CN109389161A - Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning - Google Patents

Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning Download PDF

Info

Publication number
CN109389161A
CN109389161A CN201811137406.9A CN201811137406A CN109389161A CN 109389161 A CN109389161 A CN 109389161A CN 201811137406 A CN201811137406 A CN 201811137406A CN 109389161 A CN109389161 A CN 109389161A
Authority
CN
China
Prior art keywords
rubbish
image data
spam samples
identification
samples image
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
Application number
CN201811137406.9A
Other languages
Chinese (zh)
Inventor
刘长红
杨兴鑫
张宏康
李文杰
朱亮宇
钟志鹏
程健翔
范俊宇
黄楠
陈建堂
彭绍湖
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou University
Original Assignee
Guangzhou University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangzhou University filed Critical Guangzhou University
Priority to CN201811137406.9A priority Critical patent/CN109389161A/en
Publication of CN109389161A publication Critical patent/CN109389161A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
    • G06K9/6256Obtaining sets of training patterns; Bootstrap methods, e.g. bagging, boosting
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B65CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65FGATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F1/00Refuse receptacles; Accessories therefor
    • B65F1/0033Refuse receptacles; Accessories therefor specially adapted for segregated refuse collecting, e.g. receptacles with several compartments; Combination of receptacles
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B65CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65FGATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F1/00Refuse receptacles; Accessories therefor
    • B65F1/10Refuse receptacles; Accessories therefor with refuse filling means, e.g. air-locks
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
    • G06K9/6262Validation, performance evaluation or active pattern learning techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6267Classification techniques
    • G06K9/6268Classification techniques relating to the classification paradigm, e.g. parametric or non-parametric approaches
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B65CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65FGATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F1/00Refuse receptacles; Accessories therefor
    • B65F1/0033Refuse receptacles; Accessories therefor specially adapted for segregated refuse collecting, e.g. receptacles with several compartments; Combination of receptacles
    • B65F2001/008Means for automatically selecting the receptacle in which refuse should be placed
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02WCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO WASTEWATER TREATMENT OR WASTE MANAGEMENT
    • Y02W30/00Technologies for solid waste management
    • Y02W30/10Waste collection, transportation, transfer or storage, e.g. segregated refuse collecting, electric or hybrid propulsion

Abstract

The invention discloses a kind of, and the rubbish based on deep learning identifies evolutionary learning method, apparatus, system and medium, which comprises obtains spam samples image data;Spam samples image data is pre-processed;Using pretreated spam samples image data as the input parameter of neural network, it is compared with rubbish identification model has been trained, according to comparison result, judges whether to identify successfully;It will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism;It is identified again by spam samples image data of the ResNet algorithm to recognition failures, ResNet algorithm is identified that successful spam samples image data is marked, and rubbish corresponding information is fed back into garbage classification throwing mechanism, update rubbish identification model;ResNet algorithm is identified that successful spam samples image data is transferred to user or maintenance personnel to be marked, updates rubbish identification model.The present invention greatly reduces the workload of maintenance personnel, realizes the Accurate classification to a large amount of rubbish.

Description

Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning
Technical field
The present invention relates to a kind of rubbish recognition methods, especially a kind of rubbish based on deep learning identifies evolutionary learning side Method, device, system and storage medium belong to rubbish identification technology field.
Background technique
The origin of machine learning can be traced to long long ago, and deep learning is also within the scope of machine learning, it makees A frontier for machine learning field occurred in 2006 and this several years growth momentum does not subtract.The structure of deep learning is For the learning organization of traditional shallow-layer, deep learning structure for shallow structure is more complicated, is related to Neural network the training number of plies it is more, so it can learn relative to traditional shallow-layer learning structure it is more abstract to data Feature, sensitiveer for the identification of complicated image, then the field of its application includes more extensively national defence, traffic, medical treatment etc., and is evolved Study is that the evolution of its study mechanism is realized on the basis of deep learning, optimizes its learning performance, and the dynamics of evolution is most Number is the input parameter of neural network itself.The rubbish that the invention proposes a kind of based on deep learning identify evolutionary learning into System, is related to object detection field, identifies classification aspect specific to rubbish, most of to apply in current rubbish identification technology It is not using evolution algorithm, because its training pattern is to have met after the complete computer of the neural metwork training of deep learning It is required, even if its model is changed, and artificial a small amount of change input parameter, so the identifiable rubbish type of its model Always be limited, in addition to this, the limited source of the training sample data of model be also restrict the prior art it is important because Element, because the sample data of most models needs artificial regeneration, time-consuming and effect is poor.
Summary of the invention
The first purpose of the invention is to provide a kind of, and the rubbish based on deep learning identifies evolutionary learning method, this method Current rubbish identification technology is improved, while having trained rubbish by the improvement of dynamics of evolution real-time update of the bad data of recognition effect Rubbish identification model makes that rubbish identification model has been trained to have the unexistent real-time of Other Waste recyclable device and accuracy, and And greatly reduce the workload of maintenance personnel, it realizes the Accurate classification to a large amount of rubbish, solves current rubbish identification device identification The problem of rubbish limited amount and precision deficiency is constantly independently updated and with this during the work time at the same time Optimization.
Second object of the present invention is to provide a kind of rubbish identification evolutionary learning device based on deep learning.
Third object of the present invention is to provide a kind of rubbish identification evolutionary learning system based on deep learning.
Fourth object of the present invention is to provide a kind of storage medium.
The first purpose of this invention can be reached by adopting the following technical scheme that:
Rubbish based on deep learning identifies evolutionary learning method, which comprises
Obtain spam samples image data;
Spam samples image data is pre-processed;
Using pretreated spam samples image data as the input parameter of neural network, mould is identified with rubbish has been trained Type is compared, and according to comparison result, judges whether to identify successfully;
It will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism;Wherein, the rubbish corresponding information Including rubbish type;
It is identified again by spam samples image data of the ResNet algorithm to recognition failures;
ResNet algorithm is identified that successful spam samples image data is marked, and rubbish corresponding information is fed back to Garbage classification throwing mechanism, accumulation identify successfully marked spam samples image data, update rubbish identification model;
The spam samples image data of ResNet algorithm recognition failures is transferred to user or maintenance personnel to be marked, And rubbish corresponding information is fed back into garbage classification throwing mechanism, the marked spam samples image data of recognition failures is accumulated, Update rubbish identification model.
Further, described using pretreated spam samples image data as the input parameter of neural network, and Training rubbish identification model is compared, and according to comparison result, judges whether to identify successfully, specifically:
Using pretreated spam samples image data as the input parameter of neural network, by having trained rubbish to identify The rubbish identification and evaluation system of model specification is to determine the recognition accuracy and rubbish type of rubbish, if the recognition accuracy of rubbish More than or equal to given threshold, then judge to identify successfully, if the recognition accuracy of rubbish is less than given threshold, judges that identification is lost It loses.
It is further, described that spam samples image data is pre-processed, specifically: spam samples image data is gone Mean value, normalization, PCA and whitening processing.
Second object of the present invention can be reached by adopting the following technical scheme that:
Rubbish based on deep learning identifies evolutionary learning device, and described device includes:
Module is obtained, for obtaining spam samples image data;
Preprocessing module, for being pre-processed to spam samples image data;
First identification module, for using pretreated spam samples image data as the input parameter of neural network, It is compared with rubbish identification model has been trained, according to comparison result, judges whether to identify successfully;
Feedback module, for when identifying successfully, rubbish corresponding information to be fed back to garbage classification throwing mechanism;Wherein, The rubbish corresponding information includes rubbish type;
Second identification module is used for when recognition failures, by ResNet algorithm to the spam samples image of recognition failures Data are identified again;
First update module, for by ResNet algorithm identify successful spam samples image data to be marked, and Rubbish corresponding information is fed back into garbage classification throwing mechanism, accumulation identifies successfully marked spam samples image data, more New rubbish identification model;
Second update module, for the spam samples image data of ResNet algorithm recognition failures to be transferred to user or dimension Shield personnel are marked, and rubbish corresponding information is fed back to garbage classification throwing mechanism, accumulate the marked rubbish of recognition failures Rubbish sample image data updates rubbish identification model.
Further, first identification module, specifically:
For using pretreated spam samples image data as the input parameter of neural network, by having trained rubbish The rubbish identification and evaluation system of identification model setting is to determine the recognition accuracy and rubbish type of rubbish, if the identification of rubbish is quasi- True rate is greater than or equal to given threshold, then judges to identify successfully, if the recognition accuracy of rubbish is less than given threshold, judges to know Do not fail.
Third object of the present invention can be reached by adopting the following technical scheme that:
Rubbish based on deep learning identifies evolutionary learning system, the system comprises:
Garbage classification throwing mechanism, for acquiring spam samples image data, according to the rubbish phase of message handler feedback Information is answered to carry out putting on by classification operation to rubbish;
Message handler pre-processes spam samples image data for obtaining spam samples image data, will be pre- Input parameter of the spam samples image data that treated as neural network is compared with rubbish identification model has been trained, According to comparison result, judge whether to identify successfully;It will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism; The spam samples image data of recognition failures is transferred to server;Receive the successful marked rubbish of identification that server is sent Rubbish corresponding information is fed back to garbage classification throwing mechanism by sample image data, and accumulation identifies successfully marked rubbish sample This image data updates rubbish identification model, and receives the marked spam samples image for the recognition failures that server is sent Rubbish corresponding information is fed back to garbage classification throwing mechanism by data, accumulates the marked spam samples picture number of recognition failures According to update rubbish identification model;Wherein, the rubbish corresponding information includes rubbish type;
Server will for being identified again by ResNet algorithm to the spam samples image data of recognition failures ResNet algorithm identifies that successful spam samples image data is marked, and will identify successfully marked spam samples image Data are sent to message handler;The spam samples image data of ResNet algorithm recognition failures is transferred to user or maintenance people The marked spam samples image data of recognition failures is sent to message handler to be marked by member.
Further, the system also includes human-computer interaction device, the server will by human-computer interaction device The spam samples image data of ResNet algorithm recognition failures is transferred to user and is marked, if user is not marked, Spam samples image data failed transmission is marked server to maintenance personnel.
Further, the garbage classification throwing mechanism includes camera, singulizing disc, dustbin, controller and multiple leakages Struggle against component, and the central location of dustbin is arranged in the singulizing disc, and singulizing disc has multiple rubbish placement sections, the dustbin With multiple groups rubbish dispensing portion, the funnel assemblies, rubbish placement section and rubbish dispensing portion are to correspond, and every group of rubbish is thrown Putting portion includes two rubbish dispensing portions, and each funnel assemblies are arranged above corresponding one group of rubbish dispensing portion, and including two Contrary funnel, the camera are connect with message handler, for acquiring the spam samples image on rubbish placement section Data, and it is transferred to message handler, the controller horizontally rotates and each rubbish placement section for control tactics disk Vertical rotation, and each funnel of control horizontally rotate and vertically rotate.
Further, the garbage classification throwing mechanism further includes support rod, and the support rod is fixed on dustbin, institute Camera is stated to be arranged on support rod.
Fourth object of the present invention can be reached by adopting the following technical scheme that:
Storage medium is stored with program, when described program is executed by processor, realizes above-mentioned rubbish identification evolutionary learning Method.
The present invention have compared with the existing technology it is following the utility model has the advantages that
1, the present invention, which first passes through, has trained rubbish identification model to judge to acquire whether spam samples image data identifies success, It is identified in the case where recognition failures, then through spam samples image data of the ResNet algorithm to recognition failures, it will ResNet algorithm identifies that successful spam samples image data is marked, and accumulation identifies successfully marked spam samples image Data, update rubbish identification model, and by the spam samples image data of ResNet algorithm recognition failures be transferred to user or Maintenance personnel accumulates the marked spam samples image data of recognition failures to be marked, and updates rubbish identification model, is formed Closed loop update, rubbish identification model is updated by the learning process constantly recycled, embodies trend of evolution, substantially increases Rubbish identification precision and range, solve the prior art because training sample it is limited caused by identifiable rubbish precision and The problem of lazy weight.
2, the present invention has very strong real-time, numerous for modern society's commodity, type, external form and the every spy of commodity Sign changes at any time and constantly, accomplishes automatically updating to a certain extent, and the workload of maintenance personnel also greatly reduces, while its The man-machine interaction having also contributes to being widely popularized for rubbish identification technology.
3, the present invention has certain generalization, although the present invention is directed evolutionary learning answering in rubbish identification technology With, but the system that closed loop of the invention updates can be generalized to other fields by the modification of target, accomplish this skill The large-scale popularization of art.
Detailed description of the invention
Fig. 1 is that the rubbish based on deep learning of the embodiment of the present invention 1 identifies the structural block diagram of evolutionary learning system.
Fig. 2 is the three-dimensional structure diagram of the garbage classification throwing mechanism of the embodiment of the present invention 1.
Fig. 3 is the facing structure figure of the garbage classification throwing mechanism of the embodiment of the present invention 1.
Fig. 4 is the enlarged drawing in Fig. 3 at A.
Fig. 5 is the enlarged drawing in Fig. 3 at B.
Fig. 6 is the side block diagram of the garbage classification throwing mechanism of the embodiment of the present invention 1.
Fig. 7 is that the rubbish based on deep learning of the embodiment of the present invention 1 identifies the SSD algorithm that evolutionary learning system uses Easy structure figure.
Fig. 8 is that the rubbish based on deep learning of the embodiment of the present invention 1 identifies that the rubbish of evolutionary learning system identifies evolution Learn schematic diagram.
Fig. 9 is that the rubbish based on deep learning of the embodiment of the present invention 1 identifies the work flow diagram of evolutionary learning system.
Figure 10 is that the rubbish based on deep learning of the embodiment of the present invention 2 identifies evolutionary learning method flow diagram.
Figure 11 is that the rubbish based on deep learning of the embodiment of the present invention 3 identifies evolutionary learning apparatus structure block diagram.
Wherein, 1- garbage classification throwing mechanism, 2- message handler, 3- server, 101- camera, 102- singulizing disc, 1021- rubbish placement section, 1022- collector ring, 1023- first motor, 1024- first shaft coupling, the first encoder of 1025-, 1026- cylinder, 103- dustbin, 104- controller, 105- funnel assemblies, 1051- funnel, 1052- steering engine, 1053- steering engine cloud Platform, the second motor of 1054-, 1055- second shaft coupling, 1056- second encoder, 106- support rod, 4- communication module, 5- auxiliary Work peripheral hardware, 6- human-computer interaction device, 7- user, 8- maintenance personnel.
Specific embodiment
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited In this.
Embodiment 1:
Machine learning in the field of target identification using very extensive, but its data set build it is most of by manpower, Lead to the model trained always and be defective, needs the high-frequency maintenance work of maintenance personnel.Therefore, the present embodiment provides A kind of rubbish based on deep learning identifies that evolutionary learning system, the system are considered to identify field, the class of rubbish in rubbish Type makes rapid progress, and computer identifies that the data characteristics for needing to extract when target is also changeable, so evolutionary learning mechanism is proposed, to know Not ineffective data are that dynamics of evolution real-time update improves computer aid training model, to adapt to identified characteristics of spam It is changeable, the precision and range of identification are increased substantially on the basis of existing technology.
As depicted in figs. 1 and 2, the rubbish identification evolutionary learning system based on deep learning of the present embodiment includes rubbish point Class throwing mechanism 1, message handler 2 and server 3, message handler 2 are connect with garbage classification throwing mechanism 1, information processing Device 2 is connect by communication module 4 with server 3.
The garbage classification throwing mechanism 1, for acquiring spam samples image data, according to the rubbish of message handler feedback Rubbish corresponding information carries out rubbish to put on by classification operation.
Shown in FIG. 1 to FIG. 6, the garbage classification throwing mechanism 1 include camera 101, singulizing disc 102, dustbin 103, Controller 104 and four funnel assemblies 105, the central location of dustbin 103 is arranged in singulizing disc 102, and singulizing disc 102 has Four rubbish placement sections 1021, dustbin 103 have four groups of rubbish dispensing portions 1031, funnel assemblies 105, rubbish placement section 1021 It is to correspond with rubbish dispensing portion 1031, every group of rubbish dispensing portion 1031 includes two rubbish dispensing portions 1031, general next Say that eight rubbish dispensing portions 103 can launch the rubbish of eight seed types, but under normal circumstances, the type of rubbish do not have it is so much, In the present embodiment, by wherein four rubbish dispensing portions 103 be respectively intended to launch metal, plastics, glass, papery rubbish, remain Under four rubbish dispensing portions 103 then be used to launch Other Waste, Other Waste refers to message handler 2 and server 3 all The rubbish of recognition failures obtains its corresponding information, is just classified as Other Waste, is collected into the laggard pedestrian's work point of certain amount Class, and can be identified according to updated rubbish identification model when launching same type rubbish again, each funnel assemblies 105 are arranged above corresponding one group of rubbish dispensing portion, and the funnel 1051 opposite including both direction, camera 101 and letter It ceases processor 2 to connect, the shooting higher adjustable focus camera of precision (clarity for guaranteeing image) is adopted as, for acquiring rubbish Spam samples image data on rubbish placement section 1021, and it is transferred to message handler 2, controller 104 and message handler 2 connect It connects, the single-chip microcontroller of STM series can be used, for horizontally rotating and each rubbish placement section 1021 for control tactics disk 102 Vertical rotation, and each funnel 1051 of control horizontally rotate and vertically rotate.
Specifically, singulizing disc 102 is arranged right below collector ring 1022 and first motor 1023, and first motor 1023 can be with Using double shaft step motors out, it is located at the lower section of collector ring 1022, and pass through first shaft coupling 1024 and the first encoder 1025 connections, first motor 1023 are also connect by driving circuit with controller 104, for the instruction according to controller, are driven Singulizing disc 102 horizontally rotates;A cylinder 1026 is equipped with below each rubbish placement section 1021 of singulizing disc 102, each Cylinder 1026 is connected by connector and corresponding rubbish placement section 1021, to push corresponding rubbish placement section 1021, is made The corresponding vertically rotation of rubbish placement section 1021, i.e., rubbish placement section 1021 can be around the pivot of singulizing disc 102.
Specifically, the lower section of each funnel 1051 is equipped with a steering engine 1052, and every steering engine 1052 passes through a steering engine cloud Platform 1053 is fixed on 1051 lower section of corresponding funnel, and for each funnel assemblies 105, two steering engine holders 1053 are respectively with second Motor 1054 connects, which can use stepper motor, passes through second shaft coupling 1055 and second encoder 1056 connections, the second motor 1054 is also connect by driving circuit with controller 104, for the instruction according to controller 104, band Corresponding two steering engine holders 1053 rotation is moved, to make corresponding funnel assemblies 105 and two steering engines 1052 are horizontal together to turn It is dynamic, while every steering engine 1052 can drive the corresponding vertically rotation of funnel 1051;When the rubbish of continuous several same types is thrown into When the rubbish placement section 1021 of singulizing disc 102, can shorten wait funnel assemblies 105 go back to catch rubbish placement section 1021 when Between.
The spam samples image data on rubbish placement section 1021 is acquired in order to facilitate camera 101, garbage classification is launched Mechanism further includes support rod 106, and support rod 106 is fixed on dustbin 103, and camera 101 is arranged on support rod 106, tool Body, support rod 106 is made of aluminum profile.
1 working principle of garbage classification throwing mechanism of the present embodiment is: can be succeeded with message handler 2 or server 3 Identify rubbish type for be illustrated, when people by rubbish launch to singulizing disc 102 one of rubbish placement section After 1021, camera 101 acquires spam samples image data, and is transferred to message handler 2, when controller 104 receives letter After ceasing the rubbish corresponding information that processor 2 is fed back, the rubbish corresponding information that controller 104 is fed back according to message handler 2, the rubbish Rubbish corresponding information includes the type of rubbish, and control first motor 1023 drives singulizing disc 102 to horizontally rotate, and makes the rubbish for placing rubbish Rubbish placement section 1021 horizontally rotates one group of rubbish dispensing portion, 1031 position where respective type, controls 1054 band of the second motor The funnel assemblies 105 for moving 1031 top of this group of rubbish dispensing portion rotate, and are directed at the outlet of one of funnel 1051 and place rubbish The rubbish placement section 1021 of rubbish drives rubbish placement section 1021 vertically to rotate, rubbish is allow to enter the leakage by cylinder 1026 Bucket 1051, the funnel assemblies 105 where the second motor 1054 of control drives the funnel 1051 horizontally rotate, and make 1051 water of funnel Flat to turn to 1031 position of rubbish dispensing portion where respective type, control steering engine 1052 turns the funnel 1051 vertically downward It is dynamic, rubbish is launched into the rubbish dispensing portion 1031 where respective type.
The message handler 2 is embedded system, for obtaining spam samples image data, to spam samples picture number According to being pre-processed, using pretreated spam samples image data as the input parameter of neural network, and rubbish has been trained Identification model is compared, and according to comparison result, judges whether to identify successfully;It will be identified by back work peripheral hardware 5 successful Rubbish corresponding information is output to display, and will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism 1, should Rubbish corresponding information includes rubbish type, and garbage classification throwing mechanism 1 puts on by classification rubbish according to rubbish corresponding information; The spam samples image data of recognition failures is transferred to server 3 by communication module 4, and then executes further identification behaviour Make, additionally can receive the successful marked spam samples image data of identification of the transmission of server 3, by rubbish corresponding information Garbage classification throwing mechanism 1 is fed back to, accumulation identifies successfully marked spam samples image data, updates rubbish and identifies mould Type, and the marked spam samples image data for the recognition failures that server 3 is sent is received, rubbish corresponding information is fed back to Garbage classification throwing mechanism 1 accumulates the marked spam samples image data of recognition failures, updates rubbish identification model;Wherein, The rubbish corresponding information includes rubbish type, and pretreatment includes removing mean value, normalization, PCA (Principal Component Analysis, principal component analysis) and albefaction, it is 300*300 that image, which unifies size, after pretreatment, and spam samples image data refers to The corresponding rgb value of image;Having trained rubbish identification model is that computer passes through SSD (single shot multibox Detector) algorithm train come, SSD algorithm structure as shown in fig. 7, the good enough suitable sizes of manual sorting picture sample Afterwards, by the rgb value of image, i.e. input parameter of the three-dimensional information as computer neural network, after certain training time Form suitable model;As shown in figure 8, using pretreated spam samples image data as the input parameter of neural network, It is compared with rubbish identification model has been trained, according to comparison result, judges whether to identify successfully, specifically: after pretreatment Input parameter of the spam samples image data as neural network, pass through the rubbish identification for having trained rubbish identification model to set Appraisement system is to determine the recognition accuracy (i.e. accuracy of identification) and rubbish type of rubbish, i.e. output the result is that the identification of rubbish is quasi- True rate and rubbish type judge to identify successfully, if the recognition accuracy of rubbish is greater than or equal to given threshold if the knowledge of rubbish Other accuracy rate is less than given threshold, then judges recognition failures, which, which is subordinated to, has trained the setting of rubbish identification model Rubbish identification and evaluation system, the given threshold have identified the recognition result exported after rubbish primarily directed to message handler, should The affecting parameters of given threshold include the accuracy rate of identification and the coefficient of stability of recognition result, and the coefficient of stability includes a kind of rubbish The size of the number of types and posting that identify, entire rubbish identification and evaluation system is by comparing recognition result and setting The relative size of threshold value determines the operation of next step.
The communication module 4 can use bluetooth module, Wi-Fi module or wire communication equipment, be used for message handler 2 With the communication between server 3, effect is the image transmitting by 2 recognition failures of message handler to server 3, while also will clothes Treated again in business device 3, and junk data is traveled back to message handler 2, and specifically, communication module 4 is passed to server 3 The image data of defeated recognition failures includes the not high data of recognition accuracy and discrimination lower (or completely identification less than) Data, the not high data of recognition accuracy refer to that the recognition result of output has rubbish type to identify posting, but the identification exported Accuracy rate is very low or identification mistake, the data of opposite recognition accuracy lower (or completely identification less than) refer to that identification is accurate Rate is also lower than the former, and almost nil or even recognition result identifies posting without rubbish type.
The back work peripheral hardware 5 includes power supply unit and signal output apparatus, and power supply unit is that entire work structuring mentions For energy, including garbage classification throwing mechanism 1, rubbish corresponding information is output to display screen by signal output apparatus.
The server 3 can use enterprise-level server, consider that entire rubbish identification evolutionary learning system needs to handle A large amount of junk data needs powerful computing capability, and is also required to networking and builds training set, for passing through ResNet algorithm The spam samples image data of recognition failures is identified again, ResNet algorithm is identified into successful spam samples picture number It will identify that successfully marked spam samples image data is sent to message handler according to being marked, and by communication module 4 2;The spam samples image data of ResNet algorithm recognition failures is transferred to user 7 or maintenance personnel 8 to be marked, and is led to It crosses communication module 4 and the marked spam samples image data of recognition failures is sent to message handler 2.
As shown in figure 8, the present embodiment can correct current rubbish identification model and man-machine using Other Waste identification model Interactive device (such as touch-control all-in-one machine) 6 corrects current two kinds of evolutionary learning modes of rubbish identification model, accumulates marked rubbish Rubbish sample image data (including raw refuse sample image data and its corresponding label), in the base of labeled sample data On plinth, it is integrated into training set, carries out the training of deep learning image recognition by specified threshold, update rubbish identification model;Wherein, His rubbish identification model correct current rubbish identification model be it is higher with discrimination, speed is slower, unmarked to confine position output Algorithm model, by a series of restrictions, the data identified update former rubbish as markd spam samples image data Identification model herein re-flags spam samples image data using this classifier algorithm of ResNet, specific to wrap Include by each pixel of image draw three same size different proportions posting and three same ratios it is various sizes of fixed Position frame, input parameter of the corresponding posting of each pixel as ResNet export result as the recognition result of each frame, Wherein choose optimal result as the spam samples image data re-flagged, these are identified into successfully marked spam samples Image data is sent to message handler 2, and the accumulation of message handler 2 identifies successfully marked spam samples image data, whole Compound training collection, the input parameter of the neural network as message handler 2 carry out deep learning image recognition by specified threshold Training updates rubbish identification model;If Other Waste identification model identifies rubbish failure, corrected using human-computer interaction device 6 Current rubbish identification model, human-computer interaction device 6 correct current rubbish identification model and refer to that server 3 passes through human-computer interaction device 6 transfer data to user 7 or maintenance personnel 8 removes the junk data of handmarking's discrimination lower (or completely identification less than), Specifically, human-computer interaction device 6 is first passed through to be transferred to the spam samples image data of Other Waste identification model recognition failures User 7 is to be marked, if user 7 is not marked, server 3 is again by the rubbish of Other Waste identification model recognition failures Sample image data is transferred to maintenance personnel 8 to be marked, then by the marked spam samples image data of recognition failures It is sent to message handler 2, the spam samples image data that message handler 2 is marked based on this accumulates the mark of recognition failures Remember spam samples image data, be integrated into training set, carries out the training of deep learning image recognition by specified threshold, update rubbish and know Other model.
As shown in FIG. 1 to FIG. 9, the workflow of the rubbish identification evolutionary learning system of the present embodiment is as follows:
S1, after rubbish is launched to garbage classification throwing mechanism 1, the camera 101 of garbage classification throwing mechanism 1 starts to adopt Collect spam samples image data, message handler 2 pre-processes spam samples image data, including go mean value, normalization, PCA and albefaction, the input parameter after pretreatment as the neural network of message handler 2, by having trained rubbish identification model The rubbish identification and evaluation system of setting is to determine the recognition accuracy (i.e. accuracy of identification) and rubbish type of rubbish, if the knowledge of rubbish Other accuracy rate is greater than or equal to given threshold, then judges to identify successfully, will identify successful rubbish phase by back work peripheral hardware 5 It answers information to be output to display, and will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism 1, garbage classification Throwing mechanism 1 puts on by classification rubbish according to rubbish corresponding information, if the recognition accuracy of rubbish is less than given threshold, Judge recognition failures, thens follow the steps S2.
The spam samples image data of recognition failures is transferred to server 3 by communication module 4 by S2, message handler 2, The data of recognition failures are divided into two kinds of data of the not high data of discrimination and discrimination lower (or completely identification less than), these Data can first pass through ResNet algorithm and once be identified again, if identifying successfully, server 3 identifies ResNet algorithm successful Spam samples image data is marked, and rubbish corresponding information is fed back to garbage classification throwing mechanism 1, and accumulation identifies successfully Marked spam samples image data, update rubbish identification model, enter step S4, if recognition failures, enter step S3.
S3, be by the spam samples image data of ResNet algorithm recognition failures other model recognition failures rubbish sample This image data, server 3 is transferred to user 7 by human-computer interaction device 6 to be marked, if user 7 is not marked, It is transferred to maintenance personnel 8 then to be marked, the marked spam samples image data of recognition failures is then sent to information Rubbish corresponding information is fed back to garbage classification throwing mechanism 1, and the rubbish sample based on this label by processor 2, message handler 2 This image data accumulates the marked spam samples image data of recognition failures, is integrated into training set, carries out by specified threshold deep Degree study image recognition training.
S4, garbage classification throwing mechanism 1 carry out sort operation according to the rubbish corresponding information that message handler 2 is fed back, and are The normal work of system is updated to parallel state with rubbish identification model, is independent of each other, has trained in new rubbish identification model Bi Hou, using new rubbish identification model until the data of new identification mistake occur, entire study mechanism is rendered as closing system Ring more new state.
Embodiment 2:
As shown in figure 9, the present embodiment realizes the data processing of message handler in above-described embodiment 1 by server, i.e., Server is directly connect with camera, controller, present embodiments provides a kind of rubbish identification theory of evolution based on deep learning Learning method, this method are applied in server, comprising the following steps:
S901, spam samples image data is obtained.
S902, spam samples image data is pre-processed, specifically: mean value is gone to spam samples image data, is returned One change, PCA and whitening processing.
S903, using pretreated spam samples image data as the input parameter of neural network, and trained rubbish Identification model is compared, and according to comparison result, judges whether to identify successfully, specifically:
Using pretreated spam samples image data as the input parameter of neural network, by having trained rubbish to identify The rubbish identification and evaluation system of model specification is to determine the recognition accuracy and rubbish type of rubbish, if the recognition accuracy of rubbish More than or equal to given threshold, then judge to identify successfully, if the recognition accuracy of rubbish is less than given threshold, judges that identification is lost It loses.
S904, it will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism;Wherein, the rubbish is corresponding Information includes rubbish type.
S905, it is identified again by spam samples image data of the ResNet algorithm to recognition failures.
S906, ResNet algorithm is identified that successful spam samples image data is marked, and by rubbish corresponding information Garbage classification throwing mechanism is fed back to, accumulation identifies successfully marked spam samples image data, updates rubbish identification model;
S907, the spam samples image data of ResNet algorithm recognition failures is transferred to user or maintenance personnel to carry out Label, and rubbish corresponding information is fed back into garbage classification throwing mechanism, accumulate the marked spam samples image of recognition failures Data update rubbish identification model.
Above-mentioned server includes processor, memory and the network interface connected by system bus;Wherein, processor is used It is calculated in offer and control ability, memory includes non-volatile memory medium, built-in storage, which deposits Operating system, computer program and database are contained, which is operating system and meter in non-volatile memory medium The operation of calculation machine program provides environment, when computer program is executed by third processor, realizes above-mentioned rubbish identification theory of evolution Learning method.
It will be understood by those skilled in the art that journey can be passed through by implementing the method for the above embodiments Sequence is completed to instruct relevant hardware, and corresponding program can store in computer-readable storage medium.
Embodiment 3:
As shown in Figure 10, a kind of rubbish identification evolutionary learning device based on deep learning, the dress are present embodiments provided It sets including obtaining module 1001, preprocessing module 1002, the first identification module 1003, feedback module 1004, the second identification module 1005, the first update module 1006 and the second update module 1007, the concrete function of modules are as follows:
The acquisition module 1001, for obtaining spam samples image data.
The preprocessing module 1002, for being pre-processed to spam samples image data, specifically: for rubbish Sample image data removes mean value, normalization, PCA and whitening processing.
First identification module 1003, for using pretreated spam samples image data as the defeated of neural network Enter parameter, be compared with rubbish identification model has been trained, according to comparison result, is judged whether to identify successfully, specifically:
For using pretreated spam samples image data as the input parameter of neural network, by having trained rubbish The rubbish identification and evaluation system of identification model setting is to determine the recognition accuracy and rubbish type of rubbish, if the identification of rubbish is quasi- True rate is greater than or equal to given threshold, then judges to identify successfully, if the recognition accuracy of rubbish is less than given threshold, judges to know Do not fail.
The feedback module 1004, for that will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism; Wherein, the rubbish corresponding information includes rubbish type.
Second identification module 1005, for by ResNet algorithm to the spam samples image datas of recognition failures again It is secondary to be identified.
First update module 1006, for ResNet algorithm to be identified that successful spam samples image data is marked Note, and rubbish corresponding information is fed back into garbage classification throwing mechanism, accumulation identifies successfully marked spam samples picture number According to update rubbish identification model.
Second update module 1007, for the spam samples image data of ResNet algorithm recognition failures to be transferred to Rubbish corresponding information is fed back to garbage classification throwing mechanism to be marked by user or maintenance personnel, accumulates recognition failures Marked spam samples image data, update rubbish identification model.
It should be noted that system provided in this embodiment only illustrate with the division of above-mentioned each functional module It is bright, in practical applications, it can according to need and be completed by different functional modules above-mentioned function distribution, i.e., by internal structure It is divided into different functional modules, to complete all or part of the functions described above.
Embodiment 3:
The present embodiment provides a kind of storage medium, which is computer readable storage medium, is stored with calculating Machine program when the computer program is executed by processor, realizes that the rubbish of above-described embodiment 2 identifies evolutionary learning method, such as Under:
Obtain spam samples image data.
Spam samples image data is pre-processed, specifically: to spam samples image data go mean value, normalization, PCA and whitening processing.
Using pretreated spam samples image data as the input parameter of neural network, mould is identified with rubbish has been trained Type is compared, and according to comparison result, judges whether to identify successfully, specifically:
Using pretreated spam samples image data as the input parameter of neural network, by having trained rubbish to identify The rubbish identification and evaluation system of model specification is to determine the recognition accuracy and rubbish type of rubbish, if the recognition accuracy of rubbish More than or equal to given threshold, then judge to identify successfully, if the recognition accuracy of rubbish is less than given threshold, judges that identification is lost It loses.
It will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism;Wherein, the rubbish corresponding information Including rubbish type.
It is identified again by spam samples image data of the ResNet algorithm to recognition failures.
ResNet algorithm is identified that successful spam samples image data is marked, and rubbish corresponding information is fed back to Garbage classification throwing mechanism, accumulation identify successfully marked spam samples image data, update rubbish identification model;
The spam samples image data of ResNet algorithm recognition failures is transferred to user or maintenance personnel is marked, product The marked spam samples image data of tired recognition failures, updates rubbish identification model.
Storage medium in the present embodiment can be disk, CD, computer storage, read-only memory (ROM, Read- Only Memory), random access memory (RAM, Random Access Memory), USB flash disk, the media such as mobile hard disk.
Rubbish identification model has been trained to judge to acquire whether spam samples image data is known in conclusion the present invention first passes through Not Cheng Gong, know in the case where recognition failures, then through spam samples image data of the ResNet algorithm to recognition failures Not, ResNet algorithm is identified that successful spam samples image data is marked, accumulation identifies successfully marked rubbish sample This image data updates rubbish identification model, and the spam samples image data of ResNet algorithm recognition failures is transferred to User or maintenance personnel accumulate the marked spam samples image data of recognition failures to be marked, and update rubbish and identify mould Type forms closed loop update, rubbish identification model is updated by the learning process constantly recycled, embodies trend of evolution, greatly Greatly improve rubbish identification precision and range, solve the prior art because training sample it is limited caused by identifiable rubbish Precision and the insufficient problem of quantity.
The above, only the invention patent preferred embodiment, but the scope of protection of the patent of the present invention is not limited to This, anyone skilled in the art is in the range disclosed in the invention patent, according to the present invention the skill of patent Art scheme and its inventive concept are subject to equivalent substitution or change, belong to the scope of protection of the patent of the present invention.

Claims (10)

1. the rubbish based on deep learning identifies evolutionary learning method, it is characterised in that: the described method includes:
Obtain spam samples image data;
Spam samples image data is pre-processed;
Using pretreated spam samples image data as the input parameter of neural network, with trained rubbish identification model into Row compares, and according to comparison result, judges whether to identify successfully;
It will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism;Wherein, the rubbish corresponding information includes Rubbish type;
It is identified again by spam samples image data of the ResNet algorithm to recognition failures;
ResNet algorithm is identified that successful spam samples image data is marked, and rubbish corresponding information is fed back into rubbish Mechanism is put on by classification, accumulation identifies successfully marked spam samples image data, updates rubbish identification model;
The spam samples image data of ResNet algorithm recognition failures is transferred to user or maintenance personnel to be marked, and will Rubbish corresponding information feeds back to garbage classification throwing mechanism, accumulates the marked spam samples image data of recognition failures, updates Rubbish identification model.
2. rubbish according to claim 1 identifies evolutionary learning method, it is characterised in that: described by pretreated rubbish Input parameter of the sample image data as neural network is compared with rubbish identification model has been trained, according to comparison result, Judge whether to identify successfully, specifically:
Using pretreated spam samples image data as the input parameter of neural network, by having trained rubbish identification model The rubbish identification and evaluation system of setting is to determine the recognition accuracy and rubbish type of rubbish, if the recognition accuracy of rubbish is greater than Or be equal to given threshold, then judge to identify successfully, if the recognition accuracy of rubbish is less than given threshold, judges recognition failures.
3. rubbish according to claim 1 or 2 identifies evolutionary learning method, it is characterised in that: described to spam samples figure As data are pre-processed, specifically: mean value, normalization, PCA and whitening processing are gone to spam samples image data.
4. rubbish based on deep learning identifies evolutionary learning device, it is characterised in that: described device includes:
Module is obtained, for obtaining spam samples image data;
Preprocessing module, for being pre-processed to spam samples image data;
First identification module, for using pretreated spam samples image data as the input parameter of neural network, and Training rubbish identification model is compared, and according to comparison result, judges whether to identify successfully;
Feedback module, for that will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism;Wherein, the rubbish Corresponding information includes rubbish type;
Second identification module, for being identified again by ResNet algorithm to the spam samples image data of recognition failures;
First update module, for ResNet algorithm to be identified successful spam samples image data to be marked, and by rubbish Rubbish corresponding information feeds back to garbage classification throwing mechanism, and accumulation identifies successfully marked spam samples image data, updates rubbish Rubbish identification model;
Second update module, for the spam samples image data of ResNet algorithm recognition failures to be transferred to user or maintenance people Member is marked, and rubbish corresponding information is fed back to garbage classification throwing mechanism, accumulates the marked rubbish sample of recognition failures This image data updates rubbish identification model.
5. rubbish according to claim 4 identifies evolutionary learning device, it is characterised in that: first identification module, tool Body are as follows:
For using pretreated spam samples image data as the input parameter of neural network, by having trained rubbish to identify The rubbish identification and evaluation system of model specification is to determine the recognition accuracy and rubbish type of rubbish, if the recognition accuracy of rubbish More than or equal to given threshold, then judge to identify successfully, if the recognition accuracy of rubbish is less than given threshold, judges that identification is lost It loses.
6. the rubbish based on deep learning identifies evolutionary learning system, it is characterised in that: the system comprises:
Garbage classification throwing mechanism is accordingly believed for acquiring spam samples image data according to the rubbish that message handler is fed back Breath carries out rubbish to put on by classification operation;
Message handler pre-processes spam samples image data for obtaining spam samples image data, will pre-process Input parameter of the spam samples image data afterwards as neural network is compared with rubbish identification model has been trained, according to Comparison result judges whether to identify successfully;It will identify that successful rubbish corresponding information feeds back to garbage classification throwing mechanism;It will know Not Shi Bai spam samples image data be transferred to server;Receive the successful marked spam samples of identification that server is sent Rubbish corresponding information is fed back to garbage classification throwing mechanism by image data, and accumulation identifies successfully marked spam samples figure As data, rubbish identification model is updated, and receives the marked spam samples image data for the recognition failures that server is sent, Rubbish corresponding information is fed back into garbage classification throwing mechanism, accumulates the marked spam samples image data of recognition failures, more New rubbish identification model;Wherein, the rubbish corresponding information includes rubbish type;
Server will for being identified again by ResNet algorithm to the spam samples image data of recognition failures The successful spam samples image data of ResNet algorithm identification will identify successfully marked spam samples figure to be marked As data are sent to message handler;The spam samples image data of ResNet algorithm recognition failures is transferred to user or maintenance Personnel are marked, and the marked spam samples image data of recognition failures is sent to message handler.
7. rubbish according to claim 6 identifies evolutionary learning system, it is characterised in that: the system also includes man-machine friendships The spam samples image data of ResNet algorithm recognition failures is transferred to by mutual equipment, the server by human-computer interaction device User is marked, if user is not marked, server is by spam samples image data failed transmission to maintenance personnel It is marked.
8. rubbish according to claim 6 or 7 identifies evolutionary learning system, it is characterised in that: the garbage classification is launched Mechanism includes camera, singulizing disc, dustbin, controller and multiple funnel assemblies, and the singulizing disc is arranged in dustbin Entreat position, and singulizing disc have multiple rubbish placement sections, the dustbin have multiple groups rubbish dispensing portion, the funnel assemblies, Rubbish placement section and rubbish dispensing portion are to correspond, and every group of rubbish dispensing portion includes two rubbish dispensing portions, each funnel Component is arranged above corresponding one group of rubbish dispensing portion, and the funnel opposite including both direction, the camera and information Processor connection, for acquiring the spam samples image data on rubbish placement section, and is transferred to message handler, the control Device is used for the vertical rotation of control tactics disk horizontally rotated with each rubbish placement section, and the level of each funnel of control turns Dynamic and vertical rotation.
9. rubbish according to claim 8 identifies evolutionary learning system, it is characterised in that: the garbage classification throwing mechanism It further include support rod, the support rod is fixed on dustbin, and the camera is arranged on support rod.
10. storage medium is stored with program, it is characterised in that: when described program is executed by processor, realize claim 1-3 Described in any item rubbish identify evolutionary learning method.
CN201811137406.9A 2018-09-28 2018-09-28 Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning Pending CN109389161A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811137406.9A CN109389161A (en) 2018-09-28 2018-09-28 Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811137406.9A CN109389161A (en) 2018-09-28 2018-09-28 Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning

Publications (1)

Publication Number Publication Date
CN109389161A true CN109389161A (en) 2019-02-26

Family

ID=65418180

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811137406.9A Pending CN109389161A (en) 2018-09-28 2018-09-28 Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning

Country Status (1)

Country Link
CN (1) CN109389161A (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109940024A (en) * 2019-03-22 2019-06-28 深兰科技(上海)有限公司 The method and recyclable device that a kind of pair of waste is classified
CN110378903A (en) * 2019-09-16 2019-10-25 广东电网有限责任公司佛山供电局 A kind of transmission line of electricity anti-accident measures Intelligent statistical method
CN110513120A (en) * 2019-08-17 2019-11-29 冀中能源峰峰集团有限公司 A kind of cutting head of roadheader adaptive location system and method
CN110598784A (en) * 2019-09-11 2019-12-20 北京建筑大学 Machine learning-based construction waste classification method and device
CN110653169A (en) * 2019-08-06 2020-01-07 中国长城科技集团股份有限公司 Garbage treatment method and device and terminal equipment
CN110813792A (en) * 2019-04-04 2020-02-21 苏州科睿信飞智能科技有限公司 Intelligent garbage recognition and classification method
CN111037554A (en) * 2019-12-12 2020-04-21 杭州翼兔网络科技有限公司 Garbage cleaning method, device, equipment and medium based on machine learning
CN111086796A (en) * 2019-11-05 2020-05-01 广东拜登网络技术有限公司 Perfecting method for garbage type identification and background server
CN111186656A (en) * 2020-01-10 2020-05-22 上海电力大学 Target garbage classification method and intelligent garbage can
CN111209215A (en) * 2020-02-24 2020-05-29 腾讯科技(深圳)有限公司 Application program testing method and device, computer equipment and storage medium
CN110550347B (en) * 2019-08-23 2020-08-04 珠海格力电器股份有限公司 Garbage classification recycling method, user terminal and garbage can
WO2020206862A1 (en) * 2019-04-08 2020-10-15 江西理工大学 Automatic sorting system
WO2021057004A1 (en) * 2019-09-25 2021-04-01 北京星选科技有限公司 Information processing method and apparatus, electronic device and computer readable storage medium

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106000904A (en) * 2016-05-26 2016-10-12 北京新长征天高智机科技有限公司 Automatic sorting system for household refuse
CN106067031A (en) * 2016-05-26 2016-11-02 北京新长征天高智机科技有限公司 Cooperate with the degree of depth learning network Machine Vision Recognition system based on artificial mechanism for correcting errors
CN106203498A (en) * 2016-07-07 2016-12-07 中国科学院深圳先进技术研究院 A kind of City scenarios rubbish detection method and system
CN106845408A (en) * 2017-01-21 2017-06-13 浙江联运知慧科技有限公司 A kind of street refuse recognition methods under complex environment
CN106874954A (en) * 2017-02-20 2017-06-20 佛山市络思讯科技有限公司 The method and relevant apparatus of a kind of acquisition of information
CN107153844A (en) * 2017-05-12 2017-09-12 上海斐讯数据通信技术有限公司 The accessory system being improved to flowers identifying system and the method being improved
CN107480660A (en) * 2017-09-30 2017-12-15 深圳市锐曼智能装备有限公司 Dangerous goods identifying system and its method
CN108224894A (en) * 2018-01-08 2018-06-29 合肥美的智能科技有限公司 The recognition methods of food materials freshness, device, refrigerator and medium based on deep learning

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106000904A (en) * 2016-05-26 2016-10-12 北京新长征天高智机科技有限公司 Automatic sorting system for household refuse
CN106067031A (en) * 2016-05-26 2016-11-02 北京新长征天高智机科技有限公司 Cooperate with the degree of depth learning network Machine Vision Recognition system based on artificial mechanism for correcting errors
CN106203498A (en) * 2016-07-07 2016-12-07 中国科学院深圳先进技术研究院 A kind of City scenarios rubbish detection method and system
CN106845408A (en) * 2017-01-21 2017-06-13 浙江联运知慧科技有限公司 A kind of street refuse recognition methods under complex environment
CN106874954A (en) * 2017-02-20 2017-06-20 佛山市络思讯科技有限公司 The method and relevant apparatus of a kind of acquisition of information
CN107153844A (en) * 2017-05-12 2017-09-12 上海斐讯数据通信技术有限公司 The accessory system being improved to flowers identifying system and the method being improved
CN107480660A (en) * 2017-09-30 2017-12-15 深圳市锐曼智能装备有限公司 Dangerous goods identifying system and its method
CN108224894A (en) * 2018-01-08 2018-06-29 合肥美的智能科技有限公司 The recognition methods of food materials freshness, device, refrigerator and medium based on deep learning

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109940024A (en) * 2019-03-22 2019-06-28 深兰科技(上海)有限公司 The method and recyclable device that a kind of pair of waste is classified
CN110813792A (en) * 2019-04-04 2020-02-21 苏州科睿信飞智能科技有限公司 Intelligent garbage recognition and classification method
WO2020206862A1 (en) * 2019-04-08 2020-10-15 江西理工大学 Automatic sorting system
CN110653169A (en) * 2019-08-06 2020-01-07 中国长城科技集团股份有限公司 Garbage treatment method and device and terminal equipment
CN110513120B (en) * 2019-08-17 2020-12-29 冀中能源峰峰集团有限公司 Self-adaptive positioning system and method for cutting head of heading machine
CN110513120A (en) * 2019-08-17 2019-11-29 冀中能源峰峰集团有限公司 A kind of cutting head of roadheader adaptive location system and method
CN110550347B (en) * 2019-08-23 2020-08-04 珠海格力电器股份有限公司 Garbage classification recycling method, user terminal and garbage can
CN110598784B (en) * 2019-09-11 2020-06-02 北京建筑大学 Machine learning-based construction waste classification method and device
CN110598784A (en) * 2019-09-11 2019-12-20 北京建筑大学 Machine learning-based construction waste classification method and device
CN110378903A (en) * 2019-09-16 2019-10-25 广东电网有限责任公司佛山供电局 A kind of transmission line of electricity anti-accident measures Intelligent statistical method
WO2021057004A1 (en) * 2019-09-25 2021-04-01 北京星选科技有限公司 Information processing method and apparatus, electronic device and computer readable storage medium
CN111086796A (en) * 2019-11-05 2020-05-01 广东拜登网络技术有限公司 Perfecting method for garbage type identification and background server
CN111037554A (en) * 2019-12-12 2020-04-21 杭州翼兔网络科技有限公司 Garbage cleaning method, device, equipment and medium based on machine learning
CN111186656A (en) * 2020-01-10 2020-05-22 上海电力大学 Target garbage classification method and intelligent garbage can
CN111209215A (en) * 2020-02-24 2020-05-29 腾讯科技(深圳)有限公司 Application program testing method and device, computer equipment and storage medium

Similar Documents

Publication Publication Date Title
CN104217225B (en) A kind of sensation target detection and mask method
Tian et al. Apple detection during different growth stages in orchards using the improved YOLO-V3 model
CN105160309B (en) Three lanes detection method based on morphological image segmentation and region growing
CN102855638B (en) Detection method for abnormal behavior of vehicle based on spectrum clustering
CN103258432B (en) Traffic accident automatic identification processing method and system based on videos
CN101853071B (en) Gesture identification method and system based on visual sense
CN102176228B (en) Machine vision method for identifying dial plate information of multi-pointer instrument
CN103679203B (en) Robot system and method for detecting human face and recognizing emotion
CN103745234B (en) Band steel surface defect feature extraction and classification method
CN105741316A (en) Robust target tracking method based on deep learning and multi-scale correlation filtering
CN103077407B (en) Car logo positioning and recognition method and car logo positioning and recognition system
CN104156726B (en) A kind of workpiece identification method and device based on geometric characteristic
CN107123114A (en) A kind of cloth defect inspection method and device based on machine learning
CN204479490U (en) A kind of laser marking On-line Product testing and analysis system
CN104148301B (en) Waste plastic bottle sorting equipment based on cloud computing and image recognition and method
CN103237201B (en) A kind of case video analysis method based on socialization mark
CN107016405A (en) A kind of insect image classification method based on classification prediction convolutional neural networks
CN103258213B (en) A kind of for the dynamic vehicle model recognizing method in intelligent transportation system
CN104517104A (en) Face recognition method and face recognition system based on monitoring scene
CN107341523A (en) Express delivery list information identifying method and system based on deep learning
CN104850836A (en) Automatic insect image identification method based on depth convolutional neural network
CN105740899B (en) A kind of detection of machine vision image characteristic point and match compound optimization method
Strothmann et al. Plant classification with in-field-labeling for crop/weed discrimination using spectral features and 3d surface features from a multi-wavelength laser line profile system
CN105512640A (en) Method for acquiring people flow on the basis of video sequence
CN107256246B (en) printed fabric image retrieval method based on convolutional neural 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