This study aimed to classify garbage images using the deep convolutional neural network model AlexNet. Over 150 epochs of validation were performed. Our method demonstrates how deep learning models can be used to improve and automate waste management systems. This research is novel because it examines multiple approaches for a potential fusion technique and evaluates how well-suited and improved various machine learning and image detection models are. This study is the first in a planned series of research using SAM and AlexNet models, providing a foundation for future research in this area. Through incorporating novel methodologies and capitalizing on the advantages of diverse frameworks, this study advances the creation of more effective and efficient waste management strategies. This fundamental work highlights the usefulness and suitability of cutting-edge machine learning methods for environmental sustainability initiatives by laying the groundwork for significant advancements in waste classification precision and system resilience.

The growing complexity of agricultural ecosystems increasingly requires automated systems that can support real-time, data-driven decision-making. In response to this need, this paper presents a multimodal data fusion framework aimed at enhancing intelligent farming systems by integrating heterogeneous information sources, including IoT sensor streams (soil N, P, K, pH) and historical datasets from the FAO. The proposed Two-Layer Analytics Architecture combines derived event generation with probabilistic fusion mechanisms to address challenges related to scalability, data heterogeneity, and operational responsiveness. At the core of the framework, we adopt an ensemble learning strategy based on Random Forest, XGBoost, and Support Vector Machines (SVM) to classify crop suitability and soil conditions. Experimental evaluations indicate that this fusion-based approach yields a weighted F1-score improvement of 6 % compared to single-source, single-model baselines. Furthermore, the overall system architecture is validated in an edge computing context, where it demonstrates linear scalability with a processing throughput of 10⁴ events per second while maintaining fusion latency consistently below 100 ms. These results highlight the transformative potential of multimodal data fusion in enabling high-precision, resilient, and sustainable agricultural automation.

Deep-space exploration requires autonomous systems capable of navigating environments where sensor reliability is compromised by cosmic radiation, solar flares, and hardware degradation. This research addresses the "Perception-Fusion Gap"—the failure of static AI models to adapt when environmental conditions shift dynamically. We propose a modular framework that treats data integration as a continuous state estimation problem rather than a static weighted average. At the core of this strategy is a Kalman Filter-based fusion engine [1]. By utilizing the state prediction and fusion update equations, the system models the reliability of heterogeneous sensor streams in real-time. The Kalman Gain serves as a dynamic weighting mechanism, automatically favoring the model’s internal state prediction or the incoming sensor measurement based on the calculated covariance of their respective errors. This allows the vehicle to "self-repair" its perception logic during flight without manual re-calibration. To further enhance resilience, we integrate Meta-Learning Controllers (planned for WP3) that observe past fusion errors to predict future sensor drifts. This dual approach ensures that even if a primary modality, such as LiDAR, is blinded or compromised, the system can seamlessly shift reliance to secondary modalities like multispectral imagery or inertial sensors. Experimental benchmarks against static fusion baselines demonstrate that this adaptive approach significantly improves system robustness under abrupt data shifts. By bridging the gap between raw data perception and intelligent decision-making, this methodology provides a reliable foundation for the next generation of autonomous deepspace vehicles [6]. Keywords: Autonomous Systems, Self-Adaptive Fusion, Meta-Learning, Edge Computing, Sensor Resilience, State Estimation.

Agricultural catalogues are awkward to benchmark once the records stay at the centres that produced them. We turn 75,897 bibliographic records from the International System for Agricultural Science and Technology (AGRIS), already held by 7 area-prefix clients, into a federated multi-view benchmark. Each record has four disjoint views (1,096 features), a ten class AGROVOCheuristic, publication-level grouping, a client- and class-aware 70/10/20 cut, and ablinded400-record subject-matter expert (SME) audit. Against the consensus gold the heuristic reached 94.50% accuracy and macro-F1 = 0.945 (sampling-weighted 94.91%). Grouping removes 1,132 duplicate-document groups that would otherwise cross folds. Compute-matched Federated Averaging (FedAvg) is 0.2998 ± 0.0151 macro-F1, against 0.3041 ± 0.0021 when the data are pooled and 0.2884 ± 0.0037 when they are not. Eight-bit uplink quantization keeps 0.3001 ± 0.0147 at a 4× uplink cut (1.6× on total traffic). The full pipeline is archived on Zenodo as release v1.0 (DOI: 10.5281/zenodo.22956341). This is a benchmark and evaluation framework, not a new federated-learning method.

We evaluate whether typicality and uncertainty improve multimodal fusion of winter- wheat leaf area index(LAI)and biomass increments beyond availability-only gating(availability plus agronomic context). Consecutive elementary sampling unit (ESU) visits are kept when their 20m windows overlap (n = 105 LAI, n = 141 biomass); overlap is not a repeated plot. Nested field-grouped cross-validation does not favour the full gate. A nonnegative stacker is more accurate on fully observed ∆LAI (RMSE 0.827 vs 1.069). Availability-only gating has a slightly lower observed RMSE on fully observed biomass increments (101.1 versus 102.4), with no statistically significant difference detected (field Wilcoxon p = 0.23). No consistent field-level benefit of typicality and uncertainty was established. Withholding the current-biophysics expert from the increment predictor measured yt is still used in reconstruction—gives reliability-gated fusion (SARG) the lowest pooled biomass RMSE, but not a paired win versus availability-only (p = 0.42). Concatenated walk-forward does not establish usable next-campaign skill: ridge has a small numerical ∆LAI RMSE gain over the mean-increment baseline (2.190 vs 2.234), and every concatenated R² remains negative. Reconstruction \(\widehat{y}_{t+1} = y_t + \widehat{\Delta y}\) uses the measured current state.

Sustaining human life on long-duration missions require the automation of extraterrestrial greenhouses to ensure food security and oxygen production. These biological systems are highly sensitive, demanding the fusion of heterogeneous data sources—from real-time IoT soil sensors to multispectral satellite imagery—to maintain optimal conditions. We extend the Intelligent Farming (ICARA 2026) framework to address the unique complexities of space-based agriculture. The proposed solution utilizes a Two-Layer Analytics Architecture. Layer 1 (Edge Processing) is deployed on constrained devices like the NVIDIA Jetson, ingesting high-velocity IoT logs (Soil N, P, K, pH, and TDS) to perform immediate anomaly detection. Layer 2 (Cloud/Batch Processing) fuses this real-time stream with historical datasets and multispectral inputs to classify long-term crop suitability. Instead of simple averaging, we employ a probabilistic fusion of events derived from an ensemble of Random Forest (RF), XGBoost, and Support Vector Machines (SVM). This “Derived Event Fusion” allows the system to generate high-level alerts, such as combining a “Low Nitrogen” sensor event with a “Radiation Spike” forecast, significantly improving context awareness compared to raw data analysis. Experimental validation shows a 6% improvement in Weighted F1-Score accuracy over single-model baselines. Furthermore, the system demonstrates linear scalability, successfully processing up to 10,000 events per second. By optimizing for both low latency and high accuracy, this framework provides the robust decision-support required for closed-loop life support systems, ensuring crop sustainability in the harsh conditions of microgravity.