Method for rapidly detecting quality of potatoes in low-temperature storage and normal-temperature sugar recovery processes

Abstract

The invention discloses a method for rapidly detecting the quality of potatoes in low-temperature storage and normal-temperature sugar recovery processes, and belongs to the technical field of agricultural product detection. Obtaining multi-variety colored potato samples under different storage conditions and storage time; physicochemical determination is carried out on four quality indexes of potato starch, soluble solids, moisture and hardness in the storage process; acquiring an absorption coefficient spectrum and a reduced scattering coefficient spectrum of the sample slice in a range of 400-1700nm; based on the absorption coefficient spectrum and the reduced scattering coefficient spectrum, establishing and selecting an optimal quantitative prediction model of each quality index of the potatoes through different spectrum preprocessing, spectrum dimension reduction and feature extraction algorithms and a machine learning regression algorithm; and using the optimal model to rapidly detect the corresponding quality index. The method can provide a new rapid and integrated detection means for the key physical and chemical quality of the potatoes in the low-temperature storage and normal-temperature sugar return treatment processes.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural product detection, and particularly relates to a method for rapidly detecting the quality of potatoes during low-temperature storage and normal-temperature sugar recovery processes. Background Art

[0002] Storage is an important link in maintaining the quality of agricultural products after harvest. During this process, core quality indicators such as the chemical composition and hardness of agricultural products will change significantly, and there are also significant differences in the quality change trends of agricultural products under different storage conditions. Research shows that the water content of sweet potatoes continuously increases during storage at 5°C and 15°C for 21 days, while it remains relatively stable at 30°C; at the same time, all three storage temperatures promote the significant accumulation of soluble solid content (SSC), and the lower the storage temperature, the higher the SSC increase; during the 45-day storage of potatoes at 25°C, their water content and hardness show a continuous downward trend, while the SSC shows a trend of first decreasing and then increasing; after 4 months of storage of potatoes at 7°C and 10 - 15°C, the starch and water contents both decrease to varying degrees, and the starch degradation in the low-temperature storage group is more significant. The above studies all show that there are obvious differences in the physical and chemical changes of potatoes under low-temperature and normal-temperature storage conditions. This is because potatoes are high-starch crops, and under low-temperature stress, the sugar metabolism balance in their tissues is disrupted, promoting the conversion of starch into reducing sugars. During industrial processing, to avoid the browning reaction caused by reducing sugars, potatoes stored under refrigeration are usually placed at normal temperature for rewarming treatment, so that the accumulated reducing sugars are resynthesized into starch before further processing to improve product quality.

[0003] The chemical composition in the tissue of agricultural products is related to its absorption characteristics, and the physical structure is related to the scattering characteristics. This association also provides a theoretical basis for the non-destructive detection of the quality of post-harvest agricultural products based on optical detection technology. As a commonly used optical detection technology, the double integrating sphere technology can effectively separate the absorption and scattering information of samples on the basis of the advantage of non-destructive detection, thereby establishing a correlation model between optical parameters and quality indicators. Currently, it has been widely used in the field of quality monitoring during the storage of agricultural products. However, current research mainly focuses on common fruit and vegetable categories, and there is no reported systematic research on the evolution law of the optical properties of multi-variety potatoes, especially colored potatoes rich in anthocyanins, during storage.

[0004] In view of the fact that there is little research on the quality change mechanism of multi-variety potatoes under the "low-temperature saccharification-ambient-temperature sugar recovery" composite storage mode at present, especially there is no report on dynamically monitoring the quality dynamic changes of the whole storage process by using optical detection technology. Therefore, in this invention, three potato varieties, namely "Black King Kong", "Red Beauty" and "Qingshu 9", are taken as the research objects, and a composite storage mode of 4°C low-temperature storage (28 days) and 25°C ambient-temperature storage (24 days) is adopted. The absorption and scattering characteristics of the samples are obtained through the double integrating sphere technology, and the optimal rapid quantitative prediction models of the main qualities of potatoes (starch, SSC, moisture, hardness) are established and selected. Summary of the Invention

[0005] The present invention aims at potato samples of different varieties, including but not limited to yellow (Qingshu 9), red (Red Beauty) and purple (Black King Kong) samples, and rapidly, simultaneously and non-invasively detects their key physical and chemical qualities at different time nodes during the low-temperature storage and ambient-temperature sugar recovery processes, and provides a rapid quality detection method for multi-variety colored potatoes during low-temperature storage and ambient-temperature sugar recovery based on absorption and scattering characteristics. This method belongs to the field of agricultural product detection technology and can provide a new technology for the simultaneous detection of multiple qualities during the potato storage process.

[0006] For the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A rapid quality detection method for potatoes during low-temperature storage and ambient-temperature sugar recovery, comprising:

[0008] Obtaining multi-variety colored potato samples under different storage conditions and storage times;

[0009] Physically and chemically measuring four main quality indexes of potato samples during the storage period, namely starch, soluble solids, moisture and hardness; and obtaining the absorption coefficient spectrum and reduced scattering coefficient spectrum of potato sample slices in the range of 400-1700 nm;

[0010] Based on the absorption coefficient spectrum and the reduced scattering coefficient spectrum, through the combination of different spectral preprocessing, spectral dimensionality reduction and feature extraction algorithms, and machine learning regression algorithms, establish and select the optimal quantitative prediction models of each quality index of potatoes; use the optimal models to rapidly detect the corresponding quality indexes.

[0011] As a preference of the present invention, the process of obtaining the colored potato samples includes:

[0012] 1) Obtaining colored potatoes of different varieties. For example, the varieties include purple potato "Black King Kong", red potato "Red Beauty" and yellow potato "Qingshu 9";

[0013] 2) Wash the potatoes, dry them, and store them at low temperature in a 4°C cold storage. Samples are taken on the 7th, 14th, 21st, and 28th days of cold storage respectively. A number of samples of each variety are taken as potato samples for standby;

[0014] 3) Place the samples after 28 days of low-temperature storage in a constant temperature and humidity incubator at 25°C. Samples are taken on the 4th, 8th, 12th, 16th, 20th, and 24th days of warming storage respectively. A number of samples of each variety are taken as potato samples for standby.

[0015] As a preference of the present invention, physical and chemical tests are carried out on the four main storage period qualities of starch, soluble solids, moisture, and hardness of potatoes, including:

[0016] 1) Take samples from the potato samples. After drying, grinding, and sieving the samples taken from the inner and outer medullas, add ethanol and place them in a water bath at 80 - 90°C for 25 - 35 min, preferably 30 min in a water bath at 80°C; after cooling, centrifuge to remove the supernatant. Repeat the above ethanol treatment steps for the precipitate twice, and retain the precipitate for subsequent detection of starch content; add distilled water to the precipitate after evaporation to remove ethanol and place it in a boiling water bath for 10 - 20 min, preferably 15 min; after cooling, add perchloric acid and stir for extraction for 10 - 20 min (preferably 15 min), then add distilled water and centrifuge, and collect the supernatant; after the precipitate is repeatedly extracted with perchloric acid twice, wash it with distilled water several times and centrifuge to extract the supernatant; combine all the extracted supernatants and make up the volume to 50 ml; absorb the shaken supernatant, add anthrone reagent, place it in a boiling water bath for 9 - 11 min, preferably 10 min, and measure the absorbance at 620 nm after cooling; use a glucose standard solution as a standard product to determine the starch content standard curve to calculate the starch index content;

[0017] 2) Take samples from the potato samples. Crush the samples taken from the inner and outer medullas to extract juice, and drop it into an automatic refractometer to measure the soluble solids content. Two points are taken from the inner and outer medullas of each sample for measurement, and the average value is taken as the soluble solids index;

[0018] 3) Take samples from the potato samples. Put the samples taken from the inner and outer medullas into aluminum boxes respectively. First, place them in an oven at 100 - 120°C for 20 - 40 min for fixation, preferably 30 min in an oven at 105°C; then place them in an oven at 60 - 90°C and dry to constant weight, preferably 85°C; weigh the mass of the samples before and after drying to calculate their wet basis moisture content, and take the average value of multiple results of each sample as the moisture index;

[0019] 4) Take samples from the potato samples and conduct a puncture experiment. The puncture depth is 3 - 5 mm, and the puncture speed is 1 - 3 mm / s. Multiple measurements are taken for each sample and the average value is taken as the hardness index. Preferably, the puncture depth is 4 mm and the puncture speed is 2 mm / s.

[0020] Preferably, in the present invention, the process for obtaining the absorption coefficient spectrum and reduced scattering coefficient spectrum of potato sample slices includes:

[0021] 1) After the potato samples taken at different time points during low-temperature storage and normal-temperature sugar recovery are placed at room temperature for 12 h, the potato samples are cut into thin slices with a thickness of 2-4 mm as optical characteristic measurement samples;

[0022] 2) The sample is sandwiched in the middle with a quartz slide to avoid oxidation and browning caused by contact with air. The refractive index of the quartz slide matches that of the sample. Distilled water is filled at the edge of the sample to isolate oxygen and reduce the refractive index mismatch problem at the glass-sample boundary. The thickness of the quartz slide and the sample is measured by a micrometer;

[0023] 3) The transmittance T and reflectance R spectra are collected through a double integrating sphere system;

[0024] 4) Through the inverse adding doubling (IAD) algorithm, based on the measured data of R, T, the refractive indices of the quartz slide and the sample, the thicknesses of the quartz slide and the sample, and the integrating sphere size, the absorption coefficient spectrum and reduced scattering coefficient spectrum of the potato sample slices in the range of 400-1700 nm are calculated.

[0025] Preferably, in the present invention, establishing the quantitative prediction models for each quality index of potatoes and selecting the optimal models for each quality index includes:

[0026] 1) Preprocessing the absorption coefficient spectrum and reduced scattering coefficient spectrum of potato slices by different methods;

[0027] 2) Respectively performing secondary processing on the preprocessed absorption coefficient spectrum and reduced scattering coefficient spectrum by using different spectral dimension reduction and feature selection algorithms;

[0028] 3) Based on the absorption coefficient spectra and reduced scattering coefficient spectra in the 780-1100 nm band and 1100-1670 nm band after different spectral preprocessing, different spectral dimension reduction and feature selection algorithm secondary processing, regression models for starch, soluble solids, moisture, and hardness are established by using different machine learning regression algorithms; each regression model corresponds to any one of the absorption coefficient spectrum and reduced scattering coefficient spectrum, any one of the 780-1100 nm band and 1100-1670 nm band, one or more combinations of spectral preprocessing methods, any one of the spectral dimension reduction and feature selection algorithms, any one of the machine learning regression algorithms, and any one of the quality indices;

[0029] 4) Test the performance of different regression models, obtain the optimal model corresponding to each quality index, and determine the spectral type, wavelength band, spectral preprocessing method, spectral dimensionality reduction and feature selection algorithm, and machine learning regression algorithm required for quantitative prediction of the quality index according to the optimal model.

[0030] Preferably, the preprocessing method is one or a combination of Savitzky-Golay smoothing, multiplicative scatter correction, standard normal variate transformation, and detrending correction.

[0031] Preferably, the spectral dimensionality reduction and feature selection algorithm includes principal component analysis, t-distributed stochastic neighbor embedding, competitive adaptive reweighted sampling, and successive projections algorithm.

[0032] Preferably, the machine learning regression algorithm includes partial least squares regression, random forest, backpropagation neural network, and support vector machine regression.

[0033] Preferably, in the optimal model, the optimal prediction of moisture index and starch index is achieved based on the absorption coefficient spectrum in the wavelength band of 1100-1600 nm, the optimal prediction of soluble solid index is achieved based on the absorption coefficient spectrum in the range of 780-1100 nm, and the optimal prediction of hardness index is achieved based on the reduced scattering coefficient spectrum in the range of 1100-1600 nm.

[0034] Preferably, in the optimal model of moisture index, the spectral preprocessing method is Savitzky-Golay smoothing combined with standard normal variate transformation, the spectral dimensionality reduction and feature selection algorithm is principal component analysis, and the machine learning regression algorithm is partial least squares regression;

[0035] In the optimal model of starch index, the spectral preprocessing method is Savitzky-Golay smoothing combined with multiplicative scatter correction, the spectral dimensionality reduction and feature selection algorithm is competitive adaptive reweighted sampling, and the machine learning regression algorithm is backpropagation neural network;

[0036] In the optimal model of soluble solid index, the spectral preprocessing method is Savitzky-Golay smoothing combined with standard normal variate transformation, the spectral dimensionality reduction and feature selection algorithm is t-distributed stochastic neighbor embedding, and the machine learning regression algorithm is random forest;

[0037] In the optimal model of hardness index, the spectral preprocessing method is Savitzky-Golay smoothing combined with standard normal variate transformation, the spectral dimensionality reduction and feature selection algorithm is competitive adaptive reweighted sampling, and the machine learning regression algorithm is backpropagation neural network.

[0038] The beneficial effects of the present invention are:

[0039] The present invention provides a method for rapidly detecting the quality of potatoes during low-temperature storage and normal-temperature sugar recovery, which solves the problems that there is currently a lack of analysis of the dynamic change laws of the optical and physicochemical properties of potatoes during the combined storage process, a lack of determination of the differences in the optical and physicochemical properties of different potato varieties during storage, and a lack of a multi-quality integrated and rapid detection technology during the storage process of potatoes.

[0040] This method first determines that during the low-temperature storage - normal-temperature sugar recovery process, the starch content of multiple potato varieties decreases during low-temperature storage and shows a trend of first increasing and then decreasing after being transferred to normal-temperature storage; on the contrary, the content of soluble solids (SSC) increases during the low-temperature storage period and first decreases and then increases during the transfer to normal-temperature sugar recovery; the hardness shows a trend of decreasing under low-temperature conditions, increasing first under normal-temperature conditions, and then decreasing later; and the moisture shows a continuous decreasing trend throughout the combined storage process. Secondly, the absorption coefficient spectra (μ a ) of different potato varieties show significant changes and differences during the combined storage process: μ a in the near-infrared region decreases as a whole during the low-temperature storage period due to the continuous degradation of starch, and after being transferred to normal-temperature sugar recovery, μ a of "Black King Kong" and "Red Beauty" shows a characteristic of first decreasing and then increasing, while that of "Qingshu No. 9" shows a characteristic of first increasing and then decreasing. The change of the storage environment significantly affects the reduced scattering coefficient spectrum (μ' s ): μ' s continuously increases during the low-temperature storage stage, and μ' s shows a systematic decreasing trend with the extension of the rewarming time after being transferred to normal temperature.

[0041] Finally, an optimal quantitative prediction model for four indicators, namely starch, moisture, soluble solids, and hardness, during the low-temperature storage and normal-temperature sugar recovery processes of various potatoes is established. Brief Description of the Drawings

[0042] The drawings here are incorporated into the specification and form a part of this specification, showing the embodiments in line with this application, and are used together with the specification to explain the principles of this application.

[0043] Figure 1 is a flowchart of a method for rapidly detecting the quality of potatoes during low-temperature storage and normal-temperature sugar recovery proposed by the present invention.

[0044] Figure 2 is a schematic diagram of the positions of physicochemical and spectral measurements of potato slice samples.

[0045] Figure 3 is the change trends of (a) moisture content, (b) starch content, (c) SSC content, and (d) hardness of three potato varieties at different time nodes during low-temperature storage and normal-temperature sugar recovery.

[0046] Figure 4 are the average spectra of three potato varieties in the range of 400 - 1700 nm at different time points during low - temperature storage and normal - temperature sugar - recovery processes a Average spectrum.

[0047] Figure 5 are the average spectra of μ' of three potato varieties in the range of 400 - 1700 nm at different time points during low - temperature storage and normal - temperature sugar - recovery processes s Average spectrum.

[0048] Figure 6 are the differences in average spectra of μ and μ' of the inner and outer medullas of three potato varieties in the range of 400 - 1700 nm a and μ' s Average spectrum difference.

[0049] Figure 7 are the schematic diagrams of spectra after different pretreatment methods (taking the near - infrared band μ a as an example).

[0050] Figure 8 are the visualization results of variety clustering of three potato varieties based on μ in the range of 380 - 780 nm through (a) PCA, (b) SVD, (c) t - SNE, (d) Autoencoder a

[0051] Figure 9 are the optimal prediction results of (a) moisture content, (b) starch content, (c) SSC content, (d) hardness of three potato varieties during storage

[0052] Figure 10 are the schematic diagrams of the preferred characteristic bands when CARS is used for (a) starch prediction and (b) hardness prediction Detailed implementation method

[0053] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] The present invention proposes a method for rapid detection of the quality of potatoes during low - temperature storage and normal - temperature sugar - recovery processes, as Figure 1 shown, and the implementation process is as follows:

[0055] (1) Acquisition of potato samples

[0056] Acquire multi - variety colored potato samples under different storage conditions and storage times;​

[0057] In this embodiment, fresh potatoes of three varieties, namely "Black King Kong" (purple potatoes), "Red Beauty" (red potatoes), and "Qingshu 9" (yellow potatoes), which were harvested in the current season, were purchased from Yuxi, Yunnan. For each variety, 60 potatoes with consistent maturity, uniform size, basically the same shape, no mechanical damage, no pests and diseases, no sprouting, no greening, and no cracked skin were selected.

[0058] The surface dirt was washed clean with running water, wiped dry with a clean absorbent cloth, and then stored in a cold storage (4°C). Samples were taken on the 7th, 14th, 21st, and 28th days of low-temperature storage for subsequent determination of relevant quality indicators and spectral collection.

[0059] The samples stored at 4°C for 28 days were transferred to a constant temperature and humidity incubator at 25°C for normal-temperature sugar recovery experiments. Samples were taken on the 4th, 8th, 12th, 16th, 20th, and 24th days of sugar recovery for subsequent determination of relevant quality indicators and spectral collection.

[0060] The total cycle of the composite storage experiment was 52 days (28 days of low-temperature storage and 24 days of normal-temperature sugar recovery). 10 batches of samples were taken, 6 samples were taken from each batch of each variety, and a total of 178 effective samples of potatoes of three varieties were measured (among them, 2 samples in the last batch of the "Red Beauty" variety could not be detected due to rot).

[0061] (2) Quality detection during storage

[0062] A schematic diagram of the physical and chemical quality of potatoes and the spectral collection positions of slices is as Figure 2 shown. Among them, slices with a thickness of 20 mm were used for hardness measurement, slices with a thickness of 10 mm were used for moisture and starch measurement, slices with a thickness of 10 mm were used for SSC measurement, and slices with a thickness of 3.5 mm were used for spectral collection. In accordance with the illustrated order, each sample was cut into multiple thickness blocks for subsequent determination.

[0063] The moisture content was determined by the direct drying method in GB 5009.3-2016 "National Food Safety Standard - Determination of Moisture in Foods". Three portions of potato tubers weighing 2-3 g were taken near the optical parameter measurement points and placed in aluminum boxes respectively. First, they were blanched in an oven at 105°C for 30 min, and then placed in an oven at 85°C to dry to a constant weight. The mass of the tubers before and after drying was weighed with an electronic balance with a precision of 0.0001 g, and the wet basis moisture content was calculated. The average value of the three results was taken as the final moisture content. As Figure 3As shown in (a), the moisture content of the three potato varieties showed a downward trend during storage. Among them, the moisture content of "Black King Kong", "Red Beauty" and "Qing Potato No. 9" decreased by 3.38%, 0.98% and 1.87% respectively during low-temperature storage (4°C), and the moisture content of the three potato varieties decreased by 2.72%, 3.14% and 3.73% respectively during normal-temperature storage. Generally speaking, the water loss rate during the normal-temperature sugar recovery process was higher than that during low-temperature storage. The relatively large decrease in the moisture content of "Black King Kong" during low-temperature storage may be related to the individual differences of the samples. At the same time, the water loss rates of the two colored potatoes rich in active substances ("Black King Kong" and "Red Beauty") were lower than that of the yellow potato "Qing Potato No. 9" during storage.

[0064] First, the samples were pretreated to separate the soluble sugar and starch in the samples. The specific steps were as follows: The dried samples after moisture determination were crushed and passed through a 40-mesh sieve. 0.1 g of the sieved samples were accurately weighed, 5 ml of 80% ethanol was added, and after being treated in a water bath at 80°C for 30 min and cooled to room temperature, the supernatant was removed after centrifugation at 4000 r / min for 5 min. The above ethanol treatment steps were repeated twice for the precipitate, and the precipitate was reserved for subsequent detection. Then, acid treatment was carried out to decompose the starch in the samples into glucose, and the starch content in the potato tubers was obtained by converting the glucose content. The specific steps were as follows: After the precipitate was evaporated to remove ethanol, 2 ml of distilled water was added and boiled in a water bath for 15 min. After taking out and cooling to room temperature, 2 ml of 9.2 mol / L perchloric acid was added, stirred and extracted for 15 min, 3 ml of distilled water was added, and then centrifuged at 4000 r / min for 10 min to collect the supernatant. After the precipitate was repeatedly extracted twice with 4.6 mol / L perchloric acid, it was washed twice with distilled water and centrifuged to extract the supernatant. All the supernatants were combined and fixed to 50 ml. 0.1 ml of the test solution was mixed with 3 ml of anthrone reagent, boiled in a water bath for 10 min and cooled to room temperature, and then the absorbance was measured at 620 nm using an ultraviolet spectrophotometer. Using the glucose standard solution as the standard product, the above color development experiment steps were repeated to draw a standard curve. The calculation formula for starch content is:

[0065]

[0066] Among them, C is the sugar content (μg) calculated from the standard curve; V is the total volume (ml) of the sample extract; A is the volume (ml) of the extract used during color development; W is the dry weight (g) of the sample; 0.9 is the coefficient for converting glucose to starch; 10 6 is used to convert g to μg.

[0067] As Figure 3As shown in (b), the starch content of the three potato varieties showed an obvious downward trend during low-temperature storage. This is because the starch in the potato tissue was decomposed into reducing sugars under low-temperature stress. During the early stage of sugar recovery at room temperature, sugar metabolism recovered, and the accumulated reducing sugars were resynthesized into starch through gluconeogenesis. The rising amplitudes of the starch content of the three potato varieties were 24.28%, 14.87%, and 12.49% respectively. In the later stage of temperature recovery (after 16 days of storage at room temperature), due to long-term temperature recovery and enhanced respiration, the starch content of the potatoes decreased again.

[0068] Crush the potato slices near the spectral detection point to extract juice, and drop it into an automatic refractometer to measure the soluble solid content (SSC). Two points are taken from the inner and outer medulla of each sample for measurement, and the average value is taken as the final analysis value. As Figure 3 As shown in (c), the SSC content of the three potato varieties showed an obvious upward trend during low-temperature storage. This is because the reducing sugars decomposed from starch under low-temperature conditions are the main components of SSC in potato tubers, accounting for as high as 80-90%. Therefore, with the decrease of starch content under low-temperature conditions, the SSC content gradually increases. At the same time, during the early stage of sugar recovery at room temperature, while consuming reducing sugars to resynthesize starch, the SSC content also decreases. The respiration of potato tubers also requires soluble sugars for function, resulting in a decrease in SSC content. On the 12th to 16th day of sugar recovery at room temperature, the SSC content of "Hongmeiren" and "Qingshu 9" increased suddenly, rising by 9.27% and 10.28% respectively. This is because long-term storage damaged the cell membrane of the tubers, releasing starch hydrolase to decompose starch into soluble sugar substances again. On the 24th day of sugar recovery at room temperature, due to the consumption of soluble substances during germination, the SSC content of the samples decreased again. The SSC of "Heijingang" showed a stable decreasing trend during storage at room temperature and did not germinate. This may be due to its relatively high content of antioxidant substances and overall stability.

[0069] Use TA.XTplus texture, select the p / 5 probe, and conduct a puncture experiment on potato tubers. The puncture depth is 4 mm, and the speed before and after the test is 2 mm / s. Each potato tuber sample is measured five times, and the average value is taken for analysis. As Figure 3 As shown in (d), during low-temperature storage, the hardness of the three potato varieties decreased by 1.3%, 2.38%, and 1.58% respectively. This may be due to low-temperature saccharification and slow water loss, resulting in the softening of the cell wall structure. During storage at room temperature, temperature recovery promoted the resynthesis of sugars into starch, and the cell structure recovered. The hardness of the three potato varieties increased by 2.12%, 1.45%, and 2.52% respectively. In the later stage of storage at room temperature, the germination of potato tubers, the consumption of starch, and the intensification of microbial activities led to the disintegration of tissue cells, resulting in a continuous decrease in hardness.

[0070] The above results show that during low-temperature storage and normal-temperature storage, significant dynamic changes occur in the internal quality of different varieties of potatoes. Different storage conditions also vary in their effects on potato quality. Among the three types of potatoes, the purple potato "Black King Kong" exhibits the best storage stability during storage.

[0071] Frying is the most common processing method for potatoes. A low starch content leads to high oil consumption during processing, while a high reducing sugar content is prone to the Maillard reaction, which not only causes uneven product color but also produces harmful chemicals. During low-temperature storage, potatoes undergo low-temperature saccharification, where starch decomposes and its content decreases while the reducing sugar content increases, which has a significant impact on the quality of potato chips. Table 1 shows the chromaticity values of potato chips stored at 4°C for 7 days and 28 days. It can be seen that as the low-temperature storage time increases, the brightness values (L*) of potato chips of all three varieties show a relatively obvious downward trend, while a* and b* both increase. This indicates that as the low-temperature storage time increases, the increase in the reducing sugar content in potato slices leads to browning on the surface of the chips and a decrease in brightness. At the same time, for colored potatoes, high temperatures also cause thermal degradation of anthocyanins in the tissue, which further leads to deterioration of the chip color.

[0072] Table 1 Color of potato chips at different storage times (stored at 4°C for 7 days and 28 days)

[0073]

[0074] (3) Acquisition of absorption and scattering characteristics

[0075] Samples are taken out of the cold storage and artificial climate chamber at different time points during low-temperature storage and normal-temperature sugar recovery. After being placed at room temperature for 12 h, potato tubers are cut into slices about 3.5 mm thick for the determination of absorption and scattering characteristics;

[0076] The samples are sandwiched between quartz glass slides to prevent large-area contact with air and cause oxidative browning. The refractive index of the glass slides used is 1.5, and the refractive index of the potato samples is 1.34. Trace amounts of distilled water are filled at the edges of the samples to isolate oxygen and reduce the refractive index mismatch problem at the glass-sample boundary. The thickness of the glass slides and samples is measured using a micrometer;

[0077] Transmittance (T) and reflectance (R) spectra are collected through a double integrating sphere system; the spectral collection positions are as Figure 2 shown, with three sampling points set in the middle of the slices;

[0078] Through the inverse adding-doubling algorithm (IAD), based on the measured data of R, T, refractive index, thickness of the glass slide and potato, and the size of the integrating sphere, etc., the absorption (μ a ) and reduced scattering coefficient (μ’ s ) spectra of potato slices in the range of 400 - 1700 nm are calculated; the specific numerical values of each input parameter are shown in Table 2;

[0079] Table 2 Input parameters of the IAD algorithm

[0080]

[0081] The μ a and μ’ s spectra of different potato varieties at different storage time nodes are averaged and calculated to analyze the dynamic change laws and differences of the optical properties of potatoes during low-temperature storage and normal-temperature sugar recovery processes.

[0082] Figure 4 is the change trend of μ a of three potato varieties during storage. It can be seen that there are significant differences in μ a of the three potato varieties in the ultraviolet-visible light band. The absorption spectra of fruit and vegetable tissues in the ultraviolet-visible light band are mainly caused by pigments (such as chlorophyll, anthocyanins, and carotenoids), and at the same time, polyphenolic compounds (such as chlorogenic acid and flavonols) in colored potatoes also have distinct characteristic absorption peaks in the ultraviolet-visible spectral region. Specifically, in the ultraviolet region, the absorption peaks mainly come from chlorogenic acid (320 - 330 nm), flavonols (360 - 380 nm), and carotenoids (420 - 480 nm); while in the visible light region, the absorption peaks of anthocyanins and chlorophyll are located at 500 - 550 nm and 640 - 680 nm respectively. Due to the rich anthocyanins, both colored potato varieties "Black King Kong" and "Red Beauty" have obvious absorption peaks in the 500 - 550 nm band, and the slight shift of the peak is also caused by the difference in the anthocyanin composition between the two: the anthocyanin peak of "Black King Kong" is located at 535 nm, while the anthocyanin peak of "Red Beauty" is blue-shifted to 528 nm; at the same time, the acylated structure makes the absorption band of "Black King Kong" extend to 400 - 750 nm, which is wider than that of "Red Beauty" (466 - 600 nm). In addition, the absorption peak of "Red Beauty" at 380 nm is caused by the absorption of flavonols in the tuber. Compared with other colored potatoes, purple potatoes have higher anthocyanin content, so "Black King Kong" also shows a higher absorption intensity (μ a = 0.78 - 1.23 mm -1 ) in the anthocyanin characteristic absorption peak band. The "Qingshu 9" potato with yellow flesh has weak absorption peaks only in the range of 380 - 480 nm (μ a = 0.1 - 0.01 mm-1 ) This is caused by the absorption of carotenoids and trace polyphenols in the samples. The absorption peak positions of the three types of potatoes are similar in the near-infrared band, with only slight differences in absorption intensity. The absorption peak near 970 nm is generated by the combination of O-H bonds in water and C-H bonds in SSC; the absorption peak near 1200 nm is caused by the C-H stretching vibration (second overtone) of polysaccharides such as starch and cellulose; the strong peak at 1400 - 1600 nm is a composite broad peak formed by the overlapping of vibration peaks of various substances such as water, protein, and starch.

[0083] Although there are certain differences in the absorption coefficients of the three types of potatoes during different storage periods in the ultraviolet-visible light band, no obvious change pattern is presented. Although the total anthocyanin content in fruits and vegetables will show losses or degradation during low-temperature and normal-temperature storage, most of these losses are concentrated in the epidermis, and the change in anthocyanin content in the pulp is relatively small. The relevant results show that during storage, the change in anthocyanin content in the potato epidermis is obvious, but due to the protection of the outer potato skin, the change in anthocyanin content in the potato flesh is weak.

[0084] The material metabolism and transformation of fruits and vegetables will all affect their optical absorption characteristics. During low-temperature and normal-temperature storage, the μ of potatoes in the near-infrared band (1400 - 1600 nm) a shows a dynamic change trend of first decreasing and then increasing. Specifically, the μ of the three types of potatoes continuously decreases during low-temperature storage a This is because under low-temperature stress, the starch in potato tubers continuously decomposes into reducing sugars. Although the 1400 - 1600 nm corresponds to the composite absorption peak of water and starch, during low-temperature storage, the transpiration of water is slow, and the decrease in starch content is greater, thus resulting in the continuous decrease of μ in this band. a During the normal-temperature sugar-recovery stage, the change patterns of μ of colored potatoes and common potatoes (yellow potato flesh) a are different. At the initial stage of transferring to normal temperature, "Black King Kong" and "Red Beauty" still continue the downward trend in the low-temperature stage, but the absorption coefficient starts to increase after 12 days of normal-temperature storage; the μ of "Qingshu 9" a continues to increase at the initial stage of normal-temperature storage and then decreases after 20 days of normal-temperature storage. This dynamic change is due to the increase in starch content and the enhancement of the absorption peak caused by normal-temperature sugar-recovery, while long-term storage will consume sugar substances, resulting in the attenuation of the absorption coefficient. At the same time, due to the rich antioxidant substances in colored potatoes, their rate of gluconeogenesis and the process of quality deterioration at normal temperature are slower than those of yellow varieties.

[0085] As Figure 5 shown, compared with μ a μ' sNo obvious characteristic peaks were presented. The sharp drop near 920 nm was due to the performance differences between the two spectrometers. Such experimental errors were inevitable and had been reported in previous studies. Based on Mie scattering theory, μ’ s should show a decreasing trend as the wavelength increases. However, the matrix of agricultural products is relatively complex, with diverse internal structures and components, such as containing different substances like pigments, water, starch, etc. These substances have different light scattering and absorption characteristics, which may cause μ’ s to fluctuate at specific wavelengths and not necessarily change monotonically. The experimental results show that in the ultraviolet-visible light region, μ’ s of the three potato samples all showed significant troughs. This was mainly due to the presence of strong absorption media such as anthocyanins and carotenoids, as well as the crosstalk between absorption and scattering in the double integrating sphere system. Compared with "Black King Kong" and "Red Beauty", the yellow potato "Qingshu 9" had a flatter μ’ s in the visible light band due to its lower absorption ability. In the near-infrared region, the scattering trough peak near 970 nm was attributed to the characteristic absorption of water and sugar substances. After that, it followed the Mie scattering law: μ’ s showed a decreasing trend with the increase of wavelength.

[0086] During low-temperature storage, μ’ s of the three potato varieties showed a continuous upward trend, which was closely related to cell structure and porosity. Research shows that the low-temperature environment accelerates the decomposition of starch, leading to the softening of cell walls and the change of microstructure, thus causing a regular increase in μ’ s . After transferring to normal temperature for sugar recovery, the changes in μ’ s of different varieties showed differences: μ’ s of "Black King Kong" and "Red Beauty" continued to decrease, while that of "Qingshu 9" first increased briefly and then continued to decrease. With the increase of storage time, the samples became soft, resulting in a continuous downward trend of μ’ s .

[0087] It can be seen from Figure 6 that the cross-sectional structure of potato tubers mainly consists of periderm, cortex, vascular bundle ring, and pith. The pith is divided into outer pith (cells are dense and rich in starch) and inner pith (cells are loose and rich in water). It can be seen that "Qingshu 9" did not show obvious differences between the outer and inner pith, and as a whole, it showed a radial star-like texture due to the extension of the vascular bundle ring; while the two colored potatoes "Black King Kong" and "Red Beauty" showed obvious differences between the outer and inner pith due to the enrichment of pigments (such as anthocyanins) in the outer pith. This difference could also be observed in the μ a spectrum. The μ a of the inner pith of the two colored potatoes was significantly lower than that of the outer pith, while the μ of the outer and inner pith of "Qingshu 9"a The difference is not significant.

[0088] (4) Calibration and verification of the quality quantitative model

[0089] For the μ of potato slices a and μ' s Spectra are preprocessed. The preprocessing methods include: Savitzky-Golay smoothing (SGS), multiplicative scatter correction (MSC), standard normal variate transformation (SNV), and detrending correction (DET). Schematic diagrams of spectra after different preprocessing methods or their combinations (taking the near-infrared band μ a as an example) are as Figure 7 shown.

[0090] For the preprocessed μ of potato slices a and μ' s spectra, spectral dimensionality reduction and feature selection algorithms are further used to compress and select spectral features. The methods include: principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), competitive adaptive reweighted sampling (CARS), and successive projections algorithm (SPA). Figure 8 Shows the visualization results of clustering three varieties of potatoes, namely "Black King Kong", "Red Beauty", and "Qingzhu 9", using two linear dimensionality reduction methods: PCA, singular value decomposition (SVD), and two nonlinear dimensionality reduction methods: t-SNE, autoencoder, for μ based on 380 - 780 nm. Then, the clustering ability of different dimensionality reduction methods for multi-variety potatoes is compared. It can be seen that all four methods can effectively map high-dimensional data to three-dimensional space to distinguish between varieties, but the two nonlinear dimensionality reduction methods can achieve better clustering effects. As linear dimensionality reduction techniques, PCA and SVD are used to extract the main features in the data and may retain the linear structure of the data. From a (a) and (b), it can be seen that the data points of "Red Beauty" and "Qingzhu 9" are relatively concentrated under these two methods, indicating that their changes in the main components are relatively small, while the data points of "Black King Kong" are more widely distributed, indicating that its changes in the main components are larger. And compared with the SVD clustering results, PCA has a better variety clustering trend on the X-axis. While t-SNE and autoencoder, as nonlinear dimensionality reduction methods, can effectively reveal the complex structure in the data. Figure 8 (c) shows that based on the first three features after t-SNE dimensionality reduction, there is a significant distinction between the two colored potatoes and the yellow potato on the Z-axis, and obvious, non-overlapping clusters can be formed among the three varieties. Autoencoder is a dimensionality reduction method based on neural networks that can learn the nonlinear features of the data and can be obtained from Figure 8 (c). It can be seen that based on the first three features after t-SNE dimensionality reduction, there is a significant distinction between the two colored potatoes and the yellow potato on the Z-axis, and obvious, non-overlapping clusters can be formed among the three varieties. Autoencoder is a dimensionality reduction method based on neural networks that can learn the nonlinear features of the data and can be obtained from Figure 8In (d), it is observed that the data points of "Heijingang" form a straight line, possibly because it is mapped to a relatively simple linear structure during the encoding process, while "Hongmeiren" and "Qingshu 9" retain certain non-linear characteristics and form a clustered distribution. Although there is a distinction among the three varieties, there is a certain overlap between the clusters. In view of this, before the subsequent quantitative modeling process of potato quality, the PCA and t-SNE methods, which have better clustering effects among linear and non-linear dimensionality reduction methods, are respectively used for dimensionality reduction processing of spectral data.

[0091] Based on the μ in two wavelength ranges (780 - 1100 nm, 1100 - 1670 nm) after different spectral preprocessing, different spectral dimensionality reduction, and feature selection algorithms a and μ' s features, regression models for starch, soluble solids, moisture, and hardness are established using different machine learning regression algorithms, including: partial least squares regression (PLSR), random forest (RF), backpropagation neural network (BPNN), and support vector machine regression (SVR).

[0092] The optical properties in the visible light band (400 - 780 nm) are mainly related to pigment substances in potato slices, while the optical signals in the near-infrared band can reflect the internal chemical components and their changes in tissues. Compared with the short-wave near-infrared (780 - 1100 nm), the long-wave near-infrared (1100 - 2500 nm) has stronger tissue penetration ability and can simultaneously obtain the physical structure characteristics and chemical composition information of samples. Therefore, in this invention, quantitative prediction models for moisture, starch, SSC, and hardness of three varieties of potatoes are constructed using the optical property parameters (μ a , μ' s ) in the ranges of 780 - 1100 nm and 1100 - 1670 nm. The optimal prediction results for each index are as Figure 9 shown.

[0093] Table 3 shows the prediction results of the moisture content of three kinds of potatoes during storage based on optical properties. It can be seen that based on μ a can efficiently achieve moisture prediction. Using the spectral preprocessing method of SGS combined with SNV, and extracting μ through PCA in the 1100 - 1600 nm wavelength band a , the established PLSR model can achieve high-precision prediction of moisture during storage (R 2 p = 0.931, RMSEP = 0.005, RPD = 3.819), and the RPD values of its validation set are all higher than 3, confirming that the model has good generalization ability. Further analysis shows that moisture has characteristic absorption responses in both the short-wave (about 970 nm) and long-wave (about 1450 nm) near-infrared regions. Among them, the moisture prediction model established based on the short-wave region has an R2 p > 0.75, but its detection accuracy has obvious limitations compared with the long-wave region, which may be closely related to the insufficient penetration depth of short-wave near-infrared and the spectral interference of sugar substances.

[0094] Table 3 Prediction results of moisture content during storage of three kinds of potatoes based on μ a and μ' s As shown in Table 4, the performance of the starch detection model is closely related to the selection of characteristic bands and the spectral preprocessing method. Among them, the BPNN model constructed by combining the preprocessing method of SGS combined with MSC and CARS to extract characteristics based on the μ

[0095]

[0096] in the 1100 - 1600 nm band shows the best performance in predicting the starch content of three kinds of potatoes (R a 2 p a = 0.879, RMSEP = 2.207, RPD = 2.888). The schematic diagram of the preferred characteristic bands for starch prediction by CARS is shown in Figure 10 (a). It can be seen that the 58 selected characteristic wavelengths are mainly concentrated in the 1400 - 1600 nm band, indicating that the μ a in this band has strong optical response characteristics to starch. Starch molecules show multiple characteristic absorption responses in the near-infrared region, mainly including the overtone vibration of the O-H bond at 700 - 900 nm, the first overtone peak at 920 - 980 nm, and the strong O-H absorption characteristics at 1450 - 1600 nm. The long-wave near-infrared region has stronger penetration ability and can capture the deep structural information of starch molecules more effectively, which may be an important reason for the excellent prediction performance in this band.

[0097] Table 4 Prediction results of starch content during storage of three kinds of potatoes based on μ a and μ' s As shown in Table 5, based on the μ

[0098]

[0099] in the 780 - 1100 nm band, the RF model established by combining SGS with SNV preprocessing and t-SNE feature compression can achieve the best prediction effect for SSC (R a 2 p ​= 0.860, RMSEP = 0.124, RPD = 2.681). As a new visualization technology, t-SNE can effectively reduce the dimension of complex data, and the characteristic absorption peak at 970 nm is generated by the combination of the O-H bond of water and C-H in soluble solids. Therefore, based on μ in the range of 780 - 1100 nm a The established SSC model has the best prediction effect.

[0100] Table 5 Prediction results of SSC content during the storage of three kinds of potatoes based on μ a and μ' s

[0101]

[0102]

[0103] As shown in Table 6, based on μ' in the range of 1100 - 1600 nm s , the BPNN model established by using SGS combined with SNV preprocessing and CARS for feature extraction can achieve the best prediction effect on hardness (R 2 p = 0.841, RMSEP = 27.656, RPD = 2.519). Among them, as shown in Figure 10 (b), when using CARS for hardness prediction, the number of preferred feature variables is reduced to 167, which is 48% of the full band. Since the hardness of the tissue is related to water and starch, the characteristic wavelengths of hardness are concentrated in the long-wave near-infrared band.

[0104] Table 6 Prediction results of hardness during the storage of three kinds of potatoes based on μ a and μ' s

[0105]

[0106] Since the chemical components in agricultural products are related to μ a , and the physical structure is related to μ' s , therefore, based on μ in the 1100 - 1600 nm band a can achieve the best prediction of water and starch content; based on μ in the range of 780 - 1100 nm a can achieve the best prediction of SSC; based on μ' in the range of 1100 - 1600 nm s can achieve the best prediction of hardness. The optimal calibration model for water is the spectral preprocessing method of SGS combined with SNV, and the PLSR model established by using PCA to compress μ a ; the optimal calibration model for starch is the preprocessing method of SGS combined with MSC, and CARS is used to extract μ​​a The BPNN model established based on the characteristics; the optimal correction model for soluble solids is the SGS combined with the SNV preprocessing method, and t-SNE is used to compress μ a The established RF model; the optimal correction model for hardness is the SGS combined with the SNV preprocessing method, and the CARS is used to extract μ' s The BPNN model established based on the characteristics shows that the rapid quality detection method of multi-variety colored potatoes during low-temperature storage and room-temperature sugar recovery based on absorption and scattering characteristics proposed in the present invention can accurately predict the moisture, starch, soluble solids and hardness indicators of potatoes during the composite storage process.

[0107] The above examples are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many variations are possible. All variations that can be directly derived or associated with the contents disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. A rapid quality detection method for potatoes during low-temperature storage and normal-temperature sugar recovery, characterized in that, Including: Obtaining multi-variety colored potato samples under different storage conditions and storage times; Performing physical and chemical determinations on four main storage period quality indicators of the potato samples, namely starch, soluble solids, moisture, and hardness; and obtaining the absorption coefficient spectrum and reduced scattering coefficient spectrum of the potato sample slices in the range of 400 - 1700 nm; Based on the absorption coefficient spectrum and reduced scattering coefficient spectrum, through the combination of different spectral preprocessing, spectral dimensionality reduction and feature extraction algorithms, and machine learning regression algorithms, establishing and selecting the optimal quantitative prediction model for each quality indicator of the potato; using the optimal model to quickly detect the corresponding quality indicators.

2. The rapid quality detection method for the low-temperature storage and normal-temperature sugar recovery process of potatoes according to claim 1, characterized in that, The process of obtaining the colored potato samples described above includes: 1) Obtaining different varieties of colored potatoes; 2) Washing and drying the potatoes, placing them in a 4°C cold storage for low-temperature storage, and sampling on the 7th, 14th, 21st, and 28th days of refrigeration respectively. A number of samples are taken for each variety as potato samples for standby; 3) Placing the samples after 28 days of low-temperature storage in a 25°C constant temperature and humidity incubator for storage, and sampling on the 4th, 8th, 12th, 16th, 20th, and 24th days of warming storage respectively. A number of samples are taken for each variety as potato samples for standby.

3. The rapid quality detection method for the low-temperature storage and normal-temperature sugar recovery process of potatoes according to claim 1, characterized in that The physical and chemical tests on the four main storage period qualities of starch, soluble solids, moisture, and hardness of the potatoes include: 1) Sampling from the potato samples. The samples taken from the inner and outer medulla are dried, ground, and sieved, then ethanol is added and placed in a water bath at 80 - 90°C for 25 - 35 min. After cooling, centrifugation is carried out to remove the supernatant. The above ethanol treatment steps are repeated twice for the precipitate, and the precipitate is reserved for subsequent starch content detection; Distilled water is added to the precipitate after removing ethanol by evaporation treatment and placed in a boiling water bath for 10 - 20 min. After cooling, perchloric acid is added and stirred for extraction for 10 - 20 min, and then distilled water is added and centrifuged, and the supernatant is collected; After the precipitate is repeatedly extracted with perchloric acid twice, it is washed with distilled water multiple times and centrifuged to extract the supernatant; All the extracted supernatants are combined and made up to a constant volume; The shaken supernatant is taken and anthrone reagent is added, and placed in a boiling water bath for 9 - 11 min. After cooling, the absorbance value is measured at 620 nm; A glucose standard solution is used as the standard product, and the starch content standard curve is determined to calculate the starch index content; 2) Sampling from the potato samples. The samples taken from the inner and outer medulla are crushed to extract juice, and dropped into an automatic refractometer for the determination of soluble solids content. Two points are taken from the inner and outer medulla of each sample for determination, and the average value is taken as the soluble solids index; 3) Sampling from the potato samples. The samples taken from the inner and outer medulla are respectively placed in aluminum boxes, first placed in an oven at 100 - 120°C for 20 - 40 min for fixation, and then placed in an oven at 60 - 90°C for drying to constant weight; Weigh the mass of the samples before and after drying, so as to calculate their wet basis moisture content, and the average value of multiple results is taken for each sample as the moisture index; 4) Sampling from the potato samples, performing a puncture experiment, with a puncture depth of 3 - 5 mm and a puncture speed of 1 - 3 mm / s, and the average value of multiple determinations for each sample is taken as the hardness index.

4. The rapid quality detection method for the potato low-temperature storage and normal-temperature sugar recovery process according to claim 1, characterized in that, The process of obtaining the absorption coefficient spectrum and reduced scattering coefficient spectrum of the potato sample slices includes: 1) After the potato samples taken at different time points during low-temperature storage and room-temperature sugar recovery are placed at room temperature for 12 h, the potato samples are cut into thin slices with a thickness of 2-4 mm as samples for optical property measurement; 2) The sample is sandwiched between quartz slides to avoid oxidation browning caused by contact with air. The refractive index of the quartz slide matches that of the sample. Distilled water is filled at the edge of the sample to isolate oxygen and reduce the refractive index mismatch problem at the boundary between the glass and the sample. The thickness of the quartz slide and the sample is measured by a micrometer; 3) The transmittance T and reflectance R spectra are collected by a double integrating sphere system; 4) Through the inverse adding-doubling algorithm IAD, based on the measured data of R, T, the refractive indices of the quartz slide and the sample, the thicknesses of the quartz slide and the sample, and the integrating sphere size, the absorption coefficient spectrum and reduced scattering coefficient spectrum of the potato sample slice in the range of 400-1700 nm are calculated.

5. The rapid quality detection method for the low-temperature storage and normal-temperature sugar recovery process of potatoes according to claim 1, wherein, The establishment of a quantitative prediction model for each quality index of potatoes and the selection of the optimal model for each quality index include: 1) Pretreat the absorption coefficient spectrum and reduced scattering coefficient spectrum of potato slices by different methods; 2) For the pretreated absorption coefficient spectrum and reduced scattering coefficient spectrum, different spectral dimensionality reduction and feature selection algorithms are used for secondary processing respectively; 3) Based on the absorption coefficient spectra and reduced scattering coefficient spectra in the 780-1100 nm band and 1100-1670 nm band after different spectral pretreatments, different spectral dimensionality reduction and feature selection algorithm secondary processing, regression models of starch, soluble solids, moisture and hardness are established using different machine learning regression algorithms; Each regression model corresponds to any one of the absorption coefficient spectrum and reduced scattering coefficient spectrum, any one of the 780-1100 nm band and 1100-1670 nm band, one or more combinations of spectral pretreatment methods, any one of the spectral dimensionality reduction and feature selection algorithms, any one of the machine learning regression algorithms, and any one of the quality indices; 4) Test the performance of different regression models, obtain the optimal model corresponding to each quality index, and determine the spectral type, band, spectral pretreatment method, spectral dimensionality reduction and feature selection algorithm, and machine learning regression algorithm required for quantitative prediction of the quality index according to the optimal model.

6. The rapid quality detection method for the low-temperature storage and normal-temperature sugar recovery process of potatoes according to claim 5, characterized in that, The pretreatment methods are one or more combinations of Savitzky-Golay smoothing, multiplicative scatter correction, standard normal variate transformation, and detrending correction.

7. The rapid quality detection method for the potato during low-temperature storage and normal-temperature sugar recovery according to claim 5, characterized in that, The spectral dimensionality reduction and feature selection algorithms include principal component analysis, t-distributed stochastic neighbor embedding, competitive adaptive reweighted sampling, and successive projections algorithm.

8. The rapid quality detection method for the low-temperature storage and normal-temperature sugar recovery process of potatoes according to claim 5, characterized in that, The machine learning regression algorithms include partial least squares regression, random forest, backpropagation neural network, and support vector machine regression.

9. The rapid quality detection method for the low-temperature storage and normal-temperature sugar recovery process of potatoes according to claim 5, characterized in that In the optimal model described above, the optimal prediction of moisture index and starch index is achieved based on the absorption coefficient spectrum in the 1100 - 1600 nm band, the optimal prediction of soluble solid index is achieved based on the absorption coefficient spectrum in the 780 - 1100 nm band, and the optimal prediction of hardness index is achieved based on the reduced scattering coefficient spectrum in the 1100 - 1600 nm band.

10. The rapid quality detection method for the low-temperature storage and normal-temperature sugar recovery process of potatoes according to claim 9, characterized in that In the optimal model of moisture index, the spectral preprocessing method is Savitzky - Golay smoothing combined with standard normal variate transformation, the spectral dimensionality reduction and feature selection algorithm is principal component analysis, and the machine learning regression algorithm is partial least squares regression; In the optimal model of starch index, the spectral preprocessing method is Savitzky - Golay smoothing combined with multiplicative scatter correction, the spectral dimensionality reduction and feature selection algorithm is competitive adaptive reweighted sampling, and the machine learning regression algorithm is backpropagation neural network; In the optimal model of soluble solid index, the spectral preprocessing method is Savitzky - Golay smoothing combined with standard normal variate transformation, the spectral dimensionality reduction and feature selection algorithm is t - distributed stochastic neighbor embedding, and the machine learning regression algorithm is random forest; In the optimal model of hardness index, the spectral preprocessing method is Savitzky - Golay smoothing combined with standard normal variate transformation, the spectral dimensionality reduction and feature selection algorithm is competitive adaptive reweighted sampling, and the machine learning regression algorithm is backpropagation neural network.

Images