The dynamic expressions of emotion convey both the emotional and functional states of an individual’s interactions. Recognizing the emotional states helps us understand human feelings and thoughts. Systems and frameworks designed to recognize human emotional states automatically can use various affective signals as inputs, such as visual, vocal and physiological signals. However, emotion recognition via a single modality can be affected by various sources of noise that are specific to that modality and the fact that different emotion states may be indistinguishable. This review examines the current state of multimodal emotion recognition methods that integrate visual, vocal or physiological modalities for practical emotion computing. Recent empirical evidence on deep learning methods used for fine-grained recognition is reviewed, with discussions on the robustness issues of such methods. This review elaborates on the profound learning challenges and solutions required for a high-quality emotion recognition system, emphasizing the benefits of dynamic expression analysis, which aids in detecting subtle micro-expressions, and the importance of multimodal fusion for improving emotion recognition accuracy. The literature was comprehensively searched via databases with records covering the topic of affective computing, followed by rigorous screening and selection of relevant studies. The results show that the effectiveness of current multimodal emotion recognition methods is affected by the limited availability of training data, insufficient context awareness, and challenges posed by real-world cases of noisy or missing modalities. The findings suggest that improving emotion recognition requires better representation of input data, refined feature extraction, and optimized aggregation of modalities within a multimodal framework, along with incorporating state-of-the-art methods for recognizing dynamic expressions.
This paper presents two new schemes for interpolating missing samples in satellite temperature cycles. The first scheme, referred to here as the cosine model, is an improvement model proposed by Göttsche et al. (F. M. Göttsche and F. S. Olesen, “Characterization of landsurfaces by modelling diurnal temperature waves extracted from Meteosat-IR data,” in Proceedings of the 9th conference on Satellite Meteorology and Oceanography, (Boston, MA), pp. 426-427, American Meteorological Society, 1998) and combines a cosine and exponential function for modelling the diurnal temperature cycle. The second scheme employs a Reproducing Kernel Hilbert Space interpolator (N. Aronszajn, “Theory of Reproducing Kernels,” Trans. Am. Math. Soc., vol. 68, pp. 337–404, 1950) for interpolating the missing samples. The application of reproducing kernel hilbert space interpolators to the diurnal temperature cycle interpolation problem is novel. Results obtained by means of computer experiments are presented.
A new active fire event detection algorithm for data collected with the Spinning Enhanced Visible and Infrared Imager (SE-VIRI) sensor, based on the extended Kalman filter, is introduced. Instead of using the observed temperatures of the spatial neighbours of a pixel to detect anomalous temperatures, the new algorithm only considers previous observations at the current pixel. The algorithm harnesses the Kalman filter to obtain a prediction of the expected brightness temperature at a given location, which is then compared to the actual SE-VIRI observation. An adaptive threshold is used to determine whether the observed difference is indicative of a potential fire event. Initial tests show that the performance of this method is comparable to that of the EUMETSAT FIR product.
Detection of an active fire in an image scene relies on an accurate estimation of the background temperature of the scene, which must be compared to the observed temperature, to decide on the presence of fire. The expected background temperature of a pixel is commonly derived based on spatial-contextual information that can overestimate the background temperature of a fire pixel and therefore results in the omission of a fire event. This paper proposes a method that assimilates brightness temperatures acquired from the Geostationary Earth Orbit (GEO) sensor MSG-SEVIRI into a Diurnal Temperature Cycle (DTC) model. The expected brightness temperatures are observational forecasts derived using the ensemble forecasting approach. The threshold on the difference between the observed and expected temperatures is derived under a Constant False Alarm Rate (CFAR) framework. The detection results are assessed against a reference dataset comprised of MODIS MOD14/MYD14 and EUMETSAT FIR products, and the performance is presented in terms of user’s and producer’s accuracies, and Precision-Recall and Receiver Operating Characteristic (ROC) graphs. The method has a high detection rate when the data assimilation is implemented with an Ensemble Kalman Filter (EnKF) and a Sampling Importance Resampling (SIR) particle filter, while the weak-constraint Four-Dimensional Variational Assimilation (4D-Var) has comparatively lower detection and false alarm rates according to the reference dataset. Consideration of the diurnal variation in the background temperature enables the proposed method to detect even low-power fires.
The Performance Efficiency Factor (PEF) is a key composite metric in broiler production that integrates livability, average body weight, feed conversion ratio (FCR) and age. While tree-based machine learning models have shown promising results for live-weight prediction, they often struggle with temporal dependencies and sparse mortality data. This study evaluates five tree-based algorithms, Decision Tree, Random Forest, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and Category Boosting (CatBoost) on daily sensor data from commercial broiler farms. Data were stratified into six age categories, pre-processed with outlier retention and feature engineering, and assessed using R2, RMSE, MAE, paired t-tests for statistical significance and SHapley Additive exPlanations (SHAP) for explainability. Cross-farm validation was performed to assess generalizability. The models achieved high accuracy for live-weight prediction (R2 up to 0.97, RMSE as low as 9.23 g in mid-cycle categories), with Random Forest and LightGBM performing best. Mortality prediction remained challenging (R2 −1.63 to 0.53) due to its sparse and stochastic nature. Nevertheless, Day 21 to Day 28 PEF forecasts showed relative errors of only 4.42–5.40%, as live-weight predictions dominated the PEF calculation. SHAP analysis consistently identified bird age, feed intake per bird and temperature as the most influential predictors. Tree-based models offer a robust, interpretable and computationally efficient solution for live-weight and PEF forecasting in commercial broiler production. The findings support proactive farm management and highlight the need for hybrid approaches to improve mortality prediction.