CN107993244B

CN107993244B - Automatic corn detection method

Info

Publication number: CN107993244B
Application number: CN201711440167.XA
Authority: CN
Inventors: 李维; 江淮; 韦文波; 杨世举; 孔园林
Original assignee: Hefei Asorting Intelligent Technology Co ltd
Current assignee: Hefei Asorting Intelligent Technology Co ltd
Priority date: 2017-12-27
Filing date: 2017-12-27
Publication date: 2020-09-11
Anticipated expiration: 2037-12-27
Also published as: CN107993244A

Abstract

The invention relates to an automatic corn detection method, which comprises the following steps: a) detecting the undersize; b) detecting the integral color of the corn kernels; c) detecting the breakage of corn kernels; d) detecting wormholes; e) detecting mildew; f) and (5) detecting surface cracks. The method greatly reduces the detection cost and improves the detection efficiency.

Description

Automatic corn detection method

Technical Field

The invention relates to grain safety detection, relates to corn, and particularly relates to an automatic corn detection method.

Background

Corn is a major food crop for the nation. In the process of collecting and storing the corns, the corns have the phenomena of mildew, fragmentation, wormholes and the like, which are very common phenomena. The abnormal corns can affect the health of the human body and must be detected and filtered.

The conventional method for manually detecting the corns one by one has high cost and low detection efficiency.

Disclosure of Invention

The invention aims to provide an automatic corn detection method, which greatly reduces the detection cost and improves the detection efficiency.

In order to achieve the above purpose, the invention adopts the technical scheme that: an automatic corn detection method comprises the following steps:

a) and (3) undersize detection: capturing a background image; the current image and the background image are subjected to subtraction; detecting the foreground to obtain pixel points corresponding to the corn kernels; detecting a connected domain to obtain the size of the corn kernels; filtering the small corn kernels compared to an empirical threshold;

b) and (3) detecting the integral color of the corn kernels: extracting the outer contour of the corn kernels; positioning a tip point, wherein the tip point is a position where the traversal contour is traversed and the included angle value on the corn kernels is maximum; detecting an endosperm area, wherein the endosperm area is a groove-shaped area with higher brightness value on the surface of the corn kernels; detecting a cotyledon part, wherein the cotyledon part is the residual area of the corn kernels except the endosperm part; extracting a color value;

c) corn kernel damage detection: extracting the outer contour of the corn kernels; detecting the smoothness of the outer contour; analyzing the symmetry of the corn kernels;

d) detecting wormholes: configuring the angle of a light source and the brightness of LED light to ensure that the whole surface of the corn kernels becomes white; at the hole position, the light can not be reflected, so that the light becomes a black hole; carrying out binarization on the image in a kernel region of the corn to change the black holes into a white foreground; detecting a connected component; finding all connected domains with the size larger than a threshold value; if the number of the connected domains is greater than or equal to 1, holes exist in the corn kernels, and early warning is given at the moment;

e) and (3) mildew detection: obtaining an endosperm region by using a region growing method; extracting the endosperm region; analyzing the brightness of the pixel value corresponding to the endosperm area to obtain a histogram; when a mildew point exists, two peaks appear in the luminance histogram; finding a classification threshold between two peaks; carrying out binarization on the endosperm region by using the threshold value found in the previous step; analyzing the connected domain to obtain the maximum connected domain; if the maximum connected domain width and the maximum connected domain height are both larger than the threshold value of 20 pixels, the mildew point exists, and early warning is performed at the moment;

f) and (3) surface crack detection: performing edge extraction on the surface of the corn kernels; short edges of filter length < 50 pixels; for each long edge, a gradient analysis was performed.

Further, capturing a background image in the step a) before corn kernel detection, the background image is obtained by the following steps:

a11) continuously shooting images against a background by using a camera;

a12) detecting the image of the period of time, judging whether a moving target exists, and if so, shooting again;

a13) if the image segment is still, the background image is obtained by calculating the mean value of all the images.

Further, the acquisition step of the groove-shaped region detected in the embryo area in the step b) comprises the following steps:

b31) searching areas with approximate brightness values outwards from the corn tip;

b32) adopting a recursive method to expand outwards step by step;

b33) stopping until the boundary is met by expanding;

b34) all of the expanded regions are then portions of endosperm.

Further, the method for detecting the smoothness of the outer contour in the step c) comprises the following specific steps:

c21) traversing each contour point;

c22) finding the 5 th point in front and the 5 th point in back of the current contour point;

c23) respectively calculating included angles formed by the two points and the current point;

c24) and traversing included angle values of all the points, and if a certain included angle value is less than 120 degrees, giving an early warning, wherein the contour has no smoothness.

Further, the corn kernel symmetry analysis in the step c) comprises the following specific steps:

c31) traversing the contour, and finding out the position with the maximum included angle value on the corn kernels as a tip point;

c32) sending out from the tip point, launching a straight line, and connecting with any contour point;

c33) calculating the symmetry of the image contents on the left side and the right side by taking the current connecting line as a symmetry axis;

c34) finding the symmetry axis with the largest number of pairs;

c35) if the number of the pixels matched with the maximum symmetry is less than 80% of the total number, the corn kernels are not complete, and a part of the corn kernels are incomplete.

Further, the step of performing gradient analysis on each long edge in the step f) comprises the following steps:

f31) traversing each edge point;

f32) collecting pixels along the direction vertical to the edge for analysis;

f33) recording pixel point positions of two sides of the edge, wherein the gradient value of the pixel points is smaller than 10, and using the pixel point positions as boundary points of the edge;

f34) obtaining the edge width corresponding to the current edge point according to the coordinate difference of the edge boundary points on the left side and the right side;

f35) if the edge width of all points of the current edge is less than 6 pixels, the line is considered as a crack, otherwise, the line is an endosperm edge;

f36) in case of cracks, a warning is required.

The invention has the technical effects that: the invention provides a method for automatically detecting the quality of corn kernels by using machine vision, which greatly reduces the detection cost and improves the detection efficiency.

Detailed Description

The method comprises the following specific steps:

a) and (3) undersize detection:

some corn kernels are very small in size. Such kernels are of no practical value. And therefore needs to be filtered out in advance.

In the invention, the following method is adopted to detect the small-size corn kernels:

a1, capturing background image

Before corn kernel detection is carried out, a background image needs to be acquired in advance. The background image acquisition method comprises the following steps:

a11) with the camera, images are continuously taken against the background for a period of time.

a12) And detecting the image in the period of time to determine whether a moving object exists. If so, a re-shot is required.

a2, making difference between current image and background image

a3, detecting the foreground to obtain pixel points corresponding to the corn kernels

The current map is compared with the background map pixel by pixel. If the difference between the brightness value of the current pixel point and the brightness value of the background pixel point at a certain pixel position is larger than a threshold (the threshold is defaulted to be 200), the pixel is marked as a foreground pixel.

a4, detecting a connected domain to obtain the size of the corn kernels

And (4) carrying out connected domain analysis on foreground pixels in the binary image, and extracting the connected domain with the largest size to be used as the corn kernels. Then, the minimum bounding rectangle of the connected domain is calculated, and the width and the length of the rectangle are calculated.

a5, filtering small-size corn kernels compared with empirical threshold

If the width and length of the kernel are both less than the empirical threshold (100 pixels), then it is necessary to filter as a small size kernel.

b) And (3) detecting the integral color of the corn kernels:

some corn kernels are blackened and darkened as a whole due to pathological changes, mildewing or high temperature in the processing process. The kernels are inedible and need to be filtered out in advance.

The specific method comprises the following steps:

b1, extracting the outer contour of the corn kernels.

And (4) analyzing a connected domain on the basis of the binary image, and further extracting contour points.

b2, tip point positioning.

And traversing the contour, and finding the position with the maximum included angle value on the corn kernels as a tip point.

b3, endosperm region detection.

The endosperm region is a groove-shaped region on the surface of the corn kernel with higher brightness value. Since corn blackening can often occur in the cotyledon portion, the endosperm region needs to be filtered out in advance to reduce the influence of the region on color detection.

And (3) obtaining a groove region by adopting a region growing method:

b31) starting from the corn tip, search outwards for areas with close brightness values.

b32) And adopting a recursive method to gradually expand outwards.

b33) Stopping until the boundary is reached by expansion or the brightness value differs greatly from the tip position.

b34) All of the expanded regions are then portions of endosperm.

b4, and detecting cotyledon parts.

The remaining area of the kernel, excluding the endosperm portion, is entirely the cotyledon portion.

b5, extracting color values.

In the cotyledon portion, the average color value of the kernel is calculated. At this time, if the values of the three RGB components of the average color value are all smaller than the threshold value (50), the corn kernels are considered to be blackish, and early warning is needed.

c) Corn kernel damage detection:

during the processing, the corn grains are easily damaged due to the mechanical pressure applied to the corn grains. The method for detecting the damaged corn comprises the following steps:

c1, extracting the outer contour of the corn kernels.

After the connected domain corresponding to the corn kernels is obtained, all contour points are extracted in sequence.

c2, and detecting the smoothness of the outer contour.

The whole corn kernels have better smoothness in outline. This is because the surface of the corn itself is smooth. However, the outer contour of a damaged corn tends to have sharp edges and corners. By calculating the included angle of the front and rear contour points of each point of the outer contour, the abnormality on the outer contour can be detected.

The specific method comprises the following steps:

c21) each contour point is traversed.

c22) The 5 th point in front and the 5 th point in back of the current contour point are found.

c23) And respectively calculating the included angles formed by the two points and the current point.

c24) And traversing included angle values of all the points, and if a certain included angle value is too small, giving an early warning, wherein the contour does not have smoothness.

c3, and analyzing the symmetry of the corn kernels.

The complete corn kernels have better left-right symmetry. Whether the current corn kernels are normal or not can be identified through symmetry analysis of the appearance of the corn kernels and detection of the symmetry axis.

The specific steps of symmetry analysis are as follows:

c31) and traversing the contour, and finding the position with the maximum included angle value on the corn kernels as a tip point.

c32) The straight line is emitted from the tip point and is connected with any contour point.

c33) And taking the current connecting line as a symmetry axis, and calculating the symmetry of the image contents on the left side and the right side.

The symmetry calculation method comprises the following steps: and comparing the color values of two pixel points which are symmetrical to each other, and if the difference of the color values of the three channels of RGB is less than a threshold value (20), the pixel points are considered to be paired. The total number of paired pixels is counted.

c34) The axis of symmetry with the greatest number of pairings is found.

d) Detecting wormholes:

the external shape of the wormhole is a round black hole. The invention detects the surface of the corn kernels and the existence of holes by the following method.

d1, the angle of the light source and the brightness of the LED light are configured so that the surface of the corn kernels becomes white.

d2, the hole position, because the light can not be reflected, so it becomes a black hole.

d3, in the corn kernel area, binarizing the image to change the black holes into a white foreground.

d4, connected domain detection.

d5, finding all connected domains with size larger than the threshold.

d6, if the number of connected domains is greater than or equal to 1, indicating that there are holes in the kernel. At this time, early warning can be given.

e) And (3) mildew detection:

when the corn kernels are mildewed, a small mass with darker color is formed in the endosperm area. Since the color of endosperm is usually brighter, the specific method for detecting the mildew point on the endosperm is as follows:

d1, obtaining the endosperm region by using a region growing method.

d2, extracting the endosperm area.

d3, analyzing the brightness of the pixel value corresponding to the endosperm area to obtain a histogram.

d4, two peaks appear in the luminance histogram in the presence of a mildew point.

d5, find a classification threshold between two peaks.

d6, carrying out binarization on the endosperm area by using the threshold value found in the step d 5.

d7, analyzing the connected domain to obtain the maximum connected domain.

d8, the maximum connected domain width and height are all larger than the threshold (20 pixels), this indicates the existence of a mildew point, and at this time, a warning is needed.

f) And (3) surface crack detection:

cracks appear on the surface of corn, and are also one of the common abnormal phenomena. When the crack of the surface is detected, the interference of non-crack edges (endosperm boundaries, folds of endosperm regions) on the surface of the corn kernel needs to be avoided, and the false detection is caused.

The invention adopts the following method to detect the surface cracks of the corn kernels:

f1, performing edge extraction on the surface of the corn kernels.

f2, short edge of filter (i.e. length >50)

f3, gradient analysis is performed for each long edge:

f31) each edge point is traversed.

f32) Along the direction perpendicular to the edge, pixels are collected for analysis.

f33) And recording the positions of the pixel points at the two sides of the edge, wherein the gradient value is less than 10, and taking the positions as boundary points of the edge.

f34) And obtaining the edge width corresponding to the current edge point according to the coordinate difference of the edge boundary points on the left side and the right side.

f35) If the edge width of all points of the current edge is less than 6 pixels, the line is considered a crack. Otherwise it is the endosperm edge.

f36) In case of cracks, a warning is required.

Claims

1. An automatic corn detection method comprises the following steps:

the method comprises the following specific steps of detecting the smoothness of the outer contour in the step c):

c21) traversing each contour point;

c24) traversing included angle values of all the points, and if a certain included angle value is smaller than 120 degrees, performing early warning, wherein the contour does not have smoothness;

the corn kernel symmetry analysis in the step c) comprises the following specific steps:

c34) finding the symmetry axis with the largest number of pairs;

c35) if the number of the pixels matched with the maximum symmetry is less than 80% of the total number, the corn kernels are not complete, and a certain part of the corn kernels are incomplete;

f) and (3) surface crack detection: performing edge extraction on the surface of the corn kernels; short edges of filter length < 50 pixels; performing gradient analysis on each long edge;

the step of performing gradient analysis on each long edge in the step f) comprises the following steps: