Advanced Computational Intelligence and Intelligent Informatics

Tube alignment is an important task characterized by the complexity of processing and aligning large-diameter tube segments. Traditional optoelectronic measurement systems consist of a large number of components, and a significant amount of time is spent on preliminary system calibration. In this article, a method of visual servo control using a FishEye camera and structured light is proposed to optimize the tube center position. The proposed technical vision system contains fewer elements, simplifying the system configuration and setup. Firstly, the optical axis of the technical vision system is aligned with the inner surface axis of the tube through calibration. Then, laser points belonging to the inner surface of the tube are extracted. Next, a visual servo control law is applied to control the deviation between the desired and current positions of the tube center. Finally, experiments are conducted in a developed virtual environment that replicates the real technological process to demonstrate the effectiveness and reliability of the proposed method. Real-world experiments involve measurement uncertainties, making it challenging to compare different methods. Additionally, it is difficult or impossible to estimate certain parameters in real conditions. The virtual environment helps overcome these issues. The simulation environment and testing code are available online: https://​github.​com/​kholodilinivan/​Pipe-Alignment-IBVS.

With the development of AI technology, the accuracy of speech recognition and the range of its use are advancing. With AI-based speech processing, many services such as automatic speech transcription, speech recognition, and speech summarization are now available. In this paper, a method is proposed for determining whether each utterance is important or not based on acoustic information of the utterances. As acoustic features, statistical measures of various acoustic features for each utterance are used. To determine the importance of the utterance, the importance of the transcribed text is labeled using the LLM chat system and trained as a supervised input. In this experiment, TED video speeches on YouTube are used as speech materials. A machine learning model with a random forest classifier is used to determine the importance. As a result, a model that can classify the importance for the training data is obtained. It is found that statistical measures of acoustic features related to the fundamental frequency (F0) of the utterance are frequently used as important features for classification. However, when evaluated on test data, it is found that sufficient accuracy is not achieved, and further examination is necessary.

As deep learning continues to evolve, datasets are becoming increasingly larger, leading to higher costs for manual labeling. Zero-shot learning eliminates the need for substantial labor costs to label training datasets. It even allows the model to predict new classes with slight modifications. However, the accuracy of zero-shot learning still remains relatively low and makes practical applications challenging. Some recent research has tried to improve accuracy by manually annotating features, but this approach again requires labor-intensive input. To reduce these labor costs, we employ ChatGPT, which can generate features automatically without any manual involvement. Importantly, this approach maintains high accuracy levels, surpassing other automated methods.

With the increasing demand for public safety, the field of abnormal human behavior recognition has undergone significant development. In addressing the low accuracy issue of existing abnormal behavior recognition algorithms due to factors such as environmental influences, changes in viewpoint, and scale variations, this study proposed an improved Spatial temporal graph convolutional network. By incorporating spatial attention and channel attention mechanisms at relevant positions in the network, a dynamic optimization of the skeletal structure graph of the human body was achieved. This ensured that key nodes expressing motion information in the skeletal graph received greater weight values, ultimately improving the accuracy of abnormal behavior classification. To this end, an abnormal behavior dataset was constructed and transformed into skeletal information recognizable by the proposed algorithm using OpenPose. Extensive experiments were conducted on this dataset as well as the large-scale NTU RGB + D dataset using the improved algorithm. The results demonstrate that the algorithm has achieved an increase of approximately 5% in recognition accuracy compared to its pre-improvement state, placing it among the top-performing algorithms in various comparative evaluations.

Object detection algorithms can assist in detecting the helmet-wearing status of electric bicycle riders, thereby saving regulatory manpower costs. However, there is currently a lack of standardized and publicly available datasets. Additionally, the basic YOLOv5s object detection algorithm, due to its limited feature extraction capabilities, may lead to numerous instances of both false negative and false positive. To enhance the model’s focus on critical information within the feature maps, this paper introduces the CBAM attention mechanism module into the Backbone section of YOLOv5s. This module sequentially infers attention maps from the input feature maps along both channel and spatial dimensions independently, and then multiplies these attention maps with the input feature maps to achieve adaptive feature optimization. This paper have established self-built dataset for experimental research, and the results indicate that compared to the original YOLOv5s model, the proposed method has improved the model’s overall mAP score by 1.89%.

In the production of industrial products, surface defect detection is mostly carried out through manual inspection. However, this detection method has several shortcomings, such as low efficiency, limited accuracy, and high inspection costs. To address these issues, we design an improved random sampling consistency (RANSAC) algorithm based on adaptive parameters of 3D point cloud data for plane defect detection. The main steps of our algorithm include the down sampling function which contains adaptive parameters, optimized based on KD-tree proximity substitution method. Our algorithm also includes the RANSAC segmentation and fitting plane function of adaptive parameters. Experimental results demonstrate that our algorithm can accurately identify protrusions or indentations defects of 1 mm or larger in those planes based on point clouds data, with a recognition rate more than 90%. The experimental results validate the suitability of our algorithm for industrial applications, offering an efficient and cost-effective solution for plane defect detection.

To address the challenge of determining the parameters of registration algorithms under low point clouds overlap, which hinders automatic and efficient calculations, we introduce an improved variant of the TrICP algorithm capable of automatically extracting the overlap regions. Firstly, the triangle threshold method is used to estimate the distance threshold, and the overlap region is extracted and bidirectionally merged to obtain relatively complete overlap point clouds. To mitigate the risk of getting stuck in local optimal solutions and minimize the impact of incorrectly identified point pairs in non-overlap regions, we incorporate an effectiveness factor in the calculation of Singular Value Decomposition (SVD) to weight the importance of the point pairs. To decrease the overlap point clouds extraction times and reduce the associated time costs, we implemented multiple iterations following each extraction of the overlap point clouds. We compared our algorithm with the ICP and TrICP algorithms using publicly available point clouds data and demonstrated the effectiveness of our algorithm in automatically addressing the challenge of fine registration for point clouds with low overlap.

Senile kyphosis, characterized by the curvature of the back caused by osteoporosis and muscle degeneration, can impair mobility and even lead to more severe issues such as fractures. In order to address this problem, this study proposes an intelligent furniture system based on a monocular camera, aiming to estimate the degree of kyphosis and automatically adjust the furniture height using a low-cost RGB camera to assist individuals in improving their hunchback in daily life. When using a monocular camera to capture human body posture information, the traditional algorithm obtains keypoints about the human body, including body length and angle, but lacks relevant details on the back. Typically, these algorithms only provide information on the keypoints of the shoulders and hips and do not provide insight into the condition of the back or estimate the degree of kyphosis. In order to solve this limitation, in addition to using traditional algorithms, this study analyzes and extracts the back curve. When using the back curve and other keypoints of the human body, the degree of kyphosis can be estimated. Based on these estimations, the intelligent furniture system can automatically adjust the height of the furniture. Experimental results demonstrate that our system can detect the degree of kyphosis in individuals and adjust the furniture height according to individual needs. Users can obtain more comfortable posture support through this adaptive adjustment, reducing discomfort and health issues associated with kyphosis.

Segmentation of major brain tissues from 3D medical images can contribute to improving diagnostic quality and reducing workload. This study aims to explore the proper structure to segment brain regions from MRI volumes. The dataset used was the preprocessed IXI dataset, and segmentation was performed on 46 regions from the head MRI images. Experimental results show that it is important to extract features in the first stage, when the resolution is large.

This paper presents an AI-based system designed to detect cognitive impairments in older adults using modern technologies such as image recognition and AI painting. The System aims to address the limitations of traditional cognitive assessment methods and improve the quality and accuracy of patient tests. The System has two main components: an AI-based Real-World Image Partial Redrawing System and an Automatic Test Generation System. The former employs image segmentation and AI painting techniques to redraw objects in images, creating various test content. The latter generates cognitive assessment tests by comparing the original and redrawn images and highlighting differences. The System also allows doctors to access and monitor patients’ test results remotely. The experimental results demonstrate that the System provides valuable insights into patients’ cognitive abilities and can aid in early detection of cognitive impairments. Future work involves data collection from different age groups to fine-tune the System and achieve automated and objective assessment of cognitive impairments.

Sheep hold significant economic value. While the loss of a few sheep may be insignificant on large-scale ranches, it poses an unacceptable burden on small and medium-sized ranches. Therefore, considerable manpower and time must be allocated for locating the lost sheep. In this paper, the authors focus on develop an application to help searching lost sheep. To correspond to the actual ranch environment, a drone is considered to use. The affordability of small drones, coupled with their mobility, offers an effective solution to address this challenge. The machine learning methods is used to identify sheep. Moreover, to build a lightweight, works quickly and correctly execute system, the system environment and several methods is discussed.

Emotion recognition plays an important role in understanding human behavior and psychological well-being. In this research, we propose a method to recognize high-valence and low-valence emotions in young adults through the analysis of Heart rate variability, utilizing non-deep learning techniques. The study explores the Young Adult’s Affective Data dataset, comprising physiological information from 25 volunteer participants aged between 8 and 25 years. We employ three non-deep learning classifiers: Support Vector Machine, Logistic Regression, and K-Nearest Neighbors for binary emotion classification. Our method achieved 83% accuracy in recognizing high-valence and low-valence emotions. Overall, our findings highlight the efficacy of HRV-based emotion recognition using non-deep learning techniques, offering promising potential for practical applications in mental health monitoring, affective computing, and human-computer interaction. This study contributes to advancing emotion recognition methods and understanding emotional well-being among young adults.

The objective of this research is to develop a controller that accurately tracks white lines and propose a control method for white line tracking vehicles using a camera. In this control method, a Convolutional Neural Network (CNN) is applied to the line image, and the control amount is determined based on the class probabilities obtained. To optimize the parameters, Differential Evolution is utilized. When creating the model, the same image is included in multiple classes to introduce ambiguity in the output probabilities and ensure smoother transitions between courses. Furthermore, the generated model is compared with the driving data acquired at the time of image capture to confirm its tracking performance.

Immune cells play a pivotal role in assessing the overall health status of the human body. In cases of endometriosis, a condition characterized by abnormal tissue growth outside the uterus, the activity of natural killer (NK) cells, a subset of immune cells, is found to be reduced. To efficiently analyze this phenomenon, researchers employ image processing techniques. This study adopts the YOLACT image processing framework, aiming to expedite the analysis time and evaluate the accuracy of the processing.

Modern agriculture faces a labor shortage due to aging and a decrease in new farmers. Artificial intelligence (AI) and data utilization aim to improve productivity. Crop detection is one example, where object recognition models automate the process compared to manual detection relying on farmer experience. However, the challenge lies in training data requirements and variations in label assignment. This research investigates how different label assignment methods impact object recognition. We compare the labeling conditions using the YOLO and assess their effect on accuracy. Increasing target classes in test data helps maintain precision, while reducing recall. Detailed labeling improves average precision.

In recent years, there has been growing interest in Computer-Aided Diagnosis (CAD) and the widespread application of Convolutional Neural Networks (CNNs) for image-based diagnosis. However, CNN-based diagnostic approaches have faced challenges in terms of interpretability. Previous studies have proposed the use of CycleGAN in analyzing the classification process of CNNs, as it offers the potential for enhanced interpretability. While successful in obtaining interpretability for simpler disease images, such as cardiac hypertrophy, it has been more challenging to achieve interpretability in complex tasks like gender classification in brain imaging. Therefore, this study aims to employ human facial images, a task easily comprehensible to humans, to conduct gender classification using CycleGAN and analyze the criteria and tendencies involved in gender transformation. Through the analysis of pre- and post-transformation images, as well as pixel value differences, we aim to explore the relationship between gender differences and image disparities. Findings indicate notable changes in factors such as hair length, eye size, and skin color. Moreover, the inclusion of overall image characteristics, beyond individual facial features, suggests the possibility of using external means such as wigs or cosmetics to potentially deceive CycleGAN’s gender classification.

With the rapid progress of artificial intelligence, the methods of face recognition have also achieved considerable progress. However, when multiple head poses are present, the accuracy of face recognition decreases due to angular deviation. Therefore, how to realize the enhancement of accurate precision in two-dimensional multiple head poses is still a worthy research topic. In this paper, based on the ResNet50 residual network structure, data enhancement and attention mechanism are adopted, and the Public Figures Face Database public face database is used to partition the training and test sets. This hybrid approach effectively improved the accuracy of face recognition in lateral. However, the hybrid method has low accuracy for large angle deflection face recognition. Therefore, based on the hybrid method, this article proposes a side face recognition algorithm based on data generation, incorporating a data generation algorithm. The recognition accuracy of the front face of this algorithm is as high as 96.3%. This article presents a data-generated lateral face recognition algorithm that further improves the accuracy of large angle deflection face recognition.

In the rapidly evolving landscape of courier shops and industries, the efficient dispatch, storage, control, and optimization of goods play a pivotal role in addressing the escalating demands for increased throughput. Currently, manual labor predominantly manages the relocation of goods within courier shops, with limited utilization of storage cranes and advanced control systems. To effectively tackle the multifaceted challenges associated with storage and to optimize overall operational efficiency. This paper introduces the design of an innovative storage crane that is customized for the transportation, distribution, control, and optimization of small-sized goods within courier shops. A rigorous fatigue analysis using the ANSYS system validates its capability to withstand repeated cyclic loading and dynamic system demands under precise control. The results unequivocally demonstrate that the crane's acceleration gradient and saturation limit meet stringent criteria, ensuring robust and dependable movement within an optimized framework. This groundbreaking storage crane design, coupled with an advanced control system, presents a promising solution to elevate goods handling efficiency in courier shops. It effectively reduces dependence on manual labor, optimizes operations, and adeptly meets the burgeoning industry demands for control and efficiency.

In this paper, based on the Qube-Serco2 rotating inverted pendulum system, the mathematical model and state space expression are established, and a stable pendulum control algorithm based on BP neural network is proposed. On this basis, the consistency algorithm is used to control the three inverted pendulums, and the experimental results verify the effectiveness of the algorithm.

Due to the reduction of control accuracy when the unmanned aerial vehicles (UAVs) are disturbed in motion, a dual-loop control strategy is established based on the robust model predictive control (TMPC) and the sliding mode control (SMC) in this paper. An outer-loop TMPC position controller is designed by using the Radial Basis Function (RBF) Neural Network to estimate external disturbances to reconstruct the nominal system model, which improves robustness to external disturbances while satisfying constraints and stability; Meanwhile, the inner-loop Sliding PID (SPID) controller is designed for the attitude subsystem via combining the SMC and the PID features, ensuring the stability of attitude control under disturbance, which enhances the UAVs trajectory tracking control effect under disturbance, and improves the stability and anti-disturbance capability. Simulation experiments demonstrate the effectiveness of the proposed algorithm in the UAVs control under disturbance.

In the biaxially oriented polyester film production line, the extrusion system is a very important link, and the stability of the extruder pressure directly affects the quality of the film product. In order to solve the problems of hysteresis, high overshoot and low anti-interference ability in the extruder pressure control system of the polyester film production line and make the output of the extruder pressure control system stable at the target value, a new controller is proposed. Traditional Proportional-Integral-Derivative (PID) control is a linear control, and modern control mostly adopts PID control. Therefore, based on particle swarm optimization (PSO), Back Propagation (BP) neural network and PID controller, an improved PSO-BP neural network PID controller is proposed for extruder pressure control system. The new system design combines the actual operation parameters of the twin-screw extruder and other information, and simulates the system based on MATLAB/Simulink module, and compares the control effect with the traditional PID and BP neural network PID controller. The results show that: based on IPSO-BP neural network PID control algorithm, the extruder control system makes the pressure output stable at the target value; the control system has small overshoot, short rise time, strong anti-interference ability, which can improve the quality and intelligence level of the film production line to a certain extent.

This paper focuses on the problem of finite-time stabilization-based adaptive neural control design for the synchronous generator with unknown nonlinearities. The fast finite-time practical stability criteria is used to address the control design. Neural networks are used to overcome the obstacles appeared in the unknown nonlinear functions. And a systematic finite-time excitation control design is developed by embedding the practical finite-time stability theory and adaptive backstepping technology. Through the closed-loop stability analysis, it’s shows that all the signals in the closed-loop systems are bounded. At last, the presented results are tested by the numerical model of the single-machine power system.

In order to improve the sampling accuracy and the adaptability to different loads, a constant air flow controller based on closed-loop control is designed and realized. The mathematical model of this constant air flow controller is established, and an interval type-2 fuzzy PID controller is designed. The stability of the closed-loop constant flow controller is analyzed theoretically. Constant flow control simulations and experiments using PID controller and the interval type-2 fuzzy PID controller are done. The simulation and experiment results show that the constant air flow controller based on interval type-2 fuzzy PID controller has satisfied performance and adaptability to different loads.

The paper presented in this article deals with the issue of distributed cooperative formation of multi-agent systems (MASs). It proposes the use of appropriate neural network control methods to address formation requirements. The paper considers distributed cooperative formation control using a leader-follower approach. The paper also employs neural networks to overcome control challenges while dealing with complex systems or complex conditions. The neural network model was designed and the leader-follower formation control protocol was proposed. The sufficient conditions for the system stability were derived using Lyapunov stability theory, graph theory, and state space methods. By simulating the results of this study, the main data of the formation process can be observed to analyze and verify whether the system meets the requirements. Finally, by using an example of 16 agents to generate a hexagonal formation, it is verified that the system achieves consistency, stability, reliability, and accuracy in the cooperative formation.

This paper investigates a moving-target enclosing control problem of multiple nonholonomic mobile agents subject to unknown heterogeneous input disturbances. The agents are required to move around the target and perform an expected spacing between neighboring agents. First, a distribution algorithm is used to generate desired circular trajectories, which can be referenced by the agents. Simultaneously, an adaptive observer of each agent is proposed to estimate the disturbances. Then based on the desired circular trajectories and the estimations, a dynamic controller is designed to let the agents globally converge to the desired circular formation. Finally, effectiveness of the method is verified through simulation.

In recent years, researchers have developed multi-agent reinforcement learning systems to automate luggage transfer. However, these systems often struggle to learn effectively in partially observable environments, such as POMDPs. This paper presents a novel learning approach that leverages relative vectors to address this limitation. The proposed method is compared to conventional approaches, and the results show that it can achieve better performance in POMDPs.

This paper proposes a method for estimating the pose and size of an object using binocular stereo vision. It utilizes a variable photo-model approach which only requires one shooting condition unknown photograph of the target object. The method constructs a 2D pixel model of the object’s bounding box using the pre-trained YOLOv4 weight from the MS COCO dataset, and converts it into 3D flat photo-models of varying sizes. The estimation of the object’s pose and size is achieved through model-based stereo-vision matching and the use of Genetic Algorithm (GA). This approach eliminates the need for extensive data and pre-training time, making it cost-effective and efficient. Additionally, it extends the application range of the traditional photo-model algorithm. Experimental results conducted in indoor scenes demonstrate the effectiveness of the variable photo-model method in estimating pose and size using a photograph of the same class.

Motion capture technology is an emerging comprehensive multidisciplinary field cross technology involving graphics, engineering and communication technology. In order to acquire the posture of dexterous hand, a motion capture model for intelligent perception of dexterous hand was proposed in this paper. Sensing system, which is composed of 16 six axis inertial sensors is used to obtain the acceleration and angular velocity of dexterous hand movement. The Kalman filter algorithm was used to calculate the angle after sensor calibration and fusion. The sensor data was sent and received through an acquisition board using a serial port and the collected sensor angles were transmitted through multiple threads with Socket. A buffer was written to store a large amount of data, which map the fused angle with the hand model on the Unity 3D platform, and transmit the angle information to the corresponding virtual hand joint according to the port number to achieve real-time transmission. This article mapped 15 knuckles one by one, resulting in more accurate mapping. Socket was used for data transfer, offering a faster speed compared to USB. Experimental results show that the proposed method could capture the movements of dexterous hands, and realize the interaction between virtual hands and dexterous hands in Unity 3D. Besides, the angle information is saved by the Json file.

This paper presents a new lower-limb rehabilitation machine to meet the rehabilitation needs of hemiplegic patients. First, we select a left-right-independent rotary pedaling mechanism to ease rehabilitation. And adaptation to the physical condition of a user. Then, we design and fabricate a half model of a lower-limb rehabilitation machine with ergonomics in mind. We use the combination of non-negative matrix factorization (NMF) and non-negative double singular value decomposition as an analysis tool to calculate muscle synergies for sEMG signals of walking muscles, and use cosine similarity to evaluate the similarity between walking and pedaling activities. A comparison between the experimental results of walking experiment pedaling clarifies the effectiveness of pedaling in gait rehabilitation. To further improve the similarity between walking and pedaling, we introduced the double integration of an sEMG signal and used a Fourier series to describe the relationship between the load input and the rotational angle for the first time. Experimental results showed that the similarity was improved by this method compared to other load-setting methods.