The profound impacts of severe air pollution on human health, ecological balance, and economic stability are undeniable. Precise air quality forecasting stands as a crucial necessity, enabling governmental bodies and vulnerable communities to proactively take essential measures to reduce exposure to detrimental pollutants. Previous research has primarily focused on predicting air quality using only time-series data. However, the importance of remote-sensing image data has received limited attention. This paper proposes a new multi-modal deep-learning model, Res-GCN, which integrates high spatial resolution remote-sensing images and time-series air quality data from multiple stations to forecast future air quality. Res-GCN employs two deep-learning networks, one utilizing the residual network to extract hidden visual information from remote-sensing images, and another using a dynamic spatio-temporal graph convolution network to capture spatio-temporal information from time-series data. By extracting features from two different modalities, improved predictive performance can be achieved. To demonstrate the effectiveness of the proposed model, experiments were conducted on two real-world datasets. The results show that the Res-GCN model effectively extracts multi-modal features, significantly enhancing the accuracy of multi-step predictions. Compared to the best-performing baseline model, the multi-step prediction’s mean absolute error, root mean square error, and mean absolute percentage error increased by approximately 6%, 7%, and 7%, respectively.

In order to improve the detection accuracy of printed circuit board (PCB), this paper proposes a PCB defect detection model SWS-Yolov8n based on the Yolov8 model. Firstly, the Shuffle Attention (SA) module is introduced to make the model have the ability to randomly fuse and exchange feature information at PCB defects; Secondly, it is proposed to improve the loss function to a dynamic non monotonic loss function, so that the model shifts a portion of the focus from high-quality anchor frames to normal mass; Finally, space to depth layer and non step convolutional layer are adopted to address the problem of fine-grained information on defect location is easy to be lost and the learning efficiency is low. After expanding the public dataset PKU-Market-PCB, experiments showed that the improved SWS-Yolov8n model increased the mean average accuracy (mAP) value by 4.47% compared to the original model. When applied to PCB defect detection tasks, it can effectively improve detection accuracy. In the future, this model can be further lightweight improved and deployed on embedded systems and mobile devices. Keywords: deep learning; PCB defect detection; Yolov8; attention mechanism; loss function

Rehabilitation robots have gained considerable focus in recent years, aiming to assist immobilized patients in regaining motor capabilities in their limbs. However, most current rehabilitation robots are designed specifically for either upper or lower limbs. This limits their ability to facilitate coordinated movement between upper and lower limbs and poses challenges in accurately identifying patients’ intentions for multi-limbs coordinated movement. This research presents a multi-postures upper and lower limb cooperative rehabilitation robot (U-LLCRR) to address this gap. Additionally, the study proposes a method that can be adjusted to accommodate multi-channel surface electromyographic (sEMG) signals. This method aims to accurately identify upper and lower limb coordinated movement intentions during rehabilitation training. By using genetic algorithms and dissimilarity evaluation, various features are optimized. The Sine-BWOA-LSSVM (SBL) classification model is developed using the improved Black Widow Optimization Algorithm (BWOA) to enhance the performance of the Least Squares Support Vector Machine (LSSVM) classifier. Discrete movement recognition studies are conducted to validate the exceptional precision of the SBL classification model in limb movement recognition, achieving an average accuracy of 92.87%. Ultimately, the U-LLCRR undergoes online testing to evaluate continuous motion, specifically the movements of “Marching in place with arm swinging”. The results show that the SBL classification model maintains high accuracy in recognizing continuous motion intentions, with an average identification rate of 89.25%. This indicates its potential usefulness in future rehabilitation robot-active training methods, which will be a promising tool for a wide range of applications in the fields of healthcare, sports, and beyond.

To meet the various need of stroke patients’ rehabilitation training and carry out complex task training in real scenes, the structure of a lower limb rehabilitation robot with movements in the sagittal plane and coronal plane is usually complicated. A new sitting/lying lower limb rehabilitation robot (LOBO) with a simple mechanism form is proposed, which is designed based on a 2-PRR parallel mechanism. First, the kinematics, singularity, and condition number of the 2-PRR parallel mechanism are analyzed, which provides the basis for mechanism parameter design. Then, through the proportional–derivative control principle, real-time tracking of LOBO’s designed trajectory is realized. Finally, the length parameters of volunteers’ lower limbs are collected, and experimental verification is conducted in LOBO’s passive training mode. The experimental results show the feasibility of LOBO’s movement in the human sagittal and coronal planes. LOBO will help human lower limbs realize the synchronous continuous rehabilitation training of hip, knee, and ankle joints spatially, which could drive the rehabilitation movement of patients’ lower limbs in the sagittal plane and coronal plane in future clinical research. LOBO can also be applied to muscle strength training for the elderly to combat the effects of aging.