Infrared imaging is a valuable technology for gas leakage detection due to its high sensitivity, long detection range, and high efficiency. Conventional target detection methods depend on manually extracting image features, which often leads to limited accuracy, low adaptability, and slow detection speeds. Deep learning technology offers a potential solution to these challenges; however, the increasing depth of neural networks imposes significant computational demands, posing challenges to real-time detection. This paper presents a compact and energyefficient gas detection system, implemented with a ZYNQ platform and an infrared camera. We propose a ZYNQ-based convolution accelerator to enhance gas plume detection from images captured by the infrared camera. Operating at a clock frequency of 130 MHz, the accelerator is capable of reaching a peak performance of 37.44 Gop/s, with power consumption of only 4.12 W. The system achieves a processing speed of 0.235 seconds per image, enabling real-time gas leakage detection.

Applying blockchain technology to a distributed Internet-of-Things (IoT) environment has emerged as a promising solution for enhancing system data security and mitigating the risk of privacy leakage. Meanwhile, to meet the computational demands of blockchain, IoT devices can offload their computational tasks to nearby edge nodes or even the cloud side, thereby compensating for their insufficient computational resources and available energy. This paper addresses the issue of computational task offloading for heterogeneous IoT devices in a distributed and dynamic environment. It first considers factors such as the delay in IoT execution of blockchain computations, energy consumption costs, and the payable budget required for offloading, and establishes a time-series optimization model. Subsequently, a swarm intelligence method is introduced into Multi-agent Reinforcement Learning to solve the optimization problem, effectively overcoming the challenges faced by purely Reinforcement Learning (RL)-based method, which are susceptible to hyperparameter tuning and prone to getting stuck in local optima, leading to sub-optimal solutions. Comparative experiments with several representative baseline methods shown that the proposed computational task offloading method exhibits relatively better convergence performance, solution quality across different network scales, and robustness.

The increase in mobile communication has escalated network load, necessitating more intelligent resource management in 5G and forthcoming networks. This research introduces a novel methodology to enhance network performance by integrating bird swarm optimization (BSO) with deep learning (DL). We investigate a configuration in which several users inside a single cell solicit services from proximate edge servers and remote cloud servers, employing Non-Orthogonal Multiple Access (NOMA) to optimize radio spectrum utilization. BSO, influenced by avian flocking patterns, aggregates users and distributes workloads among servers to reduce energy consumption, service latency, and operational expenses. Simultaneously, DL examines historical network data to forecast traffic and user requirements, enabling BSO to make immediate, educated decisions. We utilize queuing theory to model this system, addressing a complicated optimization problem that concurrently reduces energy, latency, and costs. In contrast to conventional solutions, our technology dynamically adjusts to fluctuating network conditions, providing an effective remedy for the requirements of 5G. Simulations demonstrate that the integrated BSO-DL approach diminishes energy usage, delays, and expenses by around 54 %, substantiating its efficacy in improving broadband performance. This research facilitates the development of more efficient and sustainable 5G networks , addressing the increasing demands of contemporary mobile communication through a scalable and intelligent approach.

Path planning is one of the key research topics in robotics. Nowadays, researchers pay more attention to reinforcement learning (RL) and deep learning (DL) because of RL's good generality, self-learning ability, and DL's super leaning ability. Deep deterministic policy gradient (DDPG) algorithm, which combines the architectures of deep Q-learning (DQN), deterministic policy gradient (DPG) and Actor-Critic (AC), is different from the traditional RL methods and is suitable for continuous action space. Therefore, TPR-DDPG based path planning algorithm for mobile robots is proposed. In the algorithm, the state is preprocessed by various normalization methods, and complete reward-functions are designed to make agents reach the target point quickly by optimal paths in complex environments. The BatchNorm layer is added to the policy network, which ensures the stability of the algorithm. Finally, experimental results of agents' reaching the target points successfully through the paths generated by the improved DDPG validate the effectiveness of the proposed algorithm.