Brain Computer Interface (BCI) has been an emerging topic in recent years. Specially, Artificial Intelligence (AI) is becoming a hot research area in recent years. However, many BCI techniques utilize invasive interfaces to brains (animal or human), which could cause potential risks for experimental subjects. EEG (Electroencephalography) technique has been used extensively as a non-invasive BCI solution for brain activity study. Many psychological work has suggested that human brains can generate some recognizable EEG signals associated with some specific activities. This paper suggests a novel EEG recognition method, i.e. Segmented EEG Graph using PLA (SEGPA), that incorporates improved Piecewise Linear Approximation (PLA) algorithm and EEG-based weighted network for EEG pattern recognition, which can be used for machinery control. The improved PLA algorithm and EEG-based weighted network technique incorporates the data sampling and segmentation method. This research proposes a potentially efficient method for recognizing human's brain activities that can be used for machinery or robot control.