Recognition Experiments Research Articles

Perinatal exposure to bisphenol A impairs cognitive function via the gamma-aminobutyric acid signaling pathway in male rat offspring.

Bisphenol A (BPA) is a common synthetic endocrine disruptor that can be utilized in the fabrication of materials such as polycarbonates and epoxy resins. Numerous studies have linked BPA to learning and memory problems, although the precise mechanism remains unknown. Gamma-aminobutyric acid (GABA) is the most abundant inhibitory neurotransmitter in the vertebrate central nervous system, and it is intimately related to learning and memory. This study aims to evaluate whether altered cognitive behavior involves the GABA signaling pathway in male offspring of rats exposed to BPA during the prenatal and early postnatal periods. Pregnant rats were orally given BPA (0, 0.04, 0.4, and 4 mg/kg body weight (BW)/day) from the first day of pregnancy to the 21st day of breastfeeding. Three-week-old male rat offspring were selected for an open-field experiment and a new object recognition experiment to evaluate the effect of BPA exposure on cognitive behavior. Furthermore, the role of GABA signaling markers in the cognition affected by BPA was investigated at the molecular level using western blotting and real-time polymerase chain reaction (RT-PCR). The research demonstrated that BPA exposure impacted the behavior and memory of male rat offspring and elevated the expression of glutamic acid decarboxylase 67 (GAD67), GABA type A receptors subunit (GABAARα1), and GABA vesicle transporter (VGAT) in the hippocampus while decreasing the expression levels of GABA transaminase (GABA-T) and GABA transporter 1 (GAT-1). These findings indicate that the alteration in the expression of GABA signaling molecules may be one of the molecular mechanisms by which perinatal exposure to BPA leads to decreased learning and memory in male rat offspring.

Turkish sign language recognition using fuzzy logic asisted ELM and CNN methods

This work aimed to develop a data glove for the real-time translation of Turkish sign language. In addition, a novel Fuzzy Logic Assisted ELM method (FLA-ELM) for hand gesture classification is proposed. In order to acquire motion information from the gloves, 12 flexibility sensors, two inertial sensors, and 10 Hall sensors were employed. The NVIDIA Jetson Nano, a small pocketable minicomputer, was used to run the recognition software. A total of 34 signal information was gathered from the sensors, and feature matrices were generated in the form of time series for each word. In addition, an algorithm based on Euclidean distance has been developed to detect end-points between adjacent words in a sentence. In addition to the proposed method, CNN and classical ANN methods, whose model was created by us, were used in sign language recognition experiments, and the results were compared. For each classified word, samples were collected from 25 different signers, and 3000 sample data were obtained for 120 words. Furthermore, the dataset’s size was reduced using PCA, and the results of the newly created datasets were compared to the reference results. In the performance tests, single words and three-word sentences were translated with an accuracy of up to 96.8% and a minimum 2.4 ms processing time.

Research on Feature Extraction of Underwater Acoustic Target Radiation Noise Based on Machine Learning Algorithm

Underwater acoustic target recognition is a very important technology in the field of underwater acoustics, with great economic and military value. Feature extraction technology for underwater acoustic target radiation noise signals is the key to achieving acoustic target recognition. This study aims at the feature extraction task of acoustic targets and extracts 10 types of 252-dimensional feature vectors from three domains: time domain, frequency domain, and auditory domain. Through 7 machine learning algorithms for classification and recognition experiments, the experimental results show that the recognition performance of the ensemble classifier is much better than that of a single classifier. For different types of features, this study combines three ensemble learning algorithms and feature selection algorithms to select the original 252-dimensional features. The feature selection experiment shows that the wrapper feature selection algorithm has the best effect, and the feature vector dimension can be reduced to 40 dimensions. The recognition accuracy rate is not less than 92.8%, which provides feature extraction guidance for acoustic target recognition based on feature extraction.

Beyond visual range maneuver intention recognition based on attention enhanced tuna swarm optimization parallel BiGRU

This paper researches the problem of Beyond Visual Range (BVR) air combat maneuver intention recognition. To achieve efficient and accurate intention recognition, an Attention enhanced Tuna Swarm Optimization-Parallel Bidirectional Gated Recurrent Unit network (A-TSO-PBiGRU) is proposed, which constructs a novel Parallel BiGRU (PBiGRU). Firstly, PBiGRU has a parallel network structure, whose proportion of forward and backward network can be adjusted by forward coefficient and backward coefficient. Secondly, to achieve object-oriented adjustment of forward and backward coefficients, the tuna swarm optimization algorithm is introduced and the negative log-likelihood estimation loss function is used as the objective function, it realizes the dynamic combination of sequence guidance and reverse correction. Finally, the attention mechanism is used to obtain more useful information to improve the recognition accuracy. Through offline recognition experiment, it is proved that A-TSO-PBiGRU can effectively improve the convergence speed and recognition accuracy compared with GRU-related networks. Compared with the other six comparison algorithms, maneuver intention recognition accuracy also has significant advantages. In the online recognition experiment, maneuver intention recognition accuracy of A-TSO-PBiGRU is 93.7%, it shows excellent maneuver intention recognition ability.

Application of micro-bolus injection and piezoelectric sensors to improve the safety of radiopharmaceuticals bolus injection

Radiopharmaceutical dynamic imaging typically necessitates intravenous injection via the bolus method. However, manual bolus injection carries the risk of handling errors as well as radiological injuries. Hence, there is potential for automated injection devices to replace manual injection methods. In this study, the effect of micro-bolus pulse injection technology was compared and verified by radioactive experiments using a programmable injection pump, and the overall bubble recognition experiment and rat tail vein simulation injection verification were performed using the piezoelectric sensor preloading method. The results showed that at the same injection peak speed, the effective flushing volume of micro-bolus pulse flushing (about 83 μL/pulse) was 49.65% lower than that of uniform injection and 25.77% lower than that of manual flushing. In order to avoid the dilution effect of long pipe on the volume of liquid, the use of piezoelectric sensor for sealing preloading detection could accurately predict the bubbles of more than 100 μL in the syringe. In the simulated injection experiment of rat tail vein, when the needle was placed in different tissues by preloading 100 μL normal saline, the piezoelectric sensor fed back a large difference in pressure attenuation rate within one second, which was 2.78% in muscle, 17.28% in subcutaneous and 54.71% in vein. Micro-bolus pulse injection method and piezoelectric sensor sealing preloading method have application potential in improving the safety of radiopharmaceutical automatic bolus injection.

The flexible and distributed pressure sensor with 64 units for on-line gait recognition analysis

With the rapid development of flexible electronic technology, human gait recognition device is also developing in the direction of flexibility, refinement, and intelligence. In this paper, a gait recognition system with the advantages of being non-invasive and flexible is proposed and prepared to collect complex pressure information of human plantar and identify different gaits accurately. A flexible and distributed pressure sensor with a structure of eight rows and eight columns is designed and fabricated by the screen printing process and micro-nano manufacturing technology. The insole shape can be placed directly into a sneaker for easy wearing, while 64 pressure sensing units allow for more accurate detection of complex plantar pressure information. In addition, based on the “voltage mirror” principle and the feature threshold analysis method, a distributed pressure acquisition circuit and an online gait recognition software are developed respectively to realize time-sharing decoupling of complex output signals from multiple pressure sensing units and online real-time display of different gaits. Finally, Different human gait recognition experiments have also been carried out. The results show that the test error of the pressure sensing unit is better than 8.28%. The gait recognition system can realize the reliable and accurate online real-time identification of slow walking, fast walking, and different leaning directions, which shows great potential and value in the fields of medical diagnosis, health monitoring, gait correction and artificial intelligence.

Hydrogen defect acoustic emission recognition by deep learning neural network

Hydrogen attack is the major failure cause of hydrogen equipment breakdown maintenance. Especially, common problems like cracks frequently occur but are challenging to find while operating, presenting an issue for safety and production. In the process of hydrogen damage evolution, a method for defect state recognition is proposed in this paper. Acoustic emission (AE) technology is used for inspecting the entire hydrogen charging process. The characteristic parameters including the counts and duration of the AE signals are first preprocessed, and the current damage states such as the dislocation propagation, and the occurrence of cracks are identified. Then, a deep learning convolutional neural network is used to create a hydrogen defect recognition (HDR) model with the input of a short-time Fourier transform for the feature vector extraction of various damage status AE signals. Finally, the hydrogen defect recognition experiment revealed that HDR is better in classification accuracy at 98.37% due to dislocation propagation and cracks identification. The study can provide an online damage recognition approach for damage state early warning and evaluation to guarantee hydrogen equipment safety operation.

The Florence multi-resolution 3D facial expression dataset

In the literature, several 3D face datasets have been collected, aiming at advancing the field of 3D face analysis from different perspectives. Data collection generally follows specific research needs, and the existing 3D face datasets all have different characteristics that are tailored for investigating different tasks, encompassing face recognition, facial expressions and emotions analysis, 3D face reconstruction. However, the majority of these datasets are either collected with high-resolution scanners, or consumer level devices, such as the Kinect, the latter being motivated by the burdensome and costly process of collecting high-quality scans. Differently from 2D imagery, the difference in resolution in 3D data represents a non negligible problem that is under-investigated, and still prevents the successful development of methods that can work in real scenarios. In this paper, we propose a new 3D face dataset, named “Florence Multi-Resolution 3D Facial Expression” (Florence 3DMRE), which aims at bridging the gap between high- and low-resolution 3D face datasets. Its peculiarity consists in (1) including high-resolution (HR) models obtained with a HR scanner, and paired samples collected with a Kinect sensor, (2) LR and HR scans are synchronized and capture extreme and asymmetric facial deformations as used in facial rehabilitation exercises. In total, our dataset consists of 14 subjects, each performing 19 complex and asymmetric expressions. For each of them, we collected a high-resolution scan, and an RGB-D sequence. Finally, to highlight the value of the dataset and the challenges it introduces, we use the collected data to perform baseline experiments for cross-resolution 3D face recognition and reconstruction. The dataset is released for research purposes only, and complies to GDPR for data treatment. The dataset can be found at this link.

Holistic and featural processing’s link to face recognition varies by individual and task

While it is generally accepted that holistic processing facilitates face recognition, recent studies suggest that poor recognition might also arise from imprecise perception of local features in the face. This study aimed to examine to what extent holistic and featural processing relates to individual differences in face recognition ability (FRA), during face learning (Experiment 1) and face recognition (Experiment 2). Participants performed two tasks: (1) The “Cambridge Face Memory Test-Chinese” which measured participants’ FRAs, and (2) an “old/new recognition memory test” encompassing whole faces (preserving holistic and featural processing) and faces revealed through a dynamic aperture (impairing holistic processing but preserving featural processing). Our results showed that participants recognised faces more accurately in conditions when holistic information was preserved, than when it is impaired. We also show that the better use of holistic processing during face learning and face recognition was associated with better FRAs. However, enhanced featural processing during recognition, but not during learning, was related to better FRAs. Together, our findings demonstrate that good face recognition depends on distinct roles played by holistic and featural processing at different stages of face recognition.

The interaction between talker head orientation, spatial separation, and extended high frequencies for speech-in-speech recognition

Speech-in-speech recognition experiments generally involve presentation of stimuli as if target and masker talkers are facing the listener. This is because speech signals are recorded with microphones located in front of the talkers. In real-world situations, masker talkers are often rotated away from the listener, facing their own conversational partners. This head orientation mismatch between the target and masker talkers provides cues that enhance speech recognition, including cues at extended high frequencies (EHFs; >8 kHz) due to the directional nature of EHFs in speech radiation. However, it is unclear how head orientation and EHF cues affect speech recognition when target and masker speech come from different spatial locations, as is the case in realistic multi-talker environments. In the present study, we investigated the EHF benefit in a complex auditory scene involving differences in target and masker head orientation and spatial location. Masker head orientation was either 0° (facing) or 90° (non-facing). Target speech location was 0° azimuth, while masker speech location was either 0° or ±45° azimuth. Preliminary analyses suggest spatial cues provide greater benefit than head orientation cues, and that the EHF benefit is diminished when talkers are spatially separated. [Work supported by NIH grant R01DC019745.]

Audio-Visual Emotion Recognition With Preference Learning Based on Intended and Multi-Modal Perceived Labels

This paper introduces a novel preference learning framework that simultaneously considers both the intended and the perceived labels while addressing the mismatches between them. Based on analyzing the discrepancies and agreements between the intended and the perceived labels in different modalities of audio-only, visual-only, and audio-visual, as well as the consistency among the perceptual ratings of all raters, we propose three sets of pair-wise ranking rules to generate multi-scale relevant scores for preference learning, scaling from sketchy manner to detailed manner. Three ranking models with support vector machine (SVM), deep neural networks (DNN), and gradient boosting decision trees (GBDT) are developed. Our results demonstrate that all three preference learning models significantly outperform the conventional classifiers baselines, and the LambdaMART model with gradient boosting decision trees achieves the best performance. The improvement from the preference learning models confirm the benefits of complementary information provided by different types of labels. We also observe additional improvement from the detailed ‘complex ranking rules’, particular with the best LambdaMART model, which suggests that we should treat intended and perceived labels in single-model & multi-modal differently. We further discuss the complementary of different ranking models, and obtain the best overall accuracy of 85.06% on CREMA-D dataset when combining the two best ranking models–LambdaMART and RankNet–together, which is significantly better than the 76.19% accuracy attained by the baseline models. Finally, we perform the cross-corpus emotion recognition experiments by training emotion rankers on CREMA-D dataset and tested the ranking-based emotion classifier on the SAVEE dataset that do not have perceived labels annotated.

Brain Functional Representation of Highly Occluded Object Recognition.

Recognizing highly occluded objects is believed to arise from the interaction between the brain's vision and cognition-controlling areas, although supporting neuroimaging data are currently limited. To explore the neural mechanism during this activity, we conducted an occlusion object recognition experiment using functional magnetic resonance imaging (fMRI). During magnet resonance examinations, 66 subjects engaged in object recognition tasks with three different occlusion degrees. Generalized linear model (GLM) analysis showed that the activation degree of the occipital lobe (inferior occipital gyrus, middle occipital gyrus, and occipital fusiform gyrus) and dorsal anterior cingulate cortex (dACC) was related to the occlusion degree of the objects. Multivariate pattern analysis (MVPA) further unearthed a considerable surge in classification precision when dACC activation was incorporated as a feature. This suggested the combined role of dACC and the occipital lobe in occluded object recognition tasks. Moreover, psychophysiological interaction (PPI) analysis disclosed that functional connectivity (FC) between the dACC and the occipital lobe was enhanced with increased occlusion, highlighting the necessity of FC between these two brain regions in effectively identifying exceedingly occluded objects. In conclusion, these findings contribute to understanding the neural mechanisms of highly occluded object recognition, augmenting our appreciation of how the brain manages incomplete visual data.

In-Vehicle Speech Recognition for Voice-Driven UAV Control in a Collaborative Environment of MAV and UAV

Most conventional speech recognition systems have mainly concentrated on voice-driven control of personal user devices such as smartphones. Therefore, a speech recognition system used in a special environment needs to be developed in consideration of the environment. In this study, a speech recognition framework for voice-driven control of unmanned aerial vehicles (UAVs) is proposed in a collaborative environment between manned aerial vehicles (MAVs) and UAVs, where multiple MAVs and UAVs fly together, and pilots on board MAVs control multiple UAVs with their voices. Standard speech recognition systems consist of several modules, including front-end, recognition, and post-processing. Among them, this study focuses on recognition and post-processing modules in terms of in-vehicle speech recognition. In order to stably control UAVs via voice, it is necessary to handle the environmental conditions of the UAVs carefully. First, we define control commands that the MAV pilot delivers to UAVs and construct training data. Next, for the recognition module, we investigate an acoustic model suitable for the characteristics of the UAV control commands and the UAV system with hardware resource constraints. Finally, two approaches are proposed for post-processing: grammar network-based syntax analysis and transaction-based semantic analysis. For evaluation, we developed a speech recognition system in a collaborative simulation environment between a MAV and an UAV and successfully verified the validity of each module. As a result of recognition experiments of connected words consisting of two to five words, the recognition rates of hidden Markov model (HMM) and deep neural network (DNN)-based acoustic models were 98.2% and 98.4%, respectively. However, in terms of computational amount, the HMM model was about 100 times more efficient than DNN. In addition, the relative improvement in error rate with the proposed post-processing was about 65%.

Does context recollection depend on the base-rate of contextual features?

Episodic recollection is defined by the re-experiencing of contextual and target details of a past event. The base-rate dependency hypothesis assumes that the retrieval of one contextual feature from an integrated episodic trace cues the retrieval of another associated feature, and that the more often a particular configuration of features occurs, the more effective this mutual cueing will be. Alternatively, the conditional probability of one feature given another feature may be neglected in memory for contextual features since they are not directly bound to one another. Three conjoint recognition experiments investigated whether memory for context is sensitive to the base-rates of features. Participants studied frequent versus infrequent configurations of features and, during the test, they were asked to recognise one of these features with (vs. without) another feature reinstated. The results showed that the context recollection parameter, representing the re-experience of contextual features in the dual-recollection model, was higher for frequent than infrequent feature configurations only when the binding of feature information was made easier and the differences in the base-rates were extreme, otherwise no difference was found. Similarly, base-rates of features influenced response guessing only in the condition with salient differences in base-rates. The Bayes factor analyses showed that the evidence from two of our experiments favoured the base-rate neglect hypothesis over the base-rate dependency hypothesis; the opposite result was obtained in the third experiment, but only when high base-rate disproportion and facilitated feature binding conditions were used.

Improved machine learning leak fault recognition for low-pressure natural gas valve

Monitoring valve operation status is very significant in saving natural gas resources and realizing sustainability of the fossil energy. At present, many machine learning algorithms have been applied to nondestructive testing (NDT) field. Among them, the twin support vector machine (TWSVM) is a representative binary classification method. However, the traditional TWSVM assigns same classification weights to all sample points, including classification boundary points and other non-boundary sample points. These same classification weights often lead to classification errors or overfitting in TWSVM. Therefore, to overcome the drawback, we propose an improved nonlinear TWSVM (I-TWSVM) to optimize the classification weights assignment processing. In new assignment processing, the boundary points are given greater classification weights than other non-boundary points, which makes the I-TWSVM becomes more sensitive than the traditional TWSVM to boundary points. The I-TWSVM has been applied to recognize the acoustic emission (AE) sample data for valve internal leak fault accident. The leak recognition experiment revealed that the proposed algorithm is better than the traditional nonlinear TWSVM in classification accuracy and sensitivity, and its calculation time is faster than that of nonlinear TWSVM, which can promote the nondestructive testing technology development for low-pressure natural gas valve.

Sound Identification Method for Gas and Coal Dust Explosions Based on MLP.

To solve the problems of backward gas and coal dust explosion alarm technology and single monitoring means in coal mines, and to improve the accuracy of gas and coal dust explosion identification in coal mines, a sound identification method for gas and coal dust explosions based on MLP in coal mines is proposed, and the distributions of the mean value of the short-time energy, zero crossing rate, spectral centroid, spectral spread, roll-off, 16-dimensional time-frequency features, MFCC, GFCC, short-time Fourier coefficients of gas explosion sound, coal dust sound, and other underground sounds were analyzed. In order to select the most suitable feature vector to characterize the sound signal, the best feature extraction model of the Relief algorithm was established, and the cross-entropy distribution of the MLP model trained with the different numbers of feature values was analyzed. In order to further optimize the feature value selection, the recognition results of the recognition models trained with the different numbers of sound feature values were compared, and the first 35-dimensional feature values were finally determined as the feature vector to characterize the sound signal. The feature vectors are input into the MLP to establish the sound recognition model of coal mine gas and coal dust explosion. An analysis of the feature extraction, optimal feature extraction, model training, and time consumption for model recognition during the model establishment process shows that the proposed algorithm has high computational efficiency and meets the requirement of the real-time coal mine safety monitoring and alarm system. From the results of recognition experiments, the sound recognition algorithm can distinguish each kind of sound involved in the experiments more accurately. The average recognition rate, recall rate, and accuracy rate of the model can reach 95%, 95%, and 95.8%, respectively, which is obviously better than the comparison algorithm and can meet the requirements of coal mine gas and coal dust explosion sensing and alarming.

Flight state recognition for UAV optical flow velocity measurement

For the visual navigation and positioning of drones, the optical flow algorithm is used to calculate the flight speed of the drone, the calculation errors of its rotation state, and the vertical take-off and landing state. Based on the characteristics of its optical flow data, this paper compares various classification calculation methods. It uses the median KNN curve of the optical flow vector to simplify its feature quantity. Finally, it realizes the motion of the UAV with the optical flow characteristics based on the SVM algorithm. State recognition experiments have proved that this method can effectively identify the motion state of the UAV, and the calculation efficiency is high.

Outlier-Robust Gaze Signal Filtering Framework Based on Eye-Movement Modality Recognition and Set-Membership Approach.

High-quality gaze signals are crucial in many biomedical fields that utilize them. However, the limited studies on gaze signal filtering can hardly address the outliers and non-Gaussian noise in gaze data simultaneously. Our objective is to design a generic filtering framework capable of reducing the noise and eliminating outliers of the gaze signal. In this study, we design an eye-movement modality-based zonotope set-membership filtering framework (EM-ZSMF) to suppress the noise and outliers of the gaze signal. This framework consists of an eye-movement modality recognition model (EG-NET), an eye-movement modality-based gaze movement model (EMGM), and a zonotope set-membership filter (ZSMF). The eye-movement modality determines the EMGM, and the ZSMF combined with the EMGM completes the filtering of the gaze signal. Moreover, this study establishes an eye-movement modality and gaze filtering dataset (ERGF) that can be utilized for the evaluation of future work integrating eye-movement modality with gaze signal filtering. The eye-movement modality recognition experiments demonstrated that our proposed EG-NET achieved the best Cohen's kappa compared with previous studies. The gaze data filtering experiments showed that the proposed EM-ZSMF reduced the gaze signal noise and eliminated outliers effectively, and achieved the best performance (RMSEs and RMS) compared with previous methods. The proposed EM-ZSMF effectively recognizes eye-movement modalities, reduces gaze signal noise and, eliminates outliers. To the best of the authors' knowledge, this is the first attempt to simultaneously solve the problem of non-Gaussian noise and outliers in gaze signals. The proposed framework has the potential for application in any eye image-based eye trackers and contributes to the development of eye-tracking technology.

Atypical scanning strategies of emotional faces for individuals with high autistic traits.

Autism has become one of the primary diseases causing disability in children, and the incidence has risen rapidly in recent years. The preclinical study on individuals with high autistic traits is extremely important to reduce genetic risks of autism because high autistic traits is the susceptibility marker of autism. However, few studies explored the face scanning pattern of people with high autistic traits in typical developing populations. In this study, we designed a facial emotion recognition experiment including four emotions (happy, neutral, sad, angry) and three angles (0°, 45°, 90°) , and informed the participants to identify the facial emotion. Forty-two college students with typical development were recruited and divided into high autistic traits (HAT) group and low autistic traits (LAT) group by the Autism-Spectrum Quotient, and we collected the eye movement data using eye-tracking technology when they performed the task. The response time, recognition accuracy, AOI based proportional fixation time and pupil diameter were computed and analyzed for both groups. HATs showed significantly lower recognition accuracy and lower pupil diameter than LATs when recognizing negative emotions (P<0.05) , indicating HATs kept poor autonomic nervous arousal. What ' s more, the proportional fixation time of HATs were significantly more in mouth area but less in eye area than that of LAT group (P<0.05) , revealed HATs had an atypical emotional faces scanning strategies that paid less attention to eyes and more attention to mouth. Our research provides a feasible objective biomarker for screening high autistic traits population.

Machine Vision Algorithm for Identifying Packaging Components of HN-3 Arterial Blood Sample Collector

The arterial blood sample collector produced in large quantities often fails to meet the requirements due to missing components in the packaging bag, and traditional manual detection methods are both inefficient and inaccurate. To solve this problem, a PyCharm-integrated development environment was used to study image processing and recognition algorithms for identifying components inside the packaging bag of the HN-3 arterial blood sample collector. The machine vision system was used to capture images of the packaging bags of the HN-3 Arterial blood sample collector. Template matching was employed to extract the packaging ROI, and the threshold segmentation method in the HSV color model was used to extract material features based on the packaging ROI. Morphological processing algorithms such as dilation or erosion were used to enhance the connectivity of the extracted features. The existence of components was determined by setting thresholds for the connected domain area or length. The results of the recognition experiment show that the false detection rate is 0.2%, the missed detection rate is 0%, and the average image processing time per product is no more than 39 ms. Compared with manual recognition methods, the efficiency and accuracy have been improved by 36.5 times and 2.3%, respectively. The experimental results confirm the effectiveness of the image processing algorithm.