Non-autistic observers often interpret autistic emotional expressions more negatively, though it is unclear whether this reflects observer bias or genuine differences in autistic people's emotional experience and expression. To examine this, 20 autistic and 20 non-autistic adults reported the intensity of their felt emotion while re-experiencing video-recorded events eliciting mild and strong happiness, sadness, and anger. A total of 379 non-autistic observers, half blind to diagnostic status, viewed the recordings and identified the emotion and its intensity. iMotions emotion recognition software also classified the emotional valence of the expressions. Overall, autistic and non-autistic participants reported comparable levels of felt emotion, although differences emerged in how their expressions were perceived. Observers more accurately identified happiness in non-autistic participants and sadness and anger in autistic participants. They also judged autistic participants as expressing sadness and anger more intensely. Informing observers of the diagnostic status of participants largely did not modulate effects. iMotions more often classified mild autistic expressions as neutral and mild non-autistic expressions as positive. Because observer and iMotion findings emerged despite autistic and non-autistic participants not differing in felt emotion, they suggest that non-autistic observers and emotion recognition algorithms differentially interpret authentic autistic and non-autistic emotional expressions, which may contribute to misinterpretations of autistic people.Lay AbstractAutistic people may express emotions in ways that differ from non-autistic people, and non-autistic people sometimes misinterpret them as flat, overly intense, or hard to read. This misunderstanding can affect how autistic people are judged in everyday life, including in job interviews, friendships, and other important situations. In this study, we wanted to know how well non-autistic people-and emotion recognition software-can identify emotions on the faces of autistic and non-autistic people when they are actually feeling emotion. To do this, autistic and non-autistic adults were videotaped while recounting personal experiences that made them feel mild and strong happiness, sadness, and anger. They rated how strongly they felt each emotion during the videotaping. Later, short video clips of their facial expressions were shown (without sound) to a large group of non-autistic viewers, who identified the emotion and rated its intensity. Some viewers were told whether the person in the video was autistic or not. We found that autistic and non-autistic people reported feeling emotions at comparable levels, but non-autistic viewers were better at recognizing happy expressions in non-autistic people compared to autistic people, and better at recognizing sad and angry expressions in autistic people compared to non-autistic people. Viewers tended to rate autistic expressions, especially sadness and anger, as more intense than those of non-autistic people, even though the computer software rated autistic expressions as more neutral compared to non-autistic participants. These results suggest that autistic people feel emotions just as deeply as non-autistic people, but differences in expressive style and non-autistic biases may lead to misinterpretation. These findings highlight the need for greater awareness of communication differences in autism and for reducing misinterpretations in how autistic people are perceived.

This study aims to analyze students’ facial expressions during reading and problem-solving tasks in a computer-assisted learning environment, focusing on their relationship with gender, academic performance, and time. Due to its exploratory nature, the research was conducted using a single-group post-test design. The study group consisted of 40 university students (23 female and 17 male). Participants were asked to read a scientific text and then solve a multiple-choice test. During the task, students’ facial expressions were recorded through the computer’s webcam and analyzed in real-time by a facial expression recognition algorithm measuring valence, arousal, and intensity. The collected data were analyzed using t-tests and regression analysis. Female students exhibited lower valence and higher intensity in their facial expressions compared to male students. No significant relationship was found between facial expressions and academic performance or reading duration. However, valence levels were observed to decrease as problem-solving time increased. These findings contribute to the literature on emotional state analysis in educational settings.

A novel CNN-BiLSTM-attention framework based on improved sparrow search algorithm for hydraulic turbine condition recognition

Improved driving intention recognition algorithm for fuel-efficient cruise control of heavy-duty trucks

SSA-LSTM-based locomotion mode recognition algorithm for the control of powered hip disarticulation prostheses

A coarse-to-fine goblet cell recognition algorithm for whole slide imaging of the lung tissue in a rat model of asthma

Smart optical imaging-assisted acoustic streaming enrichment system for targeted microalgae screening under multiple stress conditions.

Advanced image processing and pattern-matching algorithms assisted enzyme/hydrogel platform for dual-signal detection of glucose.

In this paper, an intelligent security monitoring system for the elderly living alone is designed and implemented to improve the efficiency of fall detection and emergency response for the elderly. The system uses STM32F407ZGT6 microcontroller as the core, combined with ICM20608 attitude detection module, ultrasonic sensor, GPS, SIM800C and 4G communication module to realize multi-modal data fusion and real-time monitoring. Through the fall recognition algorithm, Posture Analysis Method and heart rate and blood oxygen monitoring technology, the detection accuracy is improved. The system supports 4G and SMS phone dual-mode communication to ensure stable transmission in complex environments, and realizes the function of elderly location positioning and real-time viewing of elderly status through small programs. The test results show that the system has good real-time performance, accuracy and stability. This study provides an effective solution for the safety monitoring of the elderly living alone.

Three-dimensional temperature field analysis of CFST arch rib segment based on an improved shadow recognition algorithm

During the article a hybrid named-entity recognition (NER) algorithm for Uzbek is presented. It combines rule-based modules (transliteration, dialect normalization, morphological analysis) with modern neural network models. The study is motivated by Uzbek’s agglutinative morphology, dialect diversity and the lack of specialized resources, which hinder the direct application of named entity recognition methods developed for English or other high-resource languages. As part of the work, an annotated corpus of more than three thousand sentences in the Uzbek language was formed, including legal documents, scientific articles, news materials and informal texts from social networks. The corpus is marked up according to the BIOES scheme taking into account the specific morphological and lexical features of the Uzbek language. The developed rule-oriented algorithms (transliteration, dialect standardization, morphological analysis) are integrated into a single post-processing system that complements neural network models. As a result of experiments aimed at assessing the effectiveness of the proposed approach, it was found that the hybrid approach significantly improves the accuracy and completeness metrics of named entity recognition in different thematic domains. The practical value of the study is that the proposed system can serve as a basis for automatic processing of Uzbek texts in the tasks of searching and extracting information, dialect normalization, annotating large text data and digitalization of document flow. The theoretical significance is that the work expands approaches to named entity recognition for low-resource languages, offering methods that take into account morphological-syntactic and dialectal features.

This study delves into the vulnerabilities of deep learning-based gait recognition systems against adversarial attacks, a critical issue considering the increasing reliance on these technologies in high-security environments. We highlight a major issue concerning the susceptibility of these systems to adversarial interventions that compromise their reliability. The importance of this issue stems from the critical role of gait recognition in applications where security and accuracy are paramount. Our approach introduces an advanced methodology that integrates Proximal Policy Optimization (PPO) with Generative Adversarial Networks (GANs) to create and deploy adversarial attacks in the form of targeted adversarial patches. These patches are designed to deceive gait recognition algorithms without detection by human oversight, exploiting the models' weaknesses to induce misclassification. This methodology not only leverages the strengths of GANs to produce deceptive examples but also innovatively utilizes PPO to ascertain their optimal placements, thereby maximizing the disruption on gait recognition systems. We assess the impact of these attacks using the CASIA Gait Database: Dataset B and the OU-ISIR Treadmill Dataset B - Clothes variation-, covering both real-world and controlled environments. Our results demonstrate a significant decline in recognition accuracy post-attack, underscoring the effectiveness of our adversarial approach. These findings underscore critical security flaws and actively inform the broader discussion aimed at boosting the robustness of gait recognition systems. The impact of our research extends significantly, providing crucial insights that aid in the creation of more secure, attack-resistant biometric recognition systems, thereby enhancing the resilience of gait recognition technologies against the backdrop of advancing cyber threats.

To develop a fully automatic artificial intelligence (AI) system for diagnosing alveolar bone defects (ABDs) in anterior teeth on cone-beam computed tomography (CBCT) images. This fully automatic AI system consists of two stages: (1)2D image construction (sagittal and coronal slices) of anterior teeth root/bone morphology using a recognition algorithm, and (2)multi-perspective alveolar bone defect classification (normal, mild dehiscence, moderate dehiscence, severe dehiscence, and fenestration) based on the Hierarchical Multi-Scale Feature Fusion Network (HiFuse). In total, 300 CBCT images with 3600 anterior teeth from two clinical centers were used to train the model. The system achieved automatic and accurate construction of sagittal and coronal slices for 12 anterior teeth, with a high structural similarity index measure (SSIM) index of 0.803. The HiFuse model significantly outperformed ConvNeXt and Swin Transformer counterparts (P < 0.05), achieving an accuracy of 0.936 (95% CI: 0.906-0.959), F1-score of 0.932 (95% CI: 0.895-0.955), recall of 0.936 (95% CI: 0.906-0.959), and precision of 0.928 (95% CI: 0.896-0.952). HiFuse also effectively distinguished between ABD types and accurately located alveolar bone defects. Our proposed AI system demonstrated great performance in ABDs diagnosis of anterior teeth using original 3D CBCT images and has potential for assisting with orthodontic diagnosis and decision-making. Accurate diagnosis of ABDs in anterior teeth is essential when selecting appropriate orthodontic treatment strategies and performing bone augmentation surgery. This AI system could preliminarily achieve diagnosis of ABDs in anterior teeth, reducing manual intervention and improving the overall diagnostic workflow.

Study on the Applicability of Intelligent Recognition Algorithms to Different Types of Subgrade Defects

Soft pneumatic grippers, structures created from soft material like silicone rubber and powered by air pressure, can grip and transfer fragile items safely and stably. These grippers are widely applicable under medical situations, especially during surgery. Other applications include automated laboratory operations, underwater gripping, food delivery, extraterrestrial exploration efforts, etc. However, traditional soft grippers face serious difficulties with gripping objects of different volumes and weights. This study proposed a novel layer-jamming mechanism of tuning the stiffness of soft grippers to increase its weight capacity. The study also proposed a novel bellowed-pipe structure for soft grippers to incorporate dual-mode gripping. The process of gripping was automated through the introduction of a camera and recognition algorithms capable of computing the size and depth of the detected object. Experiments conducted in room conditions, using the variable stiffness mechanism, the gripper could grip objects 120% heavier than without using the mechanism, indicating a significant advancement from traditional soft grippers. Further research could be done with making the gripper smaller and applicable under surgical conditions.

Purpose To develop a deep learning algorithm to automatically assess the posterior fossa on first-trimester US screening scans and identify open spina bifida (OSB) and cystic posterior fossa (CPF) anomalies. Materials and Methods This is the retrospective part of an international study involving 10 fetal medicine centers. Normal and abnormal (OSB, CPF anomaly) midsagittal fetal brain US images acquired between 11 and 14 weeks of gestation (July 2009-January 2024) with confirmed diagnosis at follow-up were evaluated. Images were manually annotated to delineate the posterior fossa. The dataset was split into a training/validation (70%) and internal test (30%) set. Three convolutional neural networks were trained via threefold cross-validation on the training/validation set, with predictions on the internal test set obtained by ensemble averaging across folds. Model performance in detecting OSB and CPF anomalies was evaluated for the whole cohort and for fetuses with OSB or CPF anomalies separately. Results Images from 251 fetuses were analyzed (mean gestational age, 12.7±0.65 weeks; 150 normal, 101 abnormal: 43 OSB, 58 CPF anomalies). On the internal test, the MobileNetV3 Large Weights achieved the best performance (area under the receiver operating characteristic curve, 0.94 [95% CI: 0.88, 0.99]; accuracy, 88% (67/76); recall, 81% (25/31); specificity, 93% (42/45); precision, 89% (25/28); NPV, 88% (42/48); and F1-score, 0.85). OSB was classified more accurately (93% (52/56) vs 88% (57/65), P = .38) and with higher recall (91% (10/11) versus 75% (15/20), P = .38 although the difference was not significant. Conclusion MobileNetV3 Large Weights accurately assessed the fetal posterior fossa between 11 and 14 weeks of gestation, distinguishing normal images from those showing OSB or CPF anomalies. ©RSNA, 2026.

Green apples exhibit a coloration that closely matches their surrounding environment, leading to low recognition accuracy for existing artificial intelligence models. This paper presents a green apple recognition algorithm that integrates an improved U-shaped network (U-Net) and you only look once network (YOLO) v8 to address this challenge. First, the U-Net is enhanced via Dilated Convolution, Attention Gates, and Residual Connections to blur the background, thereby emphasizing the green apple target. Second, convolutional transformations and an attention mechanism are incorporated into YOLO v8, enabling it to focus more effectively on green apple targets within similarly colored backgrounds. Finally, the improved YOLO v8 is employed to recognize green apple targets segmented by the U-Net, with its performance compared against existing models. Research results show that the proposed algorithm achieves a precision of 92.5% and a Recall of 96.8% in green apple recognition, representing a significant improvement over classical models. To mitigate omission issues and further enhance overall performance, an improved YOLO v8 module is connected in parallel with the primary model. Based on its underlying principles, this approach is also applicable to other green fruits with colors and textures highly similar to their backgrounds, demonstrating strong robustness and generalization capabilities.

The Tree-Seed Algorithm (TSA) is a swarm intelligence algorithm inspired by the propagation relationship between trees and seeds. However, the original TSA is prone to premature convergence and becomes trapped in local optima when addressing high-dimensional, complex optimization problems, limiting its practical efficacy. To overcome these limitations, this paper proposes an Adaptive and Migration-enhanced Tree Seed Algorithm (AMTSA), which integrates three key mechanisms to significantly enhance performance in solving complex optimization tasks. First, to effectively evade local optima, an adaptive tree migration mechanism is designed to dynamically adjust the search step-size and direction based on individual fitness, thereby improving global exploration. Second, to enhance the algorithm's adaptability and efficiency across different search stages, an adaptive seed generation strategy based on the dynamic Weibull distribution is introduced. This strategy enables flexible control over the number of seeds and promotes a balanced search throughout the solution space. Third, to mitigate convergence oscillations during the global search, a nonlinear step-size adjustment function inspired by the GBO algorithm is incorporated, which effectively improves convergence stability by responding to the iteration progress. Rigorous testing on the IEEE CEC 2014 benchmark functions demonstrates that AMTSA's overall performance surpasses not only state-of-the-art optimizers like JADE and LSHADE but also recent TSA variants, including STSA, fb-TSA, and MTSA. To further validate its robustness in high-dimensional spaces, AMTSA was tested on 30 benchmark functions at 30, 50, and 100 dimensions. Results show that AMTSA ranked first in the number of functions optimized best and exhibited the fastest convergence speed among all compared algorithms. In a real-world application, AMTSA was employed to optimize multi-threshold segmentation for lung cancer CT images. The resulting AMTSA-SVM classification model achieved an accuracy of 89.5%, significantly outperforming models such as standard SVM (76.22%), DE-SVM (82%), GA-SVM (79.33%), TSA-SVM (84.44%), and JADE-SVM (89.12%). In conclusion, the proposed AMTSA, by integrating adaptive migration, dynamic seed generation, and nonlinear step-size control, successfully addresses the inherent deficiencies of the native TSA, offering a more efficient and robust tool for solving high-dimensional, complex optimization problems. The AMTSA source code will be available at www.jianhuajiang.com.

This paper presents a modulation signal classification and recognition algorithm based on data augmentation and time–frequency joint attention (DA-TFJA) for underwater acoustic (UWA) communication systems. UWA communication, as an important means of marine information transmission, plays a key role in fields such as marine engineering, military reconnaissance, and marine science research. Accurate recognition of modulated signals is a core technology for ensuring the reliability of UWA communication systems. Traditional classification and recognition methods, mostly based on pure neural network algorithms, suffer from insufficient feature representation and limited generalization performance in complex and changing UWA channel environments. They also struggle to address complex factors such as multipath, Doppler shift, and noise interference, often resulting in scarce effective training samples and inadequate classification accuracy. To overcome these limitations, the proposed DA-TFJA algorithm simulates the characteristics of real UWA channels through two novel data augmentation strategies: the adaptive time–frequency transform enhancement algorithm (ATFT) and dynamic path superposition enhancement algorithm (DPSE). An end-to-end recognition network is developed that integrates a multiscale time–frequency feature extractor (MTFE), two-layer long short-term memory (LSTM) temporal modeling, and a time–frequency joint attention mechanism (TFAM). This comprehensive architecture achieves high-precision recognition of six modulation types, including 2FSK, 4FSK, BPSK, QPSK, DSSS, and OFDM. Experimental results demonstrate that compared with existing advanced methods, DA-TFJA achieves a classification accuracy of 98.36% on the measured reservoir dataset, representing an improvement of 3.09 percentage points, which fully verifies the effectiveness and practical value of the proposed approach.

Aiming at the problems of low accuracy and slow speed of existing point cloud weld extraction algorithms in 3D vision-based robotic intelligent welding, this study proposes a novel, to our knowledge, three-stage automatic point cloud weld extraction method. In the plane segmentation stage, the random sample consensus (RANSAC) algorithm is improved: by narrowing the selection range of sampling points to the local neighborhood and optimizing neighborhood construction with dynamic curvature detection, the efficiency of plane fitting is enhanced. In the feature point extraction stage, based on plane parameters, the plane intersection line method and distance threshold method are adopted to obtain weld seam feature points. In the curve fitting stage, farthest point sampling (FPS) is used to denoise and resample the feature points, and then the weld curve is fitted to achieve high-precision contour reconstruction. Experiments show that the method exhibits high efficiency, robustness, and engineering adaptability.