Machine learning based quantity measurement method for grain silos

• Published in: Measurement : journal of the International Measurement Confederation 2020-02, Vol.152, p.107279, Article 107279
• Main Authors: Duysak, Huseyin, Yigit, Enes
• Format: Article
• Language: English
• Subjects: Algorithms; Continuous wave radar; Feature extraction; Fourier transforms; Grain boundaries; Grain quantity measurement; Grain silos; K-nearest neighbors algorithm; KNN; Machine learning; Radar level measurement; Waveform analysis
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2019.107279

•Wide beam-width EM illumination is used for level measurement in grain silos.•Scaling model is used for quantity measurement.•Radar backscattering data are gathered for various amounts of grain.•8 features are extracted from the radar backscattering data.•The amount of grain is detected by means of KNN algorithm. In this study, K-nearest neighbour (KNN) algorithm was used to determine grain quantity in the silo via thru the air radar. Thanks to the constituted stepped frequency continuous wave radar (SFCWR) system on a model silo, the back-scattering signal of different amount of grain for different stack structure were obtained. To create the training data, 5680 measurements were performed, and accuracy of the KNN was obtained with k-fold cross validation technique. Range profiles were obtained taking the inverse Fourier transform of SFCWR data and 8 features were extracted from range profiles. To found best feature combination on classification, 255 possible feature combinations were also constituted, then they were trained and tested. The best accuracy was achieved as 96,71% by using selected 5 features. These results show that, if the effective features are extracted from range profiles, the amount of the grain can be successfully determined using machine learning techniques.