Event detection algorithms in decision support for animals health surveillance

点击翻译按钮获取中文摘要

Comprehensive animal health surveillance systems that can quickly identify and characterize important disease events are essential infrastructure necessary to maintain the security, economic viability and long term sustainability of livestock industries. Failing to detect incursions of foreign diseases or emergencies of new, previously unknown diseases before they are widely dispersed can result in catastrophic disease outbreaks. The challenges to effectively detect and describe events essential to animal health are inherently visible in surveillance systems. These events, albeit being quite rare, are of paramount importance. There are high expectations of the end-user (community, decision-makers at various levels) as to the quality of the detection of the event, its clear characterization (description) and confidence level associated with the event itself and its associated quantification (e.g., its severity, spread, etc.). Users anticipate that voluminous spatial-temporal data can be effectively accessed and summarized, with "interesting" findings revealed and communicated;n forms of event descriptors that are easily understood so that one can proceed with further actions. These expectations regarding event detection constitute one of today's challenges ;n the design of intelligent surveillance systems. User-centricity of such systems becomes a genuine necessity so that they offer full benefits to stakeholders. In close collaboration with our industrial partner, AQL Management Consulting, Inc. (AQLMC), and our supporting organization, Alberta Agriculture and Rural Development, we propose the design and deployment of an innovative event detection framework exploiting the paradigm of Computational Intelligence which serves here as a comprehensive and unified design framework for intelligent systems. The proposed event detection schemes will be realized in the presence of heterogeneity of data streams where each of them come with its own intensity (amount of data being generated), nature of data (numeric time series, images, video, textual information), and -granularity of underlying information. Another interesting and important design aspect to be dealt with in the project concerns the development of efficient ways of aggregation of event detectors.