Tomato is a climacteric fresh fruit composed of multiple tissues with diverse physical and biochemical compositions relevant to defining its quality through the food chain. The constitution of these tissues undergoes significant dynamic changes throughout the maturation process and in the post-harvest phase until the consumer such as the levels of antioxidants, lycopene, ascorbic acid, phenols and free radicals (Chandra and Ramalingam, 2011) and bioactive compound [e.g. flavonoids (Tamasi et al., 2019)].

The tomato fruit maturation process is marked by tissue specialisation, which promotes biochemical and physical changes in all tissues (Moco et al., 2007). During the ripening process, the tomato changes colour from green to red, resulting in morphological and biochemical modifications. The cultivar, environmental conditions (e.g., soil, light, temperature) and agronomic practices (e.g., irrigation, fertilisation) are essential factors that contribute to the tomato maturation process. For instance, in diverse regions, variations in antioxidants and phenols have been observed (Chandra et al., 2012). Under salinity conditions, morphological aspects such as size, water content, and colour undergo changes, impacting sugar content and acidity (Pascale et al., 2015). Additionally, genetic factors exhibit divergence even under identical conditions (Toor and Savage, 2005).

Several non-destructive techniques are available in the literature for characterising fruit maturation. However, none of these techniques can provide detailed information about the internal tissues of the fruit while also predicting its biophysical and biochemical characteristics efficiently. Although similar works approached the reconstruction of fruits using other techniques, such as electrical impedance to detect the tomato level of maturation (Verma et al., 2021), computed tomography in detect bitter pit in apples (Si and Sankaran, 2016) and X-ray in predicting the sugar content in kiwi fruit (Kanno and Kuroyama, 2020) and phenotyping charcteristics of seed in predicting the seed length, width, thickness, and radius of soybean and wheat with an accuracy ranging 80-96% (Liu et al., 2020), some of these techniques are expensive and not expedite for in situ measurements. Additionally, these methods are mainly used to identify physical damage, and none of them analysed each tissue individually or presented a qualitative approach (Donis-González et al., 2014). In biomedical science, non-destructive tissue characterisation (human and animal) has been developed through Vis-NIR spectroscopy, particularly for detecting tissue anomalies (Malone et al., 2014; Dahlstrand et al., 2019) and guiding the right incision during surgery (Vo-Dinh et al., 2010; Stelzle et al., 2012), which indicate that similar techniques can be applied to vegetation tissues.

Monitoring tomato precision maturation throughout the food chain is crucial for minimising losses within the food supply chain and improving pre- and post-harvest production and utilisation (Garcia and Barrett, 2006). From the high-tech horticulture point of view, monitoring maturation is essential to improve cultural practices, such as irrigation, fertilisation and canopy management, which are directly related to the pigments, organic acids and sugars in the tomato fruit (Bertin and Génard, 2018). In addition, to ensure that the seeds are fully developed for tomato seed production (Shrestha et al., 2016). They can guarantee that all tissues are well developed and that the gustative parameters favour the final consumer.

Traditional methods like chemical assays and chromatography used to characterise tomato biochemical parameters (e.g., lycopene) in different tissues are, in most cases, destructive, very expensive, non-adapted for small amount of tissues and time-consuming, which hinders the assessment of tomato quality parameters based on the composition of each tissue. Therefore, despite the high cost and time consumption of traditional analysis, alternative non-destructive techniques have been developed to assess tomato maturation and quality parameters, including colourimetric (Gómez et al., 2001), fluorescence (Wu and Wang, 2014; Konagaya et al., 2020), and Vis-NIR techniques (Torres et al., 2015; Zhu et al., 2015).

Vis-NIR devices have demonstrated outstanding potential for estimating fruit’s biochemical and biophysical parameters like lycopene and β-carotene (Tilahun et al., 2018), water potential (Tosin et al., 2021) and sugars and acids (Martins et al., 2022). Numerous studies on tomatoes have utilised Vis-NIR techniques to estimate various biochemical and biophysical parameters. These include soluble solid content (SSC) with reported R² of 0.87 (Ecarnot et al., 2013), 0.60-0.77 (Torres et al., 2015), and 0.88-0.98 (Ding et al., 2016). The pH levels were determined with a R² of 0.80 (Huang et al., 2018a), while colour attributes showed R² values ranging from 0.91-0.99 (Ecarnot et al., 2013). Vitamin C content was estimated with a R² of 0.67 (Azadshahraki et al., 2018), total acidity with R² of 0.66-0.94 (Ding et al., 2016) and R² of 0.91-0.98 (Najjar and Abu-Khalaf, 2021), and firmness exhibited R² of 0.70-0.72 (Najjar and Abu-Khalaf, 2021). The analysis of β-carotene revealed R² values of 0.77-0.88 (Tilahun et al., 2018), phenols showed R² values of 0.76-0.98 (Ding et al., 2016), malic acid with R² of 0.27-0.42 (Torres et al., 2015), citric acid with R² of 0.66-0.94 (Ding et al., 2016) and lycopene with R² values ranging from 0.45-0.75 (Clément et al., 2015), 0.73-0.83 (Ciaccheri et al., 2018) and 0.85-0.89 (Tilahun et al., 2018). However, tomato exhibits differences in the structural and biochemical characteristics of different tissues, which leads to significant ramifications in the absorption and scattering of light inside the tomato fruit (Skolik et al., 2019). These complex structures of tomato make it challenging to investigate the inner tissues through Vis-NIR spectroscopy and how they behave during maturation.

Spectral information about the inner tissues of fruits can be acquired using Vis-NIR data (Martins et al., 2023). Martins et al. (2022) demonstrated empirically that different grape tissues influence the whole fruit during the maturation process and that the concentration of pigments changes during maturation in various tissues. Obtaining spectral information of the internal tissues of fruits through Vis-NIR requires appropriate modelling techniques. Principal component analysis (PCA) is one such technique that uses latent variables (LV) to perform an orthogonal transformation of the original dataset onto a reduced subspace that is spanned by the principal components (Dahlstrand et al., 2019). Conversely, the combination of LV creates a superset, presenting a direct relationship with the original information. LV models can deal with significant and correlated variables (Trygg and Wold, 2003). Multi-block hierarchical PCA (HPCA) and hierarchical partial least squares (HPLS) are frequently used in chemometrics to deal with spectral information from different sensors and a batch of data (Mishra et al., 2021; Martins et al., 2023). Therefore, multi-block analysis can create a bi-directional reconstruction of whole tomato fruit from the skin, pulp, and seed spectral data.

Tissue reconstruction is based on hierarchical latent relationships between the spectral patterns of the observed tissues, providing details of the internal plant structures (Martins et al., 2023). This manuscript uses the term “tomography-like” to partly define the capacities of data-driven class reconstruction using hierarchical relationships. The practicality of bidirectionality involves connecting the principal latent space derived from tomato tissues to the entire tomato spectrum and executing the reverse process, which consists of breaking down the tomato fruit’s spectra into the spectra of its tissues, namely the skin, pulp, and seed. This methodology was recently applied to grapes, as demonstrated by Tosin et al. (2023), and it is expected to work in tomato, in which the tissue composition is different from grapes. The term tomography-like is debatable in that it only implies the resolution of the tissue image; here, it is aimed to provide a median spectrum of each tissue, given the fruit spectra in a non-destructive way. Once several spectra are taken from different positions on the fruit, a 3D resolved image can be reconstructed by relating the positions [x, y, z] and spectral gradients within internal tissues (Martins et al., 2022; Tosin et al., 2023). In this sense, this work can be considered ‘tomography-like’. Data-driven reconstruction does not use the same numerical approaches and solutions as the classical approaches but is entering several research areas due to their computational efficiency (Bar-Sinai et al., 2019; Martins et al., 2022).

Furthermore, spatially resolved tissue fruit composition is yet very complex to be obtained experimentally (Tosin et al., 2023). There are still very significant constraints at the level of analytical chemistry state-of-the-art on the quantity of sample that can be used to quantify parameters considered in today’s routine analysis, such as SSC and pigments. Therefore, metabolic or compositional imaging validation is still limited to the laboratory ground truth methods.

This research used a point-of-measurement (POM), where the light enters the fruit and has internal reflections, being only able to return to the spectrometer through a centre fibre optics pinhole, meaning that all light reaching the spectrometer interacts with the inner fruit’s tissues, maximising the spectral information on all internal tissues.

In multispectral or hyperspectral imaging, light is generally illuminated outside the fruit. This demonstrates that most light reaching the imaging sensor is reflected, non-absorbed, and carries little information about internal tissue composition.

Therefore, relationships with the recorded spectra using this method are generally limited to covariant information with pigmentation, which has the danger of quantifying through correlation and not due to the causal characteristic features present in the spectra. At present, it is believed that POM devices can be of more practical application to field studies than hyperspectral cameras.

Through multi-block analysis, this research facilitates the reconstruction of hyperspectral data by utilising information from individual spectra of tomato tissues, namely skin, pulp, and seed. This method enables the decomposition of the overall tomato spectrum into its constituent tissues, offering a bi-directional relationship. This study further provides a qualitative analysis of the tomato ripening process, elucidating the maturation levels in the skin, pulp, and seed at various developmental stages. By employing a POM sensing approach that maximises spectral information from internal tissues, the research addresses the limitations of traditional destructive methods, providing a non-destructive alternative for characterising tomato biochemical parameters. This contribution to the high-tech horticulture food supply chain can be used to ensure superior quality produce, reduce waste, enhance market value, and advance agricultural practices.

Therefore, three main objectives have been established in this work: i) to reconstruct the tomato hyperspectral data using information from the skin, pulp, and seed spectra through multi-block analysis; ii) to demonstrate that the spectral information of the entire tomato can be decomposed into the skin, pulp, and seed; and iii) to provide a qualitative analysis of the dynamics of the tomato ripening process, demonstrating the maturation levels in the skin, pulp, and seed, and how these tissues behave at different stages of the maturation process.

This paper presents a disruptive methodology for the bi-directional reconstruction of whole tomato hyperspectral data and internal tissues (skin, pulp, and seeds). This non-destructive method can explain how different tomato tissues behave at various stages of the ripening process. As described in Figures 1 – 3 , this work aims to bi-directionally reconstruct the spectral information of the tomato from the data of the skin, pulp, and seed and to decompose the information of the tomato spectra into the internal tissues. Although each tomato tissue presented a particular space ( Figure 3 ), creating a superset with the scores of each tomato fraction made it possible to reproduce the entire spectral tomato ( Figure 4 ). The most important LV of each fraction formed the superset used to reconstruct the tomato spectra. Through hierarchical PLS, the tissues could predict the entire tomato spectral data. The decomposition of the whole tomato data and the same number of LV combined in the hierarchical PLS can predict the tomato tissues. The literature reports that the tomography-like approach (Martins et al., 2023) successfully worked in grapes (Tosin et al., 2023), and the results of this paper support that it can be applied to aqueous fruits like tomato. Due to the complex nature of tomato maturation and the diverse biochemical compositions of its internal tissues (Skolik et al., 2019), encompassing skin, pulp, and seed, this work presents a technique for the bi-directional spectral reconstruction of tomatoes using Vis-NIR data.

The literature offers several methodologies (Mishra et al., 2021) that could be used for spectral reconstruction. For instance, O2-PLS (Trygg and Wold, 2003) and OnPLS (Lofstedt et al., 2013) utilise spectral data’s local and global joints, bioheat models (Alzahrani and Abbas, 2019; Marin et al., 2021) could be adapted to predict the internal tomato tissues and adaptive neuro-fuzzy inference system (Abdullahi et al., 2021; Abdullahi et al., 2022), a hybrid computational model that combines the adaptive capabilities of neural networks with the interpretability of a mathematical framework that deals with uncertainty and imprecision in decision-making. However, these methods are not hierarchical and do not allow for convolution and fusion of information in a superset or deconvolution in a reverse way. Similarly, advanced approaches such as deep learning (DL) can deal with complex data (Mishra et al., 2022) and reconstruct the whole tomato with spectral information from the different tissues. Nevertheless, the decomposition of the tomato spectra into the spectral data of its tissues becomes even more complex and challenging. Nonetheless, the results presented in Table 1 show that hierarchical PCA combined with PLS can effectively perform bi-directional modelling with the directions X − Y and Y − X , and remove orthogonal data between Y and X , facilitating the reconstruction of the tomato and its internal tissues.

Owing to the multiple internal tissues that compose the tomato fruit, it faced a challenge in reconstructing the complete spectral information of the tomato ( Figure 4 ). Furthermore, separating and identifying different tissues in green tomato are difficult because green tomato presents more fibre concentration (Chandra and Ramalingam, 2011) and pectin (Moco et al., 2007; Huang et al., 2018b). As a solution, all internal tissues (except the jelly parenchyma) were considered the pulp. Nevertheless, green tomatoes have less interest when compared with more advanced stages of maturation.

Spectral data acquisition encountered a few challenges related to the position of the tomato in the system used to obtain spectral data. First, green tomato is opaque, limiting light’s ability to pass through ( Figure 4 ). This limitation requires the optical fibre probe to be well-positioned to obtain more light signals. However, searching for light using the probe may not obtain a signal from all internal structures, which limits tissue reconstruction. A similar effect was observed in matured tomato. Depending on the probe position, some internal tissues may not be assessed, or less information may be obtained. Second, the tissues considered as the pulp will affect the reconstruction of the entire tomato.

During the data acquisition of the entire tomato, almost all the internal tissues are expected to be evaluated using a fibre probe. However, to provide detailed information in the superset utilised to reconstruct the entire tomato, it may be necessary to individually assess each tissue considered as pulp and examine their specific details. Therefore, the errors observed in Table 1 and Figure 4 indicate less accuracy in the reconstruction of the entire green tomato and less in the other tomato fractions of the matured tomato. Nevertheless, this method can reconstruct the Vis-NIR spectra of tomato and internal tissues and be used for different purposes.

This work assessed the SSC, chlorophyll, lycopene, and puncture force of the tomato using the original and reconstructed spectra for each fraction of the tomato ( Table 2 ). The SSC and puncture force were assessed in the entire tomato, and the spectral data of each fraction were considered in the modelling. Chlorophyll and lycopene provided spectral information for each tissue and empirically demonstrated the concentration of each pigment in the tissues. The original and reconstructed spectra results were very similar for reconstructing the entire tomato and the reconstructed spectra ( Table 2 ).

Among the quality parameters assessed, the SSC results were the most stable and robust for the original and reconstructed spectra. Nevertheless, it is essential to highlight that the different tomato tissues present distinct SSC concentrations, and the individual tissues could have been assessed to predict the concentration of SSC in each fraction. For chlorophyll and lycopene, the ratio of the zones of the spectrum empirically demonstrated the different concentrations of pigments in the distinct tissues ( Supplementary Figures 2–5 ). Considering all the tomato tissues, the skin is the part that is indicated to have more concentration of chlorophylls. This study reveals that the skin has the highest concentrations of chlorophyll and lycopene, except during the red maturation stage. These results align with existing literature, particularly Chandra et al. (2012), highlighting the skin’s elevated lycopene levels compared to pulp and seeds. However, in terms of chlorophyll content, the skin ranks as the second lowest tissue during maturation, as observed by Moco et al. (2007). The qualitative approach to classifying internal tomato tissues throughout maturation may contribute to these variations compared to the quantitative methods in the cited literature.

Consistent with prior research (Moco et al., 2007; Chandra et al., 2012), this paper reveals that the skin exhibits a higher concentration of lycopene, as shown in Supplementary Figure 3 . Puncture force analysis indicates elevated values in green tomatoes, probably attributable to heightened fibre and pectin content. This coincides with lower levels of SSC and lycopene, alongside increased chlorophyll concentrations (Moco et al., 2007; Huang et al., 2018b).

The temporal dynamics of maturation across distinct tomato tissues are demonstrated in Figure 5 , where chlorophyll concentrations, particularly in the skin, are higher in green tomatoes ( Figure 5A ). Conversely, matured tomatoes ( Figure 5C ) tend to exhibit increased lycopene concentrations, mainly in the skin. Considering the spectral signatures of the tomato fractions ( Figure 5 ), further studies can be conducted to determine the type of information that can be extracted. Empirically assessing the full tomato spectrum ( Figure 5 ), two bands peak near 500 nm and 690 nm in green tomato, probably related to chlorophyll (Ecarnot et al., 2013; Huang et al., 2018b). In the spectral skin signature, a similar peak (near 490 nm) in the green tomato may be related to chlorophyll a. When the pulp and seed were analysed, the same peak (near 490 nm) was present but with less intensity, suggesting a lower chlorophyll concentration in those tissues. The peaks near 550 nm are related to the carotene group (Ciaccheri et al., 2018), especially the lycopene concentration. For matured tomato, these picks present more intensity in the full tomato spectra, skin, and pulp, suggesting a higher concentration of lycopene when compared with less mature tomatoes.

Vis-NIR data can enhance the efficiency and quality of crop production by providing valuable information for optimising various agricultural practices, such as irrigation, fertilisation, and pruning (Xia et al., 2021). Furthermore, by leveraging this data, crop growers can gain precise insights into the maturation process of fruits, which is influenced by a range of biotic and abiotic factors, including diseases, water availability, temperature, and light intensity.

The methodology presented in this paper has the main advantage of providing more accurate and detailed information about the internal structure of the fruit as a tomography-like system when compared to the whole-fruit measurement by using hyperspectral or multispectral data (e.g., Mishra and Woltering, 2023; Mishra et al., 2023) that obtain majority external information. The tomography-like presented in this paper allows for the assessment of individual tissues, facilitating the acquisition of more accurate and detailed information about the internal structure of the fruit (Martins et al., 2023). Figure 3 represents the feature space of the entire tomato and the skin, pulp and seed, where the different maturation stages occupy distinct feature spaces. In contrast, traditional whole-fruit measurements often lack precision in providing insights into the inner tissue of the fruit. Also, the methodology presented in this paper can lead to more accurate predictions of the internal tissue properties and better quality control. It can also enable a more specific and targeted analysis of internal tissue properties by offering comprehensive and high-dimensional data. These data provide more detailed information about the fruit’s internal structure and can also be utilised to determine additional components beyond those presented, particularly in supporting metabolomic studies. This knowledge can help to fine-tune agricultural practices and mitigate potential risks, ultimately leading to improved crop yields and higher-quality produce.

This paper presents a novel technique for determining the quality parameters SSC (%), chlorophyll (a.u.), lycopene (a.u.) and puncture force (N) of fruits using visible and near-infrared (Vis-NIR) spectroscopy of their skin, pulp, and seed. The approach builds upon a growing body of research demonstrating the ability of hyperspectral sensors to measure a wide range of quality parameters non-destructively and accurately in crops (Martins et al., 2022; Tosin et al., 2022; Tosin et al., 2023). Furthermore, by leveraging the high-dimensional data obtained for each tissue, the method showcased in this study has the potential to unlock a multitude of additional quality parameters during fruit maturation. This capacity for rapid and precise determination could enhance fruit production’s efficiency and effectiveness while elevating the final product’s overall quality. Finally, it is worth noting that the extensive dimensionality of the data obtained for each tissue opens possibilities for identifying and characterising other components beyond those currently presented, particularly in supporting metabolomic studies.

This paper proposes a tomography-like system that can predict the Vis-NIR information of the internal tissue. Applying multi-block hierarchical component analysis in conjunction with PLS enables the bi-directional reconstruction of spectral information, facilitating the prediction of internal tissue spectra (skin, pulp, and seed) and the decomposition of the overall tomato spectral information into its constituent tissues.

This novel approach allows assessing tomato maturation dynamics by analysing internal tissue characteristics, offering pertinent information for precision agricultural practices. Moreover, the method can identify physiological issues related to abiotic (e.g., water stress, high temperature) and biotic (e.g., bacterial infection). These identified stressors can be integrated into multifaceted omics techniques to understand the plant’s physiological responses.

Building on successful testing in grapes this technique, demonstrates its efficacy in the complex tissue structure of tomato. Thus, the same approach could be applied to other aqueous fruits, such as blueberries. However, further work is necessary to test the applicability of this technique in other fruits, to study the dynamic of the Vis-NIR information with the internal tissues during the maturation process, and to incorporate additional analytical data for validation.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

RT: Investigation, Methodology, Writing – original draft, Writing – review & editing. MC: Investigation, Methodology, Supervision, Writing – original draft, Writing – review & editing. FM-S: Methodology, Writing – review & editing. FS: Investigation, Writing – review & editing. TB: Methodology, Writing – review & editing. RM: Investigation, Methodology, Supervision, Writing – original draft, Writing – review & editing.