and a deep neural network-based model instead of the conventional physics-based approach that requires in-depth theories. The flight data is collected through the actual flight experiments of a UAV for training and validation. The proposed method achieves a temperature estimation error of less than$2.02^{\circ}\mathrm{C}$compared with the actual measured data of a UAV battery. The effectiveness of the proposed estimator is presented by case studies.

Electrochemical impedance spectroscopy (EIS) is a non-invasive method for analyzing battery states based on impedance measurements. With the widespread use of high-capacity lithium-ion batteries in the range of mΩ impedance, achieving highly accurate impedance measurements becomes crucial for precise battery examination. EIS systems frequently utilize a digital lock-in amplifier (DLIA) to achieve ultra-precision impedance readings, but it requires a long settling time. This study proposes an innovative EIS architecture with a two-stage Hilbert transform that significantly reduces the measurement time by widening the bandwidth of noise suppression low-pass filter while maintaining high accuracy. It achieves a substantial 66% reduction in the estimated settling time at the lower bound frequency of 1 Hz and a 57% reduction in the total measurement time across the frequency range of 1-to-1k Hz.