Identifying the Semantics of Collective Motion by Proximity Sensors and Non-negative Matrix Factorization

This paper presents a monitoring system to retrieve locomotion semantics based on a minimum configuration of devices. We designed a mechanism to encode the binary signals of few proximity sensors into semantic words that store locomotion information. Then, we fed these words to a non-negative matrix factorization (NMF) algorithm and derived basis vectors of clusters for the usage of online semantic retrieval. Through examining and analyzing the proposed system in a simulated fish pond, we demonstrated this simple system reached a significant level of accuracy and consistency on the categorization of locomotion semantics. The findings suggest that a real adaptive system based on the proposed framework would be feasible and useful in industrial fields. Thus, livestock farms, for example, can benefit from its effective performance with low production cost.