首页|Findings on Machine Learning Reported by Investigators at University of Connecticut (Leveraging Past Information and Machine Learning To Accelerate Land Disturbance Monitoring)

Findings on Machine Learning Reported by Investigators at University of Connecticut (Leveraging Past Information and Machine Learning To Accelerate Land Disturbance Monitoring)

扫码查看
By a News Reporter-Staff News Editor at Robotics & Machine Learning DailyNews Daily News - Investigators publish new report on Ma chine Learning. According to news reportingoriginating from Storrs, Connecticut , by NewsRx correspondents, research stated, “Near real -time (NRT)monitoring o f land disturbances holds great importance for delivering emergency aid, mitigat ing negativesocial and ecological impacts, and distributing resources for disas ter recovery. Many past NRT techniqueswere built upon examining the overall cha nge magnitude of a spectral anomaly with a predefined threshold,namely the unsu pervised approach.”Financial support for this research came from USGS-NASA Landsat Science Team (LS T) Program.Our news editors obtained a quote from the research from the University of Conne cticut, “However,their lack of fully considering spectral change direction, cha nge date, and pre-disturbance conditions oftenled to low detection sensitivity and high commission errors, especially when only a few satellite observationswe re available at the early disturbance stage, eventually resulting in a longer la g to produce a reliabledisturbance map. For this study, we developed a novel su pervised machine learning approach guided byhistorical disturbance datasets to accelerate land disturbance monitoring. This new approach consistedof two phase s. For the first phase, the supervised approach applied retrospective analysis o n historicalHarmonized Landsat Sentinel-2 (HLS) datasets from 2015 to 2021, com bined with several open disturbanceproducts. The disturbance model was construc ted for each condition of consecutive anomaly number,with the aim of enhancing the specificity for delineating early-stage disturbance regions. Then, these stage-based models were applied for an NRT scenario to predict disturbance probabil ities with 2022 HLSimages incrementally on a weekly basis. To demonstrate the c apability of this new approach, we developedan operational NRT system incorpora ting both the unsupervised and supervised approach. Latency andaccuracy were ev aluated against 3000 samples that were randomly selected from the five most infl uentialdisturbance events of the United States in 2022, based on labels and dis turbance dates interpreted fromdaily PlanetScope images. The evaluation showed that the supervised approach required 15 days (sincethe start of the disturbanc e event) to reach the plateau of its F1 curve (where most disturbance pixels aredetected with high confidence), seven days earlier with roughly 0.2 F1 score im provement compared to theunsupervised approach (0.733 vs. 0.546 F1 score). Furt her analysis showed the improvement was mainlydue to the substantial decrease i n commission errors (17.7% vs 44.4%). The latency com ponent analysisillustrated that the supervised approach only took an average of 4.1 days to yield the first disturbance alertat its fastest daily updating spe ed, owing to its decreased sensitivity lag.”

StorrsConnecticutUnited StatesNorth and Central AmericaCyborgsEmerging TechnologiesMachine LearningUniversity of Connecticut

2024

Robotics & Machine Learning Daily News

Robotics & Machine Learning Daily News

ISSN:
年,卷(期):2024.(MAY.6)