echemi logo
Product
  • Product
  • Supplier
  • Inquiry
    Home > Food News > Food Articles > Professor Zhou Hongping of Nanjing Forestry University: Hyperspectral imaging technology to detect the ripeness of Camellia oleifera fruit

    Professor Zhou Hongping of Nanjing Forestry University: Hyperspectral imaging technology to detect the ripeness of Camellia oleifera fruit

    • Last Update: 2022-11-26
    • Source: Internet
    • Author: User
    Search more information of high quality chemicals, good prices and reliable suppliers, visit www.echemi.com

    The picking period of Camellia oleifera fruit is greatly affected by environmental factors, and the weather, climate, accumulated temperature, soil fertility and other conditions in different regions are different, resulting in the ripening period of Camellia oleifera fruit early or late, so the picking time of Camellia oleifera fruit should be determined according to its specific ripeness
    .
    The last month before the harvest of Camellia oleifera fruit is the peak period of oil accumulation, during which the oil content of Camellia oleifera fruit increases most significantly, accompanied by the mutual transformation
    of internal nutrients.
    However, during this period, the external morphological characteristics of Camellia oleifera fruit, such as shape and size, fresh fruit quality, color, etc.
    , tended to be stable without significant changes, which brought certain difficulties
    to tea farmers in judging the ripeness and optimal picking period of Camellia oleifera fruit.
    Therefore, it is urgent to propose a method to quickly and accurately detect the ripeness of Camellia oleifera fruit to help the accurate harvesting of Camellia oleifera fruit


    Hu Yilei, Jiang Hongzhe, Zhou Hongping* from the School of Mechatronic Engineering of Nanjing Forestry University obtained the hyperspectral data of Camellia oleifera fruit by hyperspectral imaging technology and combined with stoichiometry to obtain the ripeness parameters of Camellia oleifera fruit as a reference, and used the classification algorithm to complete the discrimination of the ripeness of Camellia oleifera fruit, and verified the correct rate of the classification model, aiming to provide scientific basis
    for the judgment of the best picking period of Camellia oleifera fruit.



    1.
    Physical and chemical parameters of Camellia oleifera fruit ripening period


    It can be seen from Table 1 that the fruit height and diameter of Camellia oleifera fruit increased slightly during the whole picking period, but the increase was very small, so the standard deviation was low, only about 0.
    43 mm.
    The quality range of fresh fruit is between 25~30 g, and the quality range of fresh seed is between 10~15 g; The husk moisture content of Camellia oleifera fruit tended to decrease during the picking period, but the decrease was small.
    The seed yield of Camellia oleifera fruit fluctuates between 35%~40%, and the oil content of Camellia oleifera fruit increases with the delay of picking time, the increase reaches 59%, and its standard deviation is 5.
    55%, which is much higher than the standard deviation of other physical and chemical parameters, which shows that if Camellia oleifera fruit is picked in advance, it will cause serious loss
    of tea oil yield 。 And from the sampling time, on November 10, the shell of some Camellia oleifera fruit in the Camellia oleifera forest cracked, Camellia oleifera fruit entered the full harvest stage, and the harvest work was basically completed on November 14, most of the shells of Camellia oleifera fruit on the reserved 5 tea trees cracked, some Camellia oleifera fruit has fallen to the ground, in line with the characteristics of Camellia oleifera fruit overripening period, the Camellia oleifera fruit collected at this time was used as a control group, and it was found that its oil content remained basically stable, indicating that Camellia oleifera fruit entered the fully ripe period around November 10, and picking at this time can maximize the yield of Camellia oleifera.
    Moreover, the shell of Camellia oleifera fruit cracked and fell off, making it difficult to collect, and the overall harvest time
    of Camellia oleifera forest could no longer be delayed.

    Combined with Table 2, the quantitative criteria for maturity are given: samples with an oil content of (22.
    00±1.
    00)% about 30 days before harvest are used as maturity I; About 23 days before harvest, the samples with an oil content of (24.
    00±1.
    00)% were used as ripeness II.
    About 16 days before harvest, the samples with an oil content of (27.
    00±2.
    00)% were used as ripeness III.
    About 9 days before harvest, the samples with oil content of (31.
    50±2.
    50)% were used as ripeness IV.
    During the harvest of Camellia oleifera fruit, the samples with an oil content of (35.
    00±1.
    00)% were used as ripeness V
    .
    The ripeness of Camellia oleifera fruit samples with maturity level I~V increased sequentially, and the rape fruit with maturity V was the sample
    at the ripening stage.
    This experiment does not contain overripening samples, because the shell of Camellia oleifera in the overripening stage is easy to fall off when cracked, and it is easy to identify by the human eye without the use of hyperspectral equipment
    .
    The hyperspectral data of the above five ripeness Camellia oleifera fruit samples were used as datasets for the classification
    of ripeness.



    2.
    Pixel based hyperspectral image curvature correction of Camellia oleifera fruit


    Figure 2a is the spectral curve of 89 pixels from the center of Camellia oleifera fruit along the radius direction to the edge before correction, it can be seen that due to the influence of Camellia oleifera fruit curvature, the spectral reflection intensity at different positions varies greatly, and the spectral CV is 0.
    244 0; Figure 2b is the corrected spectral curve of the pixels at the same position, it can be found that the spectral intensity difference between each pixel is much smaller, CV is 0.
    075 3
    .
    The reflectivity of the 774 nm channel differs most from the reflectance of the background, so the grayscale map
    of the Camellia oleifera sample can be clearly displayed.
    Figure 3 shows the pseudo-color distribution of camellia oleifera reflectance before and after correction in the 774 nm band, and it can be intuitively seen that the reflectance distribution of camellia oleifera pixels after correction is more uniform than before calibration, indicating the effectiveness
    of the mean normalization method.



    3.
    Analysis of spectral characteristics of Camellia oleifera fruit with different ripeness


    The spectral curves before and after curvature correction of 939 samples are shown in Figure 4, and the corrected data distribution is more concentrated and the data characteristics are more obvious
    .
    The absorption peak at 670 nm was mainly related to anthocyanins and chlorophyll in the camellia oleifera husk, and the absorption peak at 970 nm was related
    to the water in the husk.

    The spectral data of Camellia oleifera fruits with different maturity grades were averaged to obtain the ripeness curves of Camellia oleifera fruits as shown in
    Figure 5.
    It can be found that the spectral intensity difference of Camellia oleifera fruit with different ripeness is mainly reflected in the range between
    500~630 nm and 720~970 nm.
    Between 500~630 nm, the difference in the reflection intensity of ripeness I and maturity II samples was small, with the delay of sampling time, the reflection intensity of ripeness III and maturity IV samples decreased compared with the former, and the reflection intensity of ripeness V samples was reduced to the minimum, showing the law that the spectral reflectance of Camellia oleifera decreased with the increase of
    ripeness in this band range 。 Between 720~970 nm, the reflection intensity of maturity II~V samples decreased with the increase of maturity, while the reflection intensity from maturity I~V samples underwent two processes of first increasing and then decreasing, which may be related to
    the dynamic change of endogenous hormone content in fruit shells.
    There were obvious differences in maturity curves at five different harvest time points, which provided a theoretical basis
    for maturity classification.



    4.
    Classification of camellia fruit ripeness based on full wavelength


    Five samples of Camellia oleifera fruit with different maturity grades were assigned category labels 1~5, and the maturity classification model of Camellia oleifera fruit based on full-wavelength spectral information was established by KNN, RF, SVM, PLS-DA respectively, and the parameter settings of each model are shown
    in Table 3.
    After parameter optimization of each model, the classification results of each model under the optimal parameters are obtained as shown
    in Table 3 and Figure 6.
    It can be found that SVM has the best classification effect compared with the other three models: the classification accuracy rate of the training set is 98%, the average classification accuracy of the test set is 97%, and the classification accuracy rate of maturity I samples is the highest.
    Secondly, the RF model showed that the average classification accuracy rate of Camellia oleifera fruit ripeness was 86%, and the classification accuracy rate of ripeness I and ripeness IV samples was high.
    The classification effect of PLS-DA model is average, and the correct rate of test set is 76%.
    The classification accuracy rate of the KNN model on the training and test sets is the lowest
    of all models.



    5.
    Classification of camellia fruit ripeness based on characteristic wavelength


    Table 4 shows the results
    of SVM classification models established by selecting feature variables by different dimensionality reduction methods.
    It can be found that the SVM model established using the spectral data of 28 characteristic wavelengths after CARS dimensionality reduction has the best effect: the average accuracy rate of the training set is 91%, the average accuracy rate of the test set is 82%, the recognition rate of maturity IV samples is the highest 91%, and the recognition rate of maturity II samples is 73%.

    The effect of the classification model established after SPA and GA dimensionality reduction is not much different: the classification accuracy rate of Camellia oleifera fruit on the training set is about 86%, and the recognition rate on the test set is about
    80%.
    The effect of principal component analysis (PCA)-SVM model is the worst: the classification accuracy rate on the training set is 64%, and the classification accuracy rate on the test set is only 55%, which shows that even if 99% of the information is retained by the PCA method in converting full-band spectral data into 7 PCs, the performance of the classification model established by combining SVM is still poor
    。 CARS-SPATAVM and GA-SVA-SVM are both classification models established after quadratic dimensionality reduction, and the classification effect of SVM models in the two sets of datasets is similar: the correct rate of the test set is about 75%, and both models have the highest classification accuracy rate for maturity V samples, and low classification accuracy rates for maturity II and maturity III samples
    .


    6.
    Classification of camellia fruit ripeness based on fusion of image features and spectral features


    Table 5 shows the classification results of SVM model under different input characteristics, and it can be found that the accuracy of the test set of the SVM model of Camellia oleifera fruit established using a single color feature is 79%, which is 3.
    7% lower than that of the single spectral feature modeling, and the recognition rate of the model for maturity IV and maturity V samples is high, 92% and 95%, respectively, and the recognition rate of maturity II and maturity III samples is low
    。 The accuracy rate of the SVM classification model of Camellia oleifera fruit ripeness established by using a single texture feature is less than 50%, indicating that the texture characteristic parameters of Camellia oleifera fruit do not change significantly during the ripening period of Camellia oleifera fruit, and the epidermis of Camellia oleifera fruit is smooth and the texture features are not obvious, which cannot be well used as a characterization factor
    of Camellia oleifera fruit ripeness.

    In general, the SVM classification model of Camellia oleifera fruit ripeness established by fusing color characteristics and spectral characteristics has the best effect, and the recognition rate of the model for ripeness I, IV and V oleifera fruit samples is generally higher than that of maturity II and III samples, because the characteristic parameters between maturity I, IV and V Camellia oleifera fruit samples are large and easy to resolve, while the feature differences between maturity II and III samples are relatively small
    .

    Conclusion


    The feasibility of using hyperspectral imaging technology to detect the ripeness of Camellia oleifera fruit was explored, and the optical characteristics of Camellia oleifera fruit were observed and analyzed by static measurement in the laboratory, and the accuracy
    of judging the ripeness of Camellia oleifera fruit by using the optical characteristics of Camellia oleifera fruit was analyzed on this basis.
    At present, the field experiment to discuss the feasibility of spectral detection of Camellia oleifera fruit ripeness in natural environment is in the scheme demonstration stage, which will be carried out during the ripening period (October-November) of Camellia oleifera fruit this year, laying a theoretical foundation
    for the new stage of intelligent picking of Camellia oleifera first ripe, then ripe and then picked.
    The main conclusions are as follows: 1) The maturity discrimination model of Camellia oleifera fruit based on full-band spectral data was established using KNN, RF, SVM, PLS-DA respectively, and it was found that the SVM model had the highest classification accuracy: the correct rate of the training set was 98%, and the accuracy rate of the test set was 97%.

    。 2) SPA, CARS, GA, CARS-SPA, GA-SPA, respectively used to reduce the dimensionality of the whole spectrum, and established an SVM model based on feature variables, and found that the classification model established by the feature band selected by the CARS method had the highest accuracy rate: the classification accuracy rate of the training set was 91%, and the correct rate of the test set was 82%.

    。 3) After extracting the color features and texture features in the hyperspectral image of Camellia oleifera fruit, and establishing the SVM classification model respectively, it is found that the accuracy rate of the SVM model established by fusing color features and spectral features is higher than that of the model established by using a single spectral feature (dimensionality reduction by CARS): the classification accuracy rate of the training set is 95%, and the correct rate of the test set is 93%.


    This article "Hyperspectral Imaging Technique to Detect Camellia oleifera Fruit Ripeness" is from Food Science, Vol.
    43, No.
    16, 2022, pp.
    324-331, authors: Hu Yilei, Jiang Hongzhe, Zhou Hongping, Xu Linyun, Ju Hao, Wang Ying
    .
    DOI:10.
    7506/spkx1002-6630-20210619-229
    。 Click to view information about
    the article.



    This article is an English version of an article which is originally in the Chinese language on echemi.com and is provided for information purposes only. This website makes no representation or warranty of any kind, either expressed or implied, as to the accuracy, completeness ownership or reliability of the article or any translations thereof. If you have any concerns or complaints relating to the article, please send an email, providing a detailed description of the concern or complaint, to service@echemi.com. A staff member will contact you within 5 working days. Once verified, infringing content will be removed immediately.

    Contact Us

    The source of this page with content of products and services is from Internet, which doesn't represent ECHEMI's opinion. If you have any queries, please write to service@echemi.com. It will be replied within 5 days.

    Moreover, if you find any instances of plagiarism from the page, please send email to service@echemi.com with relevant evidence.