CN109420622A - Tobacco leaf method for sorting based on convolutional neural networks - Google Patents
Tobacco leaf method for sorting based on convolutional neural networks Download PDFInfo
- Publication number
- CN109420622A CN109420622A CN201710746676.9A CN201710746676A CN109420622A CN 109420622 A CN109420622 A CN 109420622A CN 201710746676 A CN201710746676 A CN 201710746676A CN 109420622 A CN109420622 A CN 109420622A
- Authority
- CN
- China
- Prior art keywords
- tobacco leaf
- convolutional neural
- neural networks
- sundries
- sorting
- 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
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/34—Sorting according to other particular properties
-
- A—HUMAN NECESSITIES
- A24—TOBACCO; CIGARS; CIGARETTES; SIMULATED SMOKING DEVICES; SMOKERS' REQUISITES
- A24B—MANUFACTURE OR PREPARATION OF TOBACCO FOR SMOKING OR CHEWING; TOBACCO; SNUFF
- A24B3/00—Preparing tobacco in the factory
- A24B3/18—Other treatment of leaves, e.g. puffing, crimpling, cleaning
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/34—Sorting according to other particular properties
- B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
Landscapes
- Image Analysis (AREA)
- Manufacture Of Tobacco Products (AREA)
Abstract
The invention discloses a kind of tobacco leaf method for sorting based on convolutional neural networks, this method sorts tobacco leaf using convolutional neural networks, it can not only be by the discrepant sundries of color, also it can will come with the sorting of the sundries of textural characteristics and morphological feature, and since the resolution for normal tobacco leaf and sundries is to identify and then make sample input by artificial priori, as long as convolutional neural networks can reach the accuracy as manual sorting so the input of sample is reasonable.The present invention can learn tobacco leaf and sundries, and as the period used lengthens, the sample of acquisition is more, and training the parameter model come more can accurately identify sundries and tobacco leaf, reduce false detection rate and misdetection rate.
Description
Technical field
The present invention relates to tobacco leaf method for sorting, and in particular to a kind of tobacco leaf method for sorting based on convolutional neural networks.
Background technique
In existing tobacco leaf sort process, due to tobacco leaf form of diverse, no specific texture, color is also complex,
So its feature is not obvious compared to sundries, and sundries is many kinds of.Only from single features such as form, texture and colors
It judges by accident and fails to judge to distinguish to be very easy to generate.
It is mainly sorted at present using with color characteristic, this method energy in the discrepant situation of foreign matter color
Distinguish more foreign matter, but the foreign matter close for color, such as paper similar in other plant cauline leaf, the rope made of hemp, color
Fragment etc. do not identify not Chu Lai, and is still important index in tobacco leaf sorting evaluation and test for the rejecting of these sundries.
Summary of the invention
The purpose of the present invention is to provide a kind of tobacco leaf method for sorting based on convolutional neural networks.
Realize the technical solution of the object of the invention are as follows: a kind of tobacco leaf method for sorting based on convolutional neural networks, including such as
Lower step:
Step 1, tobacco leaf image is acquired, tobacco leaf is divided into two classes by manual sorting, one kind is normal tobacco leaf, and another kind of is miscellaneous
Object;
Step 2, design one include 4 feature extraction layers, full an articulamentum and an output layer neural network, often
A feature extraction layer includes a convolutional layer, a down-sampling layer and an active coating;
Step 3, the picture for being N*N by the two class images divided in step 1 difference cutting, progress label is simultaneously labelled,
Then these two types of images designed convolutional neural networks are inputted to be trained;
Step 4, the parameter model of convolutional neural networks is obtained by training, parameter is imported to the convolution for there was only propagated forward
Neural network is classified, and is decided whether to issue by threshold value differentiation and is rejected instruction;
Step 5, in the training process by the image making failed to judge and judged by accident at sample set, be added in training sample again into
Row training is to obtain neural network parameter model.
Compared with prior art, remarkable advantage of the invention are as follows:
(1) present invention greatly improves the discrimination for sundries, and sorts mould relative to present single feature
The shortcomings that formula cannot be adjusted according to different sundries, the present invention can learn tobacco leaf and sundries, with the week used
Phase lengthens, and the sample of acquisition is more, and training the parameter model come more can accurately identify sundries and tobacco leaf, reduces
False detection rate and misdetection rate;(2) using convolutional neural networks come to tobacco leaf sort can not only by the discrepant sundries of color,
Also it can will come with the sorting of the sundries of textural characteristics and morphological feature, and due to the resolution for normal tobacco leaf and sundries
It is to identify and then make sample input by artificial priori, as long as convolutional neural networks can reach so the input of sample is reasonable
Accuracy as manual sorting.
Detailed description of the invention
Fig. 1 is the structural model figure of neural network of the present invention.
Fig. 2 is the flow chart of training process of the present invention.
Fig. 3 is the flow chart of detection process of the present invention.
Specific embodiment
A kind of tobacco leaf method for sorting based on convolutional neural networks, includes the following steps:
Step 1, tobacco leaf image is acquired, tobacco leaf is divided into two classes by manual sorting, one kind is normal tobacco leaf, and another kind of is miscellaneous
Object;
Step 2, design one include 4 feature extraction layers, full an articulamentum and an output layer neural network, often
A feature extraction layer includes a convolutional layer, a down-sampling layer and an active coating, as shown in Figure 1;Convolution in neural network
The size of core is 5*5, and the step-length of down-sampling is 2;The quantity of each layer of convolution kernel is followed successively by 32,32,64 and 128, full articulamentum
Neuronal quantity be 500, the classification of output is 2 classes, and the quantity of small sample set is 100 when training.
Step 3, the small picture for being 128*128 by the two class images divided in step 1 difference cutting, carries out label and sticks
Then these two types of images are inputted designed convolutional neural networks and are trained by label;
Step 4, the parameter model of convolutional neural networks is obtained by training, parameter is imported to the convolution for there was only propagated forward
Neural network is classified, and is decided whether to issue by threshold value differentiation and is rejected instruction;
Step 5, in the training process by the image making failed to judge and judged by accident at sample set, be added in training sample again into
Row training is to obtain more accurate neural network parameter model.
Present invention is further described in detail with embodiment with reference to the accompanying drawing.
Embodiment
A kind of tobacco leaf method for sorting based on convolutional neural networks, includes the following steps:
(1) it acquires image: acquiring tobacco leaf image with image capturing system, classify for the image of acquisition, manually choose
The image of normal tobacco leaf and the image of sundries are selected, the image of acquisition is divided into two major classes.
(2) make sample: the two class images that classification is completed carry out cutting, and all equal cuttings of image are 128*128 size
Small picture, the picture then segmented to two classes carries out label and generates label.
(3) it builds neural network: after generating sample database, needing to build a convolutional neural networks system and come to sample
This is learnt.For the structure of network as shown in Figure 1, including four feature extraction layers, every layer includes convolutional layer, down-sampling layer
And active coating, each feature extraction layer export the input as next feature extraction layer.The size for inputting picture is 128*
128, the size of convolution kernel is 5*5, and the step-length of down-sampling is 2.The quantity of each layer of convolution kernel is followed successively by 32,32,64 and 128,
The neuronal quantity of full articulamentum is 500, and the classification of output is 2 classes, and the quantity of small sample set is 100 when training.
(4) with the sample set input of production, the instruction of parameter training convolutional neural networks: is carried out by convolutional neural networks
Practice, until reaching desired accuracy rate, if the maximum number of times for having reached iteration still fails to reach desired accuracy rate, again
It makes sample collection to be trained, notices that classification is accurate when sample set production.
(5) sorting processing is carried out using network: right first since the size of the image of different acquisition system acquisitions is different
Image carries out the pretreatment of size, the parameter model then generated using step (4), using the propagated forward of convolutional neural networks
Process classifies to input picture, and classification results are counted and are then adjudicated by threshold value, and then decide whether output sort mistake
Rejecting instruction in journey.
(6) it adjusts threshold value and optimizes: adjusting threshold value and the image failed to judge and judged by accident in sort process is stored, so
Step (2), (3) and (4) are repeated afterwards, and neural network is advanced optimized.
Fig. 1 is the structural model of neural network, and wherein tobacco is image input layer;conv1,conv2,conv3,
Conv4 is convolutional layer, and the quantity of each layer of convolution kernel is followed successively by 32,32,64 and 128, and convolution kernel size is 5*5;
Pool1, pool2, pool3, pool4 are down-sampling layer, and the step-length of down-sampling is 2;relu1,relu2,relu3,relu4
It is active coating, activation primitive is Relu function;Ip1 is full articulamentum;Ip2 is output layer;Loss is error layer, for anti-
To propagation.
The workflow of whole system is as shown in Figure 2 and Figure 3, and Fig. 2 is training process, and Fig. 3 is with process.Such as Fig. 2 institute
Show, image capturing system acquires image, makes sample collection, classify and add label, training image data, testing classification as a result,
Judged whether to continue to train or remake sample set according to classification results.It completes to carry out as shown in Figure 3 after training process
Testing process, decide whether to issue according to court verdict and reject instruction tobacco leaf sorted.It is same in the process of detection
When sample is acquired, the sample judged by accident and failed to judge is stored, adds the sample of acquisition after system runs some cycles
Enter sample set, be trained, improves the accuracy of identification of network and the quantity of identification type.
Claims (3)
1. a kind of tobacco leaf method for sorting based on convolutional neural networks, which comprises the steps of:
Step 1, tobacco leaf image is acquired, tobacco leaf is divided into two classes by manual sorting, one kind is normal tobacco leaf, and another kind of is sundries;
Step 2, design one include 4 feature extraction layers, full an articulamentum and an output layer neural network, Mei Gete
Levying extract layer includes a convolutional layer, a down-sampling layer and an active coating;
Step 3, the picture for being N*N by the two class images divided in step 1 difference cutting, progress label is simultaneously labelled, then
These two types of images are inputted designed convolutional neural networks to be trained;
Step 4, the parameter model of convolutional neural networks is obtained by training, parameter is imported to the convolutional Neural for there was only propagated forward
Network is classified, and is decided whether to issue by threshold value differentiation and is rejected instruction;
Step 5, it in the training process by the image making failed to judge and judged by accident at sample set, is added in training sample and is instructed again
Practice to obtain neural network parameter model.
2. the tobacco leaf method for sorting according to claim 1 based on convolutional neural networks, which is characterized in that will in step 3
Two class images distinguish the picture that cutting is 128*128.
3. the tobacco leaf method for sorting according to claim 1 or 2 based on convolutional neural networks, which is characterized in that in step 2
The size of convolution kernel is 5*5 in neural network, and the step-length of down-sampling is 2.The quantity of each layer of convolution kernel is followed successively by 32,32,64
With 128, the neuronal quantity of full articulamentum is 500, and the classification of output is 2 classes, and the quantity of small sample set is 100 when training.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710746676.9A CN109420622A (en) | 2017-08-27 | 2017-08-27 | Tobacco leaf method for sorting based on convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710746676.9A CN109420622A (en) | 2017-08-27 | 2017-08-27 | Tobacco leaf method for sorting based on convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109420622A true CN109420622A (en) | 2019-03-05 |
Family
ID=65501660
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710746676.9A Pending CN109420622A (en) | 2017-08-27 | 2017-08-27 | Tobacco leaf method for sorting based on convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109420622A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110516988A (en) * | 2019-07-08 | 2019-11-29 | 国网浙江省电力有限公司金华供电公司 | One kind being fitted Power Material method based on neural network |
CN112189877A (en) * | 2020-10-13 | 2021-01-08 | 燕山大学 | On-line detection method for tobacco shred impurities in tobacco production line |
CN113210264A (en) * | 2021-05-19 | 2021-08-06 | 江苏鑫源烟草薄片有限公司 | Method and device for removing tobacco impurities |
JP2022019569A (en) * | 2020-07-17 | 2022-01-27 | 株式会社カナヤ食品 | Teacher data generation method, foreign matter inspection device, and foreign matter inspection method |
CN114397297A (en) * | 2022-01-19 | 2022-04-26 | 河南中烟工业有限责任公司 | Rapid nondestructive testing method for starch content of flue-cured tobacco |
CN115953384A (en) * | 2023-01-10 | 2023-04-11 | 杭州首域万物互联科技有限公司 | On-line detection and prediction method for tobacco morphological parameters |
JP7498524B1 (en) | 2023-05-02 | 2024-06-12 | 株式会社カナヤ食品 | Method for generating teacher data, foreign body inspection device, foreign body inspection method, and foreign body detection program |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103971342A (en) * | 2014-05-21 | 2014-08-06 | 厦门美图之家科技有限公司 | Image noisy point detection method based on convolution neural network |
US20160358024A1 (en) * | 2015-06-03 | 2016-12-08 | Hyperverge Inc. | Systems and methods for image processing |
CN106326899A (en) * | 2016-08-18 | 2017-01-11 | 郑州大学 | Tobacco leaf grading method based on hyperspectral image and deep learning algorithm |
CN106874929A (en) * | 2016-12-28 | 2017-06-20 | 诸暨市奇剑智能科技有限公司 | A kind of pearl sorting technique based on deep learning |
CN107016413A (en) * | 2017-03-31 | 2017-08-04 | 征图新视(江苏)科技有限公司 | A kind of online stage division of tobacco leaf based on deep learning algorithm |
-
2017
- 2017-08-27 CN CN201710746676.9A patent/CN109420622A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103971342A (en) * | 2014-05-21 | 2014-08-06 | 厦门美图之家科技有限公司 | Image noisy point detection method based on convolution neural network |
US20160358024A1 (en) * | 2015-06-03 | 2016-12-08 | Hyperverge Inc. | Systems and methods for image processing |
CN106326899A (en) * | 2016-08-18 | 2017-01-11 | 郑州大学 | Tobacco leaf grading method based on hyperspectral image and deep learning algorithm |
CN106874929A (en) * | 2016-12-28 | 2017-06-20 | 诸暨市奇剑智能科技有限公司 | A kind of pearl sorting technique based on deep learning |
CN107016413A (en) * | 2017-03-31 | 2017-08-04 | 征图新视(江苏)科技有限公司 | A kind of online stage division of tobacco leaf based on deep learning algorithm |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110516988A (en) * | 2019-07-08 | 2019-11-29 | 国网浙江省电力有限公司金华供电公司 | One kind being fitted Power Material method based on neural network |
JP2022019569A (en) * | 2020-07-17 | 2022-01-27 | 株式会社カナヤ食品 | Teacher data generation method, foreign matter inspection device, and foreign matter inspection method |
JP7053075B2 (en) | 2020-07-17 | 2022-04-12 | 株式会社カナヤ食品 | Teacher data generation method and foreign matter inspection device and foreign matter inspection method |
CN112189877A (en) * | 2020-10-13 | 2021-01-08 | 燕山大学 | On-line detection method for tobacco shred impurities in tobacco production line |
CN113210264A (en) * | 2021-05-19 | 2021-08-06 | 江苏鑫源烟草薄片有限公司 | Method and device for removing tobacco impurities |
CN113210264B (en) * | 2021-05-19 | 2023-09-05 | 江苏鑫源烟草薄片有限公司 | Tobacco sundry removing method and device |
CN114397297A (en) * | 2022-01-19 | 2022-04-26 | 河南中烟工业有限责任公司 | Rapid nondestructive testing method for starch content of flue-cured tobacco |
CN114397297B (en) * | 2022-01-19 | 2024-01-23 | 河南中烟工业有限责任公司 | Rapid nondestructive testing method for starch content of flue-cured tobacco |
CN115953384A (en) * | 2023-01-10 | 2023-04-11 | 杭州首域万物互联科技有限公司 | On-line detection and prediction method for tobacco morphological parameters |
CN115953384B (en) * | 2023-01-10 | 2024-02-02 | 杭州首域万物互联科技有限公司 | Online detection and prediction method for morphological parameters of tobacco leaves |
JP7498524B1 (en) | 2023-05-02 | 2024-06-12 | 株式会社カナヤ食品 | Method for generating teacher data, foreign body inspection device, foreign body inspection method, and foreign body detection program |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109420622A (en) | Tobacco leaf method for sorting based on convolutional neural networks | |
CN107633255B (en) | Rock lithology automatic identification and classification method under deep learning mode | |
CN107948166B (en) | Deep learning-based traffic anomaly detection method and device | |
CN110222681A (en) | A kind of casting defect recognition methods based on convolutional neural networks | |
CN109239102A (en) | A kind of flexible circuit board open defect detection method based on CNN | |
CN110211097A (en) | Crack image detection method based on fast R-CNN parameter migration | |
CN104573000B (en) | Automatic call answering arrangement and method based on sequence study | |
CN107305636A (en) | Target identification method, Target Identification Unit, terminal device and target identification system | |
CN110120036A (en) | A kind of multiple dimensioned tire X-ray defect detection method | |
CN109064454A (en) | Product defects detection method and system | |
CN108520114A (en) | A kind of textile cloth defect detection model and its training method and application | |
CN106096661B (en) | The zero sample image classification method based on relative priority random forest | |
CN107944504B (en) | Board recognition and machine learning method and device for board recognition and electronic equipment | |
CN105654066A (en) | Vehicle identification method and device | |
CN108830332A (en) | A kind of vision vehicle checking method and system | |
CN104102929A (en) | Hyperspectral remote sensing data classification method based on deep learning | |
CN104156565B (en) | System mode analysis method based on offline CAN bus data and analytical equipment | |
CN110473170A (en) | A kind of artificial intelligence detection method suitable for the true and false determining defects of pcb board | |
CN103324937A (en) | Method and device for labeling targets | |
CN108805196A (en) | Auto-increment learning method for image recognition | |
CN110647875A (en) | Method for segmenting and identifying model structure of blood cells and blood cell identification method | |
CN101441721A (en) | Device and method for counting overlapped circular particulate matter | |
CN112528774B (en) | Intelligent unknown radar signal sorting system and method in complex electromagnetic environment | |
CN107220664A (en) | A kind of oil bottle vanning counting method based on structuring random forest | |
CN106960176A (en) | A kind of pedestrian's gender identification method based on transfinite learning machine and color characteristic fusion |
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: 20190305 |