CN113213016A - Garbage recognition and classification algorithm based on ViT, and device and control method thereof - Google Patents

Garbage recognition and classification algorithm based on ViT, and device and control method thereof Download PDF

Info

Publication number
CN113213016A
CN113213016A CN202110617614.4A CN202110617614A CN113213016A CN 113213016 A CN113213016 A CN 113213016A CN 202110617614 A CN202110617614 A CN 202110617614A CN 113213016 A CN113213016 A CN 113213016A
Authority
CN
China
Prior art keywords
garbage
vit
neural network
classification
vector
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.)
Granted
Application number
CN202110617614.4A
Other languages
Chinese (zh)
Other versions
CN113213016B (en
Inventor
易阳
余俊贤
朱奕吉
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.)
Nanjing Tech University
Original Assignee
Nanjing Tech 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 Nanjing Tech University filed Critical Nanjing Tech University
Priority to CN202110617614.4A priority Critical patent/CN113213016B/en
Publication of CN113213016A publication Critical patent/CN113213016A/en
Application granted granted Critical
Publication of CN113213016B publication Critical patent/CN113213016B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F1/00—Refuse receptacles; Accessories therefor
    • B65F1/14—Other constructional features; Accessories
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F1/00—Refuse receptacles; Accessories therefor
    • B65F1/0033—Refuse receptacles; Accessories therefor specially adapted for segregated refuse collecting, e.g. receptacles with several compartments; Combination of receptacles
    • B65F1/0053—Combination of several receptacles
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F1/00—Refuse receptacles; Accessories therefor
    • B65F1/12—Refuse receptacles; Accessories therefor with devices facilitating emptying
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F1/00—Refuse receptacles; Accessories therefor
    • B65F1/0033—Refuse receptacles; Accessories therefor specially adapted for segregated refuse collecting, e.g. receptacles with several compartments; Combination of receptacles
    • B65F2001/008—Means for automatically selecting the receptacle in which refuse should be placed
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F2210/00—Equipment of refuse receptacles
    • B65F2210/138—Identification means
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F2210/00—Equipment of refuse receptacles
    • B65F2210/176—Sorting means
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F2210/00—Equipment of refuse receptacles
    • B65F2210/178—Steps
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B65—CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65F—GATHERING OR REMOVAL OF DOMESTIC OR LIKE REFUSE
    • B65F2210/00—Equipment of refuse receptacles
    • B65F2210/182—Volume determining means
    • Y—GENERAL 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02W—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO WASTEWATER TREATMENT OR WASTE MANAGEMENT
    • Y02W30/00—Technologies for solid waste management
    • Y02W30/10—Waste collection, transportation, transfer or storage, e.g. segregated refuse collecting, electric or hybrid propulsion

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Mechanical Engineering (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a garbage recognition and classification algorithm based on ViT neural network, which recognizes garbage types through ViT neural network, can continuously optimize ViT neural network in practical application, and has strong adaptability and high accuracy. The invention also provides a garbage classification device based on the ViT neural network garbage recognition and classification algorithm, garbage slides into the garbage can along a preset track through the mutual cooperation of the overturning motion of the garbage dumping device and the rotating motion of the garbage falling limiter, and the whole mechanical structure is compact and the operation efficiency is high. The invention also provides a control method of the garbage classification device based on the ViT neural network garbage identification and classification algorithm, which realizes the function of garbage identification falling into the classification garbage can under the synergistic action of the control system. The automatic garbage sorting device has the advantages that automatic garbage sorting operation is completed efficiently, highly accurately and successfully, the problems of non-fixity of garbage size, non-standardization of garbage types and the like in practical production application are solved, and the practicability is high.

Description

Garbage recognition and classification algorithm based on ViT, and device and control method thereof
Technical Field
The invention relates to the technical field of garbage classification, in particular to a garbage recognition and classification algorithm based on ViT, and a device and a control method thereof.
Background
Along with the continuous development of national economy, people live more and more abundantly in material, but along with that, garbage in cities is more and more. The classified garbage is not only beneficial to the living environment of people, but also beneficial to the recycling of the garbage. Unfortunately, the people in China have a weak sense of garbage classification, which results in that garbage cannot be classified in time from the source, thereby causing a great deal of environmental pollution and waste of manpower and material resources.
In the prior art, chinese patent publication No. 110803413a discloses a household garbage sorting bin for office, which only provides five bins for sorting on one device, and the sorting still depends on the subjective judgment of the dispenser, and does not meet the requirements of the intelligent and automatic sorting device. Chinese patent publication No. 110733795a discloses a waste classification collection device, and this device has built-in a camera and has handled the rubbish image that needs to put in, shows the kind information of rubbish on the display screen, and the person of puting in puts in according to this tip information and puts in, though this device has already can automatic identification rubbish kind, can not realize putting in automatic letter sorting of rubbish.
Disclosure of Invention
The application also provides a garbage recognition and classification algorithm based on ViT neural network, a garbage classification device based on the garbage recognition and classification algorithm based on ViT neural network, and a control method of the garbage classification device based on the garbage recognition and classification algorithm based on ViT neural network, which solve the problem that the prior art can not realize automatic garbage sorting, are used for sorting garbage images captured by the camera through the garbage recognition and classification algorithm based on ViT neural network, and realize the sorting and storage of garbage through controlling a mechanical structure.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
an ViT neural network-based garbage recognition and classification algorithm comprises the following steps:
s1, searching a plurality of four types of junk pictures in a network to serve as a picture data set A, acquiring photos of the four types of junk pictures through a camera to serve as a photo data set B, converting the picture data set A and the photo data set B into a same-size picture data set A 'and a same-size photo data set B' with the same size by utilizing an interpolation algorithm, wherein each picture in the same-size picture data set A 'and the same-size photo data set B' is square, and the four types of junk pictures are respectively as follows: the garbage, kitchen garbage, other garbage and harmful garbage can be recovered;
s2: building ViT neural network frame structure, sending picture data set A' with the same size into ViT neural network for pre-training;
s3: dividing the photo data set B' with the same size into a training set B1 and an evaluation set B2;
s4: sending the training set B1 to the ViT neural network which is subjected to pre-training in S2 for fine adjustment;
s5: sending the evaluation set B2 to the ViT neural network which is subjected to fine tuning in S4 for accuracy test;
s6: and adding the pictures with the error detection in the evaluation set B2 into a training set to optimize ViT neural networks.
Further, the ViT neural network framework structure in S2 includes: a picture block module, a linearization module, a position information adding module, a plurality of multi-head self-attention layers, a coding network module and a probability classifier module which are formed by overlapping with a full connection layer,
the image partitioning module divides an image sent into the ViT neural network into a plurality of image blocks with the same size, the linearization module linearizes tensors corresponding to the image blocks into vectors, the position information adding module adds position information in the vectors corresponding to the image blocks, the coding network module converts the sent image into a characteristic quantity containing image information, the probability classifier module outputs a classification result vector P containing a classification result according to the characteristic vector of the image information, 4 elements contained in the classification result vector P respectively correspond to four types of garbage probabilities, and the largest element in the 4 elements is a garbage category to which the garbage belongs.
Further, the pre-training in S2 includes the following steps:
s2.1: dividing all pictures in the same-size picture data set a' into tiles:
cutting the pictures through a sliding window, cutting each picture into n multiplied by n picture blocks, wherein the size of the sliding window of each cut picture block is consistent with that of the picture block, and the picture blocks cut by one picture are prevented from being overlapped;
s2.2: and (3) carrying out linearization processing on each image block:
converting each image block into d1i*d2i*d3iOf the patch tensor, wherein d1i、d2i、d3iR, G, and B values in RGB of the color picture in the ith block, i ═ 1, 2, … …, n × n, respectively;
converting the tile tensor into a tile vector X, X ═ (X1, X2, …, Xi, …, Xn×n) Wherein Xi ═ (d 1)i*d2i*d3i，1)；
The image block vector X is linearly changed through the full connection layer, and the linear transformation formula is as follows:
Zi＝W*Xi+b(i＝1，2，……，n*n)
wherein Zi is a characteristic block vector, is a linear block vector after the ith block is subjected to linear transformation, W is a multiplier matrix, is a multiplier parameter which is learned and output by the same-size picture data set A 'in the ViT neural network, and b is a constant matrix, is a constant parameter which is learned and output by the same-size picture data set A' in the ViT neural network;
the full connection layer shares a parameter multiplier matrix number W and a constant matrix b;
s2.3: adding position information to the feature tile vector: firstly, the position information of each image block is encoded into a position matrix Li which has the same number of rows and columns as the number of rows and columns of the characteristic image block vector Zi through a position encoder, and then the position matrix is added to Zi to obtain a position characteristic image block vector Ci containing the position information:
Ci＝Zi+Li (2)
wherein Ci is a position feature pattern vector of the ith pattern block, and represents the feature pattern vector containing the position information of the ith pattern block, Li is a position matrix of the ith pattern block, i is 1, 2, … …, n is n;
s2.4: introducing classification information C0, adding C0 to the position feature tile vector set C, and forming a position feature tile vector set C', C ═ C0, C1, C2, … …, Cn×n)
S2.5: forming a coding network:
by the superposition of the multi-head self-attention layer and the full-connection layer, the learning effect of ViT neural networks is increased to form a coding network;
s2.6: acquiring an image feature vector:
inputting a position characteristic picture block vector set C 'containing classification information into an encoding network to obtain a position characteristic picture block encoding vector D1, wherein D1 is a final result of extracting a plurality of characteristic information of a picture by continuously performing vector transformation on C' i in the encoding network;
s2.7: and (3) judging the category and updating the network:
and inputting the position feature pattern block coding vector D1 into a probability classifier to obtain a classification result vector P, wherein 4 elements contained in the classification result vector P respectively correspond to four classes of garbage probabilities, the largest element in the 4 elements is the garbage class to which the garbage belongs, comparing a loss function between the classification result vector P and a real result, solving the parameter gradient of the loss function relative to the neural network, and updating ViT neural network parameters by using the gradient.
Preferably, the loss function in S2.7 uses the following formula:
Figure BDA0003092874490000031
wherein, loss represents a loss function, m represents the number of samples, P (xij) is the classification result of the i row and j column garbage in the classification result vector P, and represents the real probability of each garbage obtained by the garbage picture training set, q (xij) represents the prediction probability distribution of the i row and j column garbage predicted by the model, and log (q (xij)) represents the logarithm of q (xij).
The invention also provides a garbage classification device based on the garbage recognition and classification algorithm of the ViT neural network, which comprises the following components:
the garbage dumping device comprises a frame structure, wherein 4 garbage cans which are arranged in a field shape are arranged in the frame structure, a garbage falling limiter is arranged at the middle position above the garbage cans, a garbage dumping device is arranged right above the garbage falling limiter, a panel is arranged at the top of the frame structure, a touch screen is arranged on the panel, a garbage throwing opening is formed in the front side of the top of the panel, the garbage throwing opening is located right above the garbage dumping device, a camera is arranged at the garbage throwing opening, and the camera points to the garbage dumping device;
the garbage dumping device comprises a V-shaped plate, a garbage placing surface is arranged behind the V-shaped plate, the opposite surface of the garbage placing surface is arranged in front of the V-shaped plate, the area of the garbage placing surface is larger than the opening area of the garbage throwing opening, the V-shaped plate is fixedly arranged on a horizontally arranged transverse shaft through a V-shaped plate support, one side of the transverse shaft extends into a shaft support frame and freely rotates on the vertical surface in the shaft support frame by taking the transverse shaft as a rotating shaft, the shaft support frame is fixedly arranged in the frame structure, the other side of the transverse shaft is connected with a first steering engine shaft connector, and the first steering engine shaft connector is connected with a first steering engine;
the garbage falling limiter comprises a garbage limiting rail, the garbage limiting rail comprises a bottom plate with a low front part and a high rear part, baffles are arranged on two sides and the top of the bottom plate, the garbage limiting rail is supported by a rail and a vertical shaft, the other end of the vertical shaft is arranged in a second steering engine shaft connector, and the second steering engine shaft connector is controlled by a second steering engine to freely rotate in a horizontal plane by taking the center of the second steering engine shaft connector as the circle center;
the output end of the camera is connected with the input end of the main control unit, the output end of the main control unit is connected with the input ends of the touch screen and the execution unit, the output end of the execution unit is connected with the input ends of the first steering engine and the second steering engine, and the camera, the execution unit, the main control unit, the touch screen, the first steering engine and the second steering engine are connected with the power supply respectively.
Further, the upper end is equipped with ultrasonic sensor in the garbage bin, ultrasonic sensor with the master control unit links to each other, ultrasonic sensor is used for carrying out the dustbin and is full loaded the warning.
The invention also provides a control method of the garbage classification device based on the garbage recognition and classification algorithm of the ViT neural network, which comprises the following steps:
s1: starting a power supply, and initializing a system:
initializing a touch screen, and circularly playing the garbage classification propaganda film by the touch screen;
initializing a garbage dumping device to enable the opposite surface of a garbage placing surface on the V-shaped plate to be in a vertical state;
initializing a garbage falling limiter to enable the low surface of the bottom plate to face forwards and the high surface to face backwards;
initializing a first steering engine and a second steering engine to enable a first steering engine shaft connector and a second steering engine shaft connector to be in initial positions; s2: placing garbage, and identifying the garbage:
the garbage dumping device is placed in the garbage dumping device, a camera captures a garbage image on the garbage dumping device, the main control unit feeds garbage image information back to the execution unit, the execution unit identifies the garbage type based on a garbage identification and classification algorithm of ViT neural networks, the execution unit feeds the identified garbage type back to the main control unit, and the main control unit displays the identified garbage type on the touch screen;
s4: the second steering engine is started, and the garbage falling limiter points to the corresponding garbage can;
the execution unit controls the second steering engine to rotate the second steering engine shaft connector until the lower edge of the bottom plate points to the position of the garbage can corresponding to the garbage type identified in the S2;
s5: after the second steering wheel starts, first steering wheel starts:
the execution unit controls the first steering engine to start, so that the first steering engine shaft connector turns backwards to drive the garbage dumping device to turn backwards, and garbage falls onto the garbage limiting device track and further falls into the garbage can;
s6: and initializing the system, and waiting for the next garbage input.
Further, the garbage can full load reminding comprises the following steps: ultrasonic sensor lasts the distance between rubbish and the ultrasonic sensor in the measurement garbage bin, and with the distance signal of telecommunication feedback to the main control unit, the main control unit converts the signal of telecommunication received into distance data and carries out logic judgement, rubbish when the garbage bin in surpasss the rubbish capacity threshold value, the main control unit judges to reachs full load result, the main control unit sends full load signal to execution unit and touch-sensitive screen simultaneously, execution unit control second steering wheel, make rubbish tipper be in initial condition, the opposite face that face was placed to rubbish on the V template is in vertical state promptly, and simultaneously, the touch-sensitive screen shows full load information.
The invention has the beneficial effects that:
the application also provides a garbage recognition and classification algorithm based on ViT neural network, a garbage classification device based on the garbage recognition and classification algorithm based on ViT neural network, and a control method of the garbage classification device based on the garbage recognition and classification algorithm based on ViT neural network, which solve the problem that the prior art can not realize automatic garbage sorting, are used for sorting garbage images captured by the camera through the garbage recognition and classification algorithm based on ViT neural network, and realize the sorting and storage of garbage through controlling a mechanical structure.
1. According to the intelligent garbage classification device, the overturning motion of the garbage dumping device and the rotating motion of the garbage falling limiting device are matched with each other. And after the garbage falling limiting device rotates to the corresponding position, the dumping device dumps the garbage into the barrel, and the garbage slides into the barrel along the preset track. The intelligent garbage classification device has the advantages that the whole mechanical structure is compact and the operation efficiency is high due to the movement mode.
2. The garbage classification recognition algorithm based on the ViT neural network is trained by using a large number of data sets, so that the accuracy of garbage recognition is greatly improved. Meanwhile, the problem that the same garbage has different states and colors in practical application is solved, for example, pop cans in good state and in a flattened state and red and green mineral water bottles and ViT models can be learned and mastered, so that the garbage identification speed and accuracy are greatly improved; the function of AI autonomous learning is increased, and when the garbage bin runs into unknown garbage, the garbage model can be learned through user input information so as to meet the diversified demands of different users on various occasions.
3. The full-load prompt function is included, the touch screen displays the full-load information function of the garbage can, workers can be reminded of processing the corresponding garbage can in time, the working efficiency of the workers is improved, and a series of problems caused by garbage overflow are avoided.
4. The frame of the device adopts the full-aluminum section bar modular design, and is convenient for adjusting, overhauling and replacing the mechanical structures such as the size, the position and the like of the device. The service life and the function expansibility of the device are improved, and a user can modify the device in appearance and function according to own requirements so as to meet the use requirements.
5. The invention can play the function of the automatic playing garbage classification promo sheet to play a good propaganda role, can improve the garbage classification knowledge of the public to a certain extent, and improves the public garbage classification consciousness to a great extent.
Drawings
FIG. 1 is a schematic diagram of a VIT neural network algorithm provided by the present invention;
FIG. 2 is a flowchart of garbage classification control provided by the present invention;
FIG. 3 is a schematic diagram of a control system according to the present invention;
FIG. 4 is a schematic structural diagram of an apparatus provided by the present invention;
FIG. 5 is a schematic view of a portion of a garbage dumper provided by the present invention;
FIG. 6 is a schematic view of a partial structure of the garbage falling stopper according to the present invention;
FIG. 7 is a first photograph taken during a test using the apparatus provided by the present invention;
FIG. 8 is a second photograph taken during a use test of the apparatus provided by the present invention;
FIG. 9 is a photograph III taken during a use test of the apparatus provided by the present invention;
1-a frame structure, 2-a panel, 3-a garbage dumping device, 4-a garbage falling limiter, 5-a garbage can, 6-a main control unit, 7-a garbage identification control unit, 8-a touch screen, 9-a camera and 10-an ultrasonic sensor;
31-V-shaped plate, 32-transverse shaft, 33-shaft support frame, 34-V-shaped plate support frame, 35-first steering engine shaft connector and 36-first steering engine;
41-garbage limiter rail, 42-rail support, 43-vertical shaft, 44-second steering engine shaft connector and 45-second steering engine.
Detailed Description
The garbage recognition and classification algorithm based on ViT neural network of the present invention is further described in detail with reference to the accompanying drawings and the specific implementation method.
Example 1
As shown in fig. 1, a garbage recognition and classification algorithm based on ViT neural network includes the following steps:
s1, searching a plurality of four types of junk pictures in a network to serve as a picture data set A, acquiring photos of the four types of junk pictures through a camera to serve as a photo data set B, converting the picture data set A and the photo data set B into a same-size picture data set A 'and a same-size photo data set B' with the same size by utilizing an interpolation algorithm, wherein each picture in the same-size picture data set A 'and the same-size photo data set B' is square, and the four types of junk pictures are respectively as follows: the garbage, kitchen garbage, other garbage and harmful garbage can be recovered;
s2: building ViT neural network frame structure, sending picture data set A' with the same size into ViT neural network for pre-training;
s3: dividing the photo data set B' with the same size into a training set B1 and an evaluation set B2;
s4: sending the training set B1 to the ViT neural network which is subjected to pre-training in S2 for fine adjustment;
s5: sending the evaluation set B2 to the ViT neural network which is subjected to fine tuning in S4 for accuracy test;
s6: and adding the pictures with the error detection in the evaluation set B2 into a training set to optimize ViT neural networks.
Further, the ViT neural network framework structure in S2 includes: a picture block module, a linearization module, a position information adding module, a plurality of multi-head self-attention layers, a coding network module and a probability classifier module which are formed by overlapping with a full connection layer,
the image partitioning module divides an image sent into the ViT neural network into a plurality of image blocks with the same size, the linearization module linearizes tensors corresponding to the image blocks into vectors, the position information adding module adds position information in the vectors corresponding to the image blocks, the coding network module converts the sent image into a characteristic quantity containing image information, the probability classifier module outputs a classification result vector P containing a classification result according to the characteristic vector of the image information, 4 elements contained in the classification result vector P respectively correspond to four types of garbage probabilities, and the largest element in the 4 elements is a garbage category to which the garbage belongs.
Further, the pre-training in S2 includes the following steps:
s2.1: dividing all pictures in the same-size picture data set a' into tiles:
cutting the pictures through a sliding window, cutting each picture into n multiplied by n picture blocks, wherein the size of the sliding window of each cut picture block is consistent with that of the picture block, and the picture blocks cut by one picture are prevented from being overlapped;
s2.2: and (3) carrying out linearization processing on each image block:
1 converting each picture block into d1i*d2i*d3iOf the patch tensor, wherein d1i、d2i、d3iR, G, and B values in RGB of the color picture in the ith block, i ═ 1, 2, … …, n × n, respectively;
2 convert the tile tensor into tile vector X, X ═ X (X1, X2, …, Xi, …, Xn×n) Wherein Xi ═ d1i*d2i*d3i，1；
And 3, carrying out linear change on the image block vector X through the full connection layer, wherein the linear transformation formula is as follows:
Zi＝W*Xi+bi＝1，2，……，n*n 1
wherein Zi is a characteristic block vector, is a linear block vector after the ith block is subjected to linear transformation, W is a multiplier matrix, is a multiplier parameter which is learned and output by the same-size picture data set A 'in the ViT neural network, and b is a constant matrix, is a constant parameter which is learned and output by the same-size picture data set A' in the ViT neural network;
4, the full connection layer shares a parameter multiplier matrix number W and a constant matrix b;
s2.3: adding position information to the feature tile vector: firstly, the position information of each image block is encoded into a position matrix Li which has the same number of rows and columns as the number of rows and columns of the characteristic image block vector Zi through a position encoder, and then the position matrix is added to Zi to obtain a position characteristic image block vector Ci containing the position information:
Ci＝Zi+Li (2)
wherein Ci is a position feature pattern vector of the ith pattern block, and represents the feature pattern vector containing the position information of the ith pattern block, Li is a position matrix of the ith pattern block, i is 1, 2, … …, n is n;
s2.4: introducing classification information C0, adding C0 to the position feature tile vector set C to form a position feature tile vector set C ', C' ═ C0, C1, C2, … … and C containing classification informationn×n
S2.5: forming a coding network:
by the superposition of the multi-head self-attention layer and the full-connection layer, the learning effect of ViT neural networks is increased to form a coding network;
s2.6: acquiring an image feature vector:
inputting a position characteristic picture block vector set C 'containing classification information into an encoding network to obtain a position characteristic picture block encoding vector D1, wherein D1 is a final result of extracting a plurality of characteristic information of a picture by continuously performing vector transformation on the C' in the encoding network;
s2.7: and (3) judging the category and updating the network:
and inputting the position feature pattern block coding vector D1 into a probability classifier to obtain a classification result vector P, wherein 4 elements contained in the classification result vector P respectively correspond to four classes of garbage probabilities, the largest element in the 4 elements is the garbage class to which the garbage belongs, comparing a loss function between the classification result vector P and a real result, solving the parameter gradient of the loss function relative to the neural network, and updating ViT neural network parameters by using the gradient.
Preferably, the loss function in S2.7 uses the following formula:
Figure BDA0003092874490000081
wherein, loss represents a loss function, m represents the number of samples, P (xij) is the classification result of the i row and j column garbage in the classification result vector P, and represents the real probability of each garbage obtained by the garbage picture training set, q (xij) represents the prediction probability distribution of the i row and j column garbage predicted by the model, and log (q (xij)) represents the logarithm of q (xij).
Example 2
As shown in fig. 4-9, the present invention further provides a garbage classification device based on ViT neural network garbage recognition and classification algorithm, including:
the garbage dumping device comprises a frame structure 1, wherein 4 garbage cans 5 arranged in a field shape are arranged in the frame structure 1, a garbage falling limiting device 4 is arranged at the middle position above the garbage cans 5, a garbage dumping device 3 is arranged right above the garbage falling limiting device 4, a panel 2 is arranged at the top of the frame structure 1, a touch screen 8 is arranged on the panel 2, a garbage throwing opening is formed in the front side of the top of the panel 2 and is located right above the garbage dumping device 3, a camera 9 is arranged at the garbage throwing opening, and the camera 9 points to the garbage dumping device 3; frame construction 1 adopts movable aluminium alloy frame, and panel 2 and garbage bin 5 adopt the ya keli material, and rubbish tripper 3 has the function of accepting rubbish, transferring rubbish, dumping rubbish.
The garbage dumping device 3 comprises a V-shaped plate 31, a garbage surface is placed on the V-shaped plate 31 at the back, the opposite surface of the garbage surface is placed at the front, and the area of the garbage surface is larger than the opening area of the garbage throwing opening, the V-shaped plate 31 is fixedly arranged on a horizontal transverse shaft 32 through a V-shaped plate support 34, the number of the V-shaped plate supports 34 is two, one side of the transverse shaft 32 extends into a shaft support 33 and freely rotates on the vertical surface in the shaft support 33 by taking the transverse shaft 32 as a rotating shaft, the shaft support 33 is fixedly arranged in the frame structure 1, the other side of the transverse shaft 32 is connected with a first steering machine connector 35, and the first steering machine connector 35 is connected with a first steering machine 36; the V-shaped plate 31 may be turned backwards with a maximum turning angle of 120 degrees.
The garbage falling limiter 4 comprises a garbage limiter rail 41, the garbage limiter rail 41 comprises a bottom plate with a low front part and a high rear part, baffles are arranged on two sides and the top of the bottom plate, the garbage limiter rail 41 is supported by rails 42 and a vertical shaft 43, the other end of the vertical shaft 43 is arranged in a second steering engine shaft connector 44, and the second steering engine shaft connector 44 is controlled by a second steering engine 45 to freely rotate in the horizontal plane by taking the center of the second steering engine shaft connector 44 as the center of a circle; after the garbage dumping device dumps the garbage, due to the existence of the garbage limiting device track 41, the fallen garbage can slide into the garbage can according to a preset track, and the garbage sorting action is completed. The falling limiting device 4 can fix the falling rail of the garbage, so that the garbage can directly slide into the garbage can 5 along the falling limiting device rail 41, the design of bearing the falling limiting device rail 41 on the rail support is adopted, the falling impact force of the garbage is dispersed, the load of a mechanical structure is reduced, the service life and the reliability of the mechanical structure are prolonged, and meanwhile, the mechanical structure and the control are more concise and stable.
The output end of the camera 9 is connected with the input end of the main control unit 7, the output end of the main control unit 7 is connected with the input ends of the touch screen 8 and the execution unit 6, the output end of the execution unit 6 is connected with the input ends of the first steering engine 36 and the second steering engine 45, and the camera 9, the execution unit 6, the main control unit 7, the touch screen 8, the first steering engine 36 and the second steering engine 45 are respectively connected with a power supply. The main control unit 7 selects a raspberry group controller, the touch screen 8 selects a raspberry group touch screen, the camera selects a raspberry group camera and a seven-inch raspberry group touch screen module, and displayed content is divided into three parts, namely full-load detection, identified garbage information and display of a camera identification picture. The initialization state of the display screen is that full-load detection shows that the picture identified by the camera is normally displayed, and the picture is not full-loaded, the identified junk information is empty.
The execution unit 6 selects an Arduino Mega2560 control device, the Arduino Mega2560 controls the frequency and duty ratio of PWM waves emitted by an analog output pin of the Arduino Mega2560 to serve as output quantities, and the first steering engine 36 and the second steering engine 45 are controlled to rotate so as to realize garbage throwing and falling position limitation; the PWM wave frequency determines the period of a PWM wave, and the duty ratio determines the high level time of the PWM wave so as to determine the rotation angle of the steering engine; the algebraic sum of the times of pressing the micro-increase button and the micro-decrease button is used as compensation quantity, the precision of the steering engine reaching a target position is controlled by superposing the high level time of the theoretical PWM wave and the algebraic sum of the compensation quantity, positioning errors are eliminated, errors of garbage classification caused by inaccurate positioning are reduced, and the efficiency and the success rate of garbage classification are improved.
Further, the upper end is equipped with ultrasonic sensor 10 in garbage bin 5, ultrasonic sensor 10 with main control unit 7 links to each other, ultrasonic sensor 10 is used for carrying out the dustbin and is fully loaded the warning.
The rubbish just falls behind rubbish tipper 3 from rubbish throwing mouth, after discerning the rubbish type, the downward one side of bottom plate of control rubbish whereabouts stopper 4 is directional to the garbage bin of corresponding type, and both cooperate each other, and the rubbish that will discern the classification is accurately delivered to in the corresponding garbage bin. The movement mode enables the whole device to be compact in structure, fast in movement and high in efficiency.
Example 3
As shown in fig. 2-3, the present invention further provides a control method of a garbage classification device based on the garbage recognition and classification algorithm of ViT neural network, comprising the following steps:
s1: starting a power supply, and initializing a system:
initializing a touch screen 8, and circularly playing the garbage classification promo sheet by the touch screen 8;
initializing the garbage dumping device 3 to enable the opposite surface of the garbage placing surface on the V-shaped plate 31 to be in a vertical state;
initializing a garbage falling limiter 4 to enable the low surface of the bottom plate to face forwards and the high surface to face backwards;
initializing the first steering engine 36 and the second steering engine 45 so that the first steering engine shaft connector 35 and the second steering engine shaft connector 44 are in initial positions;
s2: placing garbage, and identifying the garbage:
garbage is placed into the garbage dumping device 3, a camera 9 captures a garbage image on the garbage dumping device 3, the main control unit 7 feeds garbage image information back to the execution unit 6, the execution unit 6 identifies the garbage type based on a garbage identification and classification algorithm of ViT neural networks, the execution unit 6 feeds the identified garbage type back to the main control unit 7, and the main control unit 7 displays the identified garbage type on the touch screen 8; this process takes about 200 ms;
s4: the second steering engine 45 is started, and the garbage falling limiter 4 points to the corresponding garbage can;
the execution unit 6 controls the second steering engine 45 to rotate the second steering engine shaft connector 44 to the position of the garbage can corresponding to the garbage type identified in the low-side pointing S2 of the bottom plate;
s5: after second steering wheel 45 starts to accomplish, first steering wheel 36 starts:
the execution unit 6 controls the first steering engine 36 to start, so that the first steering engine shaft connector 35 turns backwards to drive the garbage dumping device 3 to turn backwards, and garbage falls on the garbage limiting device rail 41 and then slides into the garbage can; the duration of the dumping state is 400-600ms to ensure that the garbage is successfully dumped, and the high success rate can be achieved for the light and thin garbage with small volume.
S6: and initializing the system, and waiting for the next garbage input.
Further, the garbage can full load reminding comprises the following steps: ultrasonic sensor 10 continuously measures the distance between rubbish in the garbage bin 5 and ultrasonic sensor 10, and feed back the distance signal of telecommunication to main control unit 7, main control unit 7 converts the signal of telecommunication received into distance data and carries out logic judgement, rubbish when in the garbage bin 5 surpasss the rubbish capacity threshold value, main control unit 7 judges and obtains full load result, main control unit 7 sends full load signal to execution unit 6 and touch-sensitive screen 8 simultaneously, execution unit 6 controls second steering wheel 45, make rubbish tipper 3 be in initial condition, the opposite face that the face was placed to rubbish on the V template 31 is in vertical state, and simultaneously, touch-sensitive screen 8 shows full load information.
In the description provided herein, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of this description, will appreciate that other embodiments can be devised which do not depart from the scope of the invention as described herein. Furthermore, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and may not have been selected to delineate or circumscribe the inventive subject matter. Accordingly, many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the disclosure. The disclosure of the present invention is intended to be illustrative, but not limiting, of the scope of the invention.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims (8)

1. A garbage recognition and classification algorithm based on ViT neural networks is characterized by comprising the following steps:
s1, searching a plurality of four types of junk pictures in a network to serve as a picture data set A, acquiring photos of the four types of junk pictures through a camera to serve as a photo data set B, converting the picture data set A and the photo data set B into a same-size picture data set A 'and a same-size photo data set B' with the same size by utilizing an interpolation algorithm, wherein each picture in the same-size picture data set A 'and the same-size photo data set B' is square, and the four types of junk pictures are respectively as follows: the garbage, kitchen garbage, other garbage and harmful garbage can be recovered;
s2: building ViT neural network frame structure, sending picture data set A' with the same size into ViT neural network for pre-training;
s3: dividing the photo data set B' with the same size into a training set B1 and an evaluation set B2;
s4: sending the training set B1 to the ViT neural network which is subjected to pre-training in S2 for fine adjustment;
s5: sending the evaluation set B2 to the ViT neural network which is subjected to fine tuning in S4 for accuracy test;
s6: and adding the pictures with the error detection in the evaluation set B2 into a training set to optimize ViT neural networks.
2. The ViT neural network-based spam recognition and classification algorithm of claim 1,
the ViT neural network framework structure in the S2 comprises: a picture block module, a linearization module, a position information adding module, a plurality of multi-head self-attention layers, a coding network module and a probability classifier module which are formed by overlapping with a full connection layer,
the image partitioning module divides an image sent into the ViT neural network into a plurality of image blocks with the same size, the linearization module linearizes tensors corresponding to the image blocks into vectors, the position information adding module adds position information in the vectors corresponding to the image blocks, the coding network module converts the sent image into a characteristic quantity containing image information, the probability classifier module outputs a classification result vector P containing a classification result according to the characteristic vector of the image information, 4 elements contained in the classification result vector P respectively correspond to four types of garbage probabilities, and the largest element in the 4 elements is a garbage category to which the garbage belongs.
3. The ViT neural network-based spam recognition and classification algorithm of claim 2,
the pre-training in the step S2 comprises the following steps:
s2.1: dividing all pictures in the same-size picture data set a' into tiles:
cutting the pictures through a sliding window, cutting each picture into n multiplied by n picture blocks, wherein the size of the sliding window of each cut picture block is consistent with that of the picture block, and the picture blocks cut by one picture are prevented from being overlapped;
s2.2: and (3) carrying out linearization processing on each image block:
(1) converting each image block into d1i*d2i*d3iOf the patch tensor, wherein d1i、d2i、d3iR, G, and B values in RGB of the color picture in the ith block, i ═ 1, 2, … …, n × n, respectively;
(2) converting the tile tensor into a tile vector X, X ═ (X1, X2, …, Xi, …, Xn×n) Wherein Xi ═ (d 1)i*d2i*d3i，1)；
(3) The image block vector X is linearly changed through the full connection layer, and the linear transformation formula is as follows:
Zi＝W*Xi+b(i＝1，2，……，n*n) (1)
wherein Zi is a characteristic block vector, is a linear block vector after the ith block is subjected to linear transformation, W is a multiplier matrix, is a multiplier parameter which is learned and output by the same-size picture data set A 'in the ViT neural network, and b is a constant matrix, is a constant parameter which is learned and output by the same-size picture data set A' in the ViT neural network;
(4) the full connection layer shares a parameter multiplier matrix number W and a constant matrix b;
s2.3: adding position information to the feature tile vector: firstly, the position information of each image block is encoded into a position matrix Li which has the same number of rows and columns as the number of rows and columns of the characteristic image block vector Zi through a position encoder, and then the position matrix is added to Zi to obtain a position characteristic image block vector Ci containing the position information:
Ci＝Zi+Li (2)
wherein Ci is a position feature pattern vector of the ith pattern block, and represents the feature pattern vector containing the position information of the ith pattern block, Li is a position matrix of the ith pattern block, i is 1, 2, … …, n is n;
s2.4: introducing classification information C0, adding C0 into the position feature tile vector set C to form a position feature tile vector set C ', C ═ C' containing classification information(C0，C1，C2，……，Cn×n)
S2.5: forming a coding network:
by the superposition of the multi-head self-attention layer and the full-connection layer, the learning effect of ViT neural networks is increased to form a coding network;
s2.6: acquiring an image feature vector:
inputting a position characteristic picture block vector set C 'containing classification information into an encoding network to obtain a position characteristic picture block encoding vector D1, wherein D1 is a final result of extracting a plurality of characteristic information of a picture by continuously performing vector transformation on C' i in the encoding network;
s2.7: and (3) judging the category and updating the network:
and inputting the position feature pattern block coding vector D1 into a probability classifier to obtain a classification result vector P, wherein 4 elements contained in the classification result vector P respectively correspond to four classes of garbage probabilities, the largest element in the 4 elements is the garbage class to which the garbage belongs, comparing a loss function between the classification result vector P and a real result, solving the parameter gradient of the loss function relative to the neural network, and updating ViT neural network parameters by using the gradient.
4. The ViT neural network-based spam recognition and classification algorithm of claim 3, wherein the loss function in S2.7 adopts the following formula:
Figure FDA0003092874480000031
wherein, loss represents a loss function, m represents the number of samples, P (xij) is the classification result of the i row and j column garbage in the classification result vector P, and represents the real probability of each garbage obtained by the garbage picture training set, q (xij) represents the prediction probability distribution of the i row and j column garbage predicted by the model, and log (q (xij)) represents the logarithm of q (xij).
5. The garbage classification device based on the garbage recognition and classification algorithm based on the ViT neural network as claimed in any one of claims 1-4, comprising:
the garbage dumping device comprises a frame structure (1), wherein 4 garbage cans (5) placed in a shape like a Chinese character tian are arranged in the frame structure (1), a garbage falling limiting device (4) is arranged at the middle position above each garbage can (5), a garbage dumping device (3) is arranged right above each garbage falling limiting device (4), a panel (2) is arranged at the top of the frame structure (1), a touch screen (8) is arranged on each panel (2), a garbage throwing opening is formed in the front side of the top of each panel (2), the garbage throwing opening is located right above each garbage dumping device (3), a camera (9) is arranged at the garbage throwing opening, and the camera (9) points to the garbage dumping devices (3); four garbage cans (5) are placed on the acrylic plate at the bottom of the device according to the shape of the Chinese character 'tian', and the distance between every two garbage cans is 50 mm.
The garbage dumping device (3) comprises a V-shaped plate (31), a garbage placing surface is placed behind the V-shaped plate (31), the opposite surface of the garbage placing surface is placed in front of the garbage placing surface, the area of the garbage placing surface is larger than the opening area of the garbage throwing opening, the V-shaped plate (31) is fixedly arranged on a horizontal transverse shaft (32) through a V-shaped plate support (34), one side of the transverse shaft (32) extends into a shaft support frame (33) and freely rotates on the vertical surface in the shaft support frame (33) by taking the transverse shaft (32) as a rotating shaft, the shaft support frame (33) is fixedly arranged in the frame structure (1), the other side of the transverse shaft (32) is connected with a first steering engine shaft connector (35), and the first steering engine shaft connector (35) is connected with a first steering engine (36);
the garbage falling limiter (4) comprises a garbage limiter rail (41), the garbage limiter rail (41) comprises a bottom plate which is low in front and high in back, baffles are arranged on two sides and the top of the bottom plate, the garbage limiter rail (41) is supported by rails (42) and a vertical shaft (43), the other end of the vertical shaft (43) is arranged in a second steering engine shaft connector (44), and the second steering engine shaft connector (44) is controlled by a second steering engine (45) to freely rotate in the horizontal plane by taking the center of the second steering engine shaft connector (44) as the circle center;
the output end of the camera (9) is connected with the input end of the main control unit (7), the output end of the main control unit (7) is connected with the input ends of the touch screen (8) and the execution unit (6), the output end of the execution unit (6) is connected with the input ends of the first steering engine (36) and the second steering engine (45), and the camera (9), the execution unit (6), the main control unit (7), the touch screen (8), the first steering engine (36) and the second steering engine (45) are connected with a power supply respectively.
6. The garbage classification device based on ViT neural network garbage recognition and classification algorithm according to claim 5,
the garbage bin is characterized in that an ultrasonic sensor (10) is arranged at the upper end in the garbage bin (5), the ultrasonic sensor (10) is connected with the main control unit (7), and the ultrasonic sensor (10) is used for carrying out full-load reminding on the garbage bin.
7. The method for controlling the garbage classification device based on the ViT neural network garbage recognition and classification algorithm, according to claim 5, comprises the following steps:
s1: starting a power supply, and initializing a system:
initializing a touch screen (8), and circularly playing the garbage classification promo through the touch screen (8);
initializing a garbage dumping device (3) to enable the opposite surface of a garbage placing surface on the V-shaped plate (31) to be in a vertical state;
initializing a garbage falling limiter (4) to enable the low surface of the bottom plate to face forwards and the high surface to face backwards;
initializing a first steering engine (36) and a second steering engine (45) to enable a first steering engine shaft connector (35) and a second steering engine shaft connector (44) to be in initial positions;
s2: placing garbage, and identifying the garbage:
garbage is placed into the garbage dumping device (3), a camera (9) captures a garbage image on the garbage dumping device (3), the main control unit (7) feeds information of the garbage image back to the execution unit (6), the execution unit (6) identifies the garbage type based on a garbage identification and classification algorithm of ViT neural network, the execution unit (6) feeds the identified garbage type back to the main control unit (7), and the main control unit (7) displays the identified garbage type on the touch screen (8);
s4: the second steering engine (45) is started, and the garbage falling limiter (4) points to the corresponding garbage can;
the execution unit (6) controls the second steering engine (45) to rotate the second steering engine shaft connector (44) to the position of the garbage bin corresponding to the garbage type identified in the S2 pointed by the lower edge of the bottom plate;
s5: after second steering wheel (45) start-up was accomplished, first steering wheel (36) started:
the execution unit (6) controls the first steering engine (36) to start, so that the first steering engine shaft connector (35) turns backwards to drive the garbage dumping device (3) to turn backwards, and garbage falls on the garbage limiting device rail (41) and then slides into the garbage can;
s6: and initializing the system, and waiting for the next garbage input.
8. The method for controlling a garbage classification device based on ViT neural network garbage recognition and classification algorithm as claimed in claim 6,
the garbage can full load reminding method comprises the following steps: ultrasonic sensor (10) continue to measure the rubbish in garbage bin (5) and the distance between ultrasonic sensor (10), and feed back the distance signal of telecommunication to main control unit (7), main control unit (7) convert the signal of telecommunication received into distance data and carry out logic judgement, rubbish when in garbage bin (5) exceeds rubbish capacity threshold value, main control unit (7) are judged and are reachd full load result, main control unit (7) are simultaneously to execution unit (6) and touch-sensitive screen (8) the signal of being full loaded, execution unit (6) control second steering wheel (45), make rubbish tipper (3) be in initial condition, the opposite face that the face was placed to rubbish on V template (31) is in vertical state, and simultaneously, touch-sensitive screen (8) show full load information.
CN202110617614.4A 2021-05-31 2021-05-31 Garbage recognition and classification method based on ViT, and device and control method thereof Active CN113213016B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110617614.4A CN113213016B (en) 2021-05-31 2021-05-31 Garbage recognition and classification method based on ViT, and device and control method thereof

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110617614.4A CN113213016B (en) 2021-05-31 2021-05-31 Garbage recognition and classification method based on ViT, and device and control method thereof

Publications (2)

Publication Number Publication Date
CN113213016A true CN113213016A (en) 2021-08-06
CN113213016B CN113213016B (en) 2022-08-12

Family

ID=77082487

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110617614.4A Active CN113213016B (en) 2021-05-31 2021-05-31 Garbage recognition and classification method based on ViT, and device and control method thereof

Country Status (1)

Country Link
CN (1) CN113213016B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113562356A (en) * 2021-08-20 2021-10-29 阿尔飞思(昆山)智能物联科技有限公司 Garbage classification box and classification method
CN114781499A (en) * 2022-04-08 2022-07-22 上海人工智能创新中心 Method for constructing ViT model-based intensive prediction task adapter
CN115215008A (en) * 2022-06-15 2022-10-21 江苏康乾环境安全技术有限公司 Medical waste's recovery transfer device
CN117373016A (en) * 2023-10-20 2024-01-09 农芯(南京)智慧农业研究院有限公司 Tobacco leaf baking state judging method, device, equipment and storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3078611A1 (en) * 2015-04-09 2016-10-12 Fomento de Construcciones y Contratas, S.A. Solid waste classification system
CN109261539A (en) * 2018-08-17 2019-01-25 湖北文理学院 A kind of garbage sorting system and method for view-based access control model identification and convolutional neural networks
CN109928107A (en) * 2019-04-08 2019-06-25 江西理工大学 A kind of automatic classification system
CN110589282A (en) * 2019-08-16 2019-12-20 喻鑫童 Intelligent garbage classification method based on machine learning and automatic garbage sorting device
CN112061624A (en) * 2020-07-13 2020-12-11 浙江科技学院 Garbage classification recognition device and recognition method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3078611A1 (en) * 2015-04-09 2016-10-12 Fomento de Construcciones y Contratas, S.A. Solid waste classification system
CN109261539A (en) * 2018-08-17 2019-01-25 湖北文理学院 A kind of garbage sorting system and method for view-based access control model identification and convolutional neural networks
CN109928107A (en) * 2019-04-08 2019-06-25 江西理工大学 A kind of automatic classification system
CN110589282A (en) * 2019-08-16 2019-12-20 喻鑫童 Intelligent garbage classification method based on machine learning and automatic garbage sorting device
CN112061624A (en) * 2020-07-13 2020-12-11 浙江科技学院 Garbage classification recognition device and recognition method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
SHUSENWANG: "《Vision Transformer (ViT) 用于图片分类》", 13 May 2021 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113562356A (en) * 2021-08-20 2021-10-29 阿尔飞思(昆山)智能物联科技有限公司 Garbage classification box and classification method
CN114781499A (en) * 2022-04-08 2022-07-22 上海人工智能创新中心 Method for constructing ViT model-based intensive prediction task adapter
CN114781499B (en) * 2022-04-08 2024-05-31 上海人工智能创新中心 Method for constructing ViT model-based intensive prediction task adapter
CN115215008A (en) * 2022-06-15 2022-10-21 江苏康乾环境安全技术有限公司 Medical waste's recovery transfer device
CN117373016A (en) * 2023-10-20 2024-01-09 农芯(南京)智慧农业研究院有限公司 Tobacco leaf baking state judging method, device, equipment and storage medium
CN117373016B (en) * 2023-10-20 2024-04-30 农芯(南京)智慧农业研究院有限公司 Tobacco leaf baking state judging method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN113213016B (en) 2022-08-12

Similar Documents

Publication Publication Date Title
CN113213016B (en) Garbage recognition and classification method based on ViT, and device and control method thereof
WO2019000929A1 (en) Garbage sorting and recycling method, garbage sorting equipment, and garbage sorting and recycling system
CN110598752B (en) Image classification model training method and system for automatically generating training data set
CN110116415A (en) A kind of Bottle & Can class rubbish identification sorting machine people based on deep learning
CN110991435A (en) Express waybill key information positioning method and device based on deep learning
CN110697273A (en) Intelligent household garbage identification and automatic classification system and method based on iterative learning control
CN107913892B (en) Intelligent garbage classification processing method and system
CN112827846B (en) Automatic garbage classification device and method
CN111186656A (en) Target garbage classification method and intelligent garbage can
Gyawali et al. Comparative analysis of multiple deep CNN models for waste classification
CN206546593U (en) House refuse intelligent classification reclaims cloud identifying system
CN111498331A (en) Garbage visual recognition method and device based on deep learning and classified recovery device
CN109675827A (en) A kind of building waste identification sorting device, recognition methods and its grasping means
CN113673298B (en) Overflow monitoring method based on time sequence gating circulation unit model
CN111619992A (en) Intelligent garbage classification system and method based on machine vision
CN103646251B (en) Apple postharvest field classification detection method and system based on embedded technology
CN113807347A (en) Kitchen waste impurity identification method based on target detection technology
CN113469264A (en) Construction method of automatic garbage classification model, garbage sorting method and system
CN110738131A (en) Garbage classification management method and device based on deep learning neural network
CN114170672A (en) Classroom student behavior identification method based on computer vision
CN110321803A (en) A kind of traffic sign recognition method based on SRCNN
CN111217062A (en) Garbage can garbage identification method based on edge calculation and deep learning
CN111680650A (en) Garbage classification recycling excitation method and system
CN113859803A (en) Intelligent identification trash can and intelligent identification method thereof
CN111137602B (en) Garbage classification method and garbage classification system based on artificial intelligence

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
GR01 Patent grant
GR01 Patent grant