Controller area network (CAN) protocol is an efficient standard enabling communication among electronic control units (ECUs). However, the CAN bus is vulnerable to malicious attacks because of a lack of defense features. In this article, a novel vehicle intrusion detection system (IDS) is developed. The challenge is that existing techniques of IDSs rarely consider attacks with small-batch, which are characterized by their small attack scale and concealed attack patterns, posing a significant threat to driving safety. To solve this problem, we developed an algorithm model that merges multidimensional long short-term memory (MD-LSTM) and self-attention mechanism (SAM), shortly named MULSAM. The MULSAM model was compared with other baseline models, including stacked long short-term memory (LSTM), MD-LSTM, etc. Experiments show that our approach has the best-detection accuracy (98.98%) and training stability. Further, to speed up the inference of MULSAM on edge, the hardware accelerator is implemented on FPGA devices using technologies, such as parallelization, modular, pipeline, and fixed-point quantization. Experiments show that our FPGA-based acceleration scheme has a better-energy efficiency than the CPU platform. Even with a certain degree of quantification, the acceleration model for MULSAM still displays a high-detection accuracy of 98.81% and a low latency of 1.88 ms.

The precise classification of human activities through sensor data collection and analysis addresses the broad demands in healthcare, security surveillance, and smart home applications amidst the rapid development of IoT technology. However, achieving high efficiency and accuracy remains a significant challenge for HAR algorithms. This paper proposes a HAR algorithm based on Transformer and Random Forest (Transformer-RF). The algorithm extracts and integrates multimodal features in the time domain, frequency domain, and statistical metrics, constructing one-dimensional and two-dimensional feature sets through feature transformation. The Transformer component, leveraging self-attention mechanisms, captures long-range dependencies and extracts global contextual information. Concurrently, the Random Forest component randomly selects features and samples, enhancing model diversity and improving complex human activity recognization capabilities. Experimental results demonstrate that compared with state-of-the-art algorithms, the Transformer-RF model achieves superior performance on both one-dimensional and two-dimensional feature sets, with an accuracy of up to 94.17%. The primary contribution of this paper lies in the introduction of an innovative Transformer-RF human activity recognization method, which not only ensures high accuracy but also exhibits excellent generalization capability and practical application potential. This study provides new insights and technical solutions for the field of human activity recognization, offering significant theoretical and practical value.

Human Activity Recognition has been extensively applied to fulfill tasks such as fall detection, human-computer interaction, virtual reality, etc. Existing radio frequency-based HAR methods, although overcoming limitations of wearable-, visual-, and acoustic-based sensing technology, still suffer from high costs and low efficiency, which limits their pervasive use. In this paper, we propose LoDiHAR, a low-cost, distributed HAR system leveraging Radio Frequency Identification technology. LoDiHAR employs low-cost and fully programmable commercial wireless components, providing full access to the PHY samples of the backscattered signals, in which signal phases can be extracted to infer different activities. Different from COTS RFID systems that adopt a polling interrogation scheme, LoDiHAR supports a distributed sensing scheme, which profiles human activities more efficiently. LoDiHAR addresses a series of technical challenges such as accurate phase extraction from backscattered signals, asynchronous distributed RF data fusion and insufficient training data. A Conditional Generative Adversarial Network framework combined with a Transformer model is designed for accurate time-series activity classification. LoDiHAR demonstrates pro-ficiency in recognizing eight types of human activities across diverse environments, achieving an accuracy of up to 94.9% while only costing 10% of the mainstream COTS RFID systems.

Quantized neural networks (QNNs) have become a standard operation for efficiently deploying deep learning models on hardware platforms in real application scenarios. An empirical study on German traffic sign recognition benchmark (GTSRB) dataset shows that under the three white-box adversarial attacks of fast gradient sign method, random + fast gradient sign method and basic iterative method, the accuracy of the full quantization model was only 55%, much lower than that of the full precision model (73%). This indicates the adversarial robustness of the full quantization model is much worse than that of the full precision model. To improve the adversarial robustness of the full quantization model, we have designed an adversarial attack defense platform based on field-programmable gate array (FPGA) to jointly optimize the efficiency and robustness of QNNs. Various hardware-friendly techniques such as adversarial training and feature squeezing were studied and transferred to the FPGA platform based on the designed accelerator of QNN. Experiments on the GTSRB dataset show that the adversarial training embedded on FPGA can increase the model's average accuracy by 2.5% on clean data, 15% under white-box attacks, and 4% under black-box attacks, respectively, demonstrating our methodology can improve the robustness of the full quantization model under different adversarial attacks.

As a promising way, controlling smart devices through gestures offers the benefits of non-contact interaction, efficiency and convenience. Previous researches on acoustic-based gesture recognition have mostly focused on near-field gestures within 1 meter and for a single user only. However, such a nearfield sensing scheme is inadequate to meet the growing demands for multi-person human-computer interaction in far-field spaces. In this paper, we present a novel acoustic-based room-scale gesture recognition system that is capable of recognizing gestures simultaneously performed by multi-user. Our approach achieves far-field sensing by examining the relationship between acoustic signal frame length and sensing range, and overcoming a series of practical challenges incurred by far-field sensing. To simultaneously detect and distinguish gestures of multiple persons, we divide the sensing area into multiple beamforming sub-scanning areas and apply binary search to detect multiple users, which allows for an efficient scanning process and facilitates real-time detection. Finally, we conduct a data augmentation scheme to enlarge the training data and apply a lightweight deep learning framework to classify different gestures. Extensive experiments confirm that our system enables multi-user gesture detection and can recognize nine gestures at a distance up to 7 meters.

In sub-wavelength lithography, traditional resolution enhancement techniques (e.g., OPC) cannot guarantee the optimality of the mask. In this paper, we present a novel inverse lithography method to solve the mask optimization problem. Recognizing that when formulated on a pixel-by-pixel basis with partially coherent optical models, the problem is a large-scale nonlinear optimization problem, we cast the optimization flow into a homotopy framework and apply an efficient numerical continuation technique. Compared to earlier pixel-based inverse lithography methods, our homotopy approach is not only more efficient, but also capable of naturally addressing the mask manufactureability problem. Experiment results in a state-of-the-art lithography environment show that our method generates high fidelity wafer images, and is 100× faster than previously reported inverse lithography method.