CN109794435A - Fruit quality detection system and method based on deep learning - Google Patents
Fruit quality detection system and method based on deep learning Download PDFInfo
- Publication number
- CN109794435A CN109794435A CN201910047483.3A CN201910047483A CN109794435A CN 109794435 A CN109794435 A CN 109794435A CN 201910047483 A CN201910047483 A CN 201910047483A CN 109794435 A CN109794435 A CN 109794435A
- Authority
- CN
- China
- Prior art keywords
- fruit
- conveyer
- plane
- camera
- slideway
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
Fruit quality detection system and method based on deep learning, mainly according to fruit surface color, texture, shape, size come comprehensive descision fruit quality to complete to be classified, entire detection system is by fruit conveyer system, Computer Vision Recognition system, hierarchy system composition.Compared to traditional fruit detection system, this system has the distinguishing feature of deep learning, it is mainly reflected in fruit image processing aspect, pass through the detailed analysis of network structure and convolutional neural networks training process to convolutional neural networks algorithm, construct the fruit image identifying system based on convolutional neural networks, compared to general neural network algorithm, have the characteristics that structure is simple, training parameter is few and adaptable.
Description
Technical field
The invention belongs to product quality detection fields, and in particular to a kind of fruit quality detection system based on deep learning
And method.
Background technique
In recent years, with the development of information technology, fruit quality detection has become the crucial skill of current agricultural field development
Art.But traditional fruit image-recognizing method can no longer meet modernization industry development demand, also need to identify fruit image
It is furtherd investigate, to improve the precision and efficiency of its image recognition.Traditional method is using the method for engineer to spy
Fixed classification task carries out the extraction of feature, and thus caused calculating cost is larger, and such method method considerably expends people
Power and financial resources, and there is classification task the correlation of height can not just carry out corresponding once changing application scenarios
Transfer learning.
In response to this, it needs to design a kind of deep learning algorithm, that is, image is allowed to be capable of the transformation of adaptive environment, from
It moves and removes study characteristics of image, rather than artificially design feature, additional manual intervention is not needed during training.
Deep learning allows the apish vision mechanism of computer, is learnt automatically from target by unsupervised training method
Most essential feature.Convolutional neural networks algorithm in deep learning, it can learn largely to input reflecting between output
Relationship is penetrated, convolutional network is trained with known mode, network just has the mapping ability between inputoutput pair.I.e. from
Input layer inputs a fruit image, and by the image processing algorithm of hidden layer, we can directly obtain me from output layer
Required for fruit quality judging result.
Convolutional neural networks select propagated forward to calculate output valve, the method that backpropagation adjusts weight and biasing.It compares
In general traditional neural network, CNN shows biggish advantage in terms of image recognition:
1, make image directly as the input of network, avoid complicated preprocessing process；
2, its weight, which shares network structure, reduces the complexity of network model, reduces the quantity of weight；
3, convolutional neural networks do not need manual designs and extract feature, can implicitly learn more directly using picture as input
Level characteristics, and then realize classification, each classifier is then utilized compared to currently used artificial design features, is had apparent excellent
Gesture.
Summary of the invention
The present invention aiming at the shortcomings in the prior art, provides a kind of fruit quality detection system based on deep learning and side
Method, it can realize full-automatic detection, be entered into product testing from sample, be not necessarily to manual intervention.
To achieve the above object, the invention adopts the following technical scheme:
Fruit quality detection system based on deep learning characterized by comprising slideway, guide rod device, conveyer, red
Outside line sensor, filming apparatus, lighting fixture, host computer and grading plant；
The slideway includes inclined-plane slideway and plane slideway, and oblique 30 ° of the inclined-plane slideway is put on steelframe, inclined-plane slideway two
Side is equipped with baffle, and the both ends of the plane slideway dock inclined-plane slideway and conveyer respectively, enable fruit in gravity
From inclined-plane, slideway slides to conveyer under effect；
The guide rod device includes two guide rods, and two guide rods are screwed in the opposite sides of slideway, guide rod
Between the upper opening that is formed it is wide with slideway, under shed is drawn close, and can control the size of under shed by rotary screw；
The conveyer is put on steelframe, making fruit at the uniform velocity rolls forward, and the middle part of conveyer is equipped with shooting area；
The infrared sensor is located at the symmetric position of conveyer both sides of the head, before shooting area, is used for host computer
Inductive signal is transmitted, and then triggers filming apparatus；
The filming apparatus includes first camera and four second cameras, and the first camera is fixed on liftable first support
On, first support is mounted on the slide bar of conveyer side, and the camera lens of first camera is vertical with conveyer machine side, right downwards
Quasi- shooting center；Around the installation of shooting center there are four second camera, the second camera is fixed on positioned at conveyer two sides
On second support, two adjacent the second supports are mutually in 90 °, and the camera lens of second camera is parallel with conveyer machine side；
The lighting fixture includes planar light source and four point light sources, and the planar light source is located above first camera, irradiated region
Domain covers downwards entire shooting area；Four point light sources are mounted on outside camera lens around four second cameras, are used for side polishing；
The host computer is located at conveyer lower part, convolutional neural networks algorithm is preset in host computer, to the figure of filming apparatus shooting
It is automatic to detect fruit character and export Quality estimation result as carrying out characteristic parameter extraction and classification；
The grading plant includes collection slide and fruit collecting box, and conveyer and fruit are docked in the collection slide both ends respectively
Collecting box, the two sides of collection slide relatively install that there are two axis to rotate plectrum, when initial position between two axis rotation plectrums
There is certain angle, axis rotates plectrum by the control of host computer, controls and be opened and closed according to testing result, fruit is made to roll into corresponding water
In fruit collecting box, control shaft rotation plectrum returns to initial position to fruit warp again later.
To optimize above-mentioned technical proposal, the concrete measure taken further include:
The conveyer uses black light-absorbing material transmission belt.
The planar light source uses LED annular light source, and the point light source uses LED point light source.
Further, the invention also provides using the fruit quality detection system based on deep learning as described above
Detection method, which comprises the steps of:
According to the type for the fruit of being detected, the under shed size of guide rod is adjusted, guarantees that fruit can pass sequentially through；
Fruit to be measured is placed in shooting area, starting filming apparatus is shot, according to imaging definition and the position in the shared visual field
It sets and ratio, it is appropriate to adjust the first support and the second support；
Start whole system, by fruit first from inclined-plane slideway mouth by static downslide, under the effect of gravity, glides and pass through from inclined-plane slideway
Conveyer is reached after crossing the guide rod of plane slideway；
Fruit passes through infrared sensor, and host computer obtains inductive signal, when by camera fields of view center, PC control
First camera and four second cameras are taken pictures simultaneously, obtain a fruit front elevation and four part side views；
Camera from coffret by these picture transfers to host computer, in host computer trained convolutional neural networks to picture
It is handled, extracts fruit character, obtain testing result；
The axis of PC control grading plant rotates plectrum according to testing result, and fruit is made to roll into corresponding fruit collecting box, complete
At classification.
The convolutional neural networks are as follows to the extraction process of fruit character:
1) picture for inputting convolutional neural networks is the pretreatment figure of 28 × 28 sizes, is convolutional layer C1, convolutional layer after input layer
C1 carries out convolution by convolution kernel and input picture and obtains characteristic plane, and convolution kernel size is 5 × 5, each mind on characteristic plane
It is connect through member with the local experiences visual field of 5 × 5 size of original image, convolution kernel moving step length is 1 pixel, the feature of convolutional layer C1
Plane sizes are 24 × 24, carry out convolution using 6 different convolution kernels and input picture in C1, obtain 6 kinds of different spies
Plan view is levied, all neurons on same characteristic plane figure share a convolution kernel, and the result after convolution passes through
Sigmoid activation primitive is by neural nonlinearity；
2) the sample level S1 after convolutional layer C1 is made of 6 characteristic planes, and it is flat that sample level S1 carries out part to upper one layer of characteristic pattern
It is connect with Further Feature Extraction, each neuron on characteristic plane with the mutual neighborhood not being overlapped of upper one layer of 2X2 size, into
The sampling of row mean value, the size of final each plane are 12 × 12；
3) in the convolutional layer C2 after sample level S1, convolution kernel number increases to 12, and size is 5 × 5, and each characteristic pattern is connection
All 6 or several characteristic patterns into S1, so it is 8 × 8 characteristic pattern that C2 layers, which have 12 sizes,；
4) the sample level S2 after convolutional layer C2 carries out mean value sampling to upper one layer 4 × 4 big small neighbourhood, exports 12 4 × 4 sizes
Characteristic plane；
5) two-dimensional surface of upper one layer of 12 4 × 4 size is launched into 1 × 192 size by the full articulamentum F1 after sample level S2
One-dimensional vector simultaneously inputs Softmax classifier, and output layer one shares 2 neurons, indicates good and bad two kinds of knots of fruit quality
Fruit, classifier realize the classification of fruit quality on 2 neurons of the maps feature vectors extracted to output layer.
Input for picture first carries out target and background segment to original image, then carries out gray processing, unified resolution
Rate recently enters convolutional neural networks.
The beneficial effects of the present invention are:
1, in terms of fruit image processing, have the characteristics that deep learning, after unsupervised neural metwork training, convolution mind
It can learn characteristic parameter from fruit image automatically through network algorithm, carry out a series of convolution, sampling, sort operation.It is defeated
Enter a fruit image, the algorithm process of process trained hidden layer, the judgement knot of direct available fruit quality
Fruit；
2, more abundanter than the characteristic parameter that traditional fruit quality detects, to keep quality testing more accurate.This detection system
It unites from the color of fruit, texture, size, shape carries out comprehensive Quality estimation, avoids because single detection parameters detect
The mistake classification for making mistakes and causing fruit quality improves the quality requirement of fruit detection；
3, when shooting picture, it is difficult to avoid the problem that because illumination too by force caused by fruit surface is reflective and transmission belt generates shade,
The LED circular lamp source diameter that this detection system uses is slightly larger than fruit diameter, using the soft light source of this dispersion, greatly subtracts
Small reflecting effect.
Detailed description of the invention
Fig. 1 is the top view of present system.
Fig. 2 is the side view of present system.
Fig. 3 is the perspective view of present system.
Fig. 4 is convolutional neural networks architecture diagram of the invention.
Appended drawing reference is as follows: inclined-plane slideway 1, plane slideway 2, guide rod 3, conveyer 4, infrared sensor 5, baffle 6,
Slide bar 7, the first support 8, the second support 9,10,11,12, first camera 13, second camera 14,15,16,17, planar light source 18,
Point light source 19,20,21,22, axis rotate plectrum 23, fruit collecting box 24, collection slide 25.
Specific embodiment
In conjunction with the accompanying drawings, the present invention is further explained in detail.
If Fig. 1 is to the fruit quality detection system shown in Fig. 3 based on deep learning, including with flowering structure:
Slideway: 1 oblique 30 ° of inclined-plane slideway is put on steelframe, and the fruit of no initial velocity can be only under gravity to cunning
The sliding of board bottom end, makes fruit obtain the speed of a rolls forward, can be with conveyer after reaching conveyer 4 in order to ensure fruit
4 uniform motion, angle initialization are 30 °, and baffle 6 is mounted on 1 two sides of inclined-plane slideway, prevents fruit from rolling out slideway；Plane slideway 2 is put
It is placed on steelframe, docks inclined-plane slideway 1, upper installation guide rod 3；
Guide rod device: two guide rods 3 are screwed in slideway two sides, and guide rod 3 has certain altitude, upper opening and plane
Slideway 2 is wide, and under shed is mutually drawn close, and controls under shed size by rotary screw for variety classes fruit, on the one hand protects
Card fruit successively passes sequentially through, and on the other hand for correcting the direction of fruit rolling, passes through it in camera fields of view as far as possible
The heart reduces offset；
Conveyer 4: it puts on steelframe, for adjusting fruit speed, is allowed at the uniform velocity rolls forward, is conducive to camera acquisition figure
Piece avoids forming shade when shooting using black light-absorbing material transmission belt.The fruit rolled at various speeds is tested in advance,
According to the clarity of the speed of camera shooting and clapped picture, suitable velocity of rotation is determined, select the suitable motor of power.
Infrared sensor 5: positioned at the symmetric position of the both sides of the head of conveyer 4, it is fixed on the bracket of 4 two sides of conveyer
On, before shooting area, there is certain distance in distance shooting area, for incuding the position of fruit, transmits induction letter to host computer
Number, host computer calculates the shooting of time interval controls camera.
Filming apparatus: shooting center is located at the central area of conveyer 4, and first camera 13 is fixed on liftable first
On seat 8, camera lens is vertical with transmission machine side, and camera is made to be directed at shooting central area downwards, and the first support 8 is mounted on conveyer 4
On the slide bar 7 of side, support is made up and down slidably by Unscrewed screw, so as to adjust camera heights；Several second cameras 14,
15,16,17 it is installed on parallel transmission machine 4 and on the second support 9,10,11,12 mutually in 90 °.
Lighting fixture: planar light source 18 uses annular LED lamp, on the support above top camera, carries out top and beats
Light, irradiation area can cover entire shooting area, have enough brightness and not will cause strong reflecting effect, alignment is clapped downwards
Take the photograph area；Several point light sources 19,20,21,22 use LED miniature light sources, are mounted on external camera lens around 4 cameras, carry out
Side polishing, angle are mutually in 90 °.
Host computer: being located at 4 lower part of conveyer, and connection infrared sensor 5, camera and axis rotate plectrum 23, can receive red
The inductive signal of outside line sensor 5, the plectrum that grading plant was taken pictures and controlled to triggering camera are swung.Host computer internal preset depth
The convolutional neural networks algorithm of study carries out feature extraction to the fruit image of input, obtains the testing result of fruit quality.
Grading plant: 25 top of collection slide is docked with conveyer 4, and lower part is connect with fruit collecting box 24, at two points
At the shunting of slideway head, the plectrum of an axis rotating manner is installed respectively, two plectrums have certain angle, axis rotation when beginning
Plectrum 23 is connect with host computer, and host computer controls the rotation of slideway plectrum according to quality measurements, after fruit warp, then
Control plectrum returns to original position.
First according to the type for the fruit of being detected, the lower ending opening size of guide rod 3 is adjusted, guarantees that fruit can be successively
Pass through.The first support 8 and the second support 9,10,11,12 are adjusted, alignment lens is made to shoot central area.Open planar light source 18
With point light source 19,20,21,22, planar light source 18 is LED circular lamp, and camera absolute visual field is completely covered in range of exposures enough, just
It is directed at shooting area well, it is mounted on slide bar 7 fixed.Point light source 19,20,21,22 is pacified around second camera 14,15,16,17
Outside camera lens.
1 oblique 30 ° of inclined-plane slideway is put on steelframe, and baffle 6 is located at slideway two sides, prevents fruit from leaving slideway.Fruit
Under gravity along inclined-plane slideway 1 by static downslide, fruit first passes through the upper of the guide rod 3 of 2 plate face upstream of plane slideway
Opening, goes out under shed, guarantees only to pass through one every time.
Infrared sensor 5 is located at the transmission of the somewhere before test point machine side two sides symmetric position, and fruit is from plane slideway 2
Be rolled on conveyer 4, when by infrared sensor 5, sensor generate induction, transmit a signal to host computer, then on
Position machine control camera is shot, and makes fruit just past camera fields of view center.
First camera 13 is installed on the first support 8, and camera lens is vertical with transmission machine side, and shooting height can be adjusted with support.
Second camera 14,15,16,17 is flat at an angle of 90 respectively on the second support 9,10,11,12.It is passed when camera receives host computer
The signal come, is immediately shot, gets 5 sub-pictures by coffret and be transferred to host computer.
The convolutional neural networks algorithm being pre-designed in host computer carries out characteristic parameter extraction and classification to image, final defeated
A judging result of fruit quality, host computer are controlled according to plectrum of the result to grading plant out, and fruit is made to roll into phase
In the collecting box answered.
Grading plant is made of collection slide 25, axis rotation plectrum 23, fruit collecting box 24.Two plectrums when initial position
Between have certain angle, after host computer obtains fruit testing result, control corresponding plectrum to lower swing, guide fruit to phase
The collecting box answered rolls, and after fruit passes through, host computer just controls plectrum and returns to original position.
The operation of fruit quality detection system are as follows: firstly, operation camera software, is placed in camera for certain standard fruit
Central region position, according to shooting image validity and clarity and the shared visual field position and ratio, to camera support into
Row adjustment appropriate.Then fruit is filled from inclined-plane slideway 1 under the effect of gravity by static downslide by the guiding of plane slideway 2
It postpones and reaches conveyer 4, by infrared sensor 5, generate inductive signal and be transferred to host computer, by camera fields of view centre bit
When setting, PC control camera is shot.Five cameras are taken pictures simultaneously, obtain a fruit front elevation and four part sides
Figure.Camera from coffret by these picture transfers to host computer, trained convolutional neural networks algorithm pair in host computer
Picture carries out processing and obtains testing result, and last host computer controls the plectrum of grading plant according to testing result, rolls into fruit
Corresponding fruit collecting box 24, completes classification.After sample has been recorded, fruit is detected with same method, host computer will
The data and sample data of detection, which compare, obtains quality measurements.
The training process of convolutional neural networks is divided into two stages: first stage is data from low level to high-level biography
The stage broadcast, i.e. propagated forward stage；Another stage is, when the result that propagated forward obtains is with being expected not to be consistent, by
Error carries out propagating trained stage, i.e. back-propagation phase to bottom from high-level.
Input for picture needs first to carry out target and background segment to original image, then carries out gray processing, unified point
Resolution recently enters convolutional neural networks.Since the basic goal of training convolutional neural networks is to extract the spy of different fruit
Sign, and background is not a part of target, does not provide any useful information for fruit identification, is mentioned instead to feature
It takes and interferes, so network structure can be simpler after removal ambient noise, while being also conducive to study of the network to feature.
Convolutional neural networks are as follows to the extraction process of fruit character:
1, the picture for inputting network is the pretreatment figure of 28 × 28 sizes, is convolutional layer, volume after the input layer of convolutional neural networks
Lamination carries out convolution by convolution kernel and input picture and obtains characteristic plane, and convolution kernel size is 5 × 5.Such as Fig. 4, characteristic plane
Upper each neuron is connect with the local experiences visual field of 5 × 5 size of original image, and convolution kernel moving step length is 1 pixel, therefore is rolled up
The characteristic plane size of lamination C1 is 24 × 24.It has used 6 different convolution kernels and input picture to carry out convolution in C1, has obtained
The characteristic plane figure different to 6 kinds, all neurons on same characteristic pattern share a convolution kernel.Result after convolution
It is not directly stored in C1 layers of characteristic pattern, but by activation primitive by neural nonlinearity, so that it is stronger to have network
Feature representation ability, activation primitive select Sigmoid function.
2, the sample level S1 after convolutional layer C1 is made of 6 characteristic planes, and sample level carries out part to upper one layer of characteristic pattern
Average and Further Feature Extraction.Each neuron on characteristic plane is connect with the neighborhood of upper one layer of 2X2 size not being overlapped mutually,
Mean value sampling is carried out, the size of final each plane is 12 × 12.
3, in order to promote classification accuracy, the convolution kernel number for increasing the second layer is 12, and size is still 5 × 5.So C2
It is 8 × 8 characteristic pattern that layer, which has 12 sizes,.But it should be noted that each characteristic pattern in C2 layers is attached to the institute in S1
There are 6 or several characteristic patterns.This incomplete connection type has not only broken up the symmetry of network, and makes C2 and S1
Between connection quantity be able to maintain within the scope of reasonable, different characteristic is mentioned to realize by this connection mechanism
It takes.
4, sample level S2 is as S1 layer operation, carries out mean value sampling to upper one layer 4 × 4 big small neighbourhood, and output 12 4 ×
The characteristic plane of 4 sizes.
5, the two-dimensional surface of upper one layer of 12 4 × 4 size is launched into the one-dimensional of 1 × 192 size by full articulamentum F1
Vector input Softmax classifier, output layer have altogether there are two neuron, indicate the good of fruit quality with badly two kinds as a result, dividing
Class device will be realized on two neurons of the maps feature vectors extracted to output layer classifies.
It should be noted that the term of such as "upper", "lower", "left", "right", "front", "rear" cited in invention, also
Only being illustrated convenient for narration, rather than to limit the scope of the invention, relativeness is altered or modified, in nothing
Under essence change technology contents, when being also considered as the enforceable scope of the present invention.
The above is only the preferred embodiment of the present invention, protection scope of the present invention is not limited merely to above-described embodiment,
All technical solutions belonged under thinking of the present invention all belong to the scope of protection of the present invention.It should be pointed out that for the art
For those of ordinary skill, several improvements and modifications without departing from the principles of the present invention should be regarded as protection of the invention
Range.
Claims (6)
1. the fruit quality detection system based on deep learning characterized by comprising slideway, guide rod device, conveyer
(4), infrared sensor (5), filming apparatus, lighting fixture, host computer and grading plant；
The slideway includes inclined-plane slideway (1) and plane slideway (2), and oblique 30 ° of the inclined-plane slideway (1) is put on steelframe,
Inclined-plane slideway (1) two sides are equipped with baffle (6), and inclined-plane slideway (1) and conveyer are docked in the both ends of the plane slideway (2) respectively
(4), fruit is enabled to slide to conveyer (4) from inclined-plane slideway (1) under gravity；
The guide rod device includes two guide rods (3), and two guide rods (3) are screwed in the opposite sides of slideway,
The upper opening formed between guide rod (3) is wide with slideway, and under shed is drawn close, and can control the big of under shed by rotary screw
It is small；
The conveyer (4) is put on steelframe, making fruit at the uniform velocity rolls forward, and the middle part of conveyer (4) is equipped with shooting area；
The infrared sensor (5) is located at the symmetric position of conveyer (4) both sides of the head, shooting area before, for
Host computer transmits inductive signal, and then triggers filming apparatus；
The filming apparatus includes first camera (13) and four second cameras (14) (15) (16) (17), the first camera
(13) it is fixed on liftable first support (8), first support (8) is mounted on the slide bar (7) of conveyer (4) side
On, the camera lens of first camera (13) is vertical with conveyer (4) machine side, is directed at shooting center downwards；It is equipped with around shooting center
Four second cameras (14) (15) (16) (17), the second camera (14) (15) (16) (17) are fixed on positioned at conveyer (4)
On the second support (9) (10) (11) (12) of two sides, two adjacent the second supports (9) (10) (11) (12) are mutually in 90 °, and second
The camera lens of camera (14) (15) (16) (17) is parallel with conveyer (4) machine side；
The lighting fixture includes planar light source (18) and four point light sources (19) (20) (21) (22), the planar light source (18)
Above first camera (13), irradiation area covers downwards entire shooting area；Four point light sources (19) (20) (21) (22) are enclosed
It is mounted on outside camera lens around four second cameras (14) (15) (16) (17), is used for side polishing；
The host computer is located at conveyer (4) lower part, presets convolutional neural networks algorithm in host computer, to filming apparatus shooting
Image carries out characteristic parameter extraction and classification, automatic to detect fruit character and export Quality estimation result；
The grading plant includes collection slide (25) and fruit collecting box (24), and collection slide (25) both ends are docked respectively
Conveyer (4) and fruit collecting box (24), the two sides of collection slide (25) relatively install there are two axis rotation plectrum (23), just
There is certain angle when beginning position between two axis rotation plectrum (23), axis rotates plectrum (23) by the control of host computer, according to
Testing result control opening and closing, rolls into fruit in corresponding fruit collecting box (24), and control shaft rotates plectrum to fruit warp again later
(23) initial position is returned to.
2. the fruit quality detection system based on deep learning as described in claim 1, it is characterised in that: the conveyer
(4) black light-absorbing material transmission belt is used.
3. the fruit quality detection system based on deep learning as described in claim 1, it is characterised in that: the planar light source
(18) LED annular light source is used, the point light source (19) (20) (21) (22) uses LED point light source.
4. using the detection method of the fruit quality detection system based on deep learning as described in claim 1, feature exists
In including the following steps:
According to the type for the fruit of being detected, the under shed size of guide rod (3) is adjusted, guarantees that fruit can pass sequentially through；
Fruit to be measured is placed in shooting area, starting filming apparatus is shot, according to imaging definition and the position in the shared visual field
It sets and ratio, it is appropriate to adjust the first support (8) and the second support (9) (10) (11) (12)；
Start whole system, by fruit first from inclined-plane slideway (1) mouth by static downslide, under the effect of gravity, from inclined-plane slideway (1)
It glides and reaches conveyer (4) after the guide rod (3) of plane slideway (2)；
Fruit passes through infrared sensor (5), and host computer obtains inductive signal, when by camera fields of view center, host computer
Control first camera (13) and four second cameras (14) (15) (16) (17) are taken pictures simultaneously, obtain a fruit front elevation and four
A part side view；
Camera from coffret by these picture transfers to host computer, in host computer trained convolutional neural networks to picture
It is handled, extracts fruit character, obtain testing result；
The axis of PC control grading plant rotates plectrum (23) according to testing result, and fruit is made to roll into corresponding fruit collecting box
(24), classification is completed.
5. the detection method of the fruit quality detection system based on deep learning as claimed in claim 4, it is characterised in that: institute
It is as follows to the extraction process of fruit character to state convolutional neural networks:
1) picture for inputting convolutional neural networks is the pretreatment figure of 28 × 28 sizes, is convolutional layer C1, convolutional layer after input layer
C1 carries out convolution by convolution kernel and input picture and obtains characteristic plane, and convolution kernel size is 5 × 5, each mind on characteristic plane
It is connect through member with the local experiences visual field of 5 × 5 size of original image, convolution kernel moving step length is 1 pixel, the feature of convolutional layer C1
Plane sizes are 24 × 24, carry out convolution using 6 different convolution kernels and input picture in C1, obtain 6 kinds of different spies
Plan view is levied, all neurons on same characteristic plane figure share a convolution kernel, and the result after convolution passes through
Sigmoid activation primitive is by neural nonlinearity；
2) the sample level S1 after convolutional layer C1 is made of 6 characteristic planes, and it is flat that sample level S1 carries out part to upper one layer of characteristic pattern
It is connect with Further Feature Extraction, each neuron on characteristic plane with the mutual neighborhood not being overlapped of upper one layer of 2X2 size, into
The sampling of row mean value, the size of final each plane are 12 × 12；
3) in the convolutional layer C2 after sample level S1, convolution kernel number increases to 12, and size is 5 × 5, and each characteristic pattern is connection
All 6 or several characteristic patterns into S1, so it is 8 × 8 characteristic pattern that C2 layers, which have 12 sizes,；
4) the sample level S2 after convolutional layer C2 carries out mean value sampling to upper one layer 4 × 4 big small neighbourhood, exports 12 4 × 4 sizes
Characteristic plane；
5) two-dimensional surface of upper one layer of 12 4 × 4 size is launched into 1 × 192 size by the full articulamentum F1 after sample level S2
One-dimensional vector simultaneously inputs Softmax classifier, and output layer one shares 2 neurons, indicates good and bad two kinds of knots of fruit quality
Fruit, classifier realize the classification of fruit quality on 2 neurons of the maps feature vectors extracted to output layer.
6. the detection method of the fruit quality detection system based on deep learning as claimed in claim 5, it is characterised in that: right
In the input of picture, target and background segment first are carried out to original image, gray processing, unified resolution is then carried out, recently enters
To convolutional neural networks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910047483.3A CN109794435A (en) | 2019-01-18 | 2019-01-18 | Fruit quality detection system and method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910047483.3A CN109794435A (en) | 2019-01-18 | 2019-01-18 | Fruit quality detection system and method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109794435A true CN109794435A (en) | 2019-05-24 |
Family
ID=66559601
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910047483.3A Pending CN109794435A (en) | 2019-01-18 | 2019-01-18 | Fruit quality detection system and method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109794435A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110050743A (en) * | 2019-05-30 | 2019-07-26 | 无锡复易生物科技有限公司 | A kind of sorting of cray and digital assay System and method for |
| CN110142228A (en) * | 2019-06-01 | 2019-08-20 | 哈尔滨理工大学 | Saw blade crack detection identifies sorting system and method for sorting |
| CN110328157A (en) * | 2019-07-17 | 2019-10-15 | 南京工程学院 | A kind of intelligent sorting device and its method for sorting based on machine vision |
| CN110348503A (en) * | 2019-07-02 | 2019-10-18 | 西京学院 | A kind of apple quality detection method based on convolutional neural networks |
| CN110340022A (en) * | 2019-05-29 | 2019-10-18 | 东莞市鑫品自动化科技有限公司 | A kind of sorting equipment for luggage classification |
| CN110575973A (en) * | 2019-10-21 | 2019-12-17 | 台州学院 | crop seed quality detection and screening system |
| CN110702015A (en) * | 2019-09-26 | 2020-01-17 | 中国南方电网有限责任公司超高压输电公司曲靖局 | Method and device for measuring icing thickness of power transmission line |
| CN111014080A (en) * | 2019-12-13 | 2020-04-17 | 浙江大学 | Self-adaptive light source adjusting mechanism and method based on fruit size |
| CN111097705A (en) * | 2019-12-30 | 2020-05-05 | 湖北师范大学 | Sorting equipment with anti-bumping structure for logistics |
| CN111721728A (en) * | 2020-07-16 | 2020-09-29 | 陈皓 | Fruit online detection device and use method thereof |
| CN112883842A (en) * | 2021-02-02 | 2021-06-01 | 四川省机械研究设计院(集团)有限公司 | Motorcycle engine assembling method and system based on mutual matching of parts and light source |
| CN113714125A (en) * | 2021-08-19 | 2021-11-30 | 河南麒展实业有限公司 | Cylinder sorting device and sorting method |
| CN114444622A (en) * | 2022-04-11 | 2022-05-06 | 中国科学院微电子研究所 | Fruit detection system and method based on neural network model |
| CN114852428A (en) * | 2022-06-07 | 2022-08-05 | 安徽格拉泰科机械有限公司 | Conveying system capable of rapidly switching packaging forms |
| CN110050743B (en) * | 2019-05-30 | 2024-04-19 | 无锡复易生物科技有限公司 | Crayfish sorting and digital analysis system and method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0520426A (en) * | 1991-07-15 | 1993-01-29 | Sumitomo Heavy Ind Ltd | Fruit hue judgement device using neural network |
| CN104502356A (en) * | 2015-01-05 | 2015-04-08 | 江苏大学 | Automatic detection method for defects of inner surface of sliding bearing on basis of computer vision |
| CN105772410A (en) * | 2016-03-09 | 2016-07-20 | 华南农业大学 | Intelligent illumination control and sorting system for fruit and vegetable quality detection |
| CN106238342A (en) * | 2016-09-14 | 2016-12-21 | 郑州轻工业学院 | The sorting of panoramic vision Rhizoma Solani tuber osi and defect detecting device and sorting detection method thereof |
| CN106733701A (en) * | 2016-12-23 | 2017-05-31 | 郑州轻工业学院 | Potato multi-stage combination dysnusia detecting system and detection method |
| CN107742132A (en) * | 2017-11-07 | 2018-02-27 | 江南大学 | Potato detection method of surface flaw based on convolutional neural networks |
| CN207551160U (en) * | 2017-09-07 | 2018-06-29 | 上海曹杨建筑粘合剂厂 | A kind of bagged material pipeline |
-
2019
- 2019-01-18 CN CN201910047483.3A patent/CN109794435A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0520426A (en) * | 1991-07-15 | 1993-01-29 | Sumitomo Heavy Ind Ltd | Fruit hue judgement device using neural network |
| CN104502356A (en) * | 2015-01-05 | 2015-04-08 | 江苏大学 | Automatic detection method for defects of inner surface of sliding bearing on basis of computer vision |
| CN105772410A (en) * | 2016-03-09 | 2016-07-20 | 华南农业大学 | Intelligent illumination control and sorting system for fruit and vegetable quality detection |
| CN106238342A (en) * | 2016-09-14 | 2016-12-21 | 郑州轻工业学院 | The sorting of panoramic vision Rhizoma Solani tuber osi and defect detecting device and sorting detection method thereof |
| CN106733701A (en) * | 2016-12-23 | 2017-05-31 | 郑州轻工业学院 | Potato multi-stage combination dysnusia detecting system and detection method |
| CN207551160U (en) * | 2017-09-07 | 2018-06-29 | 上海曹杨建筑粘合剂厂 | A kind of bagged material pipeline |
| CN107742132A (en) * | 2017-11-07 | 2018-02-27 | 江南大学 | Potato detection method of surface flaw based on convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| 李德毅 等: "《中国科协新一代信息技术系列丛书 人工智能导论》", 31 August 2018, 中国科学技术出版社 * |
| 董海鹰: "《智能控制理论及应用》", 30 September 2016, 中国铁道出版社 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110340022A (en) * | 2019-05-29 | 2019-10-18 | 东莞市鑫品自动化科技有限公司 | A kind of sorting equipment for luggage classification |
| CN110050743B (en) * | 2019-05-30 | 2024-04-19 | 无锡复易生物科技有限公司 | Crayfish sorting and digital analysis system and method |
| CN110050743A (en) * | 2019-05-30 | 2019-07-26 | 无锡复易生物科技有限公司 | A kind of sorting of cray and digital assay System and method for |
| CN110142228B (en) * | 2019-06-01 | 2021-06-25 | 哈尔滨理工大学 | Saw blade crack detection, identification and sorting system and sorting method |
| CN110142228A (en) * | 2019-06-01 | 2019-08-20 | 哈尔滨理工大学 | Saw blade crack detection identifies sorting system and method for sorting |
| CN110348503A (en) * | 2019-07-02 | 2019-10-18 | 西京学院 | A kind of apple quality detection method based on convolutional neural networks |
| CN110328157A (en) * | 2019-07-17 | 2019-10-15 | 南京工程学院 | A kind of intelligent sorting device and its method for sorting based on machine vision |
| CN110328157B (en) * | 2019-07-17 | 2021-06-29 | 南京工程学院 | Intelligent sorting device based on machine vision and sorting method thereof |
| CN110702015A (en) * | 2019-09-26 | 2020-01-17 | 中国南方电网有限责任公司超高压输电公司曲靖局 | Method and device for measuring icing thickness of power transmission line |
| CN110575973A (en) * | 2019-10-21 | 2019-12-17 | 台州学院 | crop seed quality detection and screening system |
| CN110575973B (en) * | 2019-10-21 | 2021-08-10 | 台州学院 | Crop seed quality detection and screening system |
| CN111014080B (en) * | 2019-12-13 | 2020-11-06 | 浙江大学 | Self-adaptive light source adjusting mechanism and method based on fruit size |
| CN111014080A (en) * | 2019-12-13 | 2020-04-17 | 浙江大学 | Self-adaptive light source adjusting mechanism and method based on fruit size |
| CN111097705A (en) * | 2019-12-30 | 2020-05-05 | 湖北师范大学 | Sorting equipment with anti-bumping structure for logistics |
| CN111721728A (en) * | 2020-07-16 | 2020-09-29 | 陈皓 | Fruit online detection device and use method thereof |
| CN111721728B (en) * | 2020-07-16 | 2023-02-21 | 陈皓 | Fruit online detection device and use method thereof |
| CN112883842A (en) * | 2021-02-02 | 2021-06-01 | 四川省机械研究设计院(集团)有限公司 | Motorcycle engine assembling method and system based on mutual matching of parts and light source |
| CN113714125A (en) * | 2021-08-19 | 2021-11-30 | 河南麒展实业有限公司 | Cylinder sorting device and sorting method |
| CN113714125B (en) * | 2021-08-19 | 2024-04-09 | 河南麒展实业有限公司 | Cylinder sorting device and sorting method |
| CN114444622A (en) * | 2022-04-11 | 2022-05-06 | 中国科学院微电子研究所 | Fruit detection system and method based on neural network model |
| CN114444622B (en) * | 2022-04-11 | 2022-06-17 | 中国科学院微电子研究所 | Fruit detection system and method based on neural network model |
| CN114852428A (en) * | 2022-06-07 | 2022-08-05 | 安徽格拉泰科机械有限公司 | Conveying system capable of rapidly switching packaging forms |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109794435A (en) | Fruit quality detection system and method based on deep learning | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| CN105069472B (en) | A kind of vehicle checking method adaptive based on convolutional neural networks | |
| CN106733701B (en) | Potato multi-stage combination dysnusia detecting system and detection method | |
| CN108510062A (en) | A kind of robot irregular object crawl pose rapid detection method based on concatenated convolutional neural network | |
| CN108229362A (en) | A kind of binocular recognition of face biopsy method based on access control system | |
| CN103808669B (en) | A kind of apple small holes caused by worms fast non-destructive detection method based on high light spectrum image-forming technology | |
| CN110276386A (en) | A kind of apple grading method and system based on machine vision | |
| CN113838034B (en) | Quick detection method for surface defects of candy package based on machine vision | |
| CN108830334A (en) | A kind of fine granularity target-recognition method based on confrontation type transfer learning | |
| CN107609597A (en) | A kind of number of people in lift car detecting system and its detection method | |
| CN109724984A (en) | A kind of defects detection identification device and method based on deep learning algorithm | |
| CN109325495A (en) | A kind of crop image segmentation system and method based on deep neural network modeling | |
| CN113077450B (en) | Cherry grading detection method and system based on deep convolutional neural network | |
| CN106529527A (en) | Object detection method and device, data processing deice, and electronic equipment | |
| CN108710909B (en) | Counting method for deformable, rotary and invariant boxed objects | |
| CN106874929B (en) | Pearl classification method based on deep learning | |
| CN108267455A (en) | plastic film printing character defect detecting device and method | |
| CN108549910A (en) | A kind of corn seed producing fruit ear image preliminary classification method based on convolutional neural networks | |
| CN109030504A (en) | The panorama potato defect detecting device and method combined based on actual situation imaging | |
| CN108932712A (en) | A kind of rotor windings quality detecting system and method | |
| CN108021926A (en) | A kind of vehicle scratch detection method and system based on panoramic looking-around system | |
| CN114529819A (en) | Household garbage image recognition method based on knowledge distillation learning | |
| CN110309767A (en) | In vivo detection equipment, recognition methods, device and storage medium | |
| CN109508756A (en) | A kind of ground cloud classification method based on multi thread multi-modal fusion depth network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190524 |