**1. Introduction**

The textural traits of fish, including gumminess, springiness, cohesiveness, resilience, hardness, brittleness, adhesiveness, and chewiness, are the most important traits in the aquaculture industry, and they affect the production process and the commercial value of fish [1–3]. Developing the fillet textual assessment method is beneficial for measuring the textual traits of fish-processed products [4]. Traditional fish textual assessment methods include measurements using a texture analyzer [5,6]. However, these methods are laborious and might destroy the integrity of the products. Therefore, there is an immediate requirement to construct an efficient and non-destructive method to detect the muscle texture of processed fish.

**Citation:** Cao, Y.-M.; Zhang, Y.; Yu, S.-T.; Wang, K.-K.; Chen, Y.-J.; Xu, Z.-M.; Ma, Z.-Y.; Chen, H.-L.; Wang, Q.; Zhao, R.; et al. Rapid and Non-Invasive Assessment of Texture Profile Analysis of Common Carp (*Cyprinus carpio* L.) Using Hyperspectral Imaging and Machine Learning. *Foods* **2023**, *12*, 3154. https://doi.org/10.3390/ foods12173154

Academic Editors: Zhiming Guo, Zhao Zhang and Dong Hu

Received: 24 July 2023 Revised: 11 August 2023 Accepted: 14 August 2023 Published: 22 August 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Recently, hyperspectral imaging (HSI) has become an alternative analytical approach that provides the benefits of rapid and non-destructive detection [4,7–11]. HSI combines image and spectral techniques to obtain both "spatial" and "spectral" information containing the sample [9–12]. Another feature of HSI is the ability to generate visual distribution maps of measured indicators to allow for the prediction and quantification of the composition of the sample and to determine their position on the sample surface [6,13]. Moreover, artificial intelligence and machine learning (ML) models can be used for prediction and modeling in the food industry [14]. With spectral images, HSI has been widely applied to evaluate the traits of meat products, including color, surface defects, damage, texture, water-holding capacity, flavor, freshness, and ripeness [4,15–21]. Ma et al. used HSI based on 400–1000 nm wavelengths to predict the different textural parameters of grass carp fillets during vacuum freeze-drying [4]. They predicted the Warner–Bratzler shear force, hardness, gumminess, and chewiness of fillets with prediction coefficients ranging from 0.79 to 0.87. ElMasry et al. predicted beef tenderness using hyperspectral imaging with a model based on partial least squares (PLS), showing a detection coefficient of 0.83 and a cross-validation narrative of 0.75 [17]. Zhou et al. predicted six texture parameters of silver carp muscle using HSI and ML methods, with coefficients ranging from 0.83 to 0.95 [8]. In addition, He et al. found that the SPA-LS-SVM prediction model and HSI had a prediction coefficient of 0.905 for the tenderness of salmon fillets [22]. These studies demonstrate that HSI and ML methods provide reliable solutions to measure processed fish textures.

In fish breeding, high-quality textures can provide fillets that are suitable for downstream processing and satisfy the consumer's taste. The traditional textual method requires the cut of fish muscle and is lethiferous [5,6]. Compared with the traditional textual method, HSI and ML methods have a non-destructive advantage, as they allow for the detection of the texture of live fish muscle. In the current literature, the majority of researchers have investigated the quality of fillets rather than intact fish using HSI, meaning that the spectra were usually obtained from the meat mass [16,23,24]. However, the application of HSI and ML methods to measure the live fish muscle has been less studied.

Common carp (*Cyprinus carpio*), an allotetraploid fish [25], is one of the most important freshwater-farmed fish in the world. Therefore, the aim of this study was to develop a noninvasive method in which skin HSI and ML are combined to detect the textual parameters of live fish muscle. We first acquired the skin HSI data of 387 scaled and live common carp with a hyperspectral imaging system at 400–1000 nm. Then, we measured the texture profiles of four corresponding muscle regions of each fish. The specific objectives of this study were to (1) utilize preprocessing methods to achieve spectral preprocessing and characteristic wavelength selection; (2) determine the optimal wavelengths that are most useful for the prediction of texture profile analysis (TPA) within the muscle of common carp; (3) determine the optimal relationship between the skin HSI data and muscle texture parameters using six ML methods and incorporate the skin hyperspectral index; and (4) apply the optimal model for the visualization of the distribution of muscle texture parameters.
