Tropical fruits and vegetables are agricultural crops that are typically grown in tropical regions where the climate is warm, with temperatures ranging from 20 to 35⁰C (Bahadur et al., 2020). Tropical regions are found amidst the tropics of Cancer and Capricorn, and encompass equatorial zones in Oceania, Asia, Africa, Central and South America, and the Caribbean (Zakaria, 2023). Crops grown naturally in such weather conditions provide essential minerals, water, fiber, and vitamins that contribute significantly to the well-being of humans by safeguarding against ailments such as diabetes, hypertension, and cancer (Emelike and Akusu, 2019).

The agricultural revolution and the adaptation of numerous tropical plants to regions outside of their natural range have muddied their classification, and little is known about what properly defines and distinguishes tropical fruits and vegetables from their temperate counterparts (Indiarto, 2020). Fernandes et al. (Fernandes et al., 2011) described crop classification according to size, acidity, seed type, and bearing. Included among alkaline crops are apples, bananas, peaches, cherries, persimmon, and litchi (Fernandes et al., 2011). Acidic crops include strawberry, orange, kiwi, pineapple, lemon, star fruit, and logan, whereas sub-acidic examples are mango, pear, blackberry, papaya, blueberry, cherimoya, and mulberry (Fernandes et al., 2011). Chakraborty et al. (Chakraborty et al., 2014) agreed and structured the classification of tropical fruits based on that of Fernandes. Sarkar et al. (Sarkar et al., 2018) reported classification system according to maturity stage by means of ethylene gas emission and respiration rate, including both climacteric and non-climacteric tropical produce (Sarkar et al., 2018). Tropical climacteric produce such as avocado, apple, pear, mango, papaya, broccoli, banana, kiwi, and tomato undergoes maturation in correlation with an escalation in their respiration rate and the release of ethylene gas (Indiarto, 2020), whereas tropical non-climacteric crops such as grape, berry, citrus, litchi, strawberry, raspberry, pumpkin, watermelon, cucumber, and pineapple do not undergo an elevation in their respiration rate as they reach maturity (Indiarto, 2020). The contrasting report of Retamales et al. (Retamales, 2011) centers around the production of temperate crops worldwide. In this report, apple, raspberry, pear, peach, kiwi, blueberry, strawberry and plum were considered as temperate fruits (Retamales, 2011). In addition, Benichou et al. (Benichou et al., 2018) have also classified temperate fruits as tree (apple, plum, pear and peach), vine (grape and kiwi), and small fruits such as raspberry, blueberry and currant (Benichou et al., 2018).

Papaya, pineapple, avocado, mango, and banana are considered to be major tropical fruits globally (Mukhametzyanov et al., 2022). According to a market review prediction for the years 2013 to 2022 by the Food and Agriculture Organization of the United Nations (FAO), the most exported tropical fruits globally from Central America and the Caribbean, South America and Asia, Africa, and others in millions of tons were papaya, pineapple, avocado and mango with 3.7, 3.2, 2.3, and 2.1, respectively (Altendorf, 2019). On the other hand, recent data have shown that global vegetable production increased by 68% between 2000 and 2021 (FAO, 2022). Because of the continuous and emergent demand for tropical fruits and vegetables worldwide, the present emphasis is on quality assurance in relation to end-user inclinations and commercial standards (Silva and Abud, 2017). The quality of tropical fruits and vegetables is characterized by both external and internal parameters (Jha and Matsuoka, 2000). External parameters namely color, defects, size and shape depend on not only the appearance of the product, but also on the standards set (Cubero et al., 2016), whereas internal parameters such as nutritional value, internal defects, flavor, and texture are subjective to physicochemical composition and climate change (Zainalabidin et al., 2019). The quality of fruits and vegetables influences consumer preference and is directly or indirectly linked with further value-addition and processing technologies (James et al., 2010).

Several studies have identified postharvest losses as the most prominent factor among the origins of crop quality deterioration (Porat et al., 2018; Etana, 2019; Ahmad et al., 2021). Adding to that, high temperature and relative humidity are mentioned in the biological and chemical degradation of produce freshness, which affects sweetness, flavor, weight, turgor, and nutritional value (Elik et al., 2019). However, past reports indicated that low-temperature cooling systems and edible coating materials can be used to maintain and monitor the quality of these crops (Mendy et al., 2019; Jodhani and Nataraj, 2021). Conventional methods relying on the quantification of different quality traits such as dry matter content, oil content, and moisture content have also been reported in the study of quality parameters of fruits and vegetables; however, these methods were found to be undesirable, destructive, time-consuming, and labor-intensive (Magwaza and Tesfay, 2015; Kyriacou and Rouphael, 2018). Therefore, the application of non-destructive bio-sensing methods as a promising alternative for evaluating the value of tropical produce has been adopted (Ndlovu et al., 2022; Okere et al., 2022).

Computer vision and popular pre-trained convolutional neural network (CNN) models have been used as recognition systems to sort and grade different fruits and vegetables, especially in supermarkets, regarding their variety and species (Dubey and Jalal, 2012). However, computer vision can only assess external quality attributes due to the lack of spectral information (Rahman and Cho, 2016; Bhargava and Bansal, 2021). Acoustic emission technology involves the mechanical destruction of produce when subjected to mechanical or thermal stimulus (Aboonajmi et al., 2015) and is not appropriate for all categories of fruits and vegetables (Adedeji et al, 2020 ). Extensive works have been published on the evaluation of fruits and vegetables by spectral measurements such as Fourier transform infrared (FTIR) spectroscopy (Egidio et al., 2009), Near-infrared (NIR), Raman spectroscopy (Pandiselvam et al., 2022), and hyperspectral imaging (HSI) (Wang and Zhai, 2018). Generally, these reports have concentrated on the utilization of spectral measurements for determining targeted quality parameters of a particular fruit or vegetable variety. For instance, visible and near-infrared spectroscopy was used to investigate the internal browning in mango fruits (Gabriëls et al., 2020). Ali et al. (Ali et al., 2023) investigated FTIR, NIR, and machine vision in the quality monitoring of pineapples. Metlenkin et al. (Metlenkin et al., 2022) distinguished Hass avocado fruits by defects using hyperspectral imaging (HSI). The question revolves around the practical utilization of these approaches and the challenges associated with improving data processing speed and in-line implementation (Cortés et al., 2019; Si et al., 2022). Quick hardware and software are required to fulfill the demands of swift analysis for extensive hyperspectral datasets (Xu et al., 2023) and machine learning algorithms, especially those relying on deep learning act as black boxes rather than using interpretability models for high-stakes decisions (Caceres-Hernandez et al., 2023).

The present review highlights the current advances in non-destructive spectral measurements for quality assessment, specifically for major tropical fruits and vegetables. The quality parameters of these tropical produces are covered first. The discussion on each of the spectral measurements, the tropical crops used, and the specific findings obtained from various studies, which are summarized in Table 1 , follows and can deliver valuable information on the capabilities and efficiency of these techniques. In addition, the merits and demerits of each of these spectral measurements, which are presented in Table 2 , will guide future researchers in selecting the proper evaluation method when evaluating the quality of tropical produces. To facilitate comprehension and quick understanding of key terminologies involved, the list of abbreviations and definitions contained in the paper is presented in Table 3 .

Quality inspection is the process of evaluating specific parameters of fruits and vegetables to ensure required quality standards (Phey et al., 2020). The intention of quality inspection is to detect any internal or external characteristics that can aid in identifying both standard quality parameters and defects or non-conformities that can affect the safety of fruits and vegetables or their usability in particular functions such as diets, trade, and industrial chains (Kirezieva et al., 2013).

Non-destructive techniques for quality monitoring of tropical fruits and vegetables refer to the process of inspecting their external and internal properties without causing damage or changing their physical and internal status (El-Mesery et al., 2019). The potential for employing spectral measurement approaches in the quality control of fruits and vegetables is growing enormously (Escárate et al., 2022). The reason is that these approaches are non-destructive, fast and accurate, capable for both quantitative and qualitative analysis, thereby requiring minimal sample preparation (Cozzolino, 2022). We divided non-destructive spectral measurements into two categories: (1) spectral-based approaches (FTIR, NIR, and Raman spectroscopy) and (2) imaging-based approaches (HSI), as shown in Figure 1 .

The rapid advancement of technology in the agricultural field has resulted in the combination of artificial intelligence with non-destructive spectral measurements for fruits and vegetables quality measurement (Hasanzadeh et al., 2022). Artificial intelligence models such as artificial neural networks (ANNs), genetic algorithms (GAs), fuzzy logic (FL), and adaptive neuro-fuzzy inference system (ANFIS) can assess multiple characteristics simultaneously (Homayoonfal et al., 2022). Salehi reviewed development of models used in the determination of fruits and vegetables quality (Salehi, 2020). ANNs, GAs, FL, and ANFIS detected defects, moisture content, and chilling injury of oranges, cherries, pomegranates, apples, peaches, avocados, button mushrooms, tomatoes, and potatoes (Salehi, 2020). Despite the fact that these models are typically constrained by normality, linearity, homogeneity, and variable independence, the ANFIS model outperforms others and can be successfully used in relevant research (Salehi, 2020).

Machine learning (ML) is a branch of artificial intelligence and an integral part of the development of many sensing technologies that are responsible for information retrieval, signal processing, and data analysis (Li et al., 2021). In recent decades, traditional algorithms such as linear discriminant analysis (LDA), support vector machines (SVMs), K-nearest neighbors (K-NN), naïve Bayes, extreme learning machines (ELMs), decision trees (DTs), and K-means clustering have been deployed (Fadchar and Dela Cruz, 2020). For instance, Rivera et al. (Rivera et al., 2014) used NIR–HSI and machine learning for the early detection of mechanical damage in mango. LDA, K-NN, naïve Bayes, ELMs, and DTs were used for categorization. Bayes failed, however (K-NN, ELM, DT, and LDA Title altered) results was more than 90%. The highest performance, achieved by K-NN, was 97.9% (Rivera et al., 2014).

The evolution of deep learning (DL) as a breakthrough machine learning method has been trending since 2017 due to the manual feature extraction of traditional machine learning methods (Yang and Xu, 2021) and limited performance of chemometrics models, such as spectral variability caused by sample and spectrometer heterogeneity, changing environmental conditions, and infrared spectral data with high noise, which hinder feature extraction using chemometrics models (Zhang et al., 2021). Deep learning is a subset of machine learning that use many neural network layers to extract complex feature representations with numerous levels of abstraction (Lecun et al., 2015). According to Kamilaris et al. (Kamilaris and Prenafeta-Boldú, 2018), convolutional neural network (CNN) and recurrent neural network (RNN) have been implemented for crop-type classification, counting produces, and locating their placement in the image using bounding boxes (Kamilaris and Prenafeta-Boldú, 2018). However, the RNN was found to perform better than the CNN because it considers not only space but also the time which helps to capture the time dimension (Kamilaris and Prenafeta-Boldú, 2018). Deep learning and machine learning technology-based spectral analysis has been used in the classification of three types of fruits (apple, lemon, and mango) by type of damage, type of goods, and whether the sample is raw in market, supermarket, wholesaler, and retailer applications (Bobde et al., 2021).

Garillos-Manliguez et al. (Garillos-Manliguez and Chiang, 2021) estimated six maturity stages of papaya fruit, from the unripe stage to the overripe stage, by feature concatenation of data obtained from visible light and HSI imaging (Garillos-Manliguez and Chiang, 2021). AlexNet, VGG16, VGG19, ResNet50, ResNeXt50, MobileNet, and MobileNetV2 architectures was then modified to apply multimodal data cubes made of RGB and hyperspectral data (Garillos-Manliguez and Chiang, 2021). Regarding classification of the six stages, these multimodal variations can reach F1 scores of up to 0.90 and a 1.45% top-2 error rate. However, due to the small size of the images and the great depth of the CNNs, resulting in highly tightly tuned training variables, overfitting may arise. On the other hand, increasing image size results in insufficient memory faults (Garillos-Manliguez and Chiang, 2021).

Banana fruit was graded by Mesa et al. (Mesa and Chiang, 2021) using multi-input deep learning model with RGB and HSI. These models were able to categorize tier-based bananas by 98.45% and an F1 score of 0.97 with only few samples (Mesa and Chiang, 2021). However, this technique is expensive and time consuming due to the use of two cameras. The next studies instead, should consider the use of more improved camera systems with features that can extract both RGB and HSI simultaneously (Mesa and Chiang, 2021). Another study by Ucat and Cruz explored the use of image processing with a deep learning to grade banana according to their specifications (Ucat and Dela Cruz, 2019). The trained, validated, and test data by CNN model was more than 90% in all four classes of bananas (). The suggested CNN grading system in the tensor flow model can be commercially developed (Ucat and Dela Cruz, 2019).

Portable spectrometers and real-time online detection devices have recently developed for fruits and vegetables quality assessment. Portable devices are handheld, light weight, compact size and they are applied for in-field measurements (Sohaib et al., 2020). The combination of portable NIR device with MSC-PCA+LDA model was used to evaluate pineapple quality. These models were recommended to be developed in mobile phone while PLS regression model provided 85% accuracy (Amuah et al., 2019). Subedi et al. (Subedi and Walsh, 2020) evaluated three hand held portable near infrared spectroscopy (F750, Micro NIR and Scio v1.2) in the detection of dry matter content (DMC) in avocado fruit. The second derivative spectra were recorded for the intact and skin removed avocado fruit for reflectance and interactance optical geometry. The best results of prediction obtained from the F750 instrument using the interactance mode at 720-975 nm with R²p of 0.71 and 0.88 for intact and skin removed fruits respectively (Subedi and Walsh, 2020). Real time monitoring device was designed as sensor which can function in all post-harvesting states to control the shelf life of fruits and vegetables such as lettuce. The device found to be the feasible for controlling the behavior of the crop during the post handling chain (Torres-Sánchez et al., 2020). Fruits and vegetables including banana, orange and apple were well sorted according to their external appearance by using real time online system with artificial intelligence (Tata et al., 2022). For quality categorization, machine learning models such as CNN and image processing were performed. This real time system was created in android and can be deployed in market robots where checking of huge number of products is required (Tata et al., 2022).

Non-destructive spectral measurement has emerged as a prominent solution in the agricultural sector. With the introduction of spectral measurements, there has been rapid progress in analyzing both the internal and external characteristics of tropical fruits and vegetables in a low-cost, accurate, real-time, and fast manner (Ali et al., 2017). Techniques based on FTIR, NIR, and Raman spectroscopy require simple steps to prepare samples prior to analysis (Abbas et al., 2020). In contrast to other imaging techniques such as computer vision, acoustic approaches, electric noses, and fluorescence, HSI uses spectral and spatial data to assess different parameters concurrently (Lu et al., 2020). The spectral measurements presented in this review have shown potential applications for a diverse range of tropical fruits and vegetables for the monitoring and detection of quality attributes such as SSC, TSS, TA, color, size, defects, and texture, which is particularly important for fruit and vegetable processors, food safety agencies, and consumer demands.

Significant advancements in non-destructive spectral measurement technology have occurred recently, including the development of portable spectrometers for real-time and field applications. The combination of spectral measurements and chemometric techniques is a powerful tool for multivariate data analysis, mainly in the improvement of models needed for classification and estimation of quality. A practical case study of Metlenkin et al. (Metlenkin et al., 2022) in the identification and classification of Hass avocado defects before and after storage by HSI and chemometrics. The PLSDA and SIMCA were selected as chemometric methods for multivariate data discrimination and classification. To increase the final model accuracy the calibration was performed by selecting the region of interest. The results revealed the high potential of SIMCA during both modelling and test validation with 100% accuracy. Furthermore, the integration of spectral measurements with deep learning and machine learning technology is rapidly expanding in order to improve quality control accuracy while overcoming the challenges associated with chemometrics such as spectral variability, spectrometer heterogeneity, changing environmental conditions, and infrared spectral data with high noise. The revolution in agriculture and the adaptation of numerous tropical plants to regions outside of their natural range have muddied their classification, and little is known about what properly defines and distinguishes tropical fruits and vegetables from their temperate counterparts. Therefore, there is confusion associated with those studies that reported the classification of tropical fruits and vegetables as an important factor to consider when examining the distinctive quality indicators of these crops. Taking into accounts all of the merits and demerits of non-destructive spectral measurements for the quality monitoring of tropical fruits and vegetables, the use of an adequate number of samples, different cultivars of the fruit and increasing the quality attributes to predict can help to develop robust models that emphasize the variability of tropical fruits and vegetables in terms of size and shape, skin thickness, and growing conditions.

Conceptualization: UA, B-KC. Methodology: UA, TB, MF, MK and IB. Investigation: UA, TB and B-KC. Writing and reviewing: UA, TB, MF and B-KC. Supervision: B-KC. All authors contributed to the article and approved the submitted version.