Method and device for predicting aroma quality of ripe fruit peel based on leaf volatile organic compounds

Abstract

This application discloses a method and apparatus for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds (VOCs), relating to the fields of fruit tree quality evaluation and artificial intelligence prediction technology. The method includes acquiring VOC abundance data from the leaves of the plant to be tested; inputting the leaf VOC abundance data into a cross-organ prediction model to output a predicted spectrum of VOCs in the mature fruit peel, and performing detection judgment and scale calibration on the model; calculating the weighted similarity with the target aroma type anchor point as the original aroma index; performing false positive penalty calibration to obtain a risk-calibrated aroma index; outputting an extrapolated risk warning based on the applicable domain score; mapping to the odor descriptor space to obtain odor semantic results; and ranking and outputting the evaluation results based on the risk-calibrated aroma index. This application can reduce the dependence on endpoint detection of mature fruit and the destructiveness of sampling, improving the efficiency of breeding material screening and germplasm resource evaluation.

Description

Technical Field

[0001] This application relates to the field of fruit tree quality evaluation and artificial intelligence prediction technology, and in particular to a method and apparatus for predicting the aroma quality of mature fruit peel based on volatile organic compounds in leaves. Background Technology

[0002] Fruit peel aroma is a crucial indicator of a fruit's sensory quality and commercial value, directly impacting its acceptability for fresh consumption, processing suitability, and market positioning. Taking citrus as an example, citrus aroma is composed of various volatile organic compounds, including terpenes, aldehydes, alcohols, esters, and ketones, with their composition and abundance varying significantly depending on the variety and developmental stage. Accurate evaluation of the aroma quality of mature fruit peel is of significant practical importance in crop breeding, germplasm resource evaluation, and the selection of superior varieties.

[0003] Leaves are easily accessible vegetative organs of fruit trees or other horticultural crops, rich in structured volatile organic compound information. Compared to the peel, leaves have a longer sampling window, allowing for multi-point sampling before fruit ripening, and causing less direct damage to the target fruit.

[0004] However, current evaluation of fruit peel aroma quality still primarily relies on endpoint testing of mature fruit. The conventional approach involves collecting peel tissue after the fruit has reached maturity, analyzing its volatile organic compound composition using instruments such as gas chromatography-mass spectrometry (GC-MS), and then determining the aroma quality accordingly. While this method can obtain the true aroma composition of mature fruit peel, it requires waiting for the fruit to mature, has a long testing cycle, and involves destructive sampling, making it difficult to pre-screen and rank aroma quality in large-scale breeding materials during the early stages of fruit development. Summary of the Invention

[0005] The purpose of this application is to provide a method and device for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds, which can realize early screening of fruit aroma quality and ranking of candidate varieties, reduce dependence on endpoint detection of mature fruit and sampling destructiveness, and improve the efficiency of breeding material screening and germplasm resource evaluation.

[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for predicting the aroma quality of mature fruit peel based on volatile organic compounds in leaves, including: Obtain the abundance data of volatile organic compounds in the leaves of the plants to be tested.

[0007] The volatile organic compound (VOC) abundance data of the leaves is input into the cross-organ prediction model, and the predicted VOC spectrum of mature fruit peel is output.

[0008] The predicted volatile organic compounds (VOCs) spectrum of the mature fruit peel is subjected to detection judgment and scale calibration to obtain the calibrated predicted VOCs spectrum of the mature fruit peel.

[0009] The weighted similarity between the calibrated predicted volatile organic compound (VOC) spectrum of the mature peel and the target aroma anchor point is calculated as the original mature peel aroma index. The target aroma anchor point is either a preset VOC composition of the mature peel of a superior variety or an artificially set ideal VOC composition.

[0010] The aroma index of the original mature fruit peel was calibrated with a false positive penalty to obtain the risk-calibrated aroma index.

[0011] Based on the distance between the plant under test and the preset reference sample set in the leaf volatile organic compound feature space, the applicable domain score is calculated, and the applicable domain score is compared with a preset threshold. When it is higher than the preset threshold, an extrapolation risk warning is output.

[0012] The calibrated predicted volatile organic compound spectrum of mature fruit peel is mapped to the odor descriptor space to obtain the odor semantic output result; the odor descriptor space is used to output the probability distribution of odor descriptors, aroma type retrieval results or sensory interpretive descriptions.

[0013] The plants to be tested are ranked based on the risk-calibrated aroma index, and the aroma quality evaluation results of the mature peel are output. The aroma quality evaluation results of the mature peel include the calibrated predicted spectrum of volatile organic compounds of the mature peel, the risk-calibrated aroma index, the applicable domain score, the extrapolation risk warning, and the odor semantic output results.

[0014] Optionally, data on the abundance of volatile organic compounds in the leaves of the plant under test can be obtained, including: Acquire the abundance data of volatile organic compounds (VOCs) in the leaves of the tested plants during the early and late developmental stages.

[0015] The volatile organic compound (VOC) abundance data of leaves in the early development stage and the volatile organic compound (VOC) abundance data of leaves in the late development stage are spliced ​​together to obtain a two-stage leaf VOC abundance data matrix.

[0016] Optionally, the cross-organ prediction model is trained based on a training sample set, which includes leaf volatile organic compound abundance data and corresponding mature fruit peel volatile organic compound target data for multiple plants at at least one developmental stage.

[0017] The cross-organ prediction model includes a detection probability sub-model and an abundance regression sub-model; the detection probability sub-model is used to predict the detection probability of volatile organic compounds in mature fruit peel, and the abundance regression sub-model is used to predict the abundance value of volatile organic compounds in mature fruit peel. The predicted spectrum of volatile organic compounds in mature fruit peel is obtained by combining the detection probability of volatile organic compounds in mature fruit peel and the abundance value of volatile organic compounds in mature fruit peel.

[0018] Optionally, the predicted volatile organic compound (VOC) spectrum of the mature pericarp is subjected to detection determination and scale calibration to obtain a calibrated predicted VOC spectrum of the mature pericarp, specifically including: Based on the preset specific detection threshold of each volatile organic compound in the predicted spectrum of volatile organic compounds in the mature fruit peel, the detection determination of each volatile organic compound in the predicted spectrum of volatile organic compounds in the mature fruit peel is performed to determine the detection status of each volatile organic compound.

[0019] Based on the preset calibration parameters of each volatile organic compound in the predicted volatile organic compound spectrum of the mature fruit peel, the predicted abundance of the detected volatile organic compounds is scale-calibrated to obtain the calibrated abundance value.

[0020] Based on the detection status and the calibrated abundance value, the calibrated predicted spectrum of volatile organic compounds in mature fruit peel is determined.

[0021] Optionally, the weights for calculating the weighted similarity are determined based on the correlation coefficients and detection performance of each volatile organic compound; the detection performance includes at least one of precision, recall, and F1 score.

[0022] Optionally, the aroma index of the original mature peel is calibrated with a false positive penalty to obtain a risk-calibrated aroma index, specifically including: Obtain the false positive risk estimate of the plant to be tested. The false positive risk estimate is determined based on at least one of the detection probability output by the cross-organ prediction model, the preset zero inflation ratio, the model prediction confidence, and the empirical value of cross-validation false positives.

[0023] Based on the false positive risk estimate and the preset false positive penalty parameter, the aroma index of the original mature fruit peel is calibrated to obtain the risk-calibrated aroma index.

[0024] Optionally, the distance is calculated using at least one of Euclidean distance, cosine distance, Mahalanobis distance, k-nearest neighbor distance, local outlier factor, prediction variance, or ensemble model uncertainty.

[0025] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds described above.

[0026] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds described above.

[0027] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described above for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds.

[0028] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method and apparatus for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds (VOCs). First, the abundance data of VOCs in the leaves of the plant to be tested is acquired. This data is input into a cross-organ prediction model, which outputs a predicted spectrum of VOCs in the mature fruit peel. This allows for the prediction of the aroma composition of the mature fruit peel using readily available leaf samples before fruit ripening. Next, the predicted spectrum of VOCs in the mature fruit peel is subjected to detection judgment and scaling calibration. The weighted similarity between the calibrated predicted spectrum and the target aroma type anchor point is calculated, transforming the cross-organ prediction result into a quantifiable raw mature fruit peel aroma index. This solves the problem that single chemical composition predictions are difficult to directly use for variety comparison, achieving a standardized measurement of aroma quality. Then, the raw mature fruit peel aroma index is further analyzed. The aroma index of the fruit peel is calibrated with a false positive penalty to obtain a risk-calibrated aroma index. Combined with the calculation of the applicable domain score and the output of extrapolation risk warning, this solves the problem of unreliable prediction results caused by false positive detection or samples outside the model training space in cross-organ prediction. It reduces the risk of high false positive materials being incorrectly ranked, and improves the robustness and reliability of candidate material ranking. Finally, by mapping the calibrated volatile organic compound prediction spectrum of mature fruit peel to the odor descriptor space and outputting the odor semantic output results, the problem of insufficient sensory interpretability of pure chemical prediction spectrum is solved. This allows the prediction results to be presented in the form of odor semantic description, enhancing the practicality and intuitiveness of the evaluation results in breeding and variety selection, and improving the efficiency of breeding material screening and germplasm resource evaluation. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is an application environment diagram of a method for predicting the aroma quality of ripe fruit peel based on leaf volatile organic compounds in one embodiment of this application. Figure 2 A flowchart illustrating a method for predicting the aroma quality of ripe fruit peel based on leaf volatile organic compounds, provided as an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0032] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] The method for predicting the aroma quality of ripe fruit peel based on leaf volatile organic compounds provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send the volatile organic compound (VOC) abundance data of the leaves of the plant to be tested to server 104. After receiving the VOC abundance data, server 104 inputs the VOC abundance data into a cross-organ prediction model and outputs a predicted spectrum of VOCs in the mature peel. The predicted spectrum of VOCs in the mature peel is then subjected to detection judgment and scaling calibration to obtain a calibrated predicted spectrum of VOCs in the mature peel. The weighted similarity between the calibrated predicted spectrum of VOCs in the mature peel and the target aroma anchor is calculated as the original aroma index of the mature peel. The target aroma anchor is either a preset composition of VOCs in the mature peel of a superior variety or an artificially set ideal VOC composition. The original aroma index of the mature peel is then subjected to false positive penalty calibration to obtain a risk calibration. The system calculates the volatile organic compound (VOC) index of the tested plant and the preset reference sample set in the leaf VOC feature space, then compares the applicable domain score with a preset threshold. If the score exceeds the preset threshold, an extrapolation risk warning is output. The calibrated mature peel VOC prediction spectrum is mapped to the odor descriptor space to obtain the odor semantic output result. The odor descriptor space is used to output the odor descriptor probability distribution, aroma type retrieval results, or sensory interpretive descriptions. Based on the risk-calibrated aroma index, the tested plants are sorted, and the mature peel aroma quality evaluation result is output. The mature peel aroma quality evaluation result includes the calibrated mature peel VOC prediction spectrum, the risk-calibrated aroma index, the applicable domain score, the extrapolation risk warning, and the odor semantic output result. The server 104 can feed back the obtained mature peel aroma quality evaluation result to the terminal 102. In addition, in some embodiments, the method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly perform the processing of predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds on the abundance data of leaf volatile organic compounds of the plant to be tested. Alternatively, the server 104 can obtain the abundance data of leaf volatile organic compounds of the plant to be tested from the data storage system and perform the processing of predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds on the abundance data of leaf volatile organic compounds of the plant to be tested.

[0034] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0035] In one exemplary embodiment, such as Figure 2 As shown, a method for predicting the aroma quality of ripe fruit peel based on leaf volatile organic compounds is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 208. Wherein: Step 201: Obtain the abundance data of volatile organic compounds (VOCs) in the leaves of the plant to be tested, which specifically includes the following steps.

[0036] Acquire the abundance data of volatile organic compounds (VOCs) in the leaves of the tested plants during the early and late developmental stages.

[0037] The volatile organic compound (VOC) abundance data of leaves in the early development stage and the volatile organic compound (VOC) abundance data of leaves in the late development stage are spliced ​​together to obtain a two-stage leaf VOC abundance data matrix.

[0038] This application uses citrus as an example, collecting leaf samples from the plant during the fruit development process. Citrus includes sweet orange, mandarin orange, pomelo, tangerine, orange, kumquat, lemon, citron, and their hybrids.

[0039] The leaf samples may be collected at at least one developmental stage, preferably at two or more developmental stages. More preferably, the leaf samples include leaf samples from early and late developmental stages.

[0040] In a preferred embodiment, the developmental stages include S1, S2, S3, and S4, where S1 is approximately one month after flowering, S2 is approximately three months after flowering, S3 is approximately five months after flowering, and S4 is approximately seven months after flowering. Preferably, leaf samples are collected from both S1 and S4 stages to form a combined input of volatile organic compound (VOC) abundance data from leaves in the early developmental stage and leaves in the late developmental stage.

[0041] During the training phase, fruit peel samples of corresponding varieties and growth stages can be collected simultaneously, especially those at maturity or near maturity, to establish a mapping relationship between leaf VOCs and mature fruit peel VOCs. In the practical application phase, only leaf samples from the plants to be tested can be collected and input into the trained model for prediction.

[0042] In practice, the abundance data of volatile organic compounds (VOCs) in leaves are obtained by detecting leaf samples from the plant under test at at least one developmental stage. Headspace solid-phase microextraction gas chromatography-mass spectrometry (HS-SPME-GC-MS) is preferably used to detect VOCs in leaves.

[0043] In a preferred embodiment, 0.02 g of leaf powder ground with liquid nitrogen was weighed and placed in a 20 mL screw-top headspace vial. 2 mL of saturated sodium chloride solution and 20 μL of 0.1% hexanol internal standard solution were added. The mixture was equilibrated at 40 °C and 250 rpm for 20 min. Extraction was then performed using a 50 / 30 μm DVB / CAR / PDMS extraction head for 20 min. Subsequently, desorption was performed at 300 °C for 5 min through the GC inlet. Here, rpm represents revolutions per minute and is used to indicate the rotational speed of the shaker or stirrer during sample equilibration, extraction, etc.

[0044] In a preferred embodiment, the gas chromatographic column is an HP-5 capillary column with dimensions of 30m × 0.25mm × 0.25μm; the carrier gas is helium with a flow rate of 1mL / min; the column temperature program is as follows: hold at 40℃ for 3min, increase to 115℃ at 3℃ / min, increase to 135℃ at 8℃ / min and hold for 2min, increase to 180℃ at 3℃ / min, and increase to 230℃ at 10℃ / min.

[0045] VOC identification can be accomplished through mass spectrometry library matching and retention index matching. The NIST (National Institute of Standards and Technology) mass spectrometry library is preferred for mass spectrometry matching, and the retention index is calculated using C8–C20 n-alkanes for confirmation. VOC quantification preferably uses the internal standard method, and the results can be expressed as μg / g FW, i.e., micrograms per gram of fresh weight of volatile organic compounds.

[0046] In addition to HS-SPME-GC-MS, the VOC detection method can also be replaced by gas chromatography-mass spectrometry (GC-MS), gas chromatography-ion mobility spectrometry (GC-IMS), proton-transfer-reaction mass spectrometry (PTR-MS), electronic nose, targeted GC-MS, or other data acquisition methods that can obtain the characteristics of volatile organic compounds in leaves.

[0047] Step 202: Input the leaf volatile organic compound abundance data into the cross-organ prediction model and output the volatile organic compound prediction spectrum of mature fruit peel.

[0048] As an optional implementation, the cross-organ prediction model is trained based on a training sample set, which includes leaf volatile organic compound abundance data and corresponding mature fruit peel volatile organic compound target data for multiple plants at at least one developmental stage.

[0049] The cross-organ prediction model includes a detection probability sub-model and an abundance regression sub-model; the detection probability sub-model is used to predict the detection probability of volatile organic compounds in mature fruit peel, and the abundance regression sub-model is used to predict the abundance value of volatile organic compounds in mature fruit peel. The predicted spectrum of volatile organic compounds in mature fruit peel is obtained by combining the detection probability of volatile organic compounds in mature fruit peel and the abundance value of volatile organic compounds in mature fruit peel.

[0050] During the model training phase, mature fruit peel samples corresponding to leaf samples are collected to construct mature fruit peel VOC target data.

[0051] In a preferred embodiment, 0.02 g of fruit peel powder ground with liquid nitrogen was weighed and placed in a 10 mL screw-top headspace vial. 2 mL of saturated sodium chloride solution and 20 μL of 0.1% hexanol internal standard solution were added. The mixture was equilibrated at 45 °C and 300 rpm for 30 min. Extraction was performed using a 50 / 30 μm DVB / CAR / PDMS extraction head for 30 min. Subsequently, desorption was performed at 250 °C for 5 min at the GC inlet.

[0052] For GC-MS of mature pericarps, an HP-5 capillary column and helium carrier gas can be used. The column temperature program can be set as follows: hold at 40℃ for 3 min, increase to 70℃ at 3℃ / min, increase to 130℃ at 1℃ / min, and then increase to 230℃ at 15℃ / min. The methods for VOC identification and quantification in mature pericarps are similar to those for VOC detection in leaves.

[0053] The target data for VOC in mature fruit peel can be the VOC composition of S4 mature fruit peel, or it can be the total VOC, specific VOC family, key aroma substances, similarity of target aroma type, or comprehensive aroma index of mature fruit peel.

[0054] Through the above steps, relatively abundant VOC characteristics were detected in the pericarp, while VOC characteristics dominated by monoterpenes were detected in the leaves. Pericarp VOCs underwent significant remodeling with developmental stages, while leaf VOCs were generally more stable but retained varietal information.

[0055] The obtained leaf VOC abundance data were organized into a leaf VOC abundance matrix. The rows of the leaf VOC abundance matrix represent the tested plants, the columns represent VOC features, and the matrix elements are the abundance values ​​of the corresponding VOCs in the tested plant samples. Undetected VOCs can be assigned a value of 0. VOC features from different organs or different stages can be aligned using a union method. For features not appearing in a certain organ or stage, zeros are used for padding.

[0056] The preprocessing includes one or more of the following: removing zero-variance features; removing VOC features with an excessively high proportion of zero values; and performing logarithmic operations on the abundance values. 10 (X+1) or log1p transformation; relative abundance normalization; Z-score standardization; aggregation by variety, stage or repetition; splicing of leaf VOC characteristics from multiple developmental stages.

[0057] In a preferred embodiment, the leaf VOC features from periods S1 and S4 are spliced ​​together to obtain a two-stage leaf VOC input matrix.

[0058] A cross-organ prediction model for VOC from leaves to mature pericarp was established using a training sample set. The training sample set included leaf VOC data from multiple plants at at least one, preferably two or more, developmental stages, as well as corresponding mature pericarp VOC data.

[0059] In a preferred embodiment, the cross-organ prediction model is a two-stage prediction model for sparse VOC data. This model may include a detection probability sub-model and an abundance regression sub-model. The detection probability sub-model is used to determine whether a certain VOC is likely to be detected in the mature peel, and the abundance regression sub-model is used to predict the abundance value of the detected VOC. The two are combined to form the mature peel VOC prediction spectrum.

[0060] The cross-organ prediction model may also employ machine learning models, ensemble learning models, gradient boosting models, random forest models, partial least squares models, ridge regression models, Least Absolute Shrinkage and Selection Operator (LASSO) models, elastic network models, support vector machine models, neural network models, graph neural network models, or combinations thereof.

[0061] Cross-organ prediction model training can employ Leave-One-Cultivar-Out (LOCO) validation. In each round of validation, one variety is retained as the test set, and the remaining varieties are used as the training set to evaluate the predictive ability of the cross-organ prediction model for the unseen varieties. Each round of validation outputs the predicted VOC value of the mature peel of the retained variety and compares it with the measured VOC value of the mature peel of that retained variety.

[0062] In a preferred embodiment, the cross-organ prediction model validation results can be compiled into a cross-validation prediction long table. The fields of the cross-validation prediction long table include leaf input combination, validation round, variety, VOC name, predicted VOC value of mature fruit peel, measured VOC value of mature fruit peel, model detection probability, prediction detection status, and prediction error.

[0063] The VOC prediction value for mature pericarps refers to the predicted value output for the excluded varieties in each round of LOCO validation. Since the excluded varieties did not participate in the cross-organ prediction model training in that round, this prediction value can be used to evaluate the predictive ability of the cross-organ prediction model for unseen varieties. The cross-validation prediction long table can serve as a unified input for subsequent mature pericarp aroma index calculation, target aroma type ranking, false positive penalty calibration, error analysis, and determination of applicable domain thresholds.

[0064] During the training and validation phase, the available VOC data from mature fruit peels can be used for model training, cross-validation, error analysis, VOC detection performance evaluation, false positive empirical value statistics, λ parameter scanning, and determination of applicable domain thresholds.

[0065] In practical applications, only leaf samples are typically collected from the plants being tested to obtain VOC input data; simultaneous VOC measurement data from mature fruit peels is not required. The VOC-specific detection thresholds, VOC-level calibration parameters, VOC weights, λ parameters, cross-validation false positive empirical values, and applicable domain thresholds used in subsequent practical applications are all derived from the model training and validation phases.

[0066] Step 203 involves detecting and scaling the predicted volatile organic compounds (VOCs) spectrum of the mature fruit peel to obtain the calibrated predicted VOCs spectrum of the mature fruit peel. This step specifically includes the following steps.

[0067] Based on the preset specific detection threshold of each volatile organic compound in the predicted spectrum of volatile organic compounds in the mature fruit peel, the detection determination of each volatile organic compound in the predicted spectrum of volatile organic compounds in the mature fruit peel is performed to determine the detection status of each volatile organic compound.

[0068] Based on the preset calibration parameters of each volatile organic compound in the predicted volatile organic compound spectrum of the mature fruit peel, the predicted abundance of the detected volatile organic compounds is scale-calibrated to obtain the calibrated abundance value.

[0069] Based on the detection status and the calibrated abundance value, the calibrated predicted spectrum of volatile organic compounds in mature fruit peel is determined.

[0070] In the specific implementation process, after obtaining the predicted VOC spectrum of mature fruit peel, further post-processing of the predicted spectrum can be performed. Specifically, detection can be determined according to the VOC specificity threshold. The VOC specificity threshold can be determined based on the low quantile of the true positive value samples of the VOC in the log1p space in the training set, for example, using the 5th quantile as the detection threshold. The true positive value samples refer to samples in the training set where the measured abundance of the VOC in mature fruit peel is greater than 0.

[0071] To reduce the interference of zero-inflated VOCs on the predicted spectrum, a minimum number of positive samples can be set. This minimum number of positive samples refers to the lower limit of the number of true non-zero samples used to calculate the detection threshold of a certain VOC, for example, no less than 3 or no less than 5 true positive samples. If the number of true positive samples for a certain VOC in the training set is lower than this lower limit, the specificity threshold of that VOC will not be estimated separately, or a global threshold, low-confidence label, reduced weights, or temporary exclusion from weighted similarity calculations will be used.

[0072] The predicted abundance can also be linearly calibrated at the VOC level in log1p space to obtain the intercept and slope parameters for each VOC, which are used to correct scale bias between different VOCs. The VOC-specific detection threshold, intercept, and slope parameters are determined during the model training and validation phases and are consistently used in practical applications.

[0073] Step 204: Calculate the weighted similarity between the calibrated predicted spectrum of volatile organic compounds in mature fruit peel and the target aroma anchor point, as the original aroma index of mature fruit peel; the target aroma anchor point is the volatile organic compound composition of mature fruit peel of a preset superior variety or an ideal volatile organic compound composition set artificially.

[0074] As an optional implementation, the weights for calculating the weighted similarity are determined based on the correlation coefficients and detection performance of each volatile organic compound; the detection performance includes at least one of precision, recall, and F1 score.

[0075] In this embodiment of the application, after processing the predicted VOC spectrum of mature fruit peel, the calibrated predicted volatile organic compound spectrum of mature fruit peel can be compared with the target aroma anchor point. The target aroma anchor point can be the VOC composition of the mature fruit peel of a superior variety, or it can be an artificially set ideal VOC composition vector.

[0076] Aroma Index of Original Mature Peel The weighted cosine similarity between the calibrated predicted VOC spectrum of mature fruit peel and the target aroma anchor point can be used. In specific implementation, for the first... For each VOC, calculate the Spearman correlation coefficient between its predicted and measured values. And calculate the positive correlation coefficient. .

[0077] .

[0078] Calculate the first based on the actual detection status and the predicted detection status. VOC detection performance indicators And calculate the original weights. .

[0079] .

[0080] For the original weights Normalization is performed to obtain the normalized weights. .

[0081] .

[0082] in, The number of VOCs included in the weighted similarity calculation, This indicates the first VOC prediction spectrum of the calibrated mature fruit peel. VOCs.

[0083] Based on normalized weights Aroma index of original ripe peel Calculated using the following weighted cosine similarity.

[0084] .

[0085] in, Indicates the first The first of the tested plants The calibrated predicted abundance of each VOC. Indicating the first of the target aroma type anchor points The abundance of each VOC.

[0086] In practical implementation, when using the S1+S4 two-stage leaf VOC input, the leaf-to-mature peel prediction framework achieved high ordination recovery ability in cross-validation with one variety retained, with a Spearman correlation coefficient of 0.775 and a cosine similarity of 0.742 between the predicted and measured mature peel VOC composition. The further assembled cross-organ prediction model, under the same validation conditions, achieved a Spearman correlation coefficient of 0.785, higher than the random forest model (0.538) and the partial least squares model (0.525).

[0087] The weighted cosine similarity is defined as the aroma index of the original mature fruit peel. Based on the aroma index of the original mature fruit peel. This allows for the preliminary sorting of the materials to be tested.

[0088] Step 205 involves performing false positive penalty calibration on the original mature fruit peel aroma index to obtain the risk-calibrated aroma index, specifically including the following steps.

[0089] Obtain the false positive risk estimate of the plant to be tested. The false positive risk estimate is determined based on at least one of the detection probability output by the cross-organ prediction model, the preset zero inflation ratio, the model prediction confidence, and the empirical value of cross-validation false positives.

[0090] Based on the false positive risk estimate and the preset false positive penalty parameter, the aroma index of the original mature fruit peel is calibrated to obtain the risk-calibrated aroma index.

[0091] To reduce the probability of high-risk candidate materials being incorrectly ranked, this application further introduces a false positive penalty calibration. The false positive penalty calibration is used to downgrade the risk of candidate materials with high false positive risk based on the target aroma type similarity ranking, thereby reducing the risk of false screening caused by too many falsely detected VOCs in the predicted spectrum.

[0092] During the training and validation phases, the false positive rate can be defined as the proportion of VOCs with a true mature fruit peel VOC value of 0 that are predicted as non-zero by the model, out of the total number of VOCs with a true VOC value of 0. By comparing the predicted and measured values ​​during the LOCO validation process, empirical values ​​for VOC-level or sample-level cross-validation can be further calculated. These empirical values ​​for cross-validation false positives can be used to identify VOCs prone to false detection, sample types easily misranked, and the stability of candidate material ranking under different penalty intensities.

[0093] In practical applications, the plants to be tested typically do not have measured VOC values ​​from mature fruit peels. Therefore, the true false positive rate is not calculated directly. Instead, the false positive risk of the plants to be tested is estimated based on the VOC detection probability output by the trained cross-organ prediction model, the proportion of zero-inflation VOCs in the training set, the model prediction confidence, the empirical value of cross-validation false positives, or a combination thereof. The proportion of zero-inflation VOCs in the training set refers to the proportion of samples in the mature fruit peel VOC target matrix of the training set that are not detected for each VOC. The model prediction confidence can be obtained by comprehensively considering indicators such as detection probability, prediction stability, cross-validation error, model ensemble variance, or applicability domain distance. The empirical value of cross-validation false positives is derived from the comparison between predicted and measured values ​​during LOCO validation, rather than being calculated after obtaining measured VOC values ​​from mature fruit peels again in the practical application stage.

[0094] In a preferred embodiment, a false positive penalty parameter λ can be set to control the impact of false positive risk on the aroma index of the original mature peel. The magnitude of the influence of the target aroma type ranking results. The larger λ is, the greater the down-regulation of candidate materials with a higher risk of false positives in the ranking; the smaller λ is, the more the ranking results depend on the original weighted similarity between the calibrated mature pericarp VOC prediction spectrum and the target aroma type anchor point.

[0095] During the model training and validation phase, the parameter λ can be scanned within the range of 0-0.40, and the optimal value is determined based on the ranking recovery ability, Top-k stability, and downregulation effect on high false positive materials. Here, Top-k refers to the top k candidate materials output after ranking according to the predicted aroma index or odor semantic similarity.

[0096] Aroma Index after Risk Calibration It can be calculated using the following formula.

[0097] .

[0098] in, The aroma index after risk calibration. This is an estimate of the false positive risk. During the training and validation phase, It can be calculated from the proportion of VOCs that are actually 0 but predicted to be non-zero; in practical applications, It is estimated from at least one of the following: model detection probability, training set VOC zero inflation ratio, model prediction confidence, and empirical value of cross-validation false positives.

[0099] In this application, the value of λ can be set to 0-0.40, preferably 0.05-0.40, and more preferably 0.15-0.25, based on experimental verification and actual needs. In a preferred embodiment, λ is set to 0.20. Through scanning the λ parameter, this application found that when λ increases from 0 to 0.15, the sorting performance improves significantly; after λ is approximately 0.20, the sorting performance enters a plateau period, and the Top-5 stability remains at a high level. Therefore, λ=0.20 can be used as a preferred risk calibration parameter.

[0100] In practice, λ=0.20 was determined through parameter scanning during the training and validation phases, and was then used as a fixed parameter directly for the aroma index of the original mature fruit peel in the actual application phase. Risk calibration for target aroma type sorting.

[0101] The false positive penalty calibration can be achieved using linear penalty, piecewise penalty, multiplicative penalty, nonlinear penalty, confidence-weighted penalty, or multi-index joint risk calibration. After false positive penalty calibration, the risk-calibrated mature peel aroma index or target aroma type ranking result can be obtained, and the Top-5, Top-10, or other Top-k candidate materials can be output accordingly.

[0102] Step 206: Based on the distance between the plant under test and the preset reference sample set in the leaf volatile organic compound feature space, calculate the applicable domain score, and compare the applicable domain score with a preset threshold. If the score is higher than the preset threshold, output an extrapolation risk warning.

[0103] As an optional implementation, the distance is calculated using at least one of Euclidean distance, cosine distance, Mahalanobis distance, k-nearest neighbor distance, local outlier factor, prediction variance, or ensemble model uncertainty.

[0104] This application further includes an application domain determination. The application domain determination is used to determine whether the plant under test is within the coverage area of ​​the model training data. The application domain determination includes two stages: threshold establishment during the training and validation phase and sample determination during the practical application phase.

[0105] During the training and validation phases, the VOC features of leaves in the training set can be transformed using log1p, and the distance between training samples can be calculated. Alternatively, in LOCO validation, the nearest neighbor distance between the excluded varieties and the training varieties can be calculated. The applicable domain threshold E can then be determined by combining the distance distribution, prediction error, or ranking performance. .

[0106] In the practical application phase, the same preprocessing method is used on the plants to be tested. During the specific implementation process, the nearest neighbor Euclidean distance between the plant and the samples in the training set in the leaf VOC feature space is calculated, and this distance is defined as the applicable domain score. If the applicable domain score is higher than the preset threshold, i.e., the applicable domain threshold E determined in the training and validation phase, then the application domain score is considered valid. If the result is negative, an extrapolation risk warning will be output, the prediction confidence level will be lowered, or the sample will be marked as a low-applicability sample in the candidate ranking report.

[0107] Step 207: Map the calibrated predicted volatile organic compound spectrum of mature fruit peel to the odor descriptor space to obtain the odor semantic output result; the odor descriptor space is used to output the probability distribution of odor descriptors, aroma type retrieval results or sensory interpretive descriptions.

[0108] The embodiments of this application can further convert the VOC chemical composition prediction results into odor semantic output. Specifically, VOC molecules are input into the odor descriptor prediction model to obtain a VOC-level odor description probability matrix; then, the sample VOC abundance vector and the VOC odor description probability matrix are weighted and aggregated to obtain a sample-level odor semantic vector.

[0109] Based on the sample-level odor semantic vector, the probability distribution of odor descriptors such as fruity, green, floral, woody, herbal, sweet, mint, rose, and dry can be output. Alternatively, a Top-k search can be performed in the mature fruit peel odor semantic database to output modeled candidate fruit peel aroma semantic profiles or aroma type search results.

[0110] The odor semantic output is a modeled odor description result derived from the molecular structure or abundance of VOCs. It is used to improve the interpretability of the prediction results, but it is not equivalent to the results of artificial sensory evaluation, nor does it replace psychophysical experiments, sensory assessments, or odor activity value (OAV) verification.

[0111] This application's embodiments reveal that, at the variety-developmental stage centroid level, leaf inputs can retrieve matching pericarp odor semantic profiles in the odor descriptor space at a higher-than-random level. This indicates that cross-organ signals between leaves and pericarps not only exist in the VOC chemical abundance space but can also be partially preserved in the odor descriptor space.

[0112] Step 208: Sort the plants to be tested based on the risk-calibrated aroma index and output the aroma quality evaluation result of the mature peel; the aroma quality evaluation result of the mature peel includes the calibrated volatile organic compound prediction spectrum of the mature peel, the risk-calibrated aroma index, the applicable domain score, the extrapolation risk warning and the odor semantic output result.

[0113] Based on the risk-calibrated aroma index, the test materials can be finally ranked, and Top-5, Top-10, or other Top-k candidate materials can be output. The ranking results of the candidate materials can be used for variety screening, germplasm resource evaluation, pre-screening of aroma-type breeding materials, priority determination of mature fruit testing, and development of target aroma-type products.

[0114] In practical applications, only leaf samples need to be collected, without the need to simultaneously obtain measured VOC data from mature fruit peels. Measured VOC data from mature fruit peels of the tested plants can be used as new training data during subsequent variety verification, model retraining, or database expansion, but it is not a necessary input for the actual prediction application in this embodiment.

[0115] By implementing steps 201 to 208, the following advantages are achieved: First, without directly collecting mature peel samples, the aroma composition of mature fruit peel can be predicted using readily available two-stage leaf sample information before fruit ripening, reducing damage to the target fruit and improving prediction stability. Second, by converting cross-organ prediction results into a quantifiable raw mature peel aroma index, the problem of single chemical composition prediction being difficult to directly use for variety comparison is solved, achieving a standardized measurement of aroma quality. Third, by calibrating the raw mature peel aroma index with false positive penalties and combining the calculation of applicable domain scores with the output of extrapolation risk warnings, the risk of high false positive materials being incorrectly ranked is reduced, improving the robustness and reliability of candidate material ranking. Finally, the odor semantic output results are output, resulting in an interpretable feature panel, which facilitates understanding of the biochemical information sources on which the model depends, enhancing the practicality and intuitiveness of the evaluation results in breeding and variety selection, and improving the efficiency of breeding material screening and germplasm resource evaluation.

[0116] In another exemplary embodiment of this application, to verify that a simplified prediction model can be constructed using a stable leaf VOC feature panel, thereby improving the generalizability and detection economy of the method, the association direction between each leaf VOC and the aroma index of the mature peel is calculated based on the leaf VOC data from step 201. Leave-one-out cross-validation or bootstrap resampling is used to calculate the sign stability of each VOC in different training folds. VOCs that maintain the same association direction in most validation folds are selected, and these VOCs are used to construct a simplified prediction model.

[0117] Using the above methods, the stable positively correlated leaf VOCs identified include α-thujene, α-pinene, γ-terpinene, β-ocimene, linalool, D-limonene, α-terpineol, terpinolene, germarene D, γ-elemene, α-guaiene, and citronellol; nerol exhibits a stable negative correlation. These characteristics are primarily enriched in terpenoid VOCs and can represent the main cross-organ association information between leaf and mature fruit peel aromas.

[0118] Therefore, the embodiments of this application can be further simplified into a method for predicting the aroma of mature fruit peel based on a stable combination of leaf VOC markers, providing a foundation for targeted GC-MS, rapid detection panels, electronic nose training, or industrialization service products.

[0119] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 3 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data on the abundance of volatile organic compounds (VOCs) in the leaves of the tested plants. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting the aroma quality of mature fruit peels based on leaf VOCs.

[0120] Figure 3The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0121] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0122] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations and be authorized by the owner of the corresponding device.

[0124] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0125] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0126] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0127] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the aroma quality of ripe fruit peel based on volatile organic compounds in leaves, characterized in that, The method includes: Obtain the abundance data of volatile organic compounds in the leaves of the plants to be tested; The leaf volatile organic compound abundance data are input into the cross-organ prediction model, and the predicted spectrum of volatile organic compounds in mature fruit peel is output. The predicted volatile organic compounds (VOCs) spectrum of the mature fruit peel is subjected to detection judgment and scale calibration to obtain the calibrated predicted VOCs spectrum of the mature fruit peel. The weighted similarity between the calibrated predicted spectrum of volatile organic compounds in mature fruit peel and the target aroma anchor point is calculated as the original aroma index of mature fruit peel; the target aroma anchor point is the volatile organic compound composition of mature fruit peel of a preset superior variety or an ideal volatile organic compound composition set artificially. The original mature fruit peel aroma index was calibrated with a false positive penalty to obtain the risk-calibrated aroma index. Based on the distance between the plant under test and the preset reference sample set in the leaf volatile organic compound feature space, the applicable domain score is calculated, and the applicable domain score is compared with a preset threshold. When it is higher than the preset threshold, an extrapolation risk warning is output. The calibrated predicted volatile organic compound spectrum of mature fruit peel is mapped to the odor descriptor space to obtain the odor semantic output result; the odor descriptor space is used to output the probability distribution of odor descriptors, aroma type retrieval results, or sensory explanatory descriptions. The plants to be tested are ranked based on the risk-calibrated aroma index, and the aroma quality evaluation results of the mature peel are output. The aroma quality evaluation results of the mature peel include the calibrated predicted spectrum of volatile organic compounds of the mature peel, the risk-calibrated aroma index, the applicable domain score, the extrapolation risk warning, and the odor semantic output results.

2. The method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds according to claim 1, characterized in that, Obtain the abundance data of volatile organic compounds in the leaves of the plants to be tested, including: Acquire the abundance data of volatile organic compounds (VOCs) in the leaves of the tested plants during the early development stage and the late development stage. The volatile organic compound (VOC) abundance data of leaves in the early development stage and the volatile organic compound (VOC) abundance data of leaves in the late development stage are spliced ​​together to obtain a two-stage leaf VOC abundance data matrix.

3. The method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds according to claim 1, characterized in that, The cross-organ prediction model is trained based on a training sample set, which includes leaf volatile organic compound abundance data and corresponding mature fruit peel volatile organic compound target data for multiple plants at at least one developmental stage. The cross-organ prediction model includes a detection probability sub-model and an abundance regression sub-model; the detection probability sub-model is used to predict the detection probability of volatile organic compounds in mature fruit peel, and the abundance regression sub-model is used to predict the abundance value of volatile organic compounds in mature fruit peel. The predicted spectrum of volatile organic compounds in mature fruit peel is obtained by combining the detection probability of volatile organic compounds in mature fruit peel and the abundance value of volatile organic compounds in mature fruit peel.

4. The method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds according to claim 1, characterized in that, The predicted volatile organic compound (VOC) spectra of the mature fruit peel are subjected to detection determination and scale calibration to obtain the calibrated predicted VOC spectra of the mature fruit peel, specifically including: Based on the preset specific detection threshold of each volatile organic compound in the predicted spectrum of volatile organic compounds in the mature fruit peel, the detection determination of each volatile organic compound in the predicted spectrum of volatile organic compounds in the mature fruit peel is performed to determine the detection status of each volatile organic compound. Based on the preset calibration parameters of each volatile organic compound in the predicted volatile organic compound spectrum of the mature fruit peel, the predicted abundance of the detected volatile organic compounds is scale-calibrated to obtain the calibrated abundance value. Based on the detection status and the calibrated abundance value, the calibrated predicted spectrum of volatile organic compounds in mature fruit peel is determined.

5. The method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds according to claim 1, characterized in that, The weights for calculating the weighted similarity are determined based on the correlation coefficients and detection performance of each volatile organic compound; the detection performance includes at least one of precision, recall, and F1 score.

6. The method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds according to claim 1, characterized in that, The aroma index of the original mature peel is calibrated for false positives to obtain a risk-calibrated aroma index, which specifically includes: Obtain the false positive risk estimate of the plant to be tested; the false positive risk estimate is determined based on at least one of the detection probability output by the cross-organ prediction model, the preset zero inflation ratio, the model prediction confidence, and the cross-validation false positive empirical value; Based on the false positive risk estimate and the preset false positive penalty parameter, the aroma index of the original mature fruit peel is calibrated to obtain the risk-calibrated aroma index.

7. The method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds according to claim 1, characterized in that, The distance is calculated using at least one of Euclidean distance, cosine distance, Mahalanobis distance, k-nearest neighbor distance, local outlier factor, prediction variance, or ensemble model uncertainty.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for predicting the aroma quality of mature fruit peel based on leaf volatile organic compounds as described in any one of claims 1-7.

Images