Video captioning refers to automatic generate natural language sentences which summarize the video contents. Inspired by the visual attention mechanism of human beings, temporal attention mechanism has been widely used in video description to selectively focus on important frames. However, most existing methods based on temporal attention mechanism suffer from the problems of recognition error and detail missing, because temporal attention mechanism cannot further catch significant regions in frames. In order to address above problems, we propose the use of a novel spatial-temporal attention mechanism (STAT) within an encoder-decoder neural network for video captioning. The proposed STAT successfully takes into account both the spatial and temporal structures in a video, so it makes the decoder to automatically select the significant regions in the most relevant temporal segments for word prediction. We evaluate our STAT on two well-known benchmarks: MSVD and MSR-VTT-10K. Experimental results show that our proposed STAT achieves the state-of-the-art performance with several popular evaluation metrics: BLEU-4, METEOR and CIDEr.

Uyghur text localization in complex background images is a challenging task and has broad application prospects on many areas. Generally, Uyghur characters in images consist of strokes which have uniform features and are different from background in several aspects, such as color, intensity and texture. Based on this fact, we propose a FASTroke keypoint extractor which is fast and stroke-specific. Compared with the commonly used MSER detector, FASTroke produces two times less components and at lest 10% more characters. Moreover, the characters in a line also have uniform features, for example size, color and stroke width. Consequently, a component similarity based clustering is presented, which does not include componentlevel classification. It brings two benefits, no extra error is imported by component-level classifier and the computation is drastically reduced. The experiments confirm that our method has achieved the best performance on UICBI-500 benchmark dataset.

This article proposes an effective stepwise nontransient method for entering and exiting the active short circuit (ASC) of permanent magnet synchronous motor (PMSM) to protect the safety of vehicle's drive system when fatal faults occur in the electric vehicles (EV) powertrain system with high motor speed. It is found that excessive transient current and overshoot voltage inevitably occur when ASC enters and exits directly, which may cause the demagnetization of permanent magnet and failure of motor controller unit (MCU). The two-phase short circuit is utilized as a transition step before entering and exiting three-phase ASC according to the rotor magnet position, and almost no transient current and overshoot voltage exist during the whole short-circuit period. Then, there are no constraints of ASC for system design and the system material cost can be reduced. This approach is useful for function safety implement of MCU for EV application with fast attenuation rate of transient current, high robustness, and limited resources requirement. The method with effective suppression of transient current and overshoot voltage is verified by the experimental results from a 15 kW PMSM test platform.

As the use of radar and radar jammers increases, a radar device is likely to face interference from jammers or other radar devices. Traditional phase-coded pulse compression radar devices are widely used, but these tools struggle to overcome jamming and mutual interference. To solve this problem, we propose a novel chaotic-encrypted pulse compression radar signal based on radar identity (ID). Each radar has its own ID, which is encrypted with different chaotic binary sequences in every pulse period. The ambiguity function calculated for the coded radar signal is thumbtack-shaped, indicating that the signal has a good resolution. The received signal is used to range and decrypt in two channels: the range channel and the radar ID channel. The signals of the two channels are analyzed separately. Analyses of anti-barrage jamming and anti-mutual interference show that both channels perform well in terms of antijamming, while the antijamming ability is influenced by the processing gain, bit error rate (BER) and correlation function. In addition, the dual-channel antijamming method further improves the radar antijamming ability. The simulation result verifies the strong antijamming ability and high range resolution of the proposed radar signal, and the proposed antijamming method performs much better than the traditional phase-coded pulse compression radar signal in the antijamming scenario.

Deep recognition models are widely vulnerable to adversarial examples, which change the model output by adding quasi-imperceptible perturbation to the image input. Recently, Segment Anything Model (SAM) has emerged to become a popular foundation model in computer vision due to its impressive generalization to unseen data and tasks. Realizing flexible attacks on SAM is beneficial for understanding the robustness of SAM in the adversarial context. To this end, this work aims to achieve a targeted adversarial attack (TAA) on SAM. Specifically, under a certain prompt, the goal is to make the predicted mask of an adversarial example resemble that of a given target image. The task of TAA on SAM has been realized in the white-box setup by assuming access to prompt and model, which is thus less practical. To address the issue of prompt dependence, we propose a simple yet effective approach by only attacking the image encoder. Moreover, we propose a novel regularization loss to enhance the cross-model transferability by increasing the feature dominance of adversarial images over random natural images. Extensive experiments verify the effectiveness of our proposed simple techniques to conduct a successful black-box TAA on SAM.