Nondestructive Discrimination of Plant-Based Patty Containing Traditional Medicinal Roots Using Visible–Near-Infrared Hyperspectral Imaging and Machine Learning Techniques

Abstract

The interest in traditional meat being replaced by plant-based food has increased throughout the years. Some agricultural products, such as root crops, could be incorporated into alternative meat products due to the health benefits. However, relevant studies have discovered that some roots are considered allergen materials, necessitating further identification to maintain consumer safety. Aside from high accuracy, the limitations offered by traditional identification methods are a reason to employ nondestructive methods. This study aimed to develop a hyperspectral imaging system measuring the 400 nm to 1000 nm spectral range for the nondestructive identification of roots in soybean-based patty. Four thin-sliced traditional medicinal roots (tianma (Gastrodia elata), balloon flower root (Platycodon grandiflorum), deodeok (Codonopsis lanceolata), and ginseng (Panax ginseng)) were incorporated in a soybean-based patty with a concentration of 5% w/w. Moreover, support vector machine (SVM) learning and one-dimensional convolutional neural networks (1D-CNN) were realized for the discrimination model in tandem with spectral data extracted from the hyperspectral image. Our study demonstrated that SVM learning effectively discriminates between original patty and patty with root addition, with an F1-score, precision, and recall beyond 96.77%. This optimum model was achieved by using the standard normal variate (SNV) spectra.