Traditional disease and pest detection technology employ cloud-based deep learning, which facing the pressures, such as high data storage and communication costs, unbalanced and insufficient data from orchards, diversity of pests and diseases, and complex detection environments. In this article, we propose a multiple pest detection technique based on federated learning (FL) and improved faster region convolutional neural network (R-CNN). As the new distributed computing model, FL can derive a shared model integrating the advantages of data from all parties without uploading local data and also reduces the communication cost. A restriction $M$ is added to the FL algorithm to ensure the convergence of the model and improve the training speed. According to the original faster R-CNN network, ResNet-101 is used instead of VGG-16 to construct the base convolutional layer to maintain the original structure of small-sized targets and improve the detection speed. Then, the multisize fusion of feature maps from different convolutional layers is performed to improve the detection accuracy of multisize multiple pests and diseases. Finally, a soft-nonmaximum suppression (NMS) algorithm is proposed to solve the apple obscured problem after the region proposal network (RPN). The experimental results show that the improved faster R-CNN can achieve an average accuracy of 90.27% on multiple pest detection, and the detection time is only 0.05 s per image. After using FL, the mean average precision (mAP) of the model reached 89.34% and the model training speed was improved by 59%.

To assist patients with restricted mobility to control wheelchair freely, this paper presents an eye-movement-controlled wheelchair prototype based on a flexible hydrogel biosensor and Wavelet Transform-Support Vector Machine (WT-SVM) algorithm. Considering the poor deformability and biocompatibility of rigid metal electrodes, we propose a flexible hydrogel biosensor made of conductive HPC/PVA (Hydroxypropyl cellulose/Polyvinyl alcohol) hydrogel and flexible PDMS (Polydimethylsiloxane) substrate. The proposed biosensor is affixed to the wheelchair user’s forehead to collect electrooculogram (EOG) and strain signals, which are the basis to recognize eye movements. The low Young’s modulus (286 KPa) and exceptional breathability (18 g m⁻² h⁻¹ of water vapor transmission rate) of the biosensor ensures a conformal and unobtrusive adhesion between it and the epidermis. To improve the recognition accuracy of eye movements (straight, upward, downward, left, and right), the WT-SVM algorithm is introduced to classify EOG and strain signals according to different features (amplitude, duration, interval). The average recognition accuracy reaches 96.3%, thus the wheelchair can be manipulated precisely.

The existing transmission line detection technologies based on cloud computing are faced with the problems such as low response speed, high communication cost, and the difficulty to obtain scattered data. This paper proposes a transmission line defect detection method based on adaptive Federated Learning (FL), which can adaptively change the data volume of the next training round according to the training energy consumption of each round. As a result, the proposed method can achieve the optimal number of communication between local training cost and global communication efficiency, which greatly improves the training speed and reduces the communication cost. Furthermore, a lightweight Mask Region-Convolutional Neural Networks (R-CNN) that adopts the MobileNet architecture to replace the original network is designed for the edge, which reduces the required number of parameters and improves the detection speed. The experimental results show that the model training efficiency of the proposed adaptive FL is 70% higher than the centralized cloud computing, and the computing cost is saved by about 40%. Compared with the original Mask R-CNN, the proposed lightweight Mask R-CNN model can achieve 1.1% higher precision, 10% shorter training time, and 40% faster detection speed.

Due to the limitation of the computing power and energy resources, an unmanned aerial vehicle (UAV) team usually offloads the inspection task to the cloud for processing when performing emergency fault inspection, which will lead to low efficiency of transmission line inspection. In order to solve the above problems, this paper proposes a task offloading strategy based on deep reinforcement learning (DRL), aiming for the application of a multi-UAV and single-edge server. First, a “device-edge-cloud” collaborative offloading architecture is constructed in the UAV edge environment. Secondly, the problem of offloading power line inspection tasks is classified as an optimization problem to obtain the minimum delay under the constraints of edge server computing and communication resources. Finally, the problem is constructed as a Markov decision, and a deep Q-network (DQN) is used to obtain the minimum delay of the system. In addition, an experience replay mechanism and a greedy algorithm are introduced in the learning process to improve the offloading accuracy. The experimental results show that the proposed offloading strategy in this paper saves 54%, 37% and 26% of the task completion time, respectively, compared with local offloading, cloud offloading and random offloading. It effectively reduces the UAV inspection delay and improves the transmission line inspection efficiency.

The existing techniques for detecting defects in photovoltaic (PV) components have some drawbacks, such as few samples, low detection accuracy, and poor real-time performance. This paper presents a cloud-edge collaborative technique for detecting the defects in PV components, based on transfer learning. The proposed cloud model is based on the YOLO v3-tiny algorithm. To increase the detection effect of small targets, we produced a third prediction layer by fusing the shallow feature information with the stitching layer in the second detection scale and introducing a residual module to achieve improvement of the YOLO v3-tiny algorithm. In order to further increase the ability of the network model to extract target features, the residual module was introduced in the YOLO v3-tiny backbone network to increase network depth and learning ability. Finally, through the model’s transfer learning and edge collaboration, the adaptability of the defect-detection algorithm to personalized applications and real-time defect detection was enhanced. The experimental results showed that the average accuracy and recall rates of the improved YOLO v3-tiny for detecting defects in PV components were 95.5% and 93.7%, respectively. The time-consumption of single panoramic image detection is 6.3 ms, whereas the consumption of the model’s memory is 64 MB. After cloud-edge learning migration, the training time for a local sample model was improved by 66%, and the accuracy reached 99.78%.

Existing short-term traffic flow prediction methods are based on centralized learning frameworks, which do not consider the diversity of data, leading to limited predictive performance. Additionally, these methods face challenges related to computational efficiency and data privacy. This paper proposes a distributed short-term traffic flow prediction method (PFL-GTCN) that integrates federated learning and TCN. In a distributed framework, multiple edge computing nodes collaborate to complete model training. First, the spectral clustering algorithm divides the spatial graph of the traffic network into multiple subgraphs with similar traffic characteristics. At the same time, G-TCN(Graph-Temporal Convolutional Network) networks are deployed on the edge computing nodes to capture the complex spatiotemporal dependencies within the traffic network. Finally, a personalized federated learning (PFL, Federated Learning with Dynamic Parameter Weights) approach with dynamic weight parameters is introduced, which dynamically adjusts the aggregation weight parameters based on the contribution of the clients. This protects data privacy and reduces communication costs. Experiments conducted on real datasets demonstrate that the proposed prediction method outperforms single models and other graph convolution networks by 1.97%-11.11%. The PFL-GTCN method outperforms traditional centralized training methods by 0.23%-11.22%.