Daily Archives: August 31, 2026

Authors: Abirami A, Miranda Lakshmi T, Martin A

Abstract: Conversational emotion recognition is inherently challenging because an utterance's emotional label does not depend on wording alone — it is shaped jointly by content, delivery, and the dialogue history surrounding it. This work introduces a multimodal deep-learning pipeline that fuses textual, acoustic, and contextual signals to classify emotion at the utterance level within conversations. Linguistic meaning is captured through pretrained BERT embeddings, broad acoustic patterns through Wav2Vec2 embeddings, and explicit vocal-affect cues through handcrafted prosodic descriptors (pitch statistics and short-time energy). Once concatenated at the utterance level, these heterogeneous features are passed into a Bidirectional Gated Recurrent Unit (BiGRU) that captures dependencies across dialogue turns, with a fully connected head performing the final classification. Because the benchmark dataset's seven-class emotion taxonomy is heavily skewed, a weighted cross-entropy loss built from inverse class-frequency statistics is applied to offset this imbalance during training. On the MELD (Multimodal EmotionLines Dataset) benchmark, the proposed pipeline reaches 77.56% overall accuracy and a weighted F1-score of 0.804, with per-class ROC-AUC spanning 0.89 to 0.96. Confusion-matrix and error analysis reveal that common emotions such as neutral and joy are classified reliably, whereas rarer emotions such as fear and disgust remain substantially harder to recognize even with class-weighted training — underlining how persistent the data-imbalance problem remains in this domain and pointing toward future work on stronger imbalance-handling and context-modelling techniques.

DOI: https://doi.org/10.5281/zenodo.22204728

Published by: vikaspatanker

Authors: T. Gopalakrishnan, Boopathi S, B. Niranjan, Mukesh P.M

Abstract: Global almond (Prunus dulcis) production leaves behind a shell fraction that exceeds the mass of the kernel it protects, and the great majority of this lignocellulosic by-product is presently burned, landfilled or sold at low value as biomass fuel. Because the shell is hard, largely cellulosic and available in very large and geographically concentrated volumes wherever almonds are processed, it is a natural candidate for the same role that walnut shell, groundnut shell, coconut shell and rice husk already occupy in the particulate-filled polymer composite literature: a low-cost, renewable, low-density filler for thermosetting resins. This review draws together what is presently known, and what is not yet known, about almond shell powder as a reinforcement for epoxy matrices. It begins with the scale and composition of the waste stream, since both bear directly on whether the material is a viable industrial filler rather than a laboratory curiosity. Processing considerations — drying, size reduction, sieving and surface treatment — are then examined, followed by the mechanical, physical, water-absorption, thermal and morphological behaviour reported for shell-powder-filled epoxy systems, read against the much larger body of work on chemically similar agro-waste fillers where almond-specific data are still sparse. The review closes by identifying the almond shell literature as considerably thinner than that for walnut or coconut shell despite comparable material properties, and names surface treatment optimisation, hybridisation with synthetic or other natural reinforcements, durability under moisture and weathering, and a proper life-cycle assessment against synthetic fillers as the most pressing gaps.

DOI: https://doi.org/10.5281/zenodo.22203936

Published by: vikaspatanker

Authors: Zachariah Munyoro, Alice Nambiro

Abstract: Different researchers have carried out Beach-litter prediction, with basis of their studies being detection and counting of existing litter. Forecasting requires integration of temporal litter patterns, with contextual conditions that influence accumulation. Continuous monitoring of beach litter is necessary within this area of study. Conventional process of monitoring provides observation to a given extent, which limits rapid or automated assessment. AI-based detection allows automatic identification and quantification, allowing monitoring to continue from detection towards prediction. Existing methods address detection, forecasting or contextual integration separately. Various predictive studies use historical litter records, contextual variables, camera observations and Machine Learning. Limited architectural integration connects AI-derived litter measurements directly with temporal and contextual information for multi-predictor variables within multi-duration dimensions. This study developed an AI-based predictive model architecture. The model connected automated litter measurement temporal construction, contextual integration, predictive modelling and forecasting generation. Images were captured through IoT-enabled cameras, subjected under RT-DETRv2, which detected, classified and counted the objects. The output of this process is category-specific quantitative litter observations. The image-level counts are aggregated into beach-week observations, forming temporal litter representation. The temporal information provides the predictive baseline. The predictive model generated category-specific forecasts at multiple future durations. These were weekly, bi-weekly and monthly forecasts. The architecture linked AI-based measurement (RT-DETRv2), data integration, prediction and monitoring output in one end-to-end flow. The model provides a framework for anticipatory beach-litter monitoring instead of detection alone. The output of the predictive model architecture provide diverse options necessary for specific management decision making as well as a feedback loop to the predictive process.

Published by: vikaspatanker