Wavelet-based cooking phase detection for time series of pot temperature

Reviewing cooking activities in a restaurant kitchen is useful for improving staff skills. This paper proposes a method for detecting the cooking phase in a time series of pot temperatures. The proposed method is based on a feature vector extracted by a wavelet transformation. A support vector machine (SVM) uses a feature vector that classifies every second as being in a "cooking" or "not-cooking" phase. Additional post processing consisting of erosion and dilation is carried out to reduce noise. An experiment with an actual dataset to evaluate the accuracy of the proposed method was conducted. The rate for correct classifications was 85%.