A feature selection method for microwave moisture content detection

Abstract

The present application belongs to the technical field of microwave detection and intelligent signal processing, and particularly relates to a feature selection method for microwave moisture content detection, comprising repeatedly sampling a same sample N times under the same environment to obtain N microwave spectrum data samples, and obtaining microwave spectrum data under an unloaded state; decomposing the collected microwave spectrum data of multiple groups of same samples into amplitude attenuation spectrum data and phase shift spectrum data; based on the amplitude attenuation spectrum data and the phase shift spectrum data of the microwave spectrum data under the unloaded state, performing abnormal value processing according to a quartile range method to obtain a final microwave spectrum data set. The present application introduces a physical constraint mechanism on the basis of traditional phase unwrapping technology, identifies an initial phase shift reference interval through a sliding window, and breaks through the technical bottleneck that an initial phase shift interval is difficult to capture due to environmental interference. The algorithm can realize high-precision phase unwrapping under a complex environment, and improve unwrapping accuracy and stability.

Description

Technical Field

[0001] This invention belongs to the field of microwave detection and intelligent signal processing technology, specifically relating to a feature selection method for microwave moisture content detection. Background Technology

[0002] Microwave moisture content detection technology, with its advantages of being non-contact, having strong penetration capabilities, and being highly sensitive to polar water molecules, has become an ideal method for detecting moisture content in various media, and has broad application prospects in fields such as agricultural product quality testing, timber drying monitoring, soil moisture monitoring, and forest fire risk early warning. Accurate and real-time acquisition of media moisture content information is of crucial practical significance for industrial production process control, precision agricultural management, and ecological environment safety monitoring. However, the multipath effect and scattering attenuation generated when microwave signals interact with complex media can lead to phase ambiguity; at the same time, the physical contradiction between the penetration depth of high-frequency signals and the resolution of low-frequency signals presents microwave moisture content detection with the dual challenges of signal distortion and information extraction.

[0003] Traditional methods for detecting moisture content have significant limitations in practical applications. For example, electrical detection methods such as capacitance and resistance methods are easily affected by the contact state between the probe and the medium and the ambient temperature, resulting in poor repeatability. Near-infrared spectroscopy (NIR), on the other hand, has limited penetration depth and can only detect the moisture content of the surface layer of the medium, failing to reflect the true internal moisture content.

[0004] In contrast, microwave transmission technology, by simultaneously acquiring amplitude attenuation and phase shift spectral data, can comprehensively capture the dielectric dispersion characteristics of the medium, providing a more sufficient information basis for accurate moisture content detection. However, existing microwave moisture content detection technologies still face the following bottlenecks in real-world, complex environments:

[0005] First, multipath effects and scattering within high-water-content media significantly reduce the signal-to-noise ratio of microwave signals at both ends of the frequency band. End-point noise interference makes it difficult for conventional phase unwrapping algorithms to capture a reliable reference, resulting in whole-cycle jumps in the intermediate stable frequency band; this phase ambiguity phenomenon severely disrupts the physical mapping relationship between microwave phase and water content.

[0006] Secondly, while broadband microwave scanning (e.g., 2.4 GHz–6 GHz) can acquire rich dielectric response information, there is severe information redundancy and multicollinearity among high-dimensional features. Furthermore, variations in dielectric thickness cause shifts in the moisture-sensitive frequency band, further increasing the difficulty of feature selection. Directly inputting full-band features into the prediction model would significantly increase computational complexity and easily introduce noise, leading to model overfitting.

[0007] Therefore, it is urgent to establish a technical system that combines broadband microwave detection with precise phase unwrapping and adaptive feature selection in order to break through the bottleneck of accurate detection of water content in complex media.

[0008] To address the aforementioned technical requirements, this invention proposes a feature selection method for microwave moisture content detection, which solves the problems of increased model prediction deviation caused by phase jumps and model overfitting caused by high-dimensional redundancy of broadband features, providing technical support for high-precision, real-time, and stable monitoring of medium moisture content in complex environments. Summary of the Invention

[0009] To address the increased model prediction bias caused by phase jumps and the model overfitting problem caused by high-dimensional redundancy of broadband features, and to provide technical support for high-precision, real-time, and stable monitoring of medium moisture content in complex environments, this invention provides a feature selection method for microwave moisture content detection, specifically including the following steps:

[0010] Under the same environment, the same sample is sampled N times to obtain N microwave spectrum data samples, and microwave spectrum data under no-load conditions is obtained.

[0011] The collected microwave spectrum data of multiple identical samples were decomposed into amplitude attenuation spectrum data and phase shift spectrum data.

[0012] Based on the amplitude attenuation spectrum data and phase shift spectrum data of microwave spectrum data under no-load conditions, outlier processing is performed according to the interquartile range method to obtain the final microwave spectrum dataset.

[0013] Compared with existing technologies, this invention introduces a physical constraint mechanism on the basis of traditional phase unwrapping technology. By identifying the initial phase offset reference interval through a sliding window, it overcomes the technical bottleneck that makes it difficult to capture the initial phase offset interval due to environmental interference. This algorithm can achieve high-precision phase unwrapping in complex environments, improving the accuracy and stability of unwrapping. Attached Figure Description

[0014] Figure 1 This is a flowchart of a feature selection method for microwave moisture content detection according to the present invention.

[0015] Figure 2 This is a schematic diagram illustrating the application of a feature selection method for microwave moisture content detection according to the present invention.

[0016] Figure 3 This is a comparison diagram of the phase shift spectrum of the present invention;

[0017] Figure 4 This is a regression diagram of the moisture content prediction results in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides a feature selection method for microwave moisture content detection, such as... Figure 1 Specifically, it includes the following steps:

[0020] Under the same environment, the same sample is sampled N times to obtain N microwave spectrum data samples, and microwave spectrum data under no-load conditions is obtained.

[0021] The collected microwave spectrum data of multiple identical samples were decomposed into amplitude attenuation spectrum data and phase shift spectrum data.

[0022] Based on the amplitude attenuation spectrum data and phase shift spectrum data of microwave spectrum data under no-load conditions, outlier processing is performed according to the interquartile range method to obtain the final microwave spectrum dataset.

[0023] Furthermore, the ATOB adaptive unwrapping algorithm is used to compensate for the phase shift spectral data of multiple identical samples that have been repeatedly sampled. Specifically, this includes the following steps:

[0024] The continuous phase offset spectrum data of each group of samples is uniformly discretized to obtain D frequency points, and one frequency point is selected from the D frequency points as a reference value.

[0025] The reference values ​​of all samples constitute the reference sample set. , Let i represent the reference value of the i-th sample, i∈{1,2,…,N}, and select the median of the reference sample as the phase offset correction benchmark.

[0026] Calculate the deviation between each sample reference value and the phase offset correction benchmark, and determine the number of whole cycles that the data set needs to be compensated based on the deviation;

[0027] The phase offset spectrum data of each sample is corrected based on the number of integer cycles that need to be compensated.

[0028] Furthermore, selecting a frequency point from the D frequency points as a reference value includes the following steps:

[0029] Starting from the j-th frequency point, W frequency points form a reference window, and D frequency points together form a D-W+1 reference window. And each reference window contains at least one frequency point;

[0030] Calculate the mean and variance of the phase offset spectrum data at each frequency point within each window. Use the ratio of the variance to the absolute value of the mean as the coefficient of variation for that window. Select the window with the smallest coefficient of variation as the reference window.

[0031] The median of the phase offset spectrum data at the frequency points within the reference window is used as the reference value.

[0032] Furthermore, based on the amplitude attenuation spectrum data and phase shift spectrum data of microwave spectrum data under no-load conditions, outlier processing is performed according to the interquartile range method, specifically including the following steps:

[0033] For a given frequency, extract the phase offset spectrum data of all samples at that frequency to form a set. , represented as , This represents the phase offset spectrum data of the current frequency point of the i-th sample, where i∈{1,2,…,N};

[0034] set The elements in the set are arranged in ascending order to determine the set. The minimum and maximum values ​​of the phase offset spectrum data in the elements are used to obtain the value at 25% of the range [minimum value, maximum value] as the first quartile and the value at 75% as the third quartile.

[0035] The difference between the third quartile and the first quartile is used as the interquartile range. Based on the third quartile, the first quartile, and the interquartile range, the lower bound and the upper bound of the normal fluctuation at the current frequency point are calculated.

[0036] If the phase offset spectrum data of N samples at the current frequency point is less than the lower bound of normal fluctuation or greater than the upper bound of normal fluctuation, then the phase offset spectrum data of the current frequency point of that sample is abnormal. This can be analyzed using set theory. Replace the current frequency point of the sample with the median of the spectrum.

[0037] This embodiment initializes the physical constraints for phase unwrapping. These constraints include: a negative constraint where the phase offset spectral data is negative; a frequency correlation constraint where the absolute value of the phase offset spectral data is positively correlated with the frequency; a thickness correlation constraint where the absolute value of the phase offset spectral data is positively correlated with the medium thickness; and a moisture content correlation constraint where the absolute value of the phase offset spectral data is positively correlated with the moisture content. In an unloaded state without the medium to be measured, microwave amplitude attenuation spectral data and phase offset spectral data collected by the microwave antenna are acquired as unloaded reference spectral data. In a state with the medium, multiple sets of repeatedly sampled phase offset spectral data and amplitude attenuation spectral data are acquired to construct a phase offset correction reference. Specifically, this includes:

[0038] S11: Establish a negative value constraint condition, where the phase offset spectrum data is negative, i.e. ;

[0039] S12: Establish frequency correlation constraints for any two frequency points. The absolute value of the phase offset spectrum data satisfies Furthermore, the absolute value of the phase shift spectrum data is positively correlated with the frequency, i.e. ;

[0040] S13: Establish thickness-related constraints, when the moisture content... When the value is constant, the absolute value of the phase shift spectrum data is positively correlated with the medium thickness, that is... ;

[0041] S14: Establish moisture content-related constraints, when the medium thickness... When the value is constant, the absolute value of the phase shift spectrum data is positively correlated with the water content, that is... .

[0042] To meet the physical constraints of phase unwrapping mentioned above, this embodiment performs uniform discretization on the continuous phase offset intervals of the multiple sets of repeatedly sampled phase offset spectrum data at a preset step size, and performs phase unwrapping on each set of repeatedly sampled phase offset spectrum data based on the physical constraints; calculates the sliding window size according to the proportion of the preset window length to the entire frequency band, and constructs a sliding window based on this sliding window size; and determines the initial phase offset reference interval by traversing the unwrapped phase offset spectrum data based on the sliding window. Specifically, the specific process of uniform discretization and determination of the initial phase offset reference interval in this embodiment includes:

[0043] S21: Perform uniform discretization on the continuous phase offset spectrum data according to the preset step size to obtain discrete phase offset spectrum data;

[0044] S22: Calculate the sliding window size based on the dimension of the discrete phase shift spectrum data and the preset window ratio. Based on the sliding window size, traverse the entire frequency band and calculate the coefficient of variation of the unwrapped phase shift spectrum data of multiple sets of repeated samples within the sliding window. Select the coefficient of variation The smallest window interval is used as the initial phase offset reference interval.

[0045] In this embodiment, the coefficient of variation is selected. The smallest window is used to find the most phase-stable interval across the entire frequency band, serving as a reference for subsequent alignment. The coefficient of variation is selected. The process of minimizing the window includes:

[0046] (1) Calculation of sliding window size:

[0047] Let the dimension (i.e., the total number of frequency points) of the discrete phase shift spectrum data acquired by microwave broadband scanning be D (e.g., D=201 in the experiment). Let the preset window ratio be... (For example If the sliding window must satisfy at least the following frequency points within the window: ,in This indicates rounding down to the nearest integer, ensuring that the window includes at least one frequency point.

[0048] (2) Calculation of the coefficient of variation (CV):

[0049] Let the starting frequency index of the sliding window during the traversal process be j (where... The continuous phase data sequence contained within this window is ; Calculate the mean and variance of the phase offset spectrum data at each frequency point within each window, i.e.:

[0050]

[0051]

[0052] in, This is the phase offset spectrum data for the k-th frequency point in the window; This represents the mean of the phase offset spectrum data of all frequency points within a window starting from the j-th frequency point; Let represent the variance of the phase offset spectrum data of all frequency points within a window starting from the j-th frequency point.

[0053] The coefficient of variation of the window starting from the j-th frequency point Defined as the ratio of the standard deviation to the absolute value of the mean, expressed as:

[0054]

[0055] (3) Determining the reference interval:

[0056] As the window step size is used to traverse the entire frequency band, the values ​​of all windows are calculated. Select The smallest window is used as the initial phase offset reference interval (which represents the interval where multipath effects and endpoint noise are weakest and the signal is most stable), and the median of the phase offset spectrum data at the frequency points within the reference window is used as the reference value.

[0057] As an alternative implementation method, in actual engineering calculations, the median of the phase offset spectrum data of the entire frequency band can be directly selected as the reference value.

[0058] To eliminate "integer-cycle jumps" caused by environmental or random noise between multiple sets of repeated samples and achieve reference alignment, this invention proposes an ATOB (Align To One Band) adaptive unwrapping algorithm to compensate for the phase offset spectral data of multiple sets of identical samples that are repeatedly sampled. The specific steps include:

[0059] (1) Establishing a calibration benchmark:

[0060] The continuous phase offset spectrum data of each group of samples is uniformly discretized to obtain D frequency points. One frequency point is selected from the D frequency points as a reference value. The reference value can be confirmed using any of the methods mentioned above.

[0061] Suppose that under the same moisture content calibration conditions, there are N sets of repeatedly sampled phase shift spectrum data. Extract the phase values ​​of these N sets of data at the aforementioned reference interval (or reference frequency point) to form a set. Calculate the statistical median of set Y as the phase offset correction benchmark. ,Right now:

[0062]

[0063] in, Let i represent the reference value of the i-th sample, where i∈{1,2,…,N}.

[0064] (2) Deviation calculation and whole cycle estimation:

[0065] Regarding the first Group repeated sampling data ( ), calculate its phase value in the reference interval. Compared with the benchmark absolute deviation :

[0066]

[0067] Because phase transitions always begin with For one period, calculate the first... Number of integer periods that need to be compensated for in a set of data (by dividing by) (and rounded to the nearest integer), that is:

[0068]

[0069] in, This represents the number of integer cycles that the i-th sample needs to compensate for; This represents the rounding function; This represents the deviation between the i-th sample reference value and the phase offset correction benchmark.

[0070] (3) Integer cycle offset correction:

[0071] Based on the number of integer cycles requiring compensation, the phase shift spectrum data of each sample is corrected, i.e.:

[0072]

[0073] in, This represents the phase offset spectrum data of the i-th sample after correction; This represents the phase offset spectrum data of the i-th sample.

[0074] Through the above steps, all repeated sampling data are aligned to a stable median benchmark, eliminating the integer periodic ambiguity jumps between groups.

[0075] like Figure 3 The left figure shows the original S21 phase data collected ten times by a microwave spectrum analyzer. The middle figure shows the spectrum obtained by unwrapping the original S21 phase data collected ten times using the existing unwrapping algorithm (Itoh algorithm). The right figure shows the effect of the method of the present invention after compensation and correction. It can be seen that after unwrapping by the existing method, the phase offset spectrum data of the same sample under the same environment are different at different sampling times. This may lead to errors when used directly for subsequent detection. However, after processing by the present invention, the sampling of the same sample under the same environment can be made more stable, which is more convenient for subsequent detection.

[0076] To eliminate extreme outliers that still exist at individual frequency points after alignment, this embodiment performs outlier processing on the amplitude attenuation spectrum data and phase shift spectrum data of microwave spectrum data under no-load conditions using the interquartile range method, specifically including the following steps:

[0077] (1) Calculation of interquartile range (IQR):

[0078] For a specific frequency point, extract the phase data of all N sets of samples at that frequency point to form a set. Sort set X in ascending order of values ​​and calculate the first quartile (i.e., the 1st quartile). Percentile (%) and third quartile (i.e., the third quartile) Percentile value), interquartile range (IQR) is calculated as follows:

[0079]

[0080] in, The interquartile range of the current frequency point; It is the third quartile; It is the first quartile; The tolerance factor is set to 1.5 in this embodiment.

[0081] (2) Determination of outlier boundaries (normal fluctuation range):

[0082] Introducing a tolerance coefficient k (preferably k=1.5 in this invention), the lower bound and upper bound of normal fluctuations for this frequency point data are calculated, namely:

[0083]

[0084]

[0085] in, This indicates the lower bound of normal fluctuations at the current frequency point; This represents the upper bound of normal fluctuations at the current frequency point.

[0086] (3) Outlier removal and replacement:

[0087] Iterate through each data point in set X ,like or If the sample exhibits an abnormal jump at the current frequency point, then the median of set X is calculated. Values ​​that are identified as anomalous jumps will be determined as such. Replace with median Med.

[0088] This embodiment also proposes an application of a feature selection method for microwave moisture content detection. In the detection of litter moisture content, a feature selection method for microwave moisture content detection is used to obtain a microwave spectrum dataset from the sample to be detected. Specifically, as shown... Figure 2 In the specific implementation process, the optimal K features are selected from the microwave spectrum dataset based on the SRFS feature selection method to form the final forward feature set for subsequent water content detection. The specific steps include:

[0089] S51: Calculate the Spearman rank correlation coefficient between each microwave spectrum data feature in the training set and the actual water content calibration value, and sort the features in descending order according to the absolute value of the correlation coefficient to obtain the descending index sequence of correlation coefficients;

[0090] S52: Select the first [number] items sequentially according to the descending index sequence of the correlation coefficients. Each feature constructs a different feature subset;

[0091] S53: For each feature subset, perform 5-fold cross-validation based on the true moisture content calibration value to ensure that multiple sample data with the same true moisture content calibration value are not simultaneously divided into the training fold and the validation fold; in each training fold of cross-validation, train a partial least squares regression model.

[0092] S54: Calculate the mean root mean square error of the validation set for each cross-validation fold to evaluate prediction performance, and the absolute value of the difference between the mean root mean square errors of the training and validation sets to evaluate fit stability; select the number of features for different forward features. The average root mean square error of the validation set and the absolute value of the difference are sorted in ascending order to obtain the corresponding ranks. The ranks of the two indicators are added together to obtain the comprehensive rank. The number of forward features selected that minimizes the comprehensive rank is selected as the optimal number of forward features selected. The corresponding feature subset is the optimal feature subset.

[0093] To verify the technical effectiveness of this invention, a forest litter microwave moisture content dataset was used for testing. Each data point in this dataset contains 201-dimensional amplitude attenuation spectrum data and 201-dimensional phase shift spectrum data, totaling 402-dimensional microwave spectrum data.

[0094] In the experimental design, test samples with significant moisture gradients were prepared under laboratory conditions, ranging from 3.45% to 64.54% and covering 32 moisture content levels. To avoid data leakage and verify the model's generalization ability, the dataset was divided into training and test sets in a 25:7 ratio based on the moisture content calibration value.

[0095] As shown in Table 1, the results indicate that the method of the present invention can effectively improve the prediction accuracy and stability of microwave moisture content detection.

[0096] Table 1 Comparison of model training performance results before and after using the SRFS feature selection method

[0097] Model Style Train RMSE Train R2 Test RMSE Test R2 All 0.5904 0.9989 1.0601 0.9960 SRFS 0.9468 0.9972 0.7801 0.9975

[0098] Specifically, Table 1 shows a comparison of model performance before (i.e., inputting all 402-dimensional features, labeled "All") and after (labeled "SRFS") using the SRFS feature selection method of this invention. Compared to directly using full-band features for modeling, after feature selection using SRFS, the model validation set determination coefficient (R²) increased from 0.9960 to 0.9975, and the root mean square error was significantly reduced. Simultaneously, the difference in root mean square error between the training and test sets was significantly reduced, effectively suppressing overfitting caused by high-dimensional redundant features.

[0099] Figure 4The moisture content prediction results after processing by the phase unwrapping and feature selection method of this invention are presented. The predicted moisture content output by the model is in high agreement with the actual observed values, with data points closely distributed near the 1:1 baseline, and the 95% confidence interval converges. These results demonstrate that the ATOB phase unwrapping algorithm proposed in this invention, combined with the SRFS forward feature selection method, can not only significantly reduce the dimensionality of online computation but also maintain high measurement accuracy and stability, meeting the requirements for moisture content detection.

[0100] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A feature selection method for microwave moisture content detection, characterized in that, Specifically, the following steps are included: Under the same environment, the same sample is sampled N times to obtain N microwave spectrum data samples, and microwave spectrum data under no-load conditions is obtained. The collected microwave spectrum data of multiple identical samples were decomposed into amplitude attenuation spectrum data and phase shift spectrum data. Based on the amplitude attenuation spectrum data and phase shift spectrum data of microwave spectrum data under no-load conditions, outlier processing is performed according to the interquartile range method to obtain the final microwave spectrum dataset.

2. The feature selection method for microwave moisture content detection according to claim 1, characterized in that, The ATOB adaptive unwrapping algorithm is used to compensate for the phase shift spectral data of multiple identical samples that have been repeatedly sampled. The specific steps include: The continuous phase offset spectrum data of each group of samples is uniformly discretized to obtain D frequency points, and one frequency point is selected from the D frequency points as a reference value. The reference values ​​of all samples constitute the reference sample set. , Let i represent the reference value of the i-th sample, i∈{1,2,…,N}, and select the median of the reference sample as the phase offset correction benchmark. Calculate the deviation between each sample reference value and the phase offset correction benchmark, and determine the number of whole cycles that the data set needs to be compensated based on the deviation; The phase offset spectrum data of each sample is corrected based on the number of integer cycles that need to be compensated.

3. The feature selection method for microwave moisture content detection according to claim 2, characterized in that, Selecting a frequency point from D frequency points as a reference value includes the following steps: Starting from the j-th frequency point, W frequency points form a reference window, and D frequency points together form a D-W+1 reference window. And each reference window contains at least one frequency point; Calculate the mean and variance of the phase offset spectrum data at each frequency point within each window. Use the ratio of the variance to the absolute value of the mean as the coefficient of variation for that window. Select the window with the smallest coefficient of variation as the reference window. The median of the phase offset spectrum data at the frequency points within the reference window is used as the reference value.

4. A feature selection method for microwave moisture content detection according to claim 2 or 3, characterized in that, The median frequency band among the D frequency points of each sample is selected as the reference value.

5. The feature selection method for microwave moisture content detection according to claim 2, characterized in that, Calculate the deviation between each sample reference value and the phase offset correction baseline, and determine the number of integer cycles that need to be compensated for this set of data based on this deviation. ； in, This represents the number of integer cycles that the i-th sample needs to compensate for; This represents the rounding function; The deviation between the i-th sample reference value and the phase offset correction benchmark is represented as: , This represents the reference value for the i-th sample. This indicates the phase offset correction reference.

6. A feature selection method for microwave moisture content detection according to claim 2 or 5, characterized in that, Based on the number of integer cycles requiring compensation, the phase shift spectrum data of each sample is corrected, i.e.: ； in, This represents the phase offset spectrum data of the i-th sample after correction; This represents the phase offset spectrum data of the i-th sample; This represents the number of integer cycles that the i-th sample needs to compensate for.

7. The feature selection method for microwave moisture content detection according to claim 1, characterized in that, Based on the amplitude attenuation spectrum data and phase shift spectrum data of microwave spectrum data under no-load conditions, outlier processing is performed according to the interquartile range method, specifically including the following steps: For a given frequency, extract the phase offset spectrum data of all samples at that frequency to form a set. , represented as , This represents the phase offset spectrum data of the current frequency point of the i-th sample, where i∈{1,2,…,N}; set The elements in the set are arranged in ascending order to determine the set. The minimum and maximum values ​​of the phase offset spectrum data in the elements are used to obtain the value at 25% of the range [minimum value, maximum value] as the first quartile and the value at 75% as the third quartile. The difference between the third quartile and the first quartile is used as the interquartile range. Based on the third quartile, the first quartile, and the interquartile range, the lower bound and the upper bound of the normal fluctuation at the current frequency point are calculated. If the phase offset spectrum data of N samples at the current frequency point is less than the lower bound of normal fluctuation or greater than the upper bound of normal fluctuation, then the phase offset spectrum data of the current frequency point of that sample is abnormal. This can be analyzed using set theory. Replace the current frequency point of the sample with the median of the spectrum.

8. The feature selection method for microwave moisture content detection according to claim 1, characterized in that, The process of calculating the lower and upper bounds of normal fluctuations at the current frequency point based on the third quartile, the first quartile, and the interquartile range includes: ； ； in, This indicates the lower bound of normal fluctuations at the current frequency point; It is the first quartile; This is the tolerance factor; The interquartile range of the current frequency point; This represents the upper bound of normal fluctuations at the current frequency point; It is the third quartile.

9. An application of a feature selection method for microwave moisture content detection, characterized in that, When detecting the moisture content of litter, a microwave spectrum dataset is obtained from the sample to be detected using a feature selection method for microwave moisture content detection as described in claim 1.

Images