Machine Learning (ML) and Internet of Things (IoT) are complementary advances: ML techniques unlock the potential of IoT with intelligence, and IoT applications increasingly feed data collected by sensors into ML models, thereby employing results to improve their business processes and services. Hence, orchestrating ML pipelines that encompass model training and implication involved in the holistic development lifecycle of an IoT application often leads to complex system integration. This article provides a comprehensive and systematic survey of the development lifecycle of ML-based IoT applications. We outline the core roadmap and taxonomy and subsequently assess and compare existing standard techniques used at individual stages.

In wearable-based human activity recognition (HAR) research, one of the major challenges is the large intra-class variability problem. The collected activity signal is often, if not always, coupled with noises or bias caused by personal, environmental, or other factors, making it difficult to learn effective features for HAR tasks, especially when with inadequate data. To address this issue, in this work, we proposed a Behaviour Pattern Disentanglement (BPD) framework, which can disentangle the behavior patterns from the irrelevant noises such as personal styles or environmental noises, etc. Based on a disentanglement network, we designed several loss functions and used an adversarial training strategy for optimization, which can disentangle activity signals from the irrelevant noises with the least dependency (between them) in the feature space. Our BPD framework is flexible, and it can be used on top of existing deep learning (DL) approaches for feature refinement. Extensive experiments were conducted on four public HAR datasets, and the promising results of our proposed BPD scheme suggest its flexibility and effectiveness. This is an open-source project, and the code can be found at http://github.com/Jie-su/BPD

Electronic Stability Program (ESP) is widely used in modern vehicles. Its safety and stability largely depend on the strength and reliability of the MEMS gyroscope. However, the tight coupling between this sensor and the environment brings significant safety hazards to the vehicle. In this study, we describe the physical vulnerability of gyroscopes to high-frequency acoustics and introduce methods for finding resonant frequencies. We devised two methods to inject the attack signal into audio files to make the acoustic attack more stealthy. The realized attack is non-intrusive and does not require tampering with the ESP hardware device, making attack detection more difficult. We also consider a neural network-based defense strategy and verify its effectiveness. The construction of the vehicle simulation system and the above experiments are completed in the co-simulation environment of Carsim and Simulink.

The rapid growth of the Internet of Things (IoT) has led to the widespread adoption of the IoT networks in numerous digital applications. To counter physical threats in these systems, automatic modulation classification (AMC) has emerged as an effective approach for identifying the modulation format of signals in noisy environments. However, identifying those threats can be particularly challenging due to the scarcity of labeled data, which is a common issue in various IoT applications, such as anomaly detection for unmanned aerial vehicles (UAVs) and intrusion detection in the IoT networks. Few-shot learning (FSL) offers a promising solution by enabling models to grasp the concepts of new classes using only a limited number of labeled samples. However, prevalent FSL techniques are primarily tailored for tasks in the computer vision domain and are not suitable for the wireless signal domain. Instead of designing a new FSL model, this work suggests a novel approach that enhances wireless signals to be more efficiently processed by the existing state-of-the-art (SOTA) FSL models. We present the semantic-consistent signal pretransformation (ScSP), a parameterized transformation architecture that ensures signals with identical semantics exhibit similar representations. ScSP is designed to integrate seamlessly with various SOTA FSL models for signal modulation recognition and supports commonly used deep learning backbones. Our evaluation indicates that ScSP boosts the performance of numerous SOTA FSL models, while preserving flexibility.

In Earth observation activities, cloud severely affects the interpretation of the high-resolution imagery, generated by optical satellites. Therefore, removing clouds from optical imagery becomes a topic of interest in the remote sensing field. Currently, most methods use auxiliary Synthetic Aperture Radar (SAR) images to reconstruct optical images by merging SAR and optical images into a deep learning network. However, the speckle noise of the SAR image is not taken into the consideration during feature fusion processing, leading to the blurry edges in the reconstructed optical images. To get fine-grained optical images, we propose a novel cloud removal framework based on the edge fusion of SAR and optical images. Firstly, the edge feature of SAR images is extracted by the GRHED. As the prior knowledge, it can provide fine-grained edge information for subsequent reconstruction work. Then channels from three modal data are stacked to guide the reconstruction of optical images by exploiting their correlations and interactions. Furthermore, a structural similarity (SSIM) loss function is introduced to optimize the training network and improve the coherence of the image structure. Experimental results confirm its advantages on the SEN12MS-CR dataset.

In recent years, deep learning (DL) techniques have been extensively utilized for specific emitter identification through the extraction of RF fingerprints. A significant challenge that DL models face in real-world scenarios is the continuous emergence of new wireless devices, such as unknown drones that suddenly appear in the sky. In these situations, the radio monitoring system must be capable of detecting these unknown devices (open-set recognition) and incrementally updating the DL model’s knowledge using only a few captured samples. This requirement presents two main challenges: 1) Incremental updates from few-shot samples can lead to catastrophic forgetting and overfitting in DL models; 2) Constructing reliable open-set thresholds for new devices with few-shot samples is difficult. To address these challenges, we propose a novel few-shot open-set incremental learning (FSOSIL) framework through meta-learning for RF fingerprint recognition, named Meta–RFF. The core idea of Meta–RFF is to simulate few-shot incremental learning and open-set recognition scenarios by constructing numerous pseudo-FSOSIL tasks and meta-training them. To enhance the open-set recognition capability, we design RF feature augmentation, open loss, and adaptive open-set thresholding modules. The algorithm’s effectiveness is validated on the large-scale aircraft recognition dataset (ADS-B), showing an improvement in closed-set accuracy and open-set AUROC of the new class by approximately 10-20% compared to other algorithms with 1-shot. We also demonstrate the algorithm’s effectiveness in a real-world test bed.