In the livestock industry, behavioral monitoring of cattle is essential for early disease detection and welfare improvement. Traditional classification models excel at predicting predefined categories but cannot generate descriptive explanations of observed behaviors. Vision-Language Models (VLMs) offer a promising alternative by generating free-form natural language responses, enabling interpretable behavioral descriptions. By finetuning InternVL3.0 with a systematic data pipeline that synthesizes natural language annotations and preserves rare behaviors through targeted sampling, we achieve substantial improvements in cattle behavior video question answering. A domain expert rates our finetuned model$3 \times$higher than the zero-shot baseline, demonstrating that domain-adapted VLMs can understand behavioral patterns and articulate observations in natural language-capabilities essential for practical livestock management systems.

Fingerprinting-based indoor localization for livestock tracking requires labor-intensive site surveys to construct radio maps, limiting practical farm deployment. This study proposes a calibration-free zone localization method using rank matching with maximum a posteriori (MAP) inference. The method exploits signal-strength rank (the relative signal strength order among anchors), which remains robust to global attenuation, enabling direct transfer of simulation-built reference maps. For noise-induced rank inversions, MAP inference via a precomputed probabilistic lookup table provides robust zone estimation. Aggregated zone-level evaluation shows F1-score above 0.80 in most zones under harsh noise. Across barn geometries, the method significantly outperforms Random Forest and heuristic baselines (Friedman test,$p<.001$) with statistically equivalent performance (TOST,$\delta=0.05$).

There are performance deficiencies that hamper the deployment of Wireless Sensor Networks (WSNs) in critical monitoring applications. Such applications are characterized by considerable network load generated as a result of sensing some characteristics of the monitored system. Excessive packet collisions lead to packet losses and retransmissions, resulting in significant overhead costs and latency. In order to address this issue, we introduce a distributed and scalable scheduling access scheme that mitigates high data loss in data-intensive sensor networks and can also handle some mobility. Our approach alleviates transmission collisions by employing virtual grids that adopt Latin Squares characteristics to time slot assignments. We show that our algorithm derives conflict-free time slot allocation schedules without incurring global overhead in scheduling. Furthermore, we verify the effectiveness of our protocol by simulation experiments. The results demonstrate that our technique can efficiently handle sensor mobility with acceptable data loss, low packet delay, and low overhead.

The spatial query has been processed in two steps, the filter step and the refinement step, due to the large volume and high complexity of the spatial data. However, this approach has been considered only in the query execution phase after completion of the query optimization phase. This paper presents query optimization strategies which take the characteristics of spatial databases into account. The first strategy is the separation of filter and refinement steps not in the query execution phase but in the query optimization phase. As the second strategy, several refinement operations can be combined in processing a complex query, and as the third strategy several filter operations can also be combined. We call the optimization technique utilizing these strategies the early separated filter and refinement (ESFAR). This paper also presents a rule-based optimization technique for ESFAR.

We propose a new object decomposition method, called DMBRs, to improve the performance of spatial query processing. This method is suitable for complex spatial objects in real-world geographic applications. The basic idea is that a polygon is recursively divided into two subpolygons by splitting its MBR until a given constraint is satisfied. To increase the efficiency of the DMBRs method, an extension of an existing spatial indexing structure is presented. Since this new structure can prune a number of false hits quickly, the performance of spatial query processing can be improved. The proposed method is compared with traditional decomposition methods by an analytical study. This comparison shows that our decomposition method outperforms the traditional decomposition methods. 1 Introduction Queries in spatial databases are usually concerned with massive volumes of data and complex spatial objects. Spatial objects are characterized by extremely irregular geometric components which do ...