Video Samples Research Articles

Thermal preference prediction through infrared thermography technology: Recognizing adaptive behaviors

The recognition of thermal adaptation behaviors is a crucial and noninvasive method for accurately and swiftly predicting the thermal preferences of occupants. This is instrumental in enhancing the energy efficiency, indoor personnel comfort, and overall productivity of office buildings. This study employed deep-learning models to analyze a thermal adaptive behavior video dataset captured by Infrared thermography (IRT) cameras to predict the thermal preferences of occupants. A large-scale infrared thermography dataset was developed for recognizing thermal adaptive behaviors using 9650 video samples. Two adaptive behavior recognition models for infrared videos were developed by leveraging neural network models: Two-Stream Inflated 3D ConvNet (I3D) and SlowFast (SF) networks. These models achieved the average prediction accuracies of 83.53 % and 88.56 %, respectively. Notably, both models exhibited a recognition time of milliseconds, facilitating the real-time recognition of thermal adaptive behaviors. This study offers vital technical and theoretical underpinnings for developing decision-making solutions for smart buildings based on noninvasive thermal preference prediction.

P-554 Artificial intelligence (AI) for embryo genetic risk assessment: a novel form of preimplantation genetic screening

Abstract Study question Can an AI-based embryo ploidy screening tool effectively provide a genetic risk assessment on Day 5 embryos? Summary answer AI screening results showed strong predictive value for early detection of at-risk embryos warranting confirmatory diagnostic testing, based only on time lapse video assessment. What is known already Euploid embryo transfer is the dogma of IVF success. Embryo evaluation relies extensively on diagnostic preimplantation genetic testing for aneuploidy (PGT-A) by means of trophectoderm biopsy. Though pivotal, this presents substantial practical and financial limitations. There is demand for an alternative genetic screening method that provides a quantitative embryo genetic risk assessment for patients who opt out of invasive diagnostic testing, or who intend to undergo fresh transfer. These results can provide information for understanding your aneuploidy risk for more precise prognosis counseling and testing. This is important since diagnostic tests may incur excess financial burden, inconclusive outcomes, and time-to-results. Study design, size, duration Time lapse videos (N = 5,000) were used for AI training. The AI core architecture was a video transformer model for spatio-temporal video sampling. Known ploidy and live-birth results were used as ground truth, with maternal age and embryo quality scores incorporated as input parameters during calibrated logistic regression. A blind test dataset (N = 708 embryos; euploid=352; aneuploid=356; mean ± SD patient age: 35.9 ± 5.4 years) was used to evaluate predictive accuracy. Participants/materials, setting, methods The AI outputs a scalar (1-99) that associates with euploidy likelihood. The AI’s screening value was determined by its specificity (correct prediction of aneuploid), sensitivity (correct prediction of euploid), false positive, and false negative rate. We determined optimal AI score thresholds that identified high aneuploidy and euploidy risk categories (AI score <33; AI score >66, respectively). Main results and the role of chance Specificity of the AI screening test was 83.0%, with a sensitivity of 56.0%. The false positive rate was 26.7% and the false negative rate was 29.3%. The AI’s high specificity relative to its sensitivity shows its clinical value to deselect embryos with the strongest risk of genetic abnormality, and for the prioritization of embryos for confirmatory PGT. Boxplot visualization of the full AI score distribution showed successful discrimination between aneuploid and euploid embryos (mean AI score for aneuploid and euploid were 51.3 and 43.5, respectively) with appropriately minimized overlapping of the interquartile range. The optimal AI score thresholds for detecting embryos with the highest and lowest likelihoods of euploidy were >66 /<33, respectively, representing 71% likelihood of euploidy for the >66 group and 72% aneuploidy for the <33 group. Limitations, reasons for caution The screening does not provide diagnostic genetic testing on the chromosome level. The influence of mosaicism on the false positive/negative rates was not assessed. Wider implications of the findings Results support the use of an AI ploidy screening test for effective decision-making and triage of at-risk embryos for transfer or confirmatory diagnostic testing via PGT-A. Trial registration number Not applicable

Automating Video-Based Two-Dimensional Motion Analysis in Sport? Implications for Gait Event Detection, Pose Estimation, and Performance Parameter Analysis.

Two-dimensional (2D) video is a common tool used during sports training and competition to analyze movement. In these videos, biomechanists determine key events, annotate joint centers, and calculate spatial, temporal, and kinematic parameters to provide performance reports to coaches and athletes. Automatic tools relying on computer vision and artificial intelligence methods hold promise to reduce the need for time-consuming manual methods. This study systematically analyzed the steps required to automate the video analysis workflow by investigating the applicability of a threshold-based event detection algorithm developed for 3D marker trajectories to 2D video data at four sampling rates; the agreement of 2D keypoints estimated by an off-the-shelf pose estimation model compared with gold-standard 3D marker trajectories projected to camera's field of view; and the influence of an offset in event detection on contact time and the sagittal knee joint angle at the key critical events of touch down and foot flat. Repeated measures limits of agreement were used to compare parameters determined by markerless and marker-based motion capture. Results highlighted that a minimum video sampling rate of 100 Hz is required to detect key events, and the limited applicability of 3D marker trajectory-based event detection algorithms when using 2D video. Although detected keypoints showed good agreement with the gold-standard, misidentification of key events-such as touch down by 20 ms resulted in knee compression angle differences of up to 20°. These findings emphasize the need for de novo accurate key event detection algorithms to automate 2D video analysis pipelines.

Mechanisms of well iron clogging in groundwater heat pump systems: Insights from video imaging, hydrogeochemical analysis, and geochemical modeling

Groundwater heat pump (GWHP) systems are increasingly popular as low-carbon and environmentally friendly technologies, but well clogging induced by iron remains a significant issue. This study investigated the clogging characteristics and biogeochemistry of three typical wells (pumping, injection, and observation wells) in an operating GWHP system using video imaging, sampling, and analysis of hydrogeochemical and microbial data. The results revealed that iron-induced well clogging is a complex process involving physical, chemical, and microbial factors. Pumping wells experience clogging due to water mixing with varying redox conditions, resulting in hematite-based iron oxide deposits. Injection wells exhibit higher clogging severity, with transformed oxidation and accumulation of reduced iron minerals at the solid-liquid interface, resulting in darker colored clogs with magnetite. Clogging in both extraction and injection wells is closely related to iron-rich aquifer sections, where severe clogging occurs. Shallow clogging due to iron oxide is limited and attributed to the oxidation of zero-valent iron in well casing material. Iron-oxidizing bacteria and iron-reducing bacteria were detected in the consolidated deposits of clogged wells, indicating their involvement in the clogging formation process. Moreover, a strong correlation was observed between the presence of nitrate-reducing bacteria in the water phase and the severity of clogging, suggesting a possible link between iron oxidation and nitrate reduction in the system. Geochemical modeling results further supported the observed clogging severity in GWHP systems and confirmed varying clogging mechanisms in different wells and depths. These findings contribute to the understanding of clogging in GWHP operations, aiding in robust water utilization and energy-saving efforts, and supporting global carbon reduction initiatives.

A deep learning model for FaceSwap and face-reenactment deepfakes detection

Recently, the higher availability of multimedia content on social websites, together with lightweight deep learning (DL) empowered tools like Generative Adversarial Networks (GANs) has caused the generation of realistic deepfakes. Such fabricated data has the potential to spread disinformation, revenge porn, initiate monetary scams, and can result in adverse immoral and illegal societal issues, etc. Hence, the accurate identification of deepfakes is mandatory to discriminate between real and manipulated content. In this work, we have presented a DL-based approach namely a unified network for FaceSwap (FS) and Face-Reenactment (FR) Deepfakes Detection (AUFF-Net). More clearly, both the spatial and temporal information from the video samples are used to detect two types of visual manipulations i.e., FS and FR. For this reason, a novel DL framework namely the Inception-Swish-ResNet-v2 model is introduced as a feature extractor for computing the information at the spatial level. While the Bi-LSTM model is utilized to measure the temporal information. Additionally, 3 dense layers are included at the last of the model structure to suggest a discriminative group of the feature vector We performed extensive experimentation on a challenging dataset namely the FaceForensic++, and attainede average accuracy values of 99.21 %, and 98.32 % for FS, or FR, respectively. Furthermore, we introduced an explainability module to show the reliable keypoints selection capability of our technique. Moreover, we have performed a cross-dataset evaluation to show the generalization power of our approach. Both the qualitative and quantitative results have confirmed the effectiveness of the suggested approach for visual manipulation categorization under the occurrence of various adversarial attacks.

Application of video adaptive sampling based on soft computing in higher education evaluation and E-learning interactive environment

The level of teaching quality largely reflects the overall strength of universities and has a direct impact on the quality of cultivating high-quality talents. In this context, higher education institutions are constantly striving to explore innovative methods, but this diverse educational environment also brings challenges to the evaluation methods of teacher teaching quality. This article explores the application of video adaptive sampling in AI data analysis of university education evaluation based on soft computing technology. In order to solve the data sampling problem in university education evaluation, this paper proposes a video adaptive sampling method based on soft computing. In this paper, an E-learning virtual interactive system is designed, multimedia teaching materials matching the requirements of E-learning environment are created, tools suitable for E-learning virtual interactive environment are selected and integrated, soft computing technology is adopted, intelligent algorithms and data analysis methods are combined. By extracting and analyzing the key information in the video, the video sampling strategy is dynamically adjusted to realize the adaptive video sampling. Modeling the content characteristics of videos through fuzzy logic, and using genetic algorithms to optimize the sampling frequency and accuracy of videos. The experimental results indicate that the video adaptive sampling method based on soft computing can effectively improve the accuracy and efficiency of AI data analysis in university education evaluation. Compared with traditional fixed sampling methods, this method can dynamically adjust according to the characteristics of video content, reduce redundant data sampling, and improve the accuracy of data analysis.

A taxonomy of didactic roles of dynamic visualization in animated mathematics videos

Abstract In recent years, a multitude of innovative educational videos focusing on mathematics and employing dynamic visualization techniques have been published on YouTube. From a sample of videos, we develop a taxonomy of didactical roles of dynamic visuals and supporting animation techniques. The taxonomy is constructed through an open coding approach, involving an iterative process of defining, analysing and refining the codes. Additionally, we conducted an interview with a video creator about the intended roles of dynamic visuals employed in his animations. The taxonomy encompasses the following categories of roles: connect objects, (co)vary objects, dynamic concept visualization, dynamic process visualization, symbol sense, connect to reality and generalization. These roles are implemented through animation techniques like moving and morphing. The nomenclature of these roles suggests their affiliation with various mathematics education research perspectives, including covariational thinking, concept–process duality, symbol sense and conversion between registers. The taxonomy serves as a foundational framework for further research into the learning outcome and effectiveness of dynamic visualization in educational mathematics videos.

Memes and Constructions of TikTok Culture in #DontPlayPlayBosku

Memes have become a medium to convey messages on TikTok. However, memes are often underestimated. The researcher has noticed that TikTok is not merely a platform as it is also a memetic, complex text. As a platform, TikTok not only mediates and facilitates the production of short videos but also becomes a culture. Through the Don't Play Play Bosku audio meme phenomenon, this study aims to provide a perspective on how meme themes are conveyed to the public to create a high engagement space and what sorts of ideology are offered by creators in the TikTok culture space. The researcher conducted digital ethnographic observations, qualitative content analysis, and used the Shifman’s three-dimension approach to analyze 10 video samples qualitatively. This study found three meme themes, including inadequacy, metaphor and a practice of playfulness, and mental satire. At the content level, memes are not only an expression of videos but also an audio control to negotiate. At the form level, memes are delivered with word twists, reactions, and effects frequently repeated. The practice of memes at the stance level indicates that the ideological position of creators in bringing up memes is an elaboration of videos and audio that give birth to these three themes in the mechanism of TikTok culture. Memes as a TikTok culture is a culture that is actively produced and continuously consumed. The construction of TikTok culture is formed from the rituals shared in algorithm-framed imitation and replication.

Myopia information on TikTok: analysis factors that impact video quality and audience engagement

BackgroundTikTok is emerging as a vital platform for health information dissemination. Despite myopia being a global public health issue, the high-quality myopia information shared by health educators often fails to go viral. It is imperative to analyze the factors influencing video quality and popularity, especially from diverse perspectives of researchers, health educators, and audiences.MethodsTikTok myopia-related videos were retrieved using TikTok’s default comprehensive search (DCS) and most liked search (MLS) strategies. Venn diagrams were employed to illustrate the relationships and commonalities between the two strategies across four sample sizes (top 200, 150, 100, and 50). Video metadata, including details such as creator information, production properties, upload time, video duration, and viewer engagement, were collected. Video quality was assessed using the DISCERN tool. Video content covering six aspects of myopia were evaluated. The impact of search strategies, video sample sizes, production properties, and myopia content on video quality and audience engagement was analyzed through single-factor or multi-factor analysis.ResultsDCS and MLS retrieval strategies, as well as varying sample sizes, resulted in differences in audience engagement for myopia videos (P < 0.039), while The DISCERN quality scores remained comparable (P > 0.221). Videos published by healthcare professionals (HCPs) and non-profit organizations (NPOs) were associated with high-quality (P ≤ 0.014) but comparatively lower popularity (P < 0.033). Videos that reported contents of risk factors, management, and outcomes showed high popularity (P < 0.018), while longer video duration (> 60s) exhibited the opposite trend (P < 0.032). Content on myopia evaluation (P ≤ 0.001) and management (P ≤ 0.022) and video duration were positively correlated with higher DISCERN quality.ConclusionVideos created by HCPs and NPOs deserve greater attention. Rather than pursuing entertaining effects, professional educators should emphasize producing concise, and high-quality myopia content that readily resonates with the audience and has the potential to go viral on the platform.

Investigation of the Vehicle Driving Trajectory During Turning at Intersectional Roads Using Deep Learning Model

Two-thirds of vehicle accidents in Malaysia occurred at the straight type of roads, followed by intersection-type roads. Despite the deployment of traffic lights on the road, accidents still occur which are caused by illegal maneuvers, speeding or misjudgment of other’s actions. Hence, motivated by the lack of previous research regarding causes of accidents on intersectional roads, this study aims to observe the pattern of the vehicles’ speed and turning angle during the right turn after the traffic stop at the intersection road. To obtain these parameters, video samples of vehicles at two types of intersections were obtained and analyzed via YOLOV7 and DeepSORT. The two road intersections researched are four-legged intersection and three-legged intersection. 153 and 35 vehicle samples were collected from these types of road intersections, respectively. It was observed that 78 and 75 vehicles exit towards the nearest and furthest lanes at four-leg controlled crossings on divided roads. While, at a single-lane to a dual carriageway road intersection, 26 and 9 vehicles exit towards the nearest and furthest lanes, respectively. From the research, 16.52 - 17.53 km/h and 67.57°-73.33° are the most optimal turning speeds and angles respectively for vehicles at four-leg controlled crossings. Whereas 14.48 - 15.51 km/h and 144.77° - 154.403° are the most optimal turning speeds and angles respectively for vehicles at a single-lane to a dual carriageway road intersection.

Decoding Emotions: Intelligent visual perception for movie image classification using sustainable AI in entertainment computing

Affective computing researchers have been studying emotion recognition (ER) based on real-world facial photos and videos for a while now, and it’s a very popular area. With the sounds produced by head posture, facial deformation, and lighting fluctuation, ER is still difficult to execute in the wild. This research employs a machine learning model for sustainable artificial intelligence (AI) in entertainment computing to perform emotion decoding for movie picture categorization. Emotion labels are produced for every video sample by the suggested model, which takes video data as input. First, we employ a face identification and selection process based on the video data to identify the most consequential face areas. Here, facial expressions from films have been gathered as input pictures, which have then been processed for noise reduction and normalisation. Then, in order to analyse the facial expressions in this image, it was segmented using the fuzzy K-means equalisation clustering model. Convolutional adversarial U-net graph neural networks have been used to classify the studied pictures for emotion decoding. Several movie-based emotion datasets are subjected to experimental analysis in order to determine accuracy, precision, recall, F-1 score, RMSE, and AUC. With excellent classification accuracy, the suggested deep learning paradigm shows promise in identifying the emotional shifts in gamers. The suggested method achieved 97% accuracy, 96% precision, 92% recall, 85% F-1 score, 79% RMSE, and 86% AUC.

Multimodal Convolutional Neural Networks for Sperm Motility and Concentration Predictions

Semen analysis is an important analysis for male infertility primary investigation and manual semen analysis is a conventional method to assess it. Manual semen analysis has been revealed with accuracy and precision limitations due to noncompliance to guidelines and procedures. Sperm motility and concentration are the main indicators for pregnancy and conception rate hence they were selected for parameters prediction. Convolutional neural network (CNN) has benefited computer vision application industry in recent years and has been widely applied in computer vision research tasks. In this paper, three-dimensional CNN (3DCNN) was designed to extract motion and temporal features, which are vital for sperm motility prediction. For sperm concentration, since two-dimensional CNN (2DCNN) is efficient in recognizing and extracting spatial features, well-established Residual Network (ResNet) architecture was adopted and customized for sperm concentration prediction. Multimodal learning approach is a technique to aggregate learnt features from different deep learning architecture that adopted other forms of modalities, which could provide deep learning model with better insights on their tasks. Hence, a multimodal learning deep learning architecture was designed to receive both image-based (frames extracted from video samples) and video-based (stacked frames pre-processed from video samples) input that could provide well-extracted spatial and temporal features for sperm parameters prediction. The results obtained using the proposed methodology have surpassed other similar research works who used deep learning approach. For sperm motility, its best achieved average mean absolute error (MAE) was 8.048, and sperm concentration obtained a competent Pearson’s correlation coefficient (RP) value of 0.853.

Physical attributes of workout instructors and appearance-related messaging in a sample of home workout videos on YouTube: A content analysis.

Given the popularity of home workouts and effects of exposure to media messaging regarding appearance ideals, the purpose of this analysis was to examine the content of YouTube home workout videos. A sample of 298 YouTube home workout videos was coded for demographic factors and body shape of instructors and the presence of body positivity messaging, appearance-related messaging, and other relevant themes. All videos in the sample included detailed instructions and demonstrations of the exercises by instructors. The majority of the videos depicted young White women with low body fat and thin/athletic bodies with visible muscles. Instructors often focused on burning fat/calories but also emphasized body functionality, exercising for fun/enjoyment, and the importance of taking care of one's body. Overall, findings demonstrate that workout instructors on YouTube conformed to unrealistic, narrowly defined appearance ideals present in Western culture, but the messaging in the videos contained many positive and affirming qualities.

Survey on Silentinterpreter : Analysis of Lip Movement and Extracting Speech using Deep Learning

Lip reading is a complex but interesting path for the growth of speech recognition algorithms. It is the ability of deciphering spoken words by evaluating visual cues from lip movements. In this study, we suggest a unique method for lip reading that converts lip motions into textual representations by using deep neural networks. Convolutional neural networks are used in the methodology to extract visual features, recurrent neural networks are used to simulate temporal context, and the Connectionist Temporal Classification loss function is used to align lip features with corresponding phonemes. The study starts with a thorough investigation of data loading methods, which include alignment extraction and video preparation. A well selected dataset with video clips and matching phonetic alignments is presented. We select relevant face regions, convert frames to grayscale, then standardize the resulting data so that it can be fed into a neural network. The neural network architecture is presented in depth, displaying a series of bidirectional LSTM layers for temporal context understanding after 3D convolutional layers for spatial feature extraction. Careful consideration of input shapes, layer combinations, and parameter selections forms the foundation of the model's design. To train the model, we align predicted phoneme sequences with ground truth alignments using the CTC loss. Dynamic learning rate scheduling and a unique callback mechanism for training visualization of predictions are integrated into the training process. After training on a sizable dataset, the model exhibits remarkable convergence and proves its capacity to understand intricate temporal correlations. Through the use of both quantitative and qualitative evaluations, the results are thoroughly assessed. We visually check the model's lip reading abilities and assess its performance using common speech recognition criteria. It is explored how different model topologies and hyperparameters affect performance, offering guidance for future research. The trained model is tested on external video samples to show off its practical application. Its accuracy and resilience in lip-reading spoken phrases are demonstrated. By providing a deep learning framework for precise and effective speech recognition, this research adds to the rapidly changing field of lip reading devices. The results offer opportunities for additional development and implementation in various fields, such as assistive technologies, audio-visual communication systems, and human-computer interaction.

Early assessment of communicative competence in children with hearing loss using the Child-Caregiver Communication Assessment through Rebesco's Evaluation (CC-CARE) method

PurposeThis work presents a new frame-by-frame video analysis method called Child-Caregiver Communication Assessment through Rebesco's Evaluation (CC-CARE), developed in the context of pediatric hearing loss as a rehabilitation tool for assessing children's early communication skills. CC-CARE stems from the commonly used Tait video analysis and extends it by including a new set of parameters aimed at disentangling between hearing-dependent and hearing-independent aspects of communication. Method: In this paper, we collected video samples of child-caregiver interactions in a group of 65 normal-hearing children and a group of 165 hearing-impaired children. For each group, we present the CC-CARE method and describe the parameters, their score distributions, correlations and we estimate the adherence of the CC-CARE scores with children's developmental trajectory. Moreover, we compare the results of CC-CARE scores between the two groups having had different development of the auditory system. Finally, a fully-data driven approach was employed to assess the consistency of the communicative efficacy index (CEI), a score aiming to capture a global result of the CC-CARE procedure. Results: Correlations among parameter scores were found in each within-group analysis, revealing CC-CARE's internal consistency in measuring associated but nonoverlapping communication dimensions. For both groups, CC-CARE scores were associated with participants' age. Differences between scores emerged for a between-group analysis, indicating CC-CARE sensitivity to extract communication differences as a function of the hearing status. For both groups, the data analysis revealed that the CEI captures large variance portions across all parameter scores of the CC-CARE method. Conclusions: Results provide the first evidence that the CC-CARE video analysis method could be a reliable tool capable of highlighting the cascading effects of hearing impairment on children's preverbal communicative efficacy. The CC-CARE method aims to support early rehabilitation of hearing loss by describing a child's communicative efficacy.

Influence of network video film and television communication based on sampling theory

The media landscape has been evolving at a dizzying pace in recent years, with short video platforms emerging as a significant public information conduit; yet, it must be acknowledged that the impact of these platforms on transmission is still rather limited. Both the width and the depth are not up to par. Improving the impact, increasing the reach and impact, and making the most of the Douyin platform for scientific communication activity is an immediate concern. Douyin, China's most popular video-sharing platform, with 400 million daily users. Douyin provides regular individuals with the resources they need to start their own video art careers; it is popular outside of tier 1 cities. There are also inventive contests on the site where individuals can try their hand at being famous overnight. In this study, we look at how social media platforms evaluate short video head accounts online. Research is conducted to separate out the communication features of the head accounts based on popular scientific themes, the length of the form, and the use of other platform functions in the operation process. This study use random theory to pick various sorts of short videos as the final effective analysis samples, with a random sampling range of 15% of the empirical sample size, in order to deal with the temporal dimension and the enormous number of online video samples. Among the topics covered by Douyin's popular scientific short video accounts, the percentages for science, technology, innovations, medical care, health, and nature stand at 18.99%, 16.59%, and 15.38%, respectively, according to the article's sample data. Online video and film may be effectively disseminated in this research's media ecosystem, which is based on sampling theory.

Health information in short videos about metabolic dysfunction-associated steatotic liver disease: Analysing quality and reliability.

Short videos, crucial for disseminating health information on metabolic dysfunction-associated steatotic liver disease (MASLD), lack a clear evaluation of quality and reliability. This study aimed to assess the quality and reliability of MASLD-related videos on Chinese platforms. Video samples were collected from three platforms (TikTok, Kwai and Bilibili) during the period from November 2019 to July 2023. Two independent reviewers evaluated the integrity of the information contained therein by scoring six key aspects of its content: definition, epidemiology, risk factors, outcomes, diagnosis and treatment. The quality and reliability of the videos were assessed using the Journal of the American Medical Association (JAMA) criteria, the Global Quality Score (GQS) and the modified DISCERN score. A total of 198 videos were included. The video content exhibited an overall unsatisfactory quality, with a primary emphasis on risk factors and treatment, while diagnosis and epidemiology were seldom addressed. Regarding the sources of the videos, the GQS and modified DISCERN scores varied significantly between the platforms (p = .003), although they had generally similar JAMA scores (p = .251). Videos created by medical professionals differed significantly in terms of JAMA scores (p = .046) compared to those created by nonmedical professionals, but there were no statistically significant differences in GQS (p = .923) or modified DISCERN scores (p = .317). The overall quality and reliability of the videos were poor and varied between platforms and uploaders. Platforms and healthcare professionals should strive to provide more reliable health-related information regarding MASLD.

A self-supervised learning network for remote heart rate measurement

The rPPG technology is a facial video-based non-contact physiological parameter detection technology, which has been widely used in the measurement of heart rate, respiration, blood oxygen and other physiological parameters. Most of the current research hotspots in this field are based on deep learning methods, which typically necessitate a significant amount of data samples for training the neural network to obtain accurate measurement results. However, the high cost required to label the data samples limits the diffusion and application of the technique. To address the above problems, we propose a new self-supervised aimed at acquiring the ability to estimate rPPG signals from facial videos, eliminating the need for labeled data. The framework expands the unlabeled video samples into multiple positive/negative samples and uses contrast learning to obtain an rPPG signal estimation network, which outputs the corresponding rPPG signals of face videos. To promote the convergence of network training, a new frequency loss function is designed. This function can effectively shorten the distance between the frequencies of similar sample signals and push the distance between the frequencies of different sample signals far away, so as to enhance the frequency consistency between sample signals and make the model easier to learn the differences between different sample pairs. Our method is evaluated on four standard rPPG datasets, the experimental results show that the accuracy is close to the current best supervised method, and is superior to the previous self-supervised method without using any labeled data.

Research on Intelligent Supervision System of Ore Pass

To prevent safety accidents caused by mining vehicles and personnel entering the operation area by mistakes, it is necessary to reduce the risk of the ore pass. However, the underground space of the mine is narrow, and factors such as dust and noise during the unloading process endanger the health of the personnel on duty in the ore pass. As such, the target detection technology based on deep learning is introduced into the underground monitoring system. The underground surveillance video samples are collected to establish a dataset for Yolov3 algorithm to identify minecarts. Through optimizing the Yolov3 model training process and algorithm, and using the dual-camera collaborative discrimination method, the influence of brightness on the recognition results when the loaders or trucks lights are turned on can be overcome. Four types of minecarts can be accurately identified from the underground surveillance video. On the basis of mining car recognition, an intelligent access control system for mine shafts based on Jetson Nano’s embedded development is developed. The on-site operation results show that the average accuracy of target vehicle recognition is within the range of 95%-100%. The system continuously recognizes the mine car 5 times from the detection program and sends the opening and closing command to complete a 90 ° rotation, which takes only 3 seconds，effectively meeting the needs of the mine for ore pass control.

Video-based working condition recognition of fused magnesium furnace with stochastic configuration networks

Working condition recognition of a fused magnesium furnace (FMF) suffers from unbalanced under-burning condition samples, inconsistent quality of training samples and difficulty in characterizing dynamic production processes with images. This paper presents a novel approach to detect the under-burning working condition in FMF based on a 3D-cycle-generative adversarial network (3D-cycle-GAN) and Video Swin Transformer-Stochastic Configuration Networks (Video Swin-SCNs). Firstly, to resolve the temporal discontinuity defect caused by Cycle-GAN through the visual appearance of video composite frames, we construct a motion consistency-based 3D-Cycle-GAN model that considers the visual appearance and temporal continuity constraints of unpaired video transitions and is designed to generate video samples of under-burning working conditions. Secondly, a reinforcement learning approach is used to assess the value of the video quality and to filter out possible low-quality samples generated. Finally, the local attention is extended from the spatial domain to the spatial-temporal domain to solve the difficulty in characterizing the dynamic production process with the static images. The spatial-temporal features from the Video Swin Transformer are fed into SCNs to classify the working conditions. The experiment results indicate the effectiveness and feasibility of the proposed method.