Determination of foreign-material content in uncleaned peanuts by microwave measurements and machine learning techniques. Issue 1 (2nd January 2022)

• Record Type: Journal Article
• Authors: Julrat, Sakol; Trabelsi, Samir
• Is Part Of: Journal of microwave power and electromagnetic energy. Volume 56:Issue 1(2022)
• Page Start: 3
• Page End: 17
• Publication Date: 2022-01-02
• Subjects: Microwave sensing -- foreign material content -- dielectric properties -- bulk density -- uncleaned peanuts -- machine learning; Microwave heating -- Periodicals; Microwaves -- Periodicals; Electromagnetic waves -- Periodicals; Electromagnetics; Microwaves; Radiation; Micro-ondes -- Périodiques; Ondes électromagnétiques -- Périodiques; Electromagnetic waves; Microwave heating; Microwaves; Electronic journals; Periodicals; Periodicals; 537.05
• DOI: 10.1080/08327823.2021.1993047
• Languages: English
• ISSNs: 0832-7823

Abstract

Abstract: Foreign-material content determination in uncleaned peanuts based on dielectric properties and bulk density measurements by microwave techniques is presented in this paper. A microwave free-space transmission technique was used at 10 GHz. Two measurement systems for measuring the dielectric properties of cleaned unshelled peanuts (nine-peanut pods) and uncleaned unshelled peanuts placed in polycarbonate sample holder (12.1 cm × 21 cm × 20.5 cm) were developed and integrated in one single measuring unit. The nine-peanut-pods system provided the cleaned unshelled peanuts moisture content which was used in the algorithms for foreign material content determination. The dielectric properties and bulk density measurements of the uncleaned unshelled peanut sample were related to the foreign-material content. These parameters, namely bulk density and dielectric properties of uncleaned peanuts and cleaned unshelled moisture content were supplied to machine learning algorithms, linear regression technique and artificial neural network algorithms. Results obtained with the artificial neural network algorithm showed the best estimate of foreign material content with a standard error of performance of 1.36% compared to that obtained with the linear regression algorithm with a standard of performance of 2.39%.