Anomaly Detection in Real-time Continuous Fruit-based Monitoring of Olive via Extensimeter

In this study, we analyze the real-time measurements collected by extensimeter (fruit gauge) in olive orchards to identify their anomalies. The field data are collected by two different kinds of extensimeters (strain gauges and variable linear resistance transducer) with hourly temporal resolution and a time span of 3.5 months in 2019, 3 months in 2020, and 2.5 months in 2021. To recognize the outliers in the sensor records, conventional statistical approaches including sliding window techniques such as Moving Mean Absolute Deviation, Median Absolute Deviation as well as one innovative method that integrates the sliding window technique with Moving T-square (SWT-T-square) are implemented. The performance of the mentioned approaches is evaluated using well-known statistical indices such as the Confusion Matrix, accuracy, sensitivity, and specificity criterion. To visually compare the models’ performance, the results of the methods are represented using plots that represent the number of outliers against window size. The results prove that the SWT-T-square integrated model outperforms others in recognition of the outliers. It is useful for acquiring more robust data and identification of sensor non-functionality or low accuracy during continuous monitoring.