Disposable paper urine pants intelligent sorting system based on computer vision
Technical field
The present invention relates to a kind of sorting of disposable paper urine pants, in particular to a kind of disposable paper urine based on computer vision
Trousers intelligent sorting system.
Background technique
Sorting operation be home-delivery center according to demand or distribution plan by specified article quickly and accurately from storage space or
Other positions pick out, and the operation process classified according to certain mode, concentrated.How sorting efficiency is improved,
Improve the emphasis that sorting accuracy rate is sorting operation.
Current goods sorting basic skills mainly has:
(1) sifting sort manual sorting: is carried out to commodity by pure manual mode.This kind of sorting mode labor intensity
Greatly, sorting efficiency is low, and accuracy rate is affected by human factors.
(2) semi-automatic sorting: being main means of delivery using mechanical (such as conveyer), by being equipped in each sort positions
Operating personnel sorts.This sorting mode can reduce labor intensity to a certain extent, improve efficiency, however, there remains
It manually treats segregating articles to be identified, and carries out sorting operation, need certain human cost.
(3) automatic sorting: refer to cargo from enter sorting system to specified position until, all operations be according to
The instruction of machine is automatically performed.Therefore, such mode sorting efficiency is higher, and the assortment of article and quantity of sorting are also very big, but
Be sorting accuracy rate it is to be improved.
The sort plan of disposable paper urine pants commodity mainly has manual sorting, mechanical sorting, barcode scanning sorting, radio frequency knowledge at present
It does not sort.
Manual sorting: which manually identifies disposable paper diaper, screens and is classified by pure.
Mechanical sorting: the program transports disposable paper diaper using means of delivery (such as: conveyer belt), needs in the sorting stage
Manual identified merchandise classification is wanted, and is sent to sorting port manually.
Barcode scanning sorting: the program obtains bar code or two dimensional code in disposable paper urine pants packaging by barcode scanning instrument, and
By the included production information of solution wanding or two dimensional code, then rear end sorting manipulator is notified to divide paper diaper
Class.
Frequency penetrates identification sorting: the program establishes connection by the label on reader and disposable paper urine pants packaging, utilizes
Radiofrequency signal reads the production information of product, is equally fulfiled assignment by rear end sorting manipulator.
Above-mentioned method for sorting is respectively provided with following problems and disadvantage:
Manual sorting: which is only applicable in the disposable paper diaper sorting that quantity is few, type is single, and sorting efficiency is by people
The influence of work factor, type of merchandize, quantity.
Mechanical sorting: its operating system of which and process are relatively easy, but sorting efficiency is equally by artifact, commodity
The influence of type, quantity.It when sorting amount, type of merchandize increase, needs to extend producing line length, increases artificial quantity, it can be significant
The cost of sorting is promoted, and reduces the efficiency and precision of sorting.
Barcode scanning sorting: which improve to a certain extent artificial, semi-automatic sorting there are the problem of, but practical operation
In the process, due to will appear the uncertainty of barcode position, so needing through detection system to laying for goods and to shooting
Angle is adjusted, or even just be can guarantee to sweep as much as possible and taken bar code or two dimension in the multiple scanning cameras of multi-faceted arrangement
Code;If bar code or two dimensional code are just in the bottom of disposable paper urine pants packaging, such method failure.
Frequency penetrates sorting: which needs to enclose radio frequency to the disposable paper urine pants that every portion is produced in implementation process
Chip (coil), and corresponding product information is write to it, it is cumbersome that process is written in information;Meanwhile writing and reading radio frequency chip
The instrument and equipment of (coil) is also more expensive.Such method higher cost needs artificial assistance operation, sorting efficiency be not high to lack
Point.
Summary of the invention
The object of the present invention is to provide a kind of disposable paper urine pants intelligent sorting systems based on computer vision, to make up
The deficiencies in the prior art.
In order to achieve the above objectives, the present invention take the specific technical proposal is:
A kind of disposable paper urine pants intelligent sorting system based on computer vision includes transmission-sorting subsystem, image
Acquisition subsystem, and the embedded Control subsystem of the above-mentioned two subsystem of control;Wherein, the transmission-sorting subsystem packet
Include DC Transmission motor, conveyer belt, encoder and sorting manipulator and servo motor;Described image acquisition subsystem includes
High speed camera, LED light source group and photoelectric sensor;The embedded Control subsystem includes image processing apparatus GPU, centre
Manage device CPU, display and storage device.
The transmission-sorting subsystem receives the speed command that embedded Control subsystem is issued, and passes through DC Transmission
Motor and encoder provide conveyer belt real time execution speed to embedded Control subsystem to control transmission speed;The biography
The sorting instruction for sending-sorting subsystem to receive the sending of embedded Control subsystem, to control corresponding sorting manipulator work, with complete
At sorting work;When commodity pass through photoelectric sensor, the transmission-sorting subsystem provides light to embedded Control subsystem
Electric switch signal;It can be after receiving sorting instruction in order to transmit-sort subsystem, commodity can be sent down in time and be passed by sorting manipulator
Band is sent, off-sorting station photogate and sorting manipulator distance are set as LGF, and record sorting manipulator and receive to instruct to dynamic
The time of work, i.e. sorting manipulator response time are tG;In order to meet Image Acquisition and sorting demand, transmission-sorting is set
Conveyer belt movement velocity is v in subsystem.
Described image acquisition subsystem adopts high speed camera according to the provided image capture instruction of embedded Control subsystem
Collect image, and is uploaded to embedded Control subsystem;Described image acquisition subsystem is provided according to embedded Control subsystem
Shutter parameter, adjustment high speed camera aperture time are tc;In order to which image acquisition subsystem can be after image capture instruction, high speed
Phase function obtains commodity image information in time, and detection zone photogate and high speed camera distance are set as LGC。
The embedded Control subsystem issues Image Acquisition to image capturing system and refers to according to off-sorting station photoelectricity gate signal
It enables;Transmission-sorting subsystem obtains commodity image information from image acquisition subsystem, and depth network VGG- is run on GPU
16, and provide the type of merchandise;When receiving off-sorting station photoelectricity gate signal, embedded Control subsystem is to transmission-sorting subsystem
System issues sorting instruction;The embedded Control subsystem shows conveyer belt real time execution speed, manipulator shape over the display
State, high speed camera aperture time, view synthesis time;16.04 ring of Ubuntu is run on the embedded Control subsystem
Border, depth network are run on TensorFlow platform;In order to enable embedded Control subsystem to handle input picture in time,
Distance is set as L between tested commodityP, image averaging processing time trAs depth network VGG-16 is handled used in single image
Average time.
The conveyer belt can be in turn divided into area, detection zone and off-sorting station to be measured, and direct current generator is located at conveyor belt two sides;Institute
High speed camera to be stated to be located at right above conveyer belt, alignment lens detection zone, the LED light source group is centered around by high speed camera camera lens,
Source alignment detection zone;The sorting manipulator is distributed in off-sorting station both sides, and passes through Serve Motor Control;The photoelectric sensing
Device is set to conveyer belt detection zone entrance and off-sorting station two sides;The embedded Control subsystem and sorting manipulator, photoelectric sensing
Device, conveyer belt encoder, high speed camera, high-performance server are connected, and are controlled.
When system starts, above-mentioned transmission-sorting subsystem, image acquisition subsystem and embedded Control subsystem three
It is both needed to start;After starting, conveyer belt is run according to setting speed, and all sorting manipulators reset, and LED light source group is opened, photoelectricity
Sensor is opened, the booting of embedded Control subsystem;Successively interval enters biography to disposable paper urine pants to be sorted at a certain distance
It is dynamic to take, and enter detection zone behind area to be measured;In the detection, it is acquired by high speed camera and photoelectric sensor wait divide
Paper diaper image is picked, wherein the photoelectric sensor of detection zone and off-sorting station, predominantly detects whether article to be sorted reaches specified area
Domain;High speed camera is located at right above conveyer belt, around with surrounding there are six LED light source group, to eliminate the light of paper diaper to be sorted
Shadow influences, and stable illumination condition is provided, to obtain relatively sharp images of items to be sorted, in the letter for receiving photoelectric sensor
After number, alignment lens detection zone acquires the high-definition image of disposable paper urine pants.
Image is sent in high-performance server from high speed camera, then by trained depth network model to primary
Property paper diaper image carries out prediction classification.
Further, above-mentioned depth network model is the convolutional neural networks VGG-16 of multilayered structure, includes 7 layers of convolution
Layer, 7 layers of pond layer and ReLu activation primitive, 3 layers of full articulamentum, finally by softmax layers of output category result, by being embedded in
GPU in formula control subsystem completes training link.
Further, above-mentioned classification results are transmitted in transmission-sorting subsystem, according to different classification results, distribution
Corresponding sorting manipulator controls signal, when disposable paper urine pants correspond to photoelectric sensor triggering by off-sorting station, transmission-point
Signal can be sorted by picking subsystem and receiving, and drive the servo motor to move according to category signal, disposable paper urine pants are divided into conjunction
Suitable region, then sorting manipulator, which resets, waits next disposable paper urine pants, is finally completed sorting.
For effective monitoring production line environment, software is controlled equipped with the visualization based on MFC in the present system, controls software
It can display real-time and be detected commodity amount on conveyer belt, commodity amount to be sorted completes classified commodity quantity;In order to increase fault
Recall rate makes a record the lower commodity of categorized but matching degree.
Advantages of the present invention and technical effect:
1. this system uses high-performance GPU in server end, training and predetermined speed of depth network are effectively improved.Make
There is powerful feature learning ability with convolutional network, the appearance information of commodity most feature can be obtained, and classify.Pass through
Training, the disposable paper urine pants type of Network Recognition can be up to hundreds of, and have quite or surmount the energy of eye recognition precision
Power, so that sorting system has good classifying quality.Network Recognition speed is also above manual sort's speed.
2. the work shape that the present invention can obtain photoelectric sensor, encoder, high speed camera by embedded Control subsystem
State, and recorded on high-performance server.It joined visualization interface simultaneously, user can intuitively grasp real each by server
Class subsystem work state, and direct monitoring production line production status on a display screen, such as detection speed, detection accuracy,
Complete urine pants quantity etc..
3. the present invention monitors tested commodity using two detection zone, off-sorting station photoelectric sensor and conveyor belt speed aspects
Position can accurately judge product locations in real time, to realize accurately sorting.
Detailed description of the invention
Fig. 1 is basic framework and work flow diagram of the invention.
Specific embodiment
The present invention is explained further and is illustrated below by way of specific embodiment and in conjunction with attached drawing.
As shown in Figure 1, a kind of disposable paper urine pants intelligent sorting system based on computer vision, includes transmission-sorting
Subsystem, image acquisition subsystem, and the embedded Control subsystem of the above-mentioned two subsystem of control;Wherein, the transmission-point
Picking subsystem includes DC Transmission motor, conveyer belt, encoder and sorting manipulator and servo motor;Described image acquisition
Subsystem includes high speed camera, LED light source group and photoelectric sensor;The embedded Control subsystem includes image processing apparatus
GPU, central processor CPU, display and storage device.
The transmission-sorting subsystem receives the speed command that embedded Control subsystem is issued, and passes through DC Transmission
Motor and encoder provide conveyer belt real time execution speed to embedded Control subsystem to control transmission speed;The biography
The sorting instruction for sending-sorting subsystem to receive the sending of embedded Control subsystem, to control corresponding sorting manipulator work, with complete
At sorting work;When commodity pass through photoelectric sensor, the transmission-sorting subsystem provides light to embedded Control subsystem
Electric switch signal;It can be after receiving sorting instruction in order to transmit-sort subsystem, commodity can be sent down in time and be passed by sorting manipulator
Band is sent, off-sorting station photogate and sorting manipulator distance are set as LGF, and record sorting manipulator and receive to instruct to dynamic
The time of work, i.e. sorting manipulator response time are tG;In order to meet Image Acquisition and sorting demand, transmission-sorting is set
Conveyer belt movement velocity is v in subsystem.
Described image acquisition subsystem adopts high speed camera according to the provided image capture instruction of embedded Control subsystem
Collect image, and is uploaded to embedded Control subsystem;Described image acquisition subsystem is provided according to embedded Control subsystem
Shutter parameter, adjustment high speed camera aperture time are tc;In order to which image acquisition subsystem can be after image capture instruction, high speed
Phase function obtains commodity image information in time, and detection zone photogate and high speed camera distance are set as LGC.
The embedded Control subsystem issues Image Acquisition to image capturing system and refers to according to off-sorting station photoelectricity gate signal
It enables;Transmission-sorting subsystem obtains commodity image information from image acquisition subsystem, and depth network VGG- is run on GPU
16, and provide the type of merchandise;When receiving off-sorting station photoelectricity gate signal, embedded Control subsystem is to transmission-sorting subsystem
System issues sorting instruction;The embedded Control subsystem shows conveyer belt real time execution speed, manipulator shape over the display
State, high speed camera aperture time, view synthesis time;16.04 ring of Ubuntu is run on the embedded Control subsystem
Border, depth network are run on TensorFlow platform;In order to enable embedded Control subsystem to handle input picture in time,
Distance is set as LP between tested commodity, and image averaging processing time tr is used in depth network VGG-16 processing single image
Average time.
The conveyer belt can be in turn divided into area, detection zone and off-sorting station to be measured, and direct current generator is located at conveyor belt two sides;Institute
High speed camera to be stated to be located at right above conveyer belt, alignment lens detection zone, the LED light source group is centered around by high speed camera camera lens,
Source alignment detection zone;The sorting manipulator is distributed in off-sorting station both sides, and passes through Serve Motor Control;The photoelectric sensing
Device is set to conveyer belt detection zone entrance and off-sorting station two sides;The embedded Control subsystem and sorting manipulator, photoelectric sensing
Device, conveyer belt encoder, high speed camera are connected, and are controlled.
When system starts, above-mentioned transmission-sorting subsystem, image acquisition subsystem and embedded Control subsystem three
It is both needed to start;After starting, conveyer belt is run according to setting speed, and all sorting manipulators reset, and LED light source group is opened, photoelectricity
Sensor is opened, embedded subsystem and high-performance server booting;Disposable paper urine pants to be sorted successively at a certain distance between
Every entering on transmission belt, and enter detection zone behind area to be measured;In the detection, pass through high speed camera and photoelectric sensor
Paper diaper image to be sorted is acquired, wherein whether the photoelectric sensor of detection zone and off-sorting station, predominantly detect article to be sorted
Reach specified region;High speed camera is located at right above conveyer belt, and around LED light source group surrounds with there are six, to be sorted to eliminate
The shadow of paper diaper influences, and provides stable illumination condition, to obtain relatively sharp images of items to be sorted, is receiving photoelectricity
After the signal of sensor, alignment lens detection zone acquires the high-definition image of disposable paper urine pants.
Image reaches embedded Control subsystem from high speed camera, then by trained depth network model to primary
Property paper diaper image carries out prediction classification.
Further, above-mentioned depth network model is the convolutional neural networks VGG-16 of multilayered structure, includes 7 layers of convolution
Layer, 7 layers of pond layer and ReLu activation primitive, 3 layers of full articulamentum, finally by softmax layers of output category result, by being embedded in
GPU in formula control system completes training link.
Further, above-mentioned classification results are transmitted in transmission-sorting subsystem, according to different classification results, distribution
Corresponding sorting manipulator controls signal, when disposable paper urine pants correspond to photoelectric sensor triggering by off-sorting station, transmission-point
Signal can be sorted by picking subsystem and receiving, and drive the servo motor to move according to category signal, disposable paper urine pants are divided into conjunction
Suitable region, then sorting manipulator, which resets, waits next disposable paper urine pants, is finally completed sorting.
For effective monitoring production line environment, software is controlled equipped with the visualization based on MFC in the present system, controls software
It can display real-time and be detected commodity amount on conveyer belt, commodity amount to be sorted completes classified commodity quantity;In order to increase fault
Recall rate makes a record the lower commodity of categorized but matching degree.
Workflow: paper diaper type n is first counted, the softmax layer neuronal quantity in sorter network VGG-16 is repaired
It is changed to n.Since every class diaper outer packaging has 6 faces, each face image of every class soap outer packing is obtained using high speed camera
SampleI is paper diaper species number, in the range of 1≤i≤n;F is outer packing surface number, in the range of 1≤f≤6.It is assumed to be
1 class paper diaper obtain the direct picture in six faces of diaper outer packaging using high speed cameraIt is so-called
Direct picture refers to that packed surface is consistent with camera perspective angle, and packed surface is without inclination.In order to promote sorter network VGG-16 classification
Performance is needed to image noise reduction and enhancement.Image denoising is filtered image using smothing filtering, eliminate noise to figure
As the influence of feature.Image enhancement will rotate image to obtain more multilevel image data, rotation angle is respectively 30 °,
60 °, 90 °, 120 °, 150 °, 180 °, direction of rotation includes clockwise and anticlockwise.It then will be by image noise reduction and enhancement
Image pattern is put into depth network VGG-16, is trained to network, and trained sorter network VGG-16 is obtained.Operation
Depth network query function goes out image averaging processing time tr.Next sorting equipment is opened, sorting is inputted into embedded subsystem
Manipulator maximum response time tr;Input detection zone photoelectric sensor and high speed camera central point distance Lgc;Input off-sorting station
Photoelectric sensor and sorting manipulator distance Lgr;Equispaced between each paper diaper of adjustment volume is LM, i.e., by paper diaper
By interval LMIt puts on conveyer belt.Setting transmission tape running speed is v, and wherein speed of service v should meet LM>v×trProtect
For card when embedded subsystem handles present image, not new paper diaper enter detection zone.Transmit tape running speed v and shutter
Time tcIt needs to meet, Lgc≈v×tc, i.e., in aperture time, paper diaper should be transmitted to high speed camera center.Conveyer belt fortune
Scanning frequency degree v should meet LGF>v×tG, i.e., within the manipulator response time, paper diaper can be pushed out conveyer belt by manipulator.Then
Transmission-sorting subsystem, image acquisition subsystem is opened, sorting system is allowed to enter working condition, allows paper diaper commodity according to setting
The distance L setMConveyer belt is sequentially entered, conveyer belt is sorted according to above-mentioned parameter commodity.