State Farm (United States)

Abstract Small mustelids are difficult to survey due to their low density and cryptic nature. Population status of North American weasels (Mustela erminea, Mustela nivalis, and Neogale frenata) are believed to be in decline, but there are no standardized monitoring protocols to evaluate their status. To support weasel monitoring, we compared the attractiveness of various combinations of baits and lures to weasels in sites located throughout the eastern and central USA. We baited a total of 122 clusters of 4 camera traps, across 14 states, with random combinations of 4 baits and 3 scent lures in the winters of 2022 and 2023. Cameras baited with meat were 3.5 times more likely to detect both short‐ and long‐tailed weasels on average (mean percentage of cameras detecting weasels: 20–30%) than those with scent lures (3–11%). Red meat was twice as effective at attracting short‐tailed weasels (50%) as chicken or cat food (20%; Z = 2.49, p < 0.01). While red meat marginally increased detections of long‐tailed weasels (21%) compared to chicken and cat food (19%), its effectiveness was influenced by whether the bait was stolen (Z = 2.08, p = 0.04). Additionally, long‐tailed weasels were detected in half the time when raw chicken was used (median days to detection: red meat = 39.5 days, raw chicken = 14.5 days). When salmon oil was added to meat bait, it increased the likelihood of detecting short‐tailed weasels and reduced the time to detection for both species. A variety of non‐target species stole meat bait during the survey, making the camera traps less effective. The addition of salmon oil may have allowed for continual attraction of weasels until stolen meat bait could be replenished. In summary, red meat was the best all‐purpose bait for weasels, although raw chicken is similarly effective for long‐tailed weasels, and the addition of salmon oil is helpful. We also recommend a specific bait enclosure design that was the most effective at minimizing theft of bait. We propose our baiting strategy can be used as a survey standard to evaluate the distribution and population status of weasels.

Hyperspectral image classification is a critical task in remote sensing, where recent transformer-based methods have shown promising improvements in spatial-spectral feature ex-traction. However, most existing transformer-based hyperspectral image classification methods rely on absolute or bias-based posi-tional embeddings that fail to preserve geometric relationships; limiting their ability to capture precise diagonal and multi-directional spatial–spectral interactions between hyperspectral pixels and resulting in degraded classification accuracy. Although Rotary Position Embedding (RoPE) and its variants have been studied in computer vision with standard RGB image input, their use in hyperspectral imaging (HSI), especially for pixel-level clas-sification, remains underexplored. We propose HyperRoPE-SST, a transformer-based architecture based on a patch-local mixed-axis 2D RoPE with learnable frequencies that preserves geometric relationships amongst the pixels within a hyperspectral patch while adapting to dataset-specific spatial–spectral characteristics. The HyperRoPE-SST pipeline employs principal component analysis (PCA) for spectral dimensionality reduction followed by the extraction of target-centered local patches and embedding of these patches with shallow convolutions. A 2D RoPE-enhanced transformer encoder subsequently fuses the spatial and spectral information via center-focused attention to produce context-aware target pixel embeddings, yielding superior classification performance. Extensive experiments on five benchmark datasets, i.e., Indian Pines (IP), Pavia University (PU), Houston 2013 (HU), Kennedy Space Center (KSC), and Salinas (SAL), demonstrate that HyperRoPE-SST consistently outperforms state-of-the-art convolutional neural network (CNN)- and transformer-based models, achieving overall accuracies of 96.81%, 95.65%, 93.38%, 99.62%, and 95.59% respectively, on the aforementioned datasets. The source code for HyperRoPE-SST will be released upon acceptance at: https://github.com/zirakkk/hyperrope-sst.

Significant tornado events have prompted a push for the development of design standards that consider tornado loading for conventional buildings and structures. One important loading parameter in the design standards is the variation in the horizontal wind speed with the height (i.e., wind speed profile) as manifested in a velocity pressure profile. Different from the atmospheric boundary layer (ABL) in which the wind speed monotonically increases with height, the average wind speed profile in tornadoes exhibits a “nose-like” profile for which the wind speed increases from the surface to a local maximum at “nose” height and then decreases above that height. A tornado task committee (TTC) was convened through the ASCE 7 Wind Load Subcommittee, in part to report on the collection, review, and analysis of tornado wind speed profile data and to propose a “design” tornado velocity pressure profile for inclusion in the new tornado load chapter of the ASCE 7-22 standard. A total of 36 tornado profiles were evaluated independent of terrain exposure or surface roughness and collected from mobile radar data. Significant variability was noted in the profiles, but many showed a peak horizontal wind speed relatively close to surface, with a median height of approximately 164 ft (50 m). A proposed tornado velocity pressure profile and associated velocity pressure exposure coefficient, KzTor, was then developed for ASCE 7-22. The proposed nominal tornado profile closely followed the median radar profile. Values of the new KzTor=1.0 between ground level and 200 ft (61 m) decrease linearly to 0.9 at a height of 328 ft (100 m) then remain constant above that height.

This research employs a series of machine learning methods to predict the direction of lane change. The response is a binary variable indicating changing the lane to the left or to the right. The employed methods include Decision Tree, Discriminant Analysis, Naïve Bayes, Support Vector Machine, k-Nearest Neighbor and Ensemble. The results are compared to the conventional logistic regression method. Both performance criteria and computational times are reported for comparison purposes. A design of experiments is run to test 25 classification methods at ratios of 25%, 50%, and 75% right to left lane change data. Moreover, samples are validated by cross and holdback validation methods. RUS Boosted trees, an ensemble method, shows improvement over logistic regression. This research provides valuable insights on lane change behaviour, including trajectories and driving styles, which falls into the field of microscopic lane change study.

Objective: There are many unknowns regarding drivers’ use and acceptance of advanced vehicle technologies. This research aimed to examine drivers’ perceptions of advanced driver assistance systems (ADAS).Methods: This research was conducted using structured interviews and focus groups of owners of vehicles with advanced technologies.Results: Drivers’ perceptions about ADAS were mixed, but generally safety was considered to be the greatest value of the systems. There was recognition that the systems may result in overreliance and thus encourage distraction behaviors or other bad driving habits, and participants generally expressed that they were ultimately responsible for the vehicle’s operation and needed to be ready to override the system if it failed.Conclusions: The findings indicate that driver characteristics and individual factors may influence perceptions, behaviors, and interactions with safety technology, and this research is a first step toward understanding any influences. Human factors issues related to automated vehicle technologies are critical for design and deployment, including those of trust, acceptance, and understanding of systems.