Zhiyu Shao

The development of artificial intelligence technology has led to the widespread utilisation of manipulators in various domains, including elderly assistance and rehabilitation, industrial picking, and family service. Their adaptive grasping ability for interactive objects has become an important research direction for intelligent operating systems. Especially in unstructured environments, the target objects have large differences in shape, size, and tactile attributes, and the traditional vision-based grasping strategy is difficult to adapt to complex scenes. In order to improve the robot’s ability to perceive and recognize object attributes, this paper proposes an adaptive grasping adjustment strategy based on the recognition of tactile attributes by combining the perceptual model and the slip detection model. In the specific implementation, the perception model is built using multilayer regression, and the slip detection model is constructed based on the Wilcoxon signed rank test, and the adaptive strategy is proposed based on the results of these two models. Through the experimental results of grasping objects with different attributes, the comprehensive grasping success rate reaches 90%, which verifies the adaptability and stability of the proposed strategy, and improves the robot’s ability to perceive and recognize object attributes.

Due to geographical dispersion and dynamic water conditions, traditional manual methods are confronted with challenges such as heavy workload, delayed flood reporting, and low data processing efficiency. This study proposes a real-time flow monitoring system for water gate stations in a certain tidal river section based on Internet of Things technology. The system adopts a hierarchical architecture, integrating the water level sensor, flowmeter and positioning module into the gate acquisition unit. By using TCP-based data transmission, reliable remote communication and cloud-based dynamic flow calculation are guaranteed. Methods such as hydraulic factors and single-tide push flow are adopted to improve the monitoring accuracy of tidal river sections. At the data application level, the visual interface dynamically displays real-time water levels, flow curves and flood alerts, meeting industry standards for data accuracy and compliance. The system achieves efficient hydrological data sharing and spatio-temporal analysis through the cloud platform and multi-terminal collaboration, forming a comprehensive monitoring chain from on-site perception to cloud processing and decision support. This provides a scientific basis for intelligent water resource management and flood control decision-making.

Analyses of teaching behaviors in physical education (PE) classrooms are critical for evaluating teaching quality. Traditional evaluation methods primarily rely on manual analysis, which suffers from complex coding procedures, low efficiency, and suboptimal accuracy, hindering long-term sustainability in teaching quality improvement. Artificial intelligence (AI) technology offers a novel approach by enabling real-time data collection, automated annotation, and in-depth analysis of teaching behaviors, thereby supporting sustainable PE teaching optimization. Leveraging the CogVLM2-Video model, the research presents a system for real-time data collection, automated annotation, and in-depth analysis of teaching behaviors. It consists of four key modules: The perception layer handles data acquisition and input providing foundational data for analysis. The platform layer manages data processing and storage, ensuring integrity and security for long-term evaluation. The model layer focuses on behavior recognition and analysis, employing advanced algorithms for precise interpretation of teaching behaviors. The application layer delivers real-time feedback and adaptive recommendations, promoting sustained teaching improvement. The system architecture was initially validated using 50 basketball lesson videos. Then, the recognition model was trained on a Kinetics-400 subset, achieving 92% accuracy and 95% consistency with manual annotations. These results demonstrate the system’s practical value and long-term applicability, offering an efficient, precise solution for PE classroom teaching behavior assessment.

Chemical oxygen demand (COD) serves as a key indicator of organic pollution in water bodies, and its rapid and accurate detection is crucial for environmental protection. Recently, ultraviolet–visible (UV–Vis) spectroscopy has gained popularity for COD detection due to its convenience and the absence of chemical reagents. Meanwhile, deep learning has emerged as an effective approach for automatically extracting spectral features and predicting COD. This paper proposes transforming one-dimensional spectra into two-dimensional spectrum images and employing convolutional neural networks (CNNs) to extract features and model automatically. However, training such deep learning models requires a vast dataset of water samples, alongside the complex task of labeling this data. To address these challenges, we introduce a transfer learning model based on VGG-16 for spectrum images. In this approach, parameters in the initial layers of the model are frozen, while those in the later layers are fine-tuned with the spectrum images. The effectiveness of this method is demonstrated through experiments conducted on our dataset, where the results indicate that it significantly enhances the accuracy of COD prediction compared to traditional methods and other deep learning methods such as partial least squares regression (PLSR), support vector machine (SVM), artificial neural network (ANN), and CNN-based methods.

Chemical oxygen demand (COD) is a critical parameter employed to assess the level of organic pollution in water. Accurate COD detection is essential for effective environmental monitoring and water quality assessment. Ultraviolet–visible (UV-Vis) spectroscopy has become a widely applied method for COD detection due to its convenience and the absence of the need for chemical reagents. This non-destructive and reagent-free approach offers a rapid and reliable means of analyzing water. Recently, deep learning has emerged as a powerful tool for automating the process of spectral feature extraction and improving COD prediction accuracy. In this paper, we propose a novel multi-scale one-dimensional convolutional neural network (MS-1D-CNN) fusion model designed specifically for spectral feature extraction and COD prediction. The architecture of the proposed model involves inputting raw UV-Vis spectra into three parallel sub-1D-CNNs, which independently process the data. The outputs from the final convolution and pooling layers of each sub-CNN are then fused into a single layer, capturing a rich set of spectral features. This fused output is subsequently passed through a Flatten layer followed by fully connected layers to predict the COD value. Experimental results demonstrate the effectiveness of the proposed method, as it was compared with three traditional methods and three deep learning methods on the same dataset. The MS-1D-CNN model showed a significant improvement in the accuracy of COD prediction, highlighting its potential for more reliable and efficient water quality monitoring.

Currently, the alarm functions of existing levee seepage monitoring systems are limited to single-parameter monitoring and lack rate-of-change alarms and correlation alarms. This can lead to false alarms, missed alarms, equipment failures, or unnecessary downtime. To enhance the intelligence of levee safety monitoring and seepage alarms, a levee seepage intelligent alarm system based on a Bidirectional Long Short-Term Memory (BILSTM) network model was designed and implemented. Firstly, data cleaning and preprocessing are carried out on the engineering safety monitoring operation data to reduce the influence of dirty data such as outliers and repetitive values on the accuracy of alarms. Secondly, for the correlation between the piezometric tube water levels of the levee and the Yangtze River water levels, a correlation analysis based on Mutual Information (MI) theory was conducted to minimize the effect of piezometric tube water level change delays on correlation. Finally, the BILSTM model was used to predict trends in these potentially abnormal data intervals. Based on engineering application requirements, alarm thresholds were established, and a multi-level alarm module was developed. Field operation test results show that the proposed method can accurately predict the piezometric tube water levels of levees, achieving intelligent alarms within the engineering safety monitoring system.

In order to effectively predict the changing trend of operating parameters in the pump unit and carry out fault diagnosis and alarm processes, a trend prediction model is proposed in this paper based on PCA-based multi-task learning (MTL) and an attention mechanism (AM). The multi-task learning method based on PCA was used to process the operating data of the pump unit to make full use of the historical data to extract the key common features reflecting the operating state of the pump unit. The attention mechanism (AM) is introduced to dynamically allocate the weight coefficient of common feature mapping for highlighting the key common features and improving the prediction accuracy of the model when predicting the trend of data change for new working conditions. The model is tested with the actual operating data of a pumping station unit, and the calculation results of different models are compared and analyzed. The results show that the introduction of multi-task learning and attention mechanisms can improve the stability and accuracy of the trend prediction model compared with traditional single-task learning and static common feature mapping weights. According to the threshold analysis of the monitoring statistical parameters of the model, a multi-stage alarm model of pump unit operation condition monitoring can be established, which provides a theoretical basis for optimizing operation and maintenance management strategy in the process of pump station management.

Pumping stations have undergone significant modernization and digitalization in recent decades. However, traditional virtual inspections often prioritize the visual experience and fail to effectively represent the haptic physical properties of devices during inspections, resulting in poor immersion and interactivity. This paper presents a novel virtual inspection system for pumping stations, incorporating virtual reality interaction and haptic force feedback technology to enhance immersion and realism. The system leverages a 3D model, crafted in 3Ds Max, to provide immersive visualizations. Multimodal feedback is achieved through a combination of haptic force feedback provided by a haptic device and visual information delivered by a VR headset. The system’s data platform integrates with external databases using Unity3D to display relevant information. The system provides immersive 3D visualizations and realistic force feedback during simulated inspections. We compared this system to a traditional virtual inspection method that demonstrated statistically significant improvements in task completion rates and a reduction in failure rates when using the multimodal feedback approach. This innovative approach holds the potential to enhance inspection safety, efficiency, and effectiveness in the pumping station industry.

In recent years, research on new technologies for expressing and exchanging information through tactile vibration has been the focus of researches. In this paper, by choosing a suitable coding scheme and a vibrating motor arrangement, we designed a device to express semantic information through vibrotactile. Three types of experiments were designed to test the usability of the encoding scheme and the device. Firstly, the vibration intensity was experimented when designing the encoding scheme, and the results showed that the encoding scheme was better with Braille units of 0.2 and 0.3 vibration intensities. In addition, the learning experiment and sentence recognition accuracy experiment were carried out to verify the usability of the device. The learning experiment results show that subjects were able to memorize Braille characters with an accuracy more than 90%, and to recognize a Chinese character (consisting of two Braille cells) with an average of 90.8% accuracy. The sentence recognition accuracy test experiment results show that the average recognition rate of the three poems used for the test was 93.33%. The device can be used for semantic information expression and touch-reading of Braille, and it can realize the reading experience of paper Braille.