The connectivity and intelligence of Internet of Things (IoT) equipment offer improved services, but several technical challenges have emerged in recent years that hinder the widespread application of IoT, e.g., security and safety. Cyber-security and privacy countermeasures are widely used in IoT equipment, and many studies have been conducted. However, an important aspect that is often overlooked in security literature is IoT equipment’s physical security and safety, namely, preventing IoT equipment from vandalism and theft. Therefore, this article provides an overview of IoT equipment’s physical security and safety to draw attention to new research opportunities in this area. Afterward, we discuss, among other aspects, antitheft and antivandalism schemes along with circuit and system design, additional sensing devices, biometry and behavior analysis, and tracking methods. Besides, we summarize the artificial intelligence solutions for the physical security and safety of IoT equipment. Finally, we conclude with four future research opportunities.

Solar Insecticidal Lamp Internet of Things (SIL-IoT) nodes are susceptible to failures due to the harsh environment, burn-in, theft, and vandalism in the agricultural setting. Most state-of-the-art research mainly focuses on fault detection without considering hardware and communication performance in terms of potential fault modes, energy consumption, and network loads. This study presents a completely decentralized solution, namely a SIL-oriented binary sliding window-based fault self-detection scheme (SILOS), which can be performed on each SIL-IoT node. The problem we are trying to answer is how to detect faults as accurately as possible while keeping the communication overhead, memory, and computational costs low. Specifically, we develop a fault dictionary concept to 1) model the faults of SIL-IoT nodes; 2) construct the fault dictionary according to the characteristics of measurements; and 3) detect faults via the fault dictionary. In addition, a binary-based sliding window (BSW) fault self-detection approach is proposed to save detection costs and reduce the false alarm rate (with only 92 B system caches). A series of experiments are performed to evaluate the performance of the proposed method. The result demonstrates that the BSW method can detect faults with an average accuracy of 99.14% with less than 1% energy consumption. By only sending the fault code, 71-B data (i.e., data transmitting and forwarding) can be reduced, saving energy consumption, and decreasing network congestion.

Sensor faults can produce abnormal and spurious observations in the solar insecticidal lamp Internet of Things (SIL-IoTs) system. Early detection and identification of the sensor node’s abnormality are critical to ensure the SIL-IoTs system’s reliability. In this study, we propose a lightweight separable 1D convolution neural network that can be implemented in SIL-IoTs nodes to identify sensor faults, reduce detecting delay, and decrease data transmission. However, the reliability of data acquired by sensors is decreased because a SIL-IoTs node releases high voltage pulse discharge (a kind of electromagnetic interference) when pests collide with its metal mesh. This kind of data fluctuation impacts fault diagnosis accuracy. Consequently, fault-related feature maps and temporal signals are characterized via a novel time and channel attention module (TCAM) method, which contributes to separating electromagnetic interference noise from sensor faults of SIL-IoTs nodes. A real-world testbed is applied to validate the effectiveness of the proposed method on sensor fault diagnosis in the SIL-IoTs system. Experimental results demonstrate that the proposed method can detect four typical sensor faults with the best trade-off between accuracy (99.9% average accuracy and 97.6% average F1-score) and efficiency (351 KB inference model size and 4.33 W average energy consumption on Raspberry Pi).

Carrying out comprehensive green prevention and control on pests and plant diseases is important for improving quality of agricultural products, reducing usage of agricultural pesticides, and controlling losses. As an important approach of physical control, Solar Insecticidal Lamp has the counting function by killing pests during its working period, further obtaining the information of pest population density in the area covered by the SIL. Motivated by the current voltage recognition and counting technology, which cannot cope with the situation of killing a pest by continuous discharge, thus cannot provide an accurate insecticidal quantity, we proposed a novel accurate insecticidal counting method from the aspect of hardware and software. In the aspect of hardware, the SIL testbed is modified by changing high voltage package, and adding voltage comparator module and sound sensor module, further collecting the data Pulse Number of Insecticidal Discharges and Pulse Number of Insecticidal Sounds. In the aspect of software, we experimentally validated the relationship between PNID and PNIS as well as insecticidal quantity by seven different machine learning algorithms. Finally, we provide compelling experimental results and discuss strong evidence that points to a future need for lightweight edge computing on SIL, as a valuable consumer electronics.

The combination of solar insecticidal lamps (SILs) and wireless sensor networks has spawned a green and efficient solution for agricultural pest control, called solar insecticidal lamps Internet of Things (SIL-IoTs). In realistic large-scale SIL-IoTs deployment scenarios, the integrated pest information collected across the network enables effective localization of pest-dense areas. However, the problem of asymmetric links caused by various factors, such as irregular wireless communication range and discharge interference of nodes, is the main obstacle to the deployment of SIL-IoTs. Motivated by this problem, the pest-dense area localization strategy (PALS) based on asymmetric links is proposed. First, the asymmetric nodes in the network are judged by analyzing the one-hop and two-hop information of SIL nodes. Subsequently, the Gabriel graph or relative neighborhood graph planarization algorithm is used to planarize the symmetric links in the network. Then, the quick rejection method and straddle experiment are used to remove the cross sections after planarization. Finally, by counting the number of SIL node discharges and facilitating the left-hand rule, PALS successfully reduces the difference between the calculated and actual pest-dense areas. The experiments showed that PALS achieves an average improvement of 15% and a maximum improvement of up to 42.2% across the four experimental settings, indicating its higher accuracy and robustness compared with the state-of-the-art algorithms.