Search

The vulnerability of deep neural networks (DNNs) has been exposed by adversarial examples. Although the adversarial perturbations can be made visually imperceptible or photorealistic on any image, they have to be added offline on pre-captured static input in order to accomplish the malicious goal. As opposed to subtle distortion, real-time misclassification on streaming images can be realized by manipulating the objects in physical world. Recently, object-contactless physical attacks, as exemplified by a translucent sticker affixed to the lens of a camera, show that a sensor-enabled edge computing platform can be an alluring target of adversarial attack. Nevertheless, success rates of reported camera-based patch attacks are not high enough to overshadow other forms of evasion attacks even when they are performed under the white-box scenario. In this paper, we present a practical and robust fault injection approach cooperated with a hardware-friendly sparse strip pattern to deceive the deployed DNN device on real-time streaming images. The strip perturbation is generated in a line-offset form by an optimization algorithm. It can be injected into camera data lane between the image sensor and the endpoint node stealthily without disturbing the data traffic through an interface bridge implemented by a tiny off-the-shelf FPGA device. We demonstrate our attack on the Raspberry Pi 4 platform with the Pi camera v2 and the Intel NCS2 inference stick. By evaluating 280 physically captured images from ten objects in 28 viewing angles, we show that the proposed attack on four ImageNet models including ResNet50, MobileNet-v2, Inception-v3 and EfficientNet-B0 can achieve 89.2% ∼ 96.1% success rates. National Research Foundation (NRF) Submitted/Accepted version This research is supported by the National Research Foundation, Singapore, under its National Cybersecurity Research & Development Programme/Cyber-Hardware Forensic & Assurance Evaluation R&D Programme (Award: CHFA-GC1-AW01).

Summary Anthocyanins protect plants from biotic and abiotic stressors and provide great health benefits to consumers. In this study, we cloned four genes (Red Lettuce Leaves 1 to 4: RLL1 to RLL4) that contribute to colour variations in lettuce. The RLL1 gene encodes a bHLH transcription factor, and a 5‐bp deletion in some cultivars abolishes its function to activate the anthocyanin biosynthesis pathway. The RLL2 gene encodes an R2R3‐MYB transcription factor, which was derived from a duplication followed by mutations in its promoter region. The RLL3 gene encodes an R2‐MYB transcription factor, which down‐regulates anthocyanin biosynthesis through competing with RLL2 for interaction with RLL1; a mis‐sense mutation compromises the capacity of RLL3 to bind RLL1. The RLL4 gene encodes a WD‐40 transcription factor, homologous to the RUP genes suppressing the UV‐B signal transduction pathway in Arabidopsis; a mis‐sense mutation in rll4 attenuates its suppressing function, leading to a high concentration of anthocyanins. Sequence analysis of the RLL1‐RLL4 genes from wild and cultivated lettuce showed that their function‐changing mutations occurred after domestication. The mutations in rll1 disrupt anthocyanin biosynthesis, while the mutations in RLL2, rll3 and rll4 activate anthocyanin biosynthesis, showing disruptive selection for leaf colour during domestication of lettuce. The characterization of multiple polymorphic genes in this study provides the necessary molecular resources for the rational breeding of lettuce cultivars with distinct levels of red pigments and green cultivars with high levels of health‐promoting flavonoids.