Audu J, Dinrifo R R, Adegbenjo A, Anyebe S P, Alonge A F
Department of Agricultural and Environmental Engineering, Federal University of Agriculture Makurdi, Nigeria.
Department of Agricultural Engineering, Lagos State Polytechnic Ikorodu, Nigeria.
Heliyon. 2023 Mar 16;9(3):e14567. doi: 10.1016/j.heliyon.2023.e14567. eCollection 2023 Mar.
Quality detection has been a major problem in the agriculture and food industries. This operation is mostly done by a subjective sensory method which is prone to high error and food destruction. Therefore, there is a need to apply artificial intelligence using a machine learning approach. This study developed two intelligent acoustic yam quality detection and classification devices using two sound-generating techniques. The software (multi-wave frequency generator) sound-generating technique generated sound from a laptop to a speaker inside a detecting chamber. This sound passes through the yam and was received on the opposite side by a microphone, into another laptop for analysis using visual analyzer software. The impact sound-generating technique used sound generated from a gentle impact of the yam on a flat surface placed inside the detection chamber. The sound produced was picked up by a microphone into a laptop for analysis. Acoustic properties considered were amplitude, frequency, sound velocity, wavelength, period and sound intensity. Discriminant analysis algorithm only was used in this first stage of the study to prove the applicability of machine learning. Three qualities (good, diseased damaged and insect-damaged) of two yam varieties (white and yellow yam) were tested. The device's performance of white yam was 79% and 68.7%, yellow yam was 82.3% and 68.7% for the software sound generation-technique and surface impact sound-generating technique, respectively. The study shows that the software sound-generating technique performed better in terms of overall yam quality detection and also proves the applicability of machine learning.
质量检测一直是农业和食品行业的一个主要问题。这项操作大多通过主观感官方法进行，这种方法容易出现高误差且会造成食品损坏。因此，有必要采用机器学习方法应用人工智能。本研究使用两种发声技术开发了两种智能声学山药质量检测与分类装置。软件（多波频率发生器）发声技术通过笔记本电脑向检测室内的扬声器发出声音。该声音穿过山药，在对面被麦克风接收，再传入另一台笔记本电脑，使用视觉分析软件进行分析。冲击发声技术利用山药轻击检测室内放置的平面产生的声音。产生的声音由麦克风采集到笔记本电脑中进行分析。考虑的声学特性有振幅、频率、声速、波长、周期和声强。在本研究的第一阶段仅使用判别分析算法来证明机器学习的适用性。对两个山药品种（白山药和黄山药）的三种质量（好、病害损伤和虫害损伤）进行了测试。对于软件发声技术和表面冲击发声技术，白山药的装置性能分别为79%和68.7%，黄山药分别为82.3%和68.7%。研究表明，软件发声技术在整体山药质量检测方面表现更好，也证明了机器学习的适用性。
