SPRAY DROPLET CHARACTERIZATION USING A PIEZOELECTRIC SENSOR THROUGH CLASSIFICATION BASED ON MACHINE LEARNING
تعیین ویژگیهای پاشش نازل با بکارگیری سنسور پیزوالکتریک از طریق طبقهبندی مبتنی بریادگیری ماشین
DOI : https://doi.org/10.35633/inmateh-59-17
Authors
(*) Corresponding authors:
Abstract
Nowadays, the indiscriminate use of pesticides for plant protection, has led to severe environmental pollution. This input accounts for a portion of the agricultural economy and should be sprayed in a way that has the highest biological efficacy and the least run-off. Therefore, real-time evaluation of spray characteristics and its classification is necessary. In the current research, a piezoelectric sensor was employed for detection of vibration signals from impaction of droplets to the active surface of the sensor. To supervised classification, the Support Vector Machine classifier as a Machine Learning model was implemented by means of extracted features from conditioned signals. By using a feature selection algorithm, six features were selected comprising mean, median, mode of signal peaks, root mean square, mean deviation and impulse factor of the signals. These features used as Support Vector Machine inputs. Model targets were spray droplet characteristics that were determined using image processing techniques on water sensitive papers. The results showed that the Linear and medium Gaussian models have the highest overall accuracy. Linear Support Vector Machine has higher accuracy and precision for training data (94.60% and 94.63%) and its model was able to predict with 92.59% accuracy. Precision of classifier model was higher than 92% for all classes. The highest miss rate of the model was approximately 15% in the separation of class C. Accurate and precise performance of linear classifier was confirmed by determining the Kappa coefficient of 0.77.
Abstract in Arabic