An Automatically Learning and Discovering Human Fishing Behaviors Scheme for CPSCN

The cyber-physical-social (CPS) computing and networking is a human centric and holistic computing framework which needs to convert the low-level data of physical, cyber, and social worlds into higher level information which can provide insights and help humans make complex decisions. Here, we focus on human fishing behavior recognition for vessel monitoring systems (VMS), an application of CPS. And the recognition of fishing behavior is the key task for studying human fishing activities, monitoring illegal fishing, and protecting fishery resources. However, VMS data basically consist of sequentially recorded position information and do not directly indicate whether a fisherman is fishing or not; thus, converting these low-level CPS data into intuitive information to humans is the primary task. In this paper, an identification model based on multi-step clustering algorithm (MSC-FBI) is proposed to automatically learn and discover fishing behaviors at sea. First, a temporal-spatial distance model is established; then, an improved multi-step clustering algorithm is used to identify human fishing behaviors, and finally, the patterns of different behaviors are extracted from the trajectory, and the unsupervised behavior learning model is established. Using this method, many experiments on different fishing trajectory data were implemented and compared with a traditional identification method based on the Gaussian mixture model (GMM-FBI). The experimental results illustrate the proposed model’s superior performance.