CN104535646A - Method for detecting imperfection of food grains - Google Patents

Abstract

The invention relates to a method for detecting the imperfection of food grains. The method comprises the following steps: acquiring collision sound information of each food grain and a specific object, extracting set characteristic information from the collision sound information, and comparing and judging imperfect grains in the food grains according to the characteristic information. The method disclosed by the invention has the following advantages that an impact excitation sound signal of the food grains and a metal plate in the free falling body process is utilized, and the imperfect grains are detected by virtue of extraction and analysis of the characteristic parameters of the signal, so that the detection method has the non-contact characteristic. In the whole detection process, the physical structures of the food grains do not need to be damaged, so that the detection method has the nondestructive testing characteristic; and moreover, according to the detection method, special requirements and limitations on the types and falling postures of the food grains are avoided, and the universality is high.

Description

A kind of grain seed imperfection detection method

Technical field

The present invention relates to a kind of grain seed imperfection detection method, belong to grain quality online test method.

Background technology

Grain quality weighs the overall target of grain quality.The factor affecting grain quality is diversified, and wherein imperfection seed too high levels is the major reason of restriction grain quality.According to standard GB/T 1351-2008, the embryo of grain seed or endosperm are subject to physical damnification or microorganism encroach, but also have the grain seed of use value to be defined as unsound grain.Unsound grain mainly comprises broken kernel, injured kernel, sprouted kernel, mouldy grain etc.Due to the existence of grain unsound grain, reduce kind quality and the edible quality of grain, also reduce technological quality and the edible quality (as mouthfeel, smell etc.) of grain simultaneously, on the other hand, unsound grain is very easily subject to the infringement of insect and mould, not easily safe storage, so seem particularly important to the accurate inspection of grain unsound grain.

At present the detection of grain unsound grain is mainly relied on according to standard GB/T/T5494-2008 and manually sampling seed is counted, but in the generality process and transportation of grain, unsound grain and normal grain mix, and both ratios are very little, therefore in order to can be objective and accurate evaluation grain quality, need sampling grain seed quantity a lot.But, up to the present, also there is no effective unsound grain detection method, normally used method comprises cursory method, x-ray method, worm's ovum are creeped and the sound-detection method of chewing, carbon dioxide detection method and Immunological Method, but these methods also existing that detection speed is slow, workload large, expensive equipment, only the defect such as quantitatively to detect in various degree.Because the physics of unsound grain its inside compared with normal grain or architectural characteristic change, therefore the present invention proposes the imperfection of collide the acoustic signal detection grain seed produced by detecting seed.

Summary of the invention

The object of this invention is to provide a kind of grain seed imperfection detection method, traditional the detection method of grain seed imperfection is also existed that detection speed is slow, workload large in order to solve, expensive equipment, only quantitatively to detect etc. the problem of defect.

For achieving the above object, the solution of the present invention comprises a kind of grain seed imperfection detection method, comprise the following steps: the collision acoustic information obtaining each grain seed and individually defined thing, extract the characteristic information of setting from collision acoustic information, compare, judge the imperfection of grain seed according to characteristic information.

Described detection method comprises the following steps:

1), for same grain variety, choosing the normal grain of S grain and T grain unsound grain respectively, by obtaining time domain and/or the frequency domain information of each seed respectively, extracting at least one characteristic parameter;

2), using the input of the characteristic parameter of each seed as three layers of BP neural network, the normal grain of described S grain and T grain unsound grain is utilized to train described neural network, to realize carrying out detection and Identification to the imperfection of grain seed.

Following parameter is obtained: the amplitude maximum of collision acoustic information and corresponding maximal value sampled point thereof from the time domain and frequency domain information of each seed; Three parameters of Weibull function: α, β, x ₀; The normalization variance yields of 8 the first short time-window functions with maximum signal amplitude value M1 in 5 the second short time-window functions, M2, M3, M4 and M5; The maximum amplitude of the discrete Fourier transformation frequency spectrum of collision acoustic information and corresponding Frequency point thereof; Spectral magnitude before and after frequency spectrum maximum amplitude corresponding to each 15 Frequency points; The maximal value of first order difference spectral function and the Frequency point of correspondence thereof; At least one characteristic parameter is chosen from parameter.

Weibull functional form is: $Y=\frac{\mathrm{\α}}{\mathrm{\β}}{\left(\frac{x-x_0}{\mathrm{\β}}\right)}^{\mathrm{\α}-1}\exp [-{\left(\frac{x-x_0}{\mathrm{\β}}\right)}^{\mathrm{\α}}],$

Again, $M_{\mathrm{1,0,0}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i,$

$M_{\mathrm{1,0,1}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i(1-\frac{i-0.35}{N}),$

$M_{\mathrm{1,0,3}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(1-\frac{i-0.35}{N})}^3,$

x ₀＝4(M _1,0,3M _1,0,0-M _1,0,1 ²)/(4M _1,0,3+M _1,0,0-4M _1,0,1)，

$\mathrm{\β}=(M_{\mathrm{1,0,0}}-x_0)/\mathrm{\Γ}[\ln \left(\frac{M_{\mathrm{1,0,0}}-{2M}_{\mathrm{1,0,1}}}{M_{\mathrm{1,0,1}}-{2M}_{\mathrm{1,0,3}}}\right)/\ln 2],$

In formula, Г is Г function, and α (>0) is form parameter, and β (>0) is scale parameter, x ₀for the starting point of Weibull function, x is Weibull argument of function, x>x ₀, Y is the amplitude that sampled point x is corresponding, and N is sampling number.

The width of described first short time-window function is 50 points, calculates the variances sigma in described first short time-window function _i ²:

${\mathrm{\σ}}_i^2=\frac{1}{49}\underset{j=1}{\overset{50}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2,$

Wherein, $\stackrel{\‾}{x}=\frac{1}{50}\underset{j=1}{\overset{50}{\mathrm{\Σ}}}{x}_j;$

According to obtain 8 normalization variance yields with

Choose the second window function that 5 width are at 10, from described maximal value sampled point, calculate the maximum amplitude in described 5 the second window functions respectively, be respectively M1, M2, M3, M4 and M5.

Be calculated as follows the first order difference spectrum of frequency spectrum function:

F ' (u)=F (u)-F (u-1), F (u) are frequency spectrum function, and F ' (u) first order difference is composed; Obtain the maximal value of first order difference spectrum and the Frequency point of correspondence thereof.

The methods such as Stepwise Discriminatory Analysis, multiple linear regression and principal component analysis (PCA) are used to choose at least one characteristic parameter from described parameter.

Tool of the present invention has the following advantages:

(1) utilize in grain seed freely falling body process and excite voice signal with the shock of sheet metal, by detecting imperfection to the extraction of this signal characteristic parameter with analysis, therefore this detection method has non-contacting feature.

(2) in whole testing process, without the need to destroying the physical arrangement of grain seed, therefore this detection method has the feature of Non-Destructive Testing.

(3) this detection method does not have special requirement and restriction, highly versatile to the kind of grain seed and whereabouts attitude.

(4) in testing process, without the need to using any chemicals, "dead" material to produce, the Acquire and process of data is completed automatically by computing machine, and therefore the method has the advantages such as pollution-free, high without the need to manual intervention, detection efficiency, real-time online.

(5) the grain seed in the present invention can be any one in the main farm produces such as paddy rice, brown rice, rice, wheat, corn, soybean, peanut, and therefore this detection method has applicability widely.

Accompanying drawing explanation

Fig. 1 is detection system structural drawing;

Fig. 2 is window function nonlinear filtering schematic diagram;

Fig. 3 is that short time-window function variance calculates schematic diagram;

Fig. 4 is that short time-window function maxima calculates schematic diagram;

Fig. 5 is grain seed imperfection overhaul flow chart.

Embodiment

Below in conjunction with accompanying drawing, the present invention will be further described in detail.

A kind of grain seed imperfection detection method, comprises the following steps: the collision acoustic information obtaining each grain seed and individually defined thing, extracts the characteristic information of setting, compare, judge the imperfection of grain seed according to characteristic information from collision acoustic information.

Based on above technical scheme, by reference to the accompanying drawings, provide with next embodiment.

Implement detection method of the present invention, adopt one based on the grain seed detection system of acoustics, comprise light emitting diode 1, photodiode 2, computing machine 3, wave filter 4, amplifier 5, microphone 6 and 7, sheet metal 8, vibratory screening apparatus 9 and charging aperture 10.Grain seed freely falling body clashes into sheet metal and produces voice signal, and this signal is connected into the input end of amplifier after microphone detection, and the signal after amplification is connected into computing machine through high-pass filtering, and the collection signal of its Computer is triggered by photodiode.System architecture as shown in Figure 1.

Three parts can be divided into: (1) signal produces by this system of function.This part is made up of vibratory screening apparatus, charging aperture, sheet metal, and its function mainly produces the acoustic signal relevant with grain seed sophistication; (2) signals collecting.This part is formed primarily of two microphones, amplifier, wave filter, and its function is mainly collected grain seed and sheet metal and collided the voice signal that produces and carry out rough handling to this signal; (3) signal transacting.This part completes primarily of computing machine, and its major function is by extracting the time-frequency of voice signal and frequency domain character parameter, by judge and then detection and indentification goes out the imperfection of grain seed.

System work process is as follows:

Grain seed is from charging aperture by grain free-falling, and light emitting diode and photodiode are with charging aperture center for benchmark, and axis of symmetry is placed, and the light-emitting area of light emitting diode is alignd in the horizontal direction with the light receiving surface of photodiode.There is no seed through out-of-date, because light path is unobstructed, so photodiode stable output photosignal, and when seed is via light path, because seed blocks effect to light, the signal that photodiode is received diminishes, and exports pulse, and this Puled input is to the trigger pip of computer sound card as sound signal collecting starting point.

When on grain seed freely falling body to sheet metal, will collide with sheet metal, and then signal of sounding.

Here sheet metal is the corrosion resistant plate of one piece of polishing, and it is of a size of 7.5 (length) × 5 (wide) × 10 (thick) cm.Compared to sheet metal, the quality of grain seed is negligible, and therefore in knockout process, the vibration of sheet metal collides relative to seed the voice signal sent is very little, ignores.In order to keep the consistance of seed and sheet metal impingement position, charging aperture is 30-50cm to the distance of sheet metal, and in order to prevent seed from producing secondary impact on a metal plate, sheet metal horizontal tilt 30 degree.

By two microphones collected sound signal simultaneously, by amplifier, this signal is amplified, then by wave filter filtering low-frequency noise, and filtered signal is delivered to computing machine and carry out processing and analyzing.

Select a kind of grain variety, choose the normal grain of S grain and T grain unsound grain respectively, the value of S, T is between 800-1000 here, and employing following steps obtain C main characteristic parameters corresponding to each seed.

The randomness of attitude when considering that the otherness of grain seed kind and seed and sheet metal are clashed into, cause the intensity difference of voice signal larger, therefore voice signal is gathered by two microphones with enlarge leadingly simultaneously, wherein the enlargement factor of microphone 6 is 1V/Pa, the enlargement factor of microphone 7 is 10V/Pa, the installation direction of two microphones is parallel with rum point normal direction, distance metal sheet surface 20-25mm, and the voice response frequency of two microphones is within the scope of 0-100KHz.The output signal of microphone is amplified further through amplifier, then by the following low-frequency noise of Hi-pass filter filtering 60KHz and DC noise, finally filtered signal is delivered to computer sound card and carry out A/D conversion, the sample frequency of sound card is 200KHz, and resolution is 16.The trigger pip of sound card is provided by the output pulse of photodiode, once receive trigger pip, computing machine gathers N number of data point (N ∈ 2000-3000) from the digital signal that sound card exports, and the number of data point is suitably chosen according to grain variety.

The selecting step of voice signal:

1. the voice signal that two microphones collect is designated as respectively f1 (x) and f2 (x), wherein the corresponding sampled point of x, be designated as x1, x2 ..., xN.The absolute value of the number of winning the confidence f1 (x) and f2 (x), is designated as | f1 (x) |, | f2 (x) |;

2. calculate respectively | f1 (x) | with | f2 (x) | the quantity of middle saturation point;

If 3. signal | f2 (x) | in (signal corresponding to the enlargement factor microphone that is 10V/Pa collects), saturated counting is less than or equal to 6, then using voice signal that this signal collides as seed, otherwise, using | f1 (x) | (signal corresponding to the enlargement factor microphone that is 1V/Pa collects) is as pending seed clash tone signal.Selected voice signal is designated as f (x), f (x) is f1 (x), f2 (x) one of them.Ask the signal after absolute value to be designated as g (x) to f (x), g (x) is | f1 (x) |, | f2 (x) | one of them.

Calculate voice signal Weibull parameter:

1. the maximum amplitude g of search signal g (x) _maxthe sampled point x of (x) and correspondence thereof _max;

2. nonlinear filtering is carried out to signal g (x), filtering can be described as: utilize a width to be window function and the signal convolution of 7, as shown in Figure 2, maximal value in window function replaces its central value, and wherein initial 3 points (corresponding x1, x2, x3) of signal remain unchanged with the signal value of contiguous 3 points (corresponding xN-2, xN-1, xN) terminated.

That is: j=i-3, i-2, i-1, i, i+1, i+2, i+3, i ∈ [4, N-3]

3. three parameter Weibull functions are used to carry out time domain matching to signal.Three parameter Weibull functional forms are:

$Y=\frac{\mathrm{\α}}{\mathrm{\β}}{\left(\frac{x-x_0}{\mathrm{\β}}\right)}^{\mathrm{\α}-1}\exp [-{\left(\frac{x-x_0}{\mathrm{\β}}\right)}^{\mathrm{\α}}]$

In formula, α (>0) is form parameter, and β (>0) is scale parameter, x ₀for the starting point of Weibull function, x is Weibull argument of function x>x ₀, Y is the amplitude that sampled point x is corresponding.Be the time-domain signal of N for sampling number, being calculated as follows of each parameter:

$M_{\mathrm{1,0,0}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i$

$M_{\mathrm{1,0,1}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i(1-\frac{i-0.35}{N})$

$M_{\mathrm{1,0,3}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(1-\frac{i-0.35}{N})}^3$

x ₀＝4(M _1,0,3M _1,0,0-M _1,0,1 ²)/(4M _1,0,3+M _1,0,0-4M _1,0,1)

$\mathrm{\β}=(M_{\mathrm{1,0,0}}-x_0)/\mathrm{\Γ}[\ln \left(\frac{M_{\mathrm{1,0,0}}-{2M}_{\mathrm{1,0,1}}}{M_{\mathrm{1,0,1}}-{2M}_{\mathrm{1,0,3}}}\right)/\ln 2]$

α＝ln2/ln[(M _1,0,0-2M _1,0,1)/2(M _1,0,1-2M _1,0,3)]

Here, Г is Г function, can obtain three parameters characterizing Weibull function: α, β and x by above formula ₀.

The process of short time-window function variance:

Choose the window function that a width is at 50, as shown in Figure 3, calculate the variances sigma of signal in this window _i ²:

${\mathrm{\σ}}_i^2=\frac{1}{49}\underset{j=1}{\overset{50}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2$

$\stackrel{\‾}{x}=\frac{1}{50}\underset{j=1}{\overset{50}{\mathrm{\Σ}}}{x}_j,$

Here x _jfor the signal amplitude of each sampled point in window, x is the average of all signals in window.Wherein the window width of first window function is 50 sampled points, and window initial point is from the sampled point x corresponding to voice signal maximum amplitude _maxbefore 40 beginnings, then with the initial point of first window function for benchmark, increase progressively with 30 point step sizes and obtain second window function, and calculate the variance of signal in this window.Therefore overlapping 20 sampled data points between adjacent two windows, by that analogy, get 8 windows altogether, calculate respectively the variance of signal in each window and the variance of 8 windows and, the normalization variance then using following formula to obtain each window is:

${\mathrm{\σ}}_{\mathrm{ni}}^2=\frac{{\mathrm{\σ}}_i^2}{{\mathrm{\Σ}}_{i=1}^8{\mathrm{\σ}}_i^2},i=\mathrm{1,2},......,8,$

8 normalization variance yields can be obtained thus with

Calculate short time-window function maxima:

From the sampled point x corresponding to the maximal value of time-domain signal _maxstart, choose the window function that a width is at 10, calculate the maximum amplitude in window function, then with 10 sampled points for step-length moving window, calculate the maximum amplitude of signal in next window, sampled point in adjacent two window functions does not have overlapping, and the maximum amplitude in Using such method Continuous plus 5 window functions, the maximum signal amplitude value of each window function is denoted as M1, M2, M3, M4 and M5 respectively.

Discrete Fourier transformation (DFT) process of voice signal:

From the sampled point x corresponding to the maximal value of time-domain signal f (x) _maxq front sampled point starts to carry out the windowing of Hamming function to signal, and add the time-continuing process that window width should be able to contain grain seed collision sheet metal, the value of Q is between 50-100 here, suitably chooses according to grain variety.Then carry out 256 Fourier transforms to the function after windowing, obtain frequency spectrum function F (u) of signal, wherein u is frequency values, u ∈ [0,100kHz], calculates the maximal value F of frequency spectrum function _maxthe Frequency point u of (u) and correspondence thereof _max.Meanwhile, u is recorded _maxthe spectral magnitude that each 15 Frequency points in front and back are corresponding, amounts to 30 range values, is designated as F _j(u), j=1,2 ..., 30.

Calculate the Difference Spectrum of signal discrete Fourier transform (DFT):

Be calculated as follows first order difference spectrum F ' (u) of frequency spectrum function F (u):

F’(u)＝F(u)-F(u-1)

Search the maximal value F ' of plain Difference Spectrum function _maxthe Frequency point u of (u) and correspondence thereof _cfmax.

It should be noted that, can individually carry out time domain or frequency domain process to the acquisition of parameter, obtain some parameters separately, and select, certainly, also time domain and frequency domain all can process, more parameter can be obtained, and select.

To sum up, according to above-mentioned time domain and frequency domain process, obtain 52 time-frequency characteristics parameters, comprising: signal amplitude maximal value g _max(x) and corresponding sampled point x thereof _max; Three parameters of Weibull function: α, β, x ₀; The normalization variance yields of 8 short time-window functions with maximum signal amplitude value M1 in 5 short time-window functions, M2, M3, M4 and M5; The maximum amplitude F of discrete Fourier transformation (DFT) frequency spectrum of signal _max(u) and corresponding Frequency point u thereof _max; Spectral magnitude F before and after frequency spectrum maximum amplitude corresponding to each 15 Frequency points _j(u), j=1,2 ..., 30; The maximal value F ' of first order difference spectral function _maxthe Frequency point u of (u) and correspondence thereof _cfmax.Use the methods such as Stepwise Discriminatory Analysis, multiple linear regression and principal component analysis (PCA) from above characteristic parameter, choose C (C >=1) individual main characteristic parameters.

Using the input of C characteristic parameter of each seed of the above-mentioned normal grain of S grain chosen and T grain unsound grain as three layers of BP neural network, this neural network comprises 1 input layer, 1 hidden layer and 1 output layer.The normal grain of S grain utilizing this to choose and T grain unsound grain are trained this neural network, obtain the detection model of grain seed imperfection.

So far, detection model completes.

When needing to carry out detection and indentification to any seed of same grain variety, utilizing the above-mentioned detection model completed to carry out detection and indentification these any seeds, grain seed can be detected.Testing process as shown in Figure 5.

Be presented above concrete embodiment, but the present invention is not limited to described embodiment.Basic ideas of the present invention are above-mentioned basic scheme, and for those of ordinary skill in the art, according to instruction of the present invention, designing the model of various distortion, formula, parameter does not need to spend creative work.The change carried out embodiment without departing from the principles and spirit of the present invention, amendment, replacement and modification still fall within the scope of protection of the present invention.

Claims (8)

1. a grain seed imperfection detection method, it is characterized in that, described detection method comprises the following steps: the collision acoustic information obtaining each grain seed and individually defined thing, extract the characteristic information of setting from described collision acoustic information, compare, judge the imperfection of described grain seed according to described characteristic information.

2. grain seed imperfection detection method according to claim 1, it is characterized in that, described detection method comprises the following steps:

1), for same grain variety, choosing the normal grain of S grain and T grain unsound grain respectively, by obtaining time domain and/or the frequency domain information of each seed respectively, extracting at least one characteristic parameter;

2), using the input of the characteristic parameter of each seed as three layers of BP neural network, the normal grain of described S grain and T grain unsound grain is utilized to train described neural network, to realize carrying out detection and Identification to the imperfection of grain seed.

3. grain seed imperfection detection method according to claim 2, is characterized in that, from the time domain and frequency domain information of each seed, obtain following parameter: the amplitude maximum of collision acoustic information and corresponding maximal value sampled point thereof; Three parameters of Weibull function: α, β, x ₀; The normalization variance yields of 8 the first short time-window functions with maximum signal amplitude value M1 in 5 the second short time-window functions, M2, M3, M4 and M5; The maximum amplitude of the discrete Fourier transformation frequency spectrum of collision acoustic information and corresponding Frequency point thereof; Spectral magnitude before and after frequency spectrum maximum amplitude corresponding to each 15 Frequency points; The maximal value of first order difference spectral function and the Frequency point of correspondence thereof; At least one characteristic parameter is chosen from parameter.

4. grain seed imperfection detection method according to claim 3, it is characterized in that, described Weibull functional form is: $Y=\frac{\mathrm{\α}}{\mathrm{\β}}{\left(\frac{x-x_0}{\mathrm{\β}}\right)}^{\mathrm{\α}-1}\exp [-{\left(\frac{x-x_0}{\mathrm{\β}}\right)}^{\mathrm{\α}}],$

Again, $M_{\mathrm{1,0,0}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i,$

$M_{\mathrm{1,0,1}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i(1-\frac{i-0.35}{N}),$

$M_{\mathrm{1,0,3}}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{x}_i{(1-\frac{i-0.35}{N})}^3,$

x ₀＝4(M _1,0,3M _1,0,0-M _1,0,1 ²)/(4M _1,0,3+M _1,0,0-4M _1,0,1)，

$\mathrm{\β}=(M_{\mathrm{1,0,0}}-x_0)/\mathrm{\Γ}[\ln \left(\frac{M_{\mathrm{1,0,0}}-2M_{\mathrm{1,0,1}}}{M_{\mathrm{1,0,1}}-2M_{\mathrm{1,0,3}}}\right)/\ln 2],$

In formula, Г is Г function, and α (>0) is form parameter, and β (>0) is scale parameter, x ₀for the starting point of Weibull function, x is Weibull argument of function, x>x ₀, Y is the amplitude that sampled point x is corresponding, and N is sampling number.

5. grain seed imperfection detection method according to claim 4, is characterized in that, the width of described first short time-window function is 50 points, calculates the variances sigma in described first short time-window function _i ²:

${\mathrm{\σ}}_i^2=\frac{1}{49}\underset{j=1}{\overset{50}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x})}^2,$

Wherein, $\stackrel{\‾}{x}=\frac{1}{50}\underset{j=1}{\overset{50}{\mathrm{\Σ}}}{x}_j;$

According to ${\mathrm{\σ}}_{\mathrm{ni}}^2=\frac{{\mathrm{\σ}}_i^2}{{\mathrm{\Σ}}_{i=1}^8{\mathrm{\σ}}_i^2},i=\mathrm{1,2},......,8,$ Obtain 8 normalization variance yields with

6. grain seed imperfection detection method according to claim 5, it is characterized in that, choose the second window function that 5 width are at 10, from described maximal value sampled point, calculate the maximum amplitude in described 5 the second window functions respectively, be respectively M1, M2, M3, M4 and M5.

7. grain seed imperfection detection method according to claim 6, is characterized in that, is calculated as follows the first order difference spectrum of frequency spectrum function:

F ' (u)=F (u)-F (u-1), F (u) are frequency spectrum function, and F ' (u) first order difference is composed; Obtain the maximal value of first order difference spectrum and the Frequency point of correspondence thereof.

8. grain seed imperfection detection method according to claim 7, is characterized in that, uses the methods such as Stepwise Discriminatory Analysis, multiple linear regression and principal component analysis (PCA) to choose at least one characteristic parameter from described parameter.