WO2019000929A1 - 垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统 - Google Patents

垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统 Download PDF

Info

Publication number
WO2019000929A1
WO2019000929A1 PCT/CN2018/073809 CN2018073809W WO2019000929A1 WO 2019000929 A1 WO2019000929 A1 WO 2019000929A1 CN 2018073809 W CN2018073809 W CN 2018073809W WO 2019000929 A1 WO2019000929 A1 WO 2019000929A1
Authority
WO
WIPO (PCT)
Prior art keywords
garbage
image
recycling
classified
sorting
Prior art date
Application number
PCT/CN2018/073809
Other languages
English (en)
French (fr)
Inventor
曾起
张丽杰
耿立华
严寒
马希通
赵天月
Original Assignee
京东方科技集团股份有限公司
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 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Priority to US16/076,451 priority Critical patent/US11446706B2/en
Publication of WO2019000929A1 publication Critical patent/WO2019000929A1/zh

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/34—Sorting according to other particular properties
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/30—Administration of product recycling or disposal
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/34—Sorting according to other particular properties
    • B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
    • B07C5/3422—Sorting according to other particular properties according to optical properties, e.g. colour using video scanning devices, e.g. TV-cameras
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/34—Sorting according to other particular properties
    • B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/36—Sorting apparatus characterised by the means used for distribution
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/36—Sorting apparatus characterised by the means used for distribution
    • B07C5/361—Processing or control devices therefor, e.g. escort memory
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/36—Sorting apparatus characterised by the means used for distribution
    • B07C5/361—Processing or control devices therefor, e.g. escort memory
    • B07C5/362—Separating or distributor mechanisms
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C2501/00—Sorting according to a characteristic or feature of the articles or material to be sorted
    • B07C2501/0054—Sorting of waste or refuse
    • 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
    • G06N3/048—Activation functions
    • 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
    • Y02W90/00—Enabling technologies or technologies with a potential or indirect contribution to greenhouse gas [GHG] emissions mitigation

Definitions

  • Embodiments of the present disclosure relate to a garbage sorting and recycling method, a garbage sorting apparatus, and a garbage sorting and recycling system.
  • the garbage is mainly disposed of by landfill and incineration.
  • the garbage that can be recycled in the garbage can be classified and recycled.
  • the method for classification and recycling of recyclable garbage mainly includes manual sorting, wind selection, etc., which has high labor cost, slow processing speed and low efficiency.
  • At least one embodiment of the present disclosure provides a garbage classification and recovery method, including: acquiring a detection image of the garbage to be classified; processing the detection image using a deep learning neural network to determine whether the garbage to be classified belongs to recyclable garbage, and if And sending a first control signal to control the garbage to be sorted into the recovery area; if not, issuing a second control signal to control the garbage to be classified into the non-recycling area.
  • the detected image includes a plurality of images taken from different angles
  • the garbage sorting and recycling method further includes: processing the detecting by using the deep learning neural network.
  • the image is combined with the processing result of the image taken at different angles to determine whether the garbage to be classified belongs to the recyclable garbage, and if so, the first control signal is sent to control the garbage to be classified into the office.
  • Said recovery zone if not, issuing said second control signal to control delivery of said waste to be collected into said non-recovery zone.
  • the garbage sorting and recycling method further includes: acquiring an identification number of the garbage sorting device corresponding to the collected detection image; and selecting the depth corresponding to the identification number according to the identification number. Learn the training parameters of the neural network.
  • the garbage sorting and recycling method provided by an embodiment of the present disclosure further includes: in the case of determining that the garbage to be classified belongs to the recyclable garbage, statistically recovering the quantity; and the predetermined quantity of the collected quantity exceeding the predetermined area of the recycling area In the case of quantity, a recovery control signal is issued to prompt the recycling center to recycle the garbage in the recycling area.
  • the recyclable garbage includes at least one type of recyclable waste
  • the recovery area includes at least one sub-recovery area
  • the first control signal includes at least a sub-control signal
  • the garbage classification and recovery method further includes: determining whether the garbage to be classified belongs to one of the at least one type of recyclable garbage, and if so, issuing the corresponding sub-control signal to control And sending the garbage to be classified into the corresponding sub-recovery area; if not, issuing the second control signal to control the garbage to be classified into the non-recycling area.
  • determining whether the to-be-classified garbage belongs to the recyclable garbage includes: calculating a matching ratio between the to-be-classified garbage and the recyclable garbage; Whether the matching rate exceeds the first preset matching rate threshold, and if yes, determining that the to-be-classified garbage belongs to the recyclable garbage; if not, determining that the to-be-classified garbage does not belong to the recyclable garbage.
  • the garbage classification and recovery method provided by the embodiment of the present disclosure further includes: determining whether the matching rate exceeds a second preset matching rate threshold, if the matching rate is lower than the first preset matching rate threshold. And if yes, storing the detection image; if not, deleting the detection image, wherein the second preset matching rate threshold is smaller than the first preset matching rate threshold.
  • the garbage sorting and recycling method further includes: determining whether the stored garbage in the detected image belongs to the recyclable garbage, and if yes, adding the stored detection image.
  • the garbage classification and recovery method provided by an embodiment of the present disclosure further includes: after adding the stored detection image to the sample image library, retraining the deep learning neural network by using the training image in the sample image library. And modifying the training parameters of the deep learning neural network according to the training result.
  • the deep learning neural network is a convolutional neural network.
  • At least one embodiment of the present disclosure also provides a garbage sorting apparatus, including: a sorting structure, an image collecting device, and a terminal controller.
  • the image acquisition device is configured to acquire a detection image of the garbage to be classified;
  • the terminal controller is configured to transmit the detection image, and is further configured to receive a control signal and control the classification structure according to the control signal.
  • the garbage sorting apparatus further includes a box, the box includes a recovery area and a non-recycling area, and the classification structure is configured to collect the garbage to be classified under the control of the control signal. Feeded into the recovery zone or the non-recovery zone.
  • At least one embodiment of the present disclosure also provides a garbage sorting and recycling system, comprising: a control device and the garbage sorting device according to any one of the above.
  • the control device includes a processor and a memory, the memory storing a computer program adapted to be executed by the processor, the computer program being executed by the processor to perform the steps of: acquiring the detected image; using The deep learning neural network processes the detected image to determine whether the garbage to be classified belongs to recyclable garbage, and if yes, sends a first control signal to the garbage sorting device to control the garbage to be sorted into the a recycling zone; if not, issuing a second control signal to the garbage sorting device to control the feeding of the waste to be sorted into the non-recycling zone.
  • the detected image includes a plurality of images taken from different angles
  • the computer program when executed by the processor, performing the following steps: using the The deep learning neural network processes the detection image and combines the processing results of the images taken at different angles to determine whether the garbage to be classified belongs to the recyclable garbage, and if so, issues the first control signal; if not, Then issuing the second control signal.
  • the following step is further performed: acquiring an identification number of the garbage sorting device corresponding to the collected detection image. And selecting, according to the identification number, a training parameter of the deep learning neural network corresponding to the identification number.
  • the following steps are further performed: in a case where it is determined that the garbage to be classified belongs to the recyclable garbage, The amount of recovery is counted; in the case where the amount of recovery exceeds the predetermined number of carriers of the recovery zone, a recovery control signal is issued to prompt the recycling center to recycle the garbage of the recovery zone.
  • the following step is further performed: calculating a matching ratio between the to-be-classified garbage and the recyclable garbage And determining whether the matching rate exceeds a first preset matching rate threshold, and if yes, determining that the to-be-classified garbage belongs to the recyclable garbage; if not, determining that the to-be-classified garbage does not belong to the recyclable garbage.
  • the following step is further performed: when the matching rate is lower than the first preset matching rate threshold In the case, it is determined whether the matching rate exceeds a second preset matching rate threshold, and if yes, storing the detection image; if not, deleting the detection image, wherein the second preset matching rate threshold is less than the The first preset matching rate threshold.
  • the following step is further performed: determining, by the processor, whether the garbage to be classified in the stored detected image belongs to The recyclable garbage, if so, the stored detection image is added to the sample image library of the deep learning neural network; if not, the stored detection image is deleted.
  • the following step is further performed: after the stored detection image is added to the sample image library, The training image in the sample image library retrains the deep learning neural network, and corrects the training parameters of the deep learning neural network according to the training result.
  • At least one embodiment of the present disclosure provides a garbage sorting and recycling method, a garbage sorting device, and a garbage sorting and recycling system.
  • the method for deep learning neural network is used for real-time detection and identification of classified garbage, and is automatically classified and recovered.
  • FIG. 1A is a schematic flowchart of an example of a garbage sorting and recycling method according to an embodiment of the present disclosure
  • FIG. 2 is a schematic flowchart of a training process and a detection process of a deep learning neural network according to an embodiment of the present disclosure
  • FIG. 3A is a schematic block diagram of a garbage sorting device according to an embodiment of the present disclosure.
  • FIG. 3B is a schematic structural diagram of a garbage sorting device according to an embodiment of the present disclosure.
  • FIG. 4 is a schematic block diagram of a garbage sorting and recycling system according to an embodiment of the present disclosure
  • FIG. 5 is a schematic diagram of a garbage sorting and recycling system according to an embodiment of the present disclosure.
  • Garbage bins are widely used in various areas such as beaches, stations, libraries, schools, etc. In different areas, trash cans receive different garbage, so that different types of recyclable garbage can be sorted and recycled according to different areas. For example, in areas such as libraries and schools, it is necessary to recycle waste paper such as waste paper; in areas such as beaches and stations, it is necessary to recycle recyclable garbage such as plastic bottles. In order to reduce the classification and recycling work of the garbage collection center staff, the recyclable garbage can be automatically classified in the garbage bin to improve the efficiency of garbage separation and recycling.
  • At least one embodiment of the present disclosure provides a garbage sorting and recycling method, including: acquiring a detected image of the garbage to be classified; processing the detected image by using a deep learning neural network to determine whether the garbage to be classified belongs to the recyclable garbage, and if so, issuing the first a control signal to control the garbage to be sorted into the recovery area; if not, a second control signal is sent to control the garbage to be sorted into the non-recycling area.
  • the garbage sorting and recycling method adopts the method of deep learning neural network to detect and identify the classified garbage in real time, and automatically classifies and recycles, thereby improving the recognition accuracy of the garbage to be sorted, reducing the classification and recycling work of the garbage recycling processing center staff, and improving the garbage.
  • the efficiency of sorting and recycling reduces the cost of waste sorting and recycling.
  • FIG. 1A is a schematic flowchart of an example of a garbage sorting and recycling method provided by an embodiment of the present disclosure
  • FIG. 1B is a schematic flowchart of another example of a garbage sorting and recycling method provided by an embodiment of the present disclosure.
  • the garbage sorting and recycling method may include the following operations:
  • S2 processing the detection image by using a deep learning neural network to determine whether the garbage to be classified belongs to recyclable garbage;
  • perform operation S3 issue a first control signal to control the garbage to be classified into the recovery area
  • operation S4 is performed: a second control signal is issued to control the garbage to be sorted into the non-recycling area.
  • the garbage sorting and recycling method provided by the embodiments of the present disclosure adopts a deep learning neural network method to perform real-time detection and identification of classified garbage, and automatically classifies and recycles, thereby improving the recognition accuracy of the garbage to be classified, and reducing the classification of the garbage collection processing center staff. Recycling work, improve the efficiency of waste separation and recycling, and reduce the cost of waste separation and recycling.
  • the garbage sorting method can be performed on the server side (or the cloud side).
  • the server end or the cloud
  • the garbage sorting device may send an image acquisition signal, and transmit the image acquisition signal to the garbage sorting device, and the garbage sorting device collects the detection image of the garbage to be classified according to the received image acquisition signal, and the detection image is Transfer to the server (or cloud).
  • the garbage sorting device may also periodically transmit the detected image of the garbage to be classified to the server (or the cloud).
  • the server side processes the received detection image using the deep learning neural network to determine whether the garbage to be classified in the detection image belongs to the recyclable garbage, and according to the processing result, the server end (or the cloud end) issues a control signal, and the The control signal is transmitted to the garbage sorting device to control the garbage sorting device to send the garbage to be classified into the designated area.
  • the server or the cloud
  • the server can automatically send out control signals; or the background user can manually control the server (or the cloud) to send control signals.
  • the garbage sorting and recycling method provided by the embodiment of the present disclosure only needs to collect the detected image of the garbage to be classified at the garbage sorting device, and transmits the collected detected image to the server end (or the cloud end), and the image is performed by the server end (or the cloud end). Identification and detection, so there is no need to set special hardware and complex software at the garbage sorting device, which is easy to maintain and popularize the garbage sorting equipment.
  • garbage sorting and recycling method on the server side (or the cloud side) as an example. It should be understood by those skilled in the art that the garbage sorting and recycling method provided by the present disclosure may also be performed only on the garbage sorting device end, and the implementation of the present disclosure is not limited thereto.
  • the detected image may include a plurality of images taken from different angles.
  • the detected image may include a first detected image and a second detected image.
  • the garbage sorting and recycling method shown in FIG. 1B is similar to the garbage sorting and recycling method shown in FIG. 1A, and the difference is:
  • S100 Acquire a first detection image and a second detection image of the garbage to be classified
  • S200 The first detection image and the second detection image are processed by using a deep learning neural network to determine whether the garbage to be classified belongs to recyclable garbage.
  • the steps after the operation S200 are the same as the steps after the operation S2 in FIG. 1A, that is, the operations S3 and S4 are performed. That is, in the example shown in FIG. 1B, in operation S200, the first detection image and the second detection image may be separately processed using the depth learning neural network, in combination with the processing result of the first detection image and the second detection image. The result is processed to determine whether the waste to be classified belongs to recyclable waste. If so, operation S3 is performed; if not, operation S4 is performed.
  • step S200 it may be separately determined whether the garbage to be classified in the first detected image and the garbage to be classified in the second detected image belong to the recyclable garbage; or may be combined with the first detected image and the second detected image.
  • the characteristic information of the garbage to be classified extracted is judged whether it belongs to the recyclable garbage.
  • the photographing angles of the first detected image and the second detected image are different.
  • using multiple detection images with different shooting angles to detect and identify the classified garbage more feature information of the garbage to be classified can be extracted, thereby improving the recognition accuracy and reducing the false positive rate. .
  • the operation of acquiring the first detection image and the operation of the second detection image may be performed in parallel, or may be performed in time and in time.
  • the deep learning neural network may include a convolutional neural network (CNN), a stack self-encoding network, a sparse coding network, a cyclic neural network, a deep belief network, etc., and the deep learning neural network may adopt one of the above neural networks. Training combinations of recyclable garbage can be trained in several combinations.
  • FIG. 2 is a schematic flowchart of a training process and a detection process of a deep learning neural network according to an embodiment of the present disclosure.
  • the training process includes the following operations.
  • the training image may include images of recyclable trash at different shooting angles.
  • operation S03 After preprocessing the training image, operation S03 is performed: the training image is randomly selected and initialized. Initializing the training image can transform the training image into a data signal that the deep learning neural network can process to facilitate subsequent operations.
  • operation S04 convolution and sampling is performed.
  • the training image can be subjected to multiple convolution and sampling processing, and the convolution processing can extract features of the training image, and the sampling processing can reduce the scale of the training data and reduce the amount of calculation.
  • multiple convolution kernels are applied to the training image to obtain multiple feature maps to obtain different features of the recyclable garbage.
  • Each feature map for example, extracts a feature of recyclable trash.
  • the sampling process can process the training image by means of average combining, maximum combining, and random combining.
  • each set of pixels (eg, four pixels, etc.) in the feature map obtained after the convolution process may be summed, multiplied by a weight value, and offset,
  • a new feature map can be obtained.
  • the weighting value and the bias control the linearity of the sigmoid function. If the weighting value is small, the operation of the sigmoid function approximates the linear operation, and the sampling process is equivalent to the blurred image; if the weighting value is large, the sampling process according to the magnitude of the offset It can be equivalent to a noisy OR operation or a noisy AND operation.
  • operation S05 full connection is performed.
  • the last sampling layer or convolutional layer is connected to one or more fully connected layers.
  • the fully connected layer is configured to synthesize the characteristics of the recyclable garbage extracted after the convolution and sampling processing and output the training parameters and feature models of the recyclable garbage.
  • the feature model is an abstract feature representation of recyclable garbage.
  • processing the first detected image and/or the second detected image using the deep learning neural network may include the following operations:
  • operation S11 After acquiring the first detection image and/or the second detection image of the garbage to be classified, operation S11 is performed: convolution and sampling.
  • the first detection image and/or the second detection image may be convoluted and sampled by using the convolution training parameters and the sampling training parameters acquired by the above training process, thereby obtaining the first detection image and/or the second detection image.
  • the characteristics of the garbage to be classified are described.
  • operation S12 full connection is performed.
  • the fully connected layer is configured to synthesize various features of the garbage to be classified and output a feature model of the garbage to be classified.
  • S13 Detection. For example, comparing the characteristic model of the garbage to be classified with the characteristic model of the recyclable garbage obtained by training to determine whether the garbage to be classified belongs to the recyclable garbage.
  • each sample image library includes training images of the same type of recyclable garbage, and the training images may include images of recyclable garbage at different angles and different forms to more fully acquire the characteristics of the recyclable garbage.
  • the training image may include a base view of a recyclable garbage such as a front view, a rear view, a bottom view, a top view, a left view, and a right view.
  • the sample image library of the deep learning neural network, the training model parameters, and the like may be deployed in the form of a database on the background server side, or may be deployed on a server side of a local area network or a wide area network (for example, a cloud) to be, for example, a background server side.
  • the background server can be set up in the monitoring room and other places for remote monitoring.
  • the first detection image and/or the second detection image may include an image of garbage to be classified, and may also include multiple images of garbage to be classified.
  • the number of the first detection image and/or the second detection image may be preset, or may be randomly generated by the controller or the server end (or the cloud) at the garbage sorting device when performing garbage sorting and recycling.
  • the first detected image includes only one image of the garbage to be classified
  • the second detected image also includes only one image of the garbage to be classified.
  • the detected image may be a grayscale image or a color image.
  • the detected image may be a photo, or may be a composite image of one frame, multiple frames, or multiple frames in the video.
  • the detected image may be pre-processed to facilitate extracting feature information of the garbage to be classified in the detected image, thereby improving reliability of feature extraction.
  • the pre-processing may include a process of scaling the photo, gamma correction, image enhancement, or noise reduction filtering, and in the case where the detected image is acquired from the video, the pre-processing may include Extract key frames of the video, etc.
  • the pre-processing can be performed before processing the detected image using the deep learning neural network, that is, before performing operation S11.
  • the first original image and/or the second original image when the image capturing area in the garbage sorting device is not placed with any object may be stored in advance.
  • the photographing angle of the first original image is the same as the first detected image
  • the photographing angle of the second original image is the same as the second detected image.
  • the image capturing area is monitored in real time by an image collecting device (for example, a camera), and the detected image is acquired from the video acquired by the image capturing device at a timing (for example, every 10 seconds, 30 seconds, or 1 minute) to obtain a first image and/or a second image, the first image having the same shooting angle as the first detected image, the second image having the same shooting angle as the second detected image; and then comparing the first image with the first original image, and Or, comparing the second image with the second original image, when the similarity ratio of the first image to the first original image is lower than a predetermined first similarity threshold, and/or the second image and the first original image In the case where the similarity ratio is lower than the predetermined second similarity threshold, it is determined that the image collection area is placed with the garbage to be classified, and the first image is taken as the first detected image and the second image is taken as the second detected image.
  • an image collecting device for example, a camera
  • the detected image is acquired from the video acquired by the image capturing device at
  • the acquired first detection image and/or the second detection image are transmitted to the server end (or the cloud), and the server end (or the cloud end) processes the first detection image and/or the second detection image, and sends the control according to the processing result.
  • the signal is sent to the garbage sorting device to control the garbage sorting device to send the garbage to be classified into the designated area.
  • the image collection device may stop collecting the detection image of the garbage to be classified, and when the garbage classification device receives the server end ( Or the cloud transmits the control signal, and after the garbage to be classified is sent to the designated area, the image collecting device re-times the detected image in the video. Therefore, it is possible to prevent the first detection image and/or the second detection image of the same garbage to be classified from being repeatedly collected, thereby reducing processing time and improving work efficiency.
  • the garbage sorting device may set a timer or a timing program, and the timer or the timing program may periodically trigger the image collecting device to collect the detected image of the garbage to be classified.
  • the timer or the timer program stops working, so that the image collection device stops collecting the detection image of the garbage to be classified, and when the garbage classification device receives the transmission from the server (or the cloud)
  • the timer or timer program performs the clear operation and re-times.
  • the garbage sorting device can also set a sensor as needed.
  • the sensor is configured to sense whether there is garbage to be classified in the image collection area, and if yes, the image collection device collects the detection image of the garbage to be classified, and then performs subsequent operations on the detection image; if not, the image acquisition device does not perform any operation to Save power.
  • the garbage sorting and recycling method further includes: acquiring an identification number of the garbage sorting device corresponding to the collected detection image; and selecting a training parameter of the deep learning neural network corresponding to the identification number according to the identification number.
  • the training parameters may include convolution training parameters, sampling training parameters, etc., and may also include parameters such as feature models.
  • the identification number of the garbage sorting device in different areas can be set in advance. Based on the identification number of each garbage sorting device, the training parameters of the deep learning neural network corresponding to the identification number are selected, so that different types of recyclable garbage are detected and recovered according to different regions.
  • garbage sorting equipment installed in school buildings, libraries, etc. can be used to recycle waste paper such as waste paper; garbage sorting equipment installed in beaches, stations, basketball courts, etc. can be used for recycling plastic bottles, etc. recycle rubbish.
  • each waste sorting device can recycle one type of recyclable waste or recycle multiple different types of recyclable waste.
  • the identification number of the garbage sorting device may correspond to a plurality of different training parameters of the deep learning neural network to achieve recovery of a plurality of different types of recyclable waste.
  • the recyclable waste includes at least one type of recyclable waste
  • the reclaimed area of each refuse sorting device includes at least one sub-recovery zone
  • the first control signal includes at least one sub-control signal.
  • the recyclable trash can include plastic articles (eg, including plastic bottles, etc.), paper products (eg, including A4 paper, books, etc.), metal products (eg, including cans, etc.), and glass articles, and the like.
  • the recovery area may include a plastic product sub-recovery area, a paper product sub-recovery area, a metal product sub-recovery area, and a glass product sub-recovery area
  • the first control signal may also include a plastic product sub-control signal and a paper product sub-control signal. , metal product sub-control signals and glass product sub-control signals.
  • the garbage sorting and recycling method may further include: determining whether the garbage to be classified belongs to one of the at least one type of recyclable garbage, and if so, issuing a corresponding sub-control signal to control the garbage to be classified. It is sent to the corresponding sub-recovery area; if not, a second control signal is sent to control the garbage to be sorted into the non-recycling area.
  • the recyclable garbage includes a plurality of types of recyclable garbage
  • determining whether the garbage to be classified belongs to one of the plurality of types of recyclable garbage may include: first, acquiring a plurality of different training parameters corresponding to the identification number; and then utilizing The first detection image and/or the second detection image are separately processed by each of the plurality of different training parameters, thereby obtaining a processing result of the plurality of first detection images and/or a processing result of the plurality of second detection images; finally, Combining the processing results of the plurality of first detection images and/or the processing results of the plurality of second detection images, determining whether the garbage to be classified belongs to one of a plurality of types of recyclable garbage.
  • the number of different training parameters corresponding to the identification number may be the same as the number of multiple types of recyclable garbage.
  • the identification number of the garbage sorting device can be in various forms and can include different types of information.
  • the identification number may be an identification code (for example, a character string), and the identifier may be used to obtain one or more kinds of information such as a training parameter, a feature model, a type of recyclable garbage, and the like by the identification code; for example, the identification information It may be a compliance code, for example, including both an identification code and a geographical location (longitude, latitude) information of the garbage sorting device.
  • the identification numbers may be stored in one database in a centralized manner and deployed on one or more servers for querying, which is not limited by the embodiments of the present disclosure.
  • the garbage sorting and recycling method further includes the following operations: counting the amount of recycling in the case where it is determined that the garbage to be classified belongs to the recyclable garbage; and issuing the recycling in the case that the collected quantity exceeds the predetermined carrying amount of the recycling area A control signal to prompt the recycling center to recycle the garbage in the recycling area.
  • the predetermined number of bearers may be set in advance according to the size of the recovery area and the type of recyclable garbage.
  • the garbage sorting device can be provided with a counter or counting program.
  • the garbage sorting device receives the first control signal and sends the garbage to be recycled into the recycling area, that is, in the case where it is determined that the garbage to be classified belongs to the recyclable garbage
  • the counter or the counting program counts the collected quantity, in the counter or the counting program If the counted amount of the collection exceeds the predetermined number of bearers in the recovery area, the server side (or the cloud) sends a recovery control signal to prompt the recycling center to recycle the garbage in the collection area.
  • the garbage sorting device when the garbage sorting device receives the second control signal and sends the garbage to be recycled into the non-recycling area, that is, when it is determined that the garbage to be classified does not belong to the recyclable garbage, the counter or the counting program may count the non-recycling The amount of garbage in the recycling area.
  • the server or cloud may issue a garbage collection control signal to prompt the recycling center or the garbage disposal center to collect the non-recycling area. Trash.
  • the counter or counting program can also be set on the server side (or the cloud side).
  • the disclosure does not limit this.
  • the recovery control signal may include information such as the type of recyclable garbage, the location of the garbage sorting device, and the like.
  • determining whether the garbage to be classified belongs to the recyclable garbage includes the following operations: calculating a matching ratio between the garbage to be classified and the recyclable garbage; determining whether the matching rate exceeds a threshold of the first preset matching rate, and if so, It is determined that the garbage to be classified belongs to recyclable garbage; if not, it is determined that the garbage to be classified is not recyclable garbage.
  • the detection image may also be stored as a sample of the subsequent deep learning training.
  • the detection image is stored while the first control signal is being issued.
  • the matching rate can be the detection result of the deep learning neural network output.
  • the first preset matching rate threshold may be 90%, that is, when the matching rate of the garbage to be classified and the recyclable garbage exceeds 90%, it may be determined that the garbage to be classified belongs to the recyclable garbage.
  • the garbage classification and recovery method further includes: if the matching rate is lower than the first preset matching rate threshold, determining whether the matching rate exceeds a second preset matching rate threshold, and if so, storing detection Image; if not, delete the detected image.
  • the second preset matching rate threshold is smaller than the first preset matching rate threshold.
  • the second preset matching rate threshold may be 80%, that is, when the matching rate of the garbage to be classified and the recyclable garbage exceeds 80% but is less than 90%, the detection image may be stored as a sample of the subsequent deep learning training.
  • first preset matching rate threshold and the second preset matching rate threshold may also be other values, as long as the second preset matching rate threshold is less than the first preset matching rate threshold.
  • the garbage sorting and recycling method further includes the following operations: determining whether the garbage to be classified in the stored detected image belongs to the recyclable garbage, and if so, adding the stored detection image to the sample image library of the deep learning neural network. If not, delete the stored test image.
  • determining whether the garbage to be classified in the stored detection image belongs to the recyclable garbage can prevent misjudgment and increase the training sample of the deep learning, thereby dynamically adjusting the training parameters of the deep learning neural network in real time.
  • the method of judging whether the garbage to be classified belongs to the recyclable garbage may be different from the method of the previous judgment, for example, a statistical method (ie, a decision theory method), a syntax recognition method, a neural network method, a template matching method, or a geometric transformation method may be used.
  • the detection image is re-detected and identified by one or more of the combination to determine whether the garbage to be classified in the stored detection image belongs to recyclable garbage.
  • the background user it is also possible for the background user to manually check whether the garbage to be classified in the stored detection image belongs to the recyclable garbage, and control the stored detection image to be added to the sample image library or delete the stored detection image according to the input instruction of the user.
  • the garbage sorting and recycling method further includes the following operations: after adding the detected image to the sample image library, retraining the deep learning neural network by using the training image in the sample image library, and modifying the deep learning neural network according to the training result. Training parameters.
  • the garbage sorting and recycling method provided by the example can expand the sample image library in time, cyclically train the sample images in the sample image library, and correct the training parameters, thereby further improving the recognition accuracy and reducing the false positive rate.
  • FIG. 3A is a schematic block diagram of a garbage sorting apparatus according to an embodiment of the present disclosure
  • FIG. 3B is a schematic structural diagram of a garbage sorting apparatus according to an embodiment of the present disclosure.
  • the garbage sorting apparatus 10 may include an image capture device 11, a cabinet 12, a sorting structure 13, an image collection area 14, and a terminal controller 15.
  • image capture device 11 may include one or more cameras.
  • the terminal controller 15 can be implemented by hardware, software, firmware, and any feasible combination thereof.
  • the image acquisition area 14 is configured to place the garbage to be classified
  • the image collection device 11 is configured to collect the detection image of the garbage to be classified.
  • the detected image may include a plurality of images taken from different angles.
  • the detected image includes the first detected image and the second detected image, and the photographing angles of the first detected image and the second detected image are different.
  • the garbage sorting device 10 may include two image capturing devices 11 and are respectively disposed at the top and the side of the image capturing area 14, and the image capturing device 11 at the top may collect the first detected image, that is, the first detected image.
  • the image taken by the X direction; the image capturing device 11 located on the side may acquire the second detected image, that is, the second detected image is an image taken from the Y direction, and the X direction and the Y direction may be perpendicular to each other.
  • the X direction may be a vertical direction and the Y direction may be a horizontal direction.
  • the image capturing device 11 may be a network camera, a digital camera, a color dome camera, an infrared camera or an integrated camera, etc., to capture the image capturing area 14 in real time, and then periodically collect the detected image from the video image captured by the image capturing device 11. .
  • the image capture device 11 may also include a camera to periodically take a picture of the image capture area 14 to acquire a detected image.
  • the detected image may be stored, for example, in the image capture device 11 to be used by other components (for example, the terminal controller 15 or the like) in the garbage sorting device 10 as needed.
  • the timer may be a pulse type timer, an on-delay type timer, an off-delay type timer, or the like.
  • the terminal controller 15 may store a timing program, and when a timing operation is required, the terminal controller 15 may directly run the timing program to implement a timing function.
  • the garbage sorting device 10 can also include sensors.
  • the sensor is configured to sense whether the image collection area 14 has garbage to be classified. If yes, the image collection device 11 collects the detection image of the garbage to be classified, and then performs subsequent operations on the detection image; if not, the image acquisition device 11 does not perform any operation. Operate to save power.
  • the terminal controller 15 is configured to transmit a detection image to the server side (or the cloud), and is also configured to receive a control signal from the server side (or the cloud side) and control the classification structure 13 according to the control signal.
  • the terminal controller 15 can be disposed on the side of the image acquisition area 14.
  • the detected image may be transmitted to the server (or the cloud), and the server (or cloud) processes the detected image, and generates a control signal according to the processing result, and the control signal may be transmitted to the terminal controller 15 for example, and the terminal controls
  • the controller 15 controls the control classification structure 13 to deliver the garbage to be classified into the designated area according to the control signal.
  • the terminal controller 15 is further configured to acquire the identification number of the garbage sorting device 10 and transmit the identification number to the server side (or the cloud side).
  • the identification number refer to the related description in the embodiment of the garbage classification and recovery method, and details are not described herein again.
  • the terminal controller 15 may include components such as a terminal processor, a communication device, a power module, and the like.
  • the terminal processor may be a micro control unit (MCU) or the like.
  • the power module can provide a stable power source for the various components in the terminal controller 15, and can also provide a stable power source for the image capture device 11.
  • the power module can be an external DC or AC power source, or can be a battery, such as a primary battery or a secondary battery.
  • the communication device may include a wired network interface or the like, that is, it uses a wired transmission mode such as a twisted pair cable, a coaxial cable, or an optical fiber to transmit information; the communication device may also include a Bluetooth module, a wireless network card (ie, a WiFi module), etc. It uses 3G/4G/5G mobile communication network, Bluetooth, Zigbee or WiFi to transmit information.
  • the tank 12 may include a recovery area 120 and a non-recovery area 121.
  • the recycling area 120 is for storing recyclable garbage
  • the non-recycling area 121 is for storing non-recyclable garbage.
  • the classification structure 13 is configured to feed the waste to be sorted into the recovery area 120 or the non-recovery area 121 under the control of the control signal.
  • the sorting structure 13 includes a motor 130 and a shutter 131.
  • the terminal processor can control the rotation direction of the motor 130 according to the control signal, and the motor 130 can drive the shutter 131 to rotate in at least two directions, thereby feeding the garbage to be classified into a designated area (for example, the recovery area 120 and the non-recovery area 121). Complete the classification of the garbage to be classified.
  • the recovery area 120, the non-recycling area 121, and the image collection area 14 have a certain accommodation space.
  • the recovery zone 120, the non-recycling zone 121, and the image acquisition zone 14 may be polyhedrons, cylinders, spheres, and the like.
  • the shape of the recovery zone 120 and the non-recovery zone 121 may be the same or different.
  • FIG. 4 is a schematic block diagram of a garbage sorting and recycling system provided by an embodiment of the present disclosure
  • FIG. 5 is a schematic diagram of a garbage sorting and recycling system provided by an embodiment of the present disclosure.
  • the garbage sorting and recycling system includes the control device 20 and the garbage sorting device 10 described in any of the above embodiments.
  • the control device 20 is provided at the server side (or the cloud side), that is, the control device 20 is a remote control device of the garbage sorting device 10.
  • the control device 20 may be disposed at the garbage sorting device 10.
  • the garbage sorting apparatus 10 may include an image capture device 11, a cabinet 12, a sorting structure 13, an image collection area 14, and a terminal controller 15.
  • the tank 12 may include a recovery area 120 and a non-recovery area 121.
  • Control device 20 may include at least one processor 21 and at least one memory 22. Components such as image capture device 11, terminal controller 15, processor 21, and memory 22 are interconnected by a bus system and/or other form of connection mechanism (not shown). It should be noted that the components and structures of the garbage sorting and recycling system shown in FIG. 4 are merely exemplary and not limiting, and the garbage sorting and recycling system may have other components and structures as needed.
  • the garbage sorting device 10 and the control device 20 can communicate via wired or wireless network signals, that is, through wired or wireless networks.
  • processor 21 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and/or program execution capabilities, such as an image processing unit (GPU), field programmable gate array (FPGA), or tensor A processing unit (TPU) or the like, the processor 21 can control other components in the server side to perform desired functions.
  • the central processing unit (CPU) may be an X86 or ARM architecture or the like.
  • memory 22 can include any combination of one or more computer program products, which can include various forms of computer readable storage media, such as volatile memory and/or nonvolatile memory.
  • Volatile memory can include, for example, random access memory (RAM) and/or cache (cache) and the like.
  • the non-volatile memory may include, for example, a read only memory (ROM), a hard disk, an erasable programmable read only memory (EPROM), a portable compact disk read only memory (CD-ROM), a USB memory, a flash memory, and the like.
  • One or more computer programs can be stored on a computer readable storage medium, and the processor 21 can execute the computer programs to perform various functions.
  • Various applications and various data may also be stored in the computer readable storage medium, such as training parameters of a deep learning neural network, a sample image library, and various data used and/or generated by the application.
  • control device 20 may also include display 23.
  • the display 23 is for displaying a first detected image and/or a second detected image or the like.
  • the display 23 can be, for example, a liquid crystal display, an organic light emitting diode display, or the like.
  • each component in the garbage sorting device 10 can be referred to the relevant part of the embodiment of the garbage sorting device, and details are not described herein again.
  • the computer program can be executed by the processor 21 to perform the steps of: acquiring a detected image; processing the detected image using the deep learning neural network to determine whether the garbage to be classified belongs to the recyclable garbage, and if so, issuing the first control signal to the garbage sorting
  • the device 10 controls the garbage to be sorted into the recovery area 120; if not, sends a second control signal to the garbage sorting device 10 to control the garbage to be sorted into the non-recovery area 121.
  • the detected image includes a plurality of images taken from different angles.
  • the computer program may also be executed by the processor 21 to perform the steps of: processing the detected image using the deep learning neural network and combining the processing results of the images taken at different angles to determine whether the garbage to be classified belongs to the recyclable garbage, and if so, issuing the first a control signal; if not, a second control signal is issued.
  • the computer program may be further executed by the processor 21 to perform the steps of: acquiring an identification number of the garbage sorting device 10 corresponding to the collected detected image; and selecting a deep learning nerve corresponding to the identification number according to the identification number Training parameters of the network.
  • the computer program can also be executed by the processor 21 to perform the steps of: counting the amount of recycling in the case where it is determined that the garbage to be classified belongs to the recyclable garbage; and the number of the predetermined bearing in the recycling area 120 In the case, a recovery control signal is issued to prompt the recycling center 30 to recycle the garbage of the recovery area 120.
  • the garbage sorting device 10 or the control device 20 may be provided with a counter or counting program.
  • the operation mode of the counter or the counting program (the operation of counting and clearing, etc.) can be referred to the description of the relevant part in the embodiment of the garbage sorting and recycling method, and the repetitions are not described herein again.
  • the counter can be an addition counter, a reversible counter, or the like.
  • the end recycling signal may be sent out by the control device 20 to control the counter or the counting program to perform the clearing operation.
  • each waste sorting device 10 can recycle one type of recyclable waste or a plurality of different types of recyclable waste.
  • the recyclable waste includes at least one type of recyclable waste
  • the reclaimed area 120 of each refuse sorting device 10 includes at least one sub-recovery zone, the first control signal including at least one sub-control signal .
  • the computer program can also be executed by the processor 21 to perform the steps of: determining whether the garbage to be classified belongs to one of the at least one type of recyclable garbage, and if so, issuing a corresponding sub-control signal to control The garbage to be classified is sent to the corresponding sub-recovery area; if not, a second control signal is sent to control the garbage to be sorted into the non-recycling area.
  • the computer program may be further executed by the processor 21 to perform the steps of: calculating a matching ratio between the garbage to be classified and the recyclable garbage; determining whether the matching rate exceeds a first preset matching rate threshold, if Determine that the waste to be classified is recyclable; if not, determine that the waste to be classified is not recyclable.
  • the computer program can also be executed by the processor 21 to perform the step of storing the detected image.
  • the computer program may be further executed by the processor 21 to perform the step of determining whether the matching rate exceeds the second preset matching rate threshold if the matching rate is lower than the first preset matching rate threshold. If so, the detected image is stored; if not, the detected image is deleted.
  • the computer program may be further executed by the processor 21 to perform the steps of: determining again whether the garbage to be classified in the stored detected image belongs to recyclable garbage, and if so, adding the stored detection image to the deep learning nerve The sample image library of the network; if not, delete the stored detected image.
  • statistical methods ie, decision theory method
  • syntax recognition method ie, syntax recognition method
  • neural network method ie, neural network method
  • template matching method ie.g., template matching method
  • geometric transformation method may be used to re-determine whether the garbage to be classified in the stored detection image belongs to recyclable garbage.
  • the background user may manually check whether the garbage to be classified in the stored detection image belongs to the recyclable garbage.
  • the stored detected image can be displayed on the display 23 for viewing by the background user.
  • the computer program can also be executed by the processor 21 to perform the steps of: re-training the deep learning neural network with the training image in the sample image library after adding the detected image to the sample image library, and based on the training result Correct the training parameters of the deep learning neural network.
  • the identifier number, the predetermined number of bearers, the deep learning neural network and its training parameters, the matching rate, the first preset matching rate threshold, the second preset matching rate threshold, the sample image library, the sub-recovery area, and the child For a description of the control signal, the type of the recyclable garbage, and the like, reference may be made to the related description in the embodiment of the garbage sorting and recycling method, and the repetitions are not described herein again.
  • Embodiments of the present disclosure also provide a storage medium storing a computer program adapted to be executed by a processor.
  • the storage medium may be applied to the garbage sorting and recycling system described in any of the above embodiments, for example, it may be a memory in a control device in the garbage sorting and recycling system.
  • the computer program can be executed by the processor to perform the steps of: acquiring a detected image; processing the detected image using a deep learning neural network to determine whether the garbage to be classified belongs to recyclable garbage, and if so, issuing a first control signal to control The garbage to be classified is sent to the recycling area; if not, a second control signal is sent to control the garbage to be classified into the non-recycling area.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Sustainable Development (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统。该垃圾分类回收方法包括：获取待分类垃圾的检测图像(S1)；使用深度学习神经网络处理检测图像以判断待分类垃圾是否属于可回收垃圾(S2)，如果是，则发出第一控制信号，以控制将待分类垃圾送入回收区(S3)；如果不是，则发出第二控制信号，以控制将待分类垃圾送入非回收区(S4)。采用深度学习神经网络的方法对待分类垃圾进行实时检测识别，并自动分类回收。

Description

垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统
本申请要求于2017年06月30日递交的中国专利申请第201710558594.1号的优先权，在此全文引用上述中国专利申请公开的内容以作为本申请的一部分。
技术领域
本公开的实施例涉及一种垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统。
背景技术
随着社会的进步，人们的生活水平和质量逐渐提高，能够消费的东西也日益增多，因此产生的垃圾也越来越多，垃圾主要采用填埋和焚烧等办法进行处理。
为了有效地减少垃圾的处理量，减缓对地球资源的消耗，可以对垃圾中可回收的垃圾进行分类回收再利用。目前，可回收垃圾的分类回收方法主要包括人工拣选、风选等，其人工成本高、处理速度慢、效率低。
发明内容
本公开至少一实施例提供一种垃圾分类回收方法，其包括：获取待分类垃圾的检测图像；使用深度学习神经网络处理所述检测图像以判断所述待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号，以控制将所述待分类垃圾送入回收区；如果不是，则发出第二控制信号，以控制将所述待分类垃圾送入非回收区。
例如，在本公开一实施例提供的垃圾分类回收方法中，所述检测图像包括多张从不同角度拍摄的图像，所述垃圾分类回收方法还包括：使用所述深度学习神经网络处理所述检测图像并结合不同角度拍摄的图像的处理结果，以判断所述待分类垃圾是否属于所述可回收垃圾，如果是，则发出所述第一控制信号，以控制将所述待分类垃圾送入所述回收区；如果不是，则发出所述第二控制信号，以控制将所述待分类垃圾送入所述非回收区。
例如，本公开一实施例提供的垃圾分类回收方法还包括：获取与采集所述 检测图像相对应的垃圾分类设备的标识号；根据所述标识号选择与所述标识号相对应的所述深度学习神经网络的训练参数。
例如，本公开一实施例提供的垃圾分类回收方法还包括：在确定所述待分类垃圾属于所述可回收垃圾的情况下，统计回收数量；在所述回收数量超过所述回收区的预定承载数量的情况下，发出回收控制信号，以提示回收中心回收所述回收区的垃圾。
例如，在本公开一实施例提供的垃圾分类回收方法中，所述可回收垃圾包括至少一种类型的可回收垃圾，所述回收区包括至少一个子回收区，所述第一控制信号包括至少一个子控制信号，所述垃圾分类回收方法还包括：判断所述待分类垃圾是否属于所述至少一种类型的可回收垃圾之一，如果是，则发出相应的所述子控制信号，以控制将所述待分类垃圾送入相应的所述子回收区；如果不是，则发出所述第二控制信号，以控制将所述待分类垃圾送入所述非回收区。
例如，在本公开一实施例提供的垃圾分类回收方法中，判断所述待分类垃圾是否属于所述可回收垃圾包括：计算所述待分类垃圾与所述可回收垃圾之间的匹配率；判断所述匹配率是否超过第一预设匹配率阈值，如果是，确定所述待分类垃圾属于所述可回收垃圾；如果不是，确定所述待分类垃圾不属于所述可回收垃圾。
例如，本公开一实施例提供的垃圾分类回收方法还包括：在所述匹配率低于所述第一预设匹配率阈值的情况下，判断所述匹配率是否超过第二预设匹配率阈值，如果是，存储所述检测图像；如果不是，删除所述检测图像，其中，所述第二预设匹配率阈值小于所述第一预设匹配率阈值。
例如，本公开一实施例提供的垃圾分类回收方法还包括：再次判断存储的所述检测图像中的所述待分类垃圾是否属于所述可回收垃圾，如果是，将存储的所述检测图像加入所述深度学习神经网络的样本图像库；如果不是，删除存储的所述检测图像。
例如，本公开一实施例提供的垃圾分类回收方法还包括：在将存储的所述检测图像加入所述样本图像库后，利用所述样本图像库中的训练图像重新训练所述深度学习神经网络，并根据训练结果修正所述深度学习神经网络的训练参数。
例如，在本公开一实施例提供的垃圾分类回收方法中，所述深度学习神经 网络为卷积神经网络。
本公开至少一实施例还提供一种垃圾分类设备，其包括：分类结构、图像采集装置以及终端控制器。所述图像采集装置被配置为采集待分类垃圾的检测图像；所述终端控制器被配置为发送所述检测图像，且还被配置为接收控制信号并根据所述控制信号控制所述分类结构。
例如，在本公开一实施例提供的垃圾分类设备中，所述分类结构包括电机和挡板，所述终端控制器用于根据所述控制信号控制所述电机的转动方向，以驱动所述挡板向至少两个方向转动。
例如，本公开一实施例提供的垃圾分类设备还包括箱体，所述箱体包括回收区和非回收区，所述分类结构被配置为在所述控制信号的控制下将所述待分类垃圾送入所述回收区或所述非回收区。
本公开至少一实施例还提供一种垃圾分类回收系统，其包括：控制装置和上述任一项所述的垃圾分类设备。所述控制装置包括：处理器和存储器，所述存储器存储有适于由所述处理器运行的计算机程序，所述计算机程序由所述处理器运行以执行如下步骤：获取所述检测图像；使用深度学习神经网络处理所述检测图像以判断所述待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号至所述垃圾分类设备，以控制将所述待分类垃圾送入所述回收区；如果不是，则发出第二控制信号至所述垃圾分类设备，以控制将所述待分类垃圾送入所述非回收区。
例如，在本公开一实施例提供的垃圾分类回收系统中，所述检测图像包括多张从不同角度拍摄的图像，在所述计算机程序由所述处理器运行时还执行如下步骤：使用所述深度学习神经网络处理所述检测图像并结合不同角度拍摄的图像的处理结果，以判断所述待分类垃圾是否属于所述可回收垃圾，如果是，则发出所述第一控制信号；如果不是，则发出所述第二控制信号。
例如，在本公开一实施例提供的垃圾分类回收系统中，在所述计算机程序由所述处理器运行时还执行如下步骤：获取与采集所述检测图像对应的所述垃圾分类设备的标识号；根据所述标识号选择与所述标识号相对应的所述深度学习神经网络的训练参数。
例如，在本公开一实施例提供的垃圾分类回收系统中，在所述计算机程序由所述处理器运行时还执行如下步骤：在确定所述待分类垃圾属于所述可回收垃圾的情况下，统计回收数量；在所述回收数量超过所述回收区的预定承载数 量的情况下，发出回收控制信号，以提示回收中心回收所述回收区的垃圾。
例如，在本公开一实施例提供的垃圾分类回收系统中，在所述计算机程序由所述处理器运行时还执行如下步骤：计算所述待分类垃圾与所述可回收垃圾之间的匹配率；判断所述匹配率是否超过第一预设匹配率阈值，如果是，确定所述待分类垃圾属于所述可回收垃圾；如果不是，确定所述待分类垃圾不属于所述可回收垃圾。
例如，在本公开一实施例提供的垃圾分类回收系统中，在所述计算机程序由所述处理器运行时还执行如下步骤：在所述匹配率低于所述第一预设匹配率阈值的情况下，判断所述匹配率是否超过第二预设匹配率阈值，如果是，存储所述检测图像；如果不是，删除所述检测图像，其中，所述第二预设匹配率阈值小于所述第一预设匹配率阈值。
例如，在本公开一实施例提供的垃圾分类回收系统中，在所述计算机程序由所述处理器运行时还执行如下步骤：再次判断存储的所述检测图像中的所述待分类垃圾是否属于所述可回收垃圾，如果是，将存储的所述检测图像加入所述深度学习神经网络的样本图像库；如果不是，删除存储的所述检测图像。
例如，在本公开一实施例提供的垃圾分类回收系统中，在所述计算机程序由所述处理器运行时还执行如下步骤：在将存储的所述检测图像加入所述样本图像库后，利用所述样本图像库中的训练图像重新训练所述深度学习神经网络，并根据训练结果修正所述深度学习神经网络的训练参数。
例如，在本公开一实施例提供的垃圾分类回收系统中，所述控制装置为所述垃圾分类设备的远程控制装置，且所述垃圾分类设备和所述控制装置通过有线或无线网络进行通信。
本公开至少一实施例提供一种垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统，采用深度学习神经网络的方法对待分类垃圾进行实时检测识别，并自动分类回收。
附图说明
为了更清楚地说明本公开实施例的技术方案，下面将对实施例的附图作简单地介绍，显而易见地，下面描述中的附图仅仅涉及本公开的一些实施例，而非对本公开的限制。
图1A为本公开一实施例提供的一种垃圾分类回收方法的一个示例的示意 性流程图；
图1B为本公开一实施例提供的一种垃圾分类回收方法的另一个示例的示意性流程图；
图2为本公开一实施例提供的一种深度学习神经网络的训练过程以及检测过程的示意性流程图；
图3A为本公开一实施例提供的一种垃圾分类设备的示意性框图；
图3B为本公开一实施例提供的一种垃圾分类设备的结构示意图；
图4为本公开一实施例提供的一种垃圾分类回收系统的示意性框图；
图5为本公开一实施例提供的一种垃圾分类回收系统的示意图。
具体实施方式
为了使得本公开实施例的目的、技术方案和优点更加清楚，下面将结合本公开实施例的附图，对本公开实施例的技术方案进行清楚、完整地描述。显然，所描述的实施例是本公开的一部分实施例，而不是全部的实施例。基于所描述的本公开的实施例，本领域普通技术人员在无需创造性劳动的前提下所获得的所有其他实施例，都属于本公开保护的范围。
除非另外定义，本公开使用的技术术语或者科学术语应当为本公开所属领域内具有一般技能的人士所理解的通常意义。本公开中使用的“第一”、“第二”以及类似的词语并不表示任何顺序、数量或者重要性，而只是用来区分不同的组成部分。“包括”或者“包含”等类似的词语意指出现该词前面的元件或者物件涵盖出现在该词后面列举的元件或者物件及其等同，而不排除其他元件或者物件。“连接”或者“相连”等类似的词语并非限定于物理的或者机械的连接，而是可以包括电性的连接，不管是直接的还是间接的。“上”、“下”、“左”、“右”等仅用于表示相对位置关系，当被描述对象的绝对位置改变后，则该相对位置关系也可能相应地改变。为了保持本公开实施例的以下说明清楚且简明，本公开省略了已知功能和已知部件的详细说明。
在生活中，人们每天都会产生大量的生活垃圾，为了对资源进行回收再利用，可以对垃圾中可回收的垃圾进行分类回收。通常，需要将所有垃圾运送到垃圾站，再由垃圾回收中心的工作人员对垃圾进行人工分类回收，因此垃圾分类回收效率低、成本高。
垃圾桶广泛应用于沙滩、车站、图书馆、学校等各种不同的区域，在不同 的区域，垃圾桶接收不同的垃圾，从而根据不同的区域需要对不同类型的可回收垃圾进行分类回收。例如，在图书馆、学校等区域，需要回收废纸等可回收垃圾；而在沙滩、车站等区域，需要回收塑料瓶等可回收垃圾。为了减少垃圾回收中心的工作人员的分类回收工作，可以在垃圾桶处对可回收垃圾进行自动分类，提高垃圾分类回收的效率。
本公开至少一实施例提供一种垃圾分类回收方法，其包括：获取待分类垃圾的检测图像；使用深度学习神经网络处理检测图像以判断待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号，以控制将待分类垃圾送入回收区；如果不是，则发出第二控制信号，以控制将待分类垃圾送入非回收区。该垃圾分类回收方法采用深度学习神经网络的方法对待分类垃圾进行实时检测识别，并自动分类回收，从而提高待分类垃圾的识别准确率，减少垃圾回收处理中心的工作人员的分类回收工作，提高垃圾分类回收的效率，降低垃圾分类回收的成本。
下面对本公开的几个实施例进行详细说明，但是本公开并不限于这些具体的实施例。
本公开实施例提供一种垃圾分类回收方法。图1A示出了本公开实施例提供的垃圾分类回收方法的一个示例的示意性流程图，图1B示出了本公开实施例提供的垃圾分类回收方法的另一个示例的示意性流程图。
例如，如图1A所示，在一个示例中，本公开实施例提供的垃圾分类回收方法可以包括以下操作：
S1：获取待分类垃圾的检测图像；
S2：使用深度学习神经网络处理检测图像以判断待分类垃圾是否属于可回收垃圾；
如果是，执行操作S3：发出第一控制信号，以控制将待分类垃圾送入回收区；
如果不是，执行操作S4：发出第二控制信号，以控制将待分类垃圾送入非回收区。
本公开实施例提供的垃圾分类回收方法采用深度学习神经网络的方法对待分类垃圾进行实时检测识别，并自动分类回收，从而提高待分类垃圾的识别准确率，减少垃圾回收处理中心的工作人员的分类回收工作，提高垃圾分类回收的效率，降低垃圾分类回收的成本。
例如，该垃圾分类回收方法可以在服务器端(或云端)执行。例如，服务器端(或云端)可以发出图像采集信号，并将该图像采集信号传输至垃圾分类设备，垃圾分类设备根据所接收到的图像采集信号采集待分类垃圾的检测图像，并将该检测图像传输至服务器端(或云端)。或者，垃圾分类设备也可以定时向服务器端(或云端)传输待分类垃圾的检测图像。然后服务器端(或云端)使用深度学习神经网络处理所接收的检测图像以判断检测图像中的待分类垃圾是否属于可回收垃圾，根据处理结果，服务器端(或云端)发出控制信号，并将该控制信号传输至垃圾分类设备，以控制垃圾分类设备将待分类垃圾送入指定区域。需要说明的是，服务器端(或云端)可以自动发出控制信号；或者，后台用户也可以人工控制服务器端(或云端)发出控制信号。
本公开实施例提供的垃圾分类回收方法在垃圾分类设备处仅需采集待分类垃圾的检测图像，并将采集到的检测图像传输到服务器端(或云端)，由服务器端(或云端)进行图像识别与检测，因此无需在垃圾分类设备处设置特殊的硬件和复杂的软件，易于垃圾分类设备的维护和普及。
需要说明的是，本公开实施例提供的垃圾分类回收方法也可以在垃圾分类设备端执行，即，垃圾分类设备在采集检测图像后，使用深度学习神经网络处理采集的检测图像以判断待分类垃圾是否属于可回收垃圾。
下面，本公开的实施例以垃圾分类回收方法在服务器端(或云端)执行为例进行说明。本领域的技术人员应当明白，本公开提供的垃圾分类回收方法也可以仅在垃圾分类设备端执行，本公开的实施对此不作限制。
例如，检测图像可以包括多张从不同角度拍摄的图像。如图1B所示，在一个示例中，检测图像可以包括第一检测图像和第二检测图像。图1B所示的垃圾分类回收方法与图1A所示的垃圾分类回收方法相似，不同之处在于：
S100：获取待分类垃圾的第一检测图像和第二检测图像；
S200：使用深度学习神经网络处理第一检测图像和第二检测图像，以判断待分类垃圾是否属于可回收垃圾。
需要说明的是，在图1B所示的示例中，操作S200后的步骤与图1A中操作S2之后的步骤相同，即执行操作S3和操作S4。也就是说，在图1B所示的示例中，在操作S200中，可以使用深度学习神经网络分别处理第一检测图像和第二检测图像，结合第一检测图像的处理结果和第二检测图像的处理结果，以判断待分类垃圾是否属于可回收垃圾。如果是，执行操作S3；如果不是，则 执行操作S4。
例如，在步骤S200中，可以分别判断第一检测图像中的待分类垃圾和第二检测图像中的待分类垃圾是否属于可回收垃圾；或者，也可以结合从第一检测图像和第二检测图像中提取的待分类垃圾的特征信息判断其是否属于可回收垃圾。
例如，第一检测图像和第二检测图像的拍摄角度不同。相比于单独使用一张检测图像进行检测识别，利用拍摄角度不同的多张检测图像对待分类垃圾进行检测识别，可以提取待分类垃圾的更多特征信息，从而提高识别准确率，降低误判率。
需要说明的是，获取第一检测图像的操作和第二检测图像的操作可以并行执行，也可以按顺序分时执行。
例如，深度学习神经网络可以包括卷积神经网络(CNN)、栈式自编码网络、稀疏编码网络、循环神经网络、深度信念网络等神经网络，深度学习神经网络可以采用上述神经网络中的一种或几种的组合对可回收垃圾的训练图像进行训练。
图2为本公开实施例提供的一种深度学习神经网络的训练过程以及检测过程的示意性流程图。例如，如图2所示，训练过程包括以下操作。
S01：获取样本图像库中可回收垃圾的训练图像。训练图像可以包括可回收垃圾在不同拍摄角度下的图像。
S02：对训练图像进行预处理。预处理可以消除训练图像中的无关信息，便于提取训练图像中的可回收垃圾的特征信息，提高特征提取的可靠性。
对训练图像进行预处理后，执行操作S03：随机选取训练图像并进行初始化。对训练图像进行初始化可以将训练图像转换为深度学习神经网络可以处理的数据信号，以便于进行后续操作。
对训练图像进行初始化后，执行操作S04：卷积和抽样。例如，可以对训练图像进行多次卷积和抽样处理，卷积处理可以提取训练图像的特征，抽样处理可以缩减训练数据的规模，减少计算量。
例如，在卷积处理过程中，对训练图像应用多个卷积核，得到多个特征映射图，从而获取可回收垃圾的不同特征。每个特征映射图例如提取可回收垃圾的一种特征。
例如，抽样处理可以采用平均值合并、最大值合并以及随机合并等方法对 训练图像进行处理。例如，在抽样处理过程中，可以对进行卷积处理后得到的特征映射图中的每组像素(例如，四个像素等)进行求和，然后乘以加权值，再加上偏置，其结果通过一个sigmoid函数后，即可得到新的特征映射图。加权值和偏置控制着sigmoid函数的线性程度，如果加权值较小，则sigmoid函数的运算近似于线性运算，抽样过程相当于模糊图像；如果加权值较大，根据偏置的大小，抽样过程可以相当于有噪声的“或”运算或有噪声的“与”运算。
对训练图像进行卷积和抽样处理后，则执行操作S05：全连接。在操作S04中，最后一个抽样层或卷积层连接到一个或多个全连接层。全连接层被配置为将卷积和抽样处理后提取到的可回收垃圾的特征进行综合并输出可回收垃圾的训练参数和特征模型。特征模型为可回收垃圾的一个抽象特征表达。
S06：是否满足结束条件。对全连接层输出的特征模型进行判断，当特征模型满足结束条件时，即，特征模型与预设的标准特征模型相匹配，则执行操作S07：输出特征模型。当特征模型不满足结束条件时，即，特征模型与预设的标准特征模型不匹配，则执行操作S08：反向传播调整权矩阵。在训练过程中，若输出的特征模型与标准特征模型之间存在误差，则通过反向传播将误差信息沿原来的路径反传，从而修正各层(例如，卷积层和抽样层)的训练参数，训练参数例如可以包括加权值和偏置，然后利用修正后的卷积层和抽样层重新对训练图像进行卷积和抽样处理，直到特征模型满足结束条件为止。
需要说明的是，虽然图2中仅示出两次卷积和两次抽样操作，但不限于此，可以对训练图像进行多次卷积和多次抽样。
例如，在操作S2和操作S200中，使用深度学习神经网络处理第一检测图像和/或第二检测图像可以包括以下操作：
获取待分类垃圾的第一检测图像和/或第二检测图像后，执行操作S11：卷积和抽样。例如，可以利用上述训练过程获取的卷积训练参数和抽样训练参数对第一检测图像和/或第二检测图像进行卷积和抽样处理，从而得到第一检测图像和/或第二检测图像中的待分类垃圾的特征。
对训练图像进行多次卷积和抽样处理后，执行操作S12：全连接。全连接层被配置为可以将待分类垃圾的各种特征进行综合并输出待分类垃圾的特征模型。
S13：检测。例如，将待分类垃圾的特性模型与训练得到的可回收垃圾的特征模型进行对比，以判断待分类垃圾是否属于可回收垃圾。
S14：输出检测结果。
然后，根据输出的检测结果执行操作S3或S4。
例如，可以预先利用大量的训练图像针对不同类型的可回收垃圾的进行深度学习训练，从而得到不同类型的可回收垃圾的训练参数和特征模型。例如，每个样本图像库包括同一类型的可回收垃圾的训练图像，训练图像可以包括可回收垃圾在不同角度、不同形态下的图像，以更全面地获取可回收垃圾的特征。例如，训练图像可以包括可回收垃圾的主视图、后视图、仰视图、俯视图、左视图、右视图等基本视图。
例如，深度学习神经网络的样本图像库、训练模型参数等可以以数据库的形式部署在后台服务器端，或者，也可以部署在局域网或广域网(例如云端)的服务器端上以被例如后台服务器端等读取。后台服务器端可以设置在监控室等地方，以进行远程监控。
例如，第一检测图像和/或第二检测图像可以包括一张待分类垃圾的图像，也可以包括多张待分类垃圾的图像。
例如，第一检测图像和/或第二检测图像的数量可以预先设定，也可以在进行垃圾分类回收时，由垃圾分类设备处的控制器或服务器端(或云端)随机生成。例如，可以预先设定第一检测图像仅包括一张待分类垃圾的图像，第二检测图像也仅包括一张待分类垃圾的图像。
例如，检测图像可以为灰度图像，也可以为彩色图像。又例如，检测图像可以是照片，也可以是视频中的一帧、多帧或多帧合成的图像。
例如，可以对检测图像进行预处理，以有利于提取检测图像中的待分类垃圾的特征信息，提高特征提取的可靠性。例如，在检测图像是照片的情况下，预处理可以包括对照片进行缩放、Gamma校正、图像增强或降噪滤波等处理，在检测图像为从视频中采集而获取的情况下，预处理可以包括提取视频的关键帧等。预处理可以在使用深度学习神经网络处理检测图像前进行，即在执行操作S11前进行。
例如，预处理可以在垃圾分类设备处进行，也可以在服务器端(或云端)处进行。
例如，可以预先存储垃圾分类设备中的图像采集区没有放置任何物体时的第一原始图像和/或第二原始图像。例如，第一原始图像的拍摄角度与第一检测图像相同，第二原始图像的拍摄角度与第二检测图像相同。在进行垃圾分类回 收时，通过图像采集装置(例如摄像头)实时监测图像采集区，并定时(例如每间隔10秒、30秒或1分)从图像采集装置获取的视频中采集检测图像，以获取第一图像和/或第二图像，第一图像的拍摄角度与第一检测图像相同，第二图像的拍摄角度与第二检测图像相同；然后将第一图像与第一原始图像进行对比，和/或，将第二图像与第二原始图像进行对比，当第一图像与第一原始图像的相似度比率低于预定的第一相似度阈值，和/或第二图像与第一原始图像的相似度比率低于预定的第二相似度阈值的情况下，确定图像采集区放置有待分类垃圾，并将第一图像作为第一检测图像，将第二图像作为第二检测图像。然后，将获取的第一检测图像和/或第二检测图像传输至服务器端(或云端)，服务器端(或云端)处理第一检测图像和/或第二检测图像，并根据处理结果发送控制信号至垃圾分类设备，以控制垃圾分类设备将待分类垃圾送入指定区域。
例如，可以预先设置第一相似度阈值和/或第二相似度阈值，且第一相似度阈值和第二相似度阈值可以相同，也可以不同。例如，第一相似度阈值和第二相似度阈值可以均为95％。对此不作限制。
需要说明的是，在将第一检测图像和/或第二检测图像传输至服务器端(或云端)后，图像采集装置可以停止采集待分类垃圾的检测图像，当垃圾分类设备接收到服务器端(或云端)传输的控制信号，并将待分类垃圾送入指定区域后，图像采集装置重新定时采集视频中的检测图像。从而可以防止重复采集同一个待分类垃圾的第一检测图像和/或第二检测图像，减少处理时间，提高工作效率。
例如，垃圾分类设备可以设置定时器或定时程序，定时器或定时程序可以定时触发图像采集装置采集待分类垃圾的检测图像。在将检测图像传输至服务器端(或云端)后，定时器或定时程序则停止工作，从而图像采集装置停止采集待分类垃圾的检测图像，当垃圾分类设备接收到服务器端(或云端)传输的控制信号，并将待分类垃圾送入指定区域后，定时器或定时程序进行清零操作并重新计时。
又例如，根据需要，垃圾分类设备也可以设置传感器。传感器用于感测图像采集区是否存在待分类垃圾，如果是，则图像采集装置采集待分类垃圾的检测图像，然后对检测图像进行后续操作；如果不是，则图像采集装置不进行任何操作，以节省功耗。
例如，在一个示例中，该垃圾分类回收方法还包括：获取与采集检测图像 相对应的垃圾分类设备的标识号；根据标识号选择与标识号相对应的深度学习神经网络的训练参数。
例如，训练参数可以包括卷积训练参数、抽样训练参数等，还可以包括特征模型等参数。
例如，可以预先对不同区域的垃圾分类设备设置标识号。基于每个垃圾分类设备的标识号，选择与标识号相对应的深度学习神经网络的训练参数，从而根据不同区域对不同类型的可回收垃圾进行检测并回收。例如，设置在学校的教学楼、图书馆等区域的垃圾分类设备可以用于回收废纸等可回收垃圾；设置在沙滩、车站、篮球场等区域的垃圾分类设备可以用于回收塑料瓶等可回收垃圾。
例如，每个垃圾分类设备可以回收一种类型的可回收垃圾，也可以回收多种不同类型的可回收垃圾。例如，垃圾分类设备的标识号可以与深度学习神经网络的多个不同的训练参数相对应，以实现回收多种不同类型的可回收垃圾。
例如，在一个示例中，可回收垃圾包括至少一种类型的可回收垃圾，相应地，每个垃圾分类设备的回收区包括至少一个子回收区，第一控制信号包括至少一个子控制信号。例如，可回收垃圾可以包括塑料制品(例如，包括塑料瓶等)、纸制品(例如，包括A4纸、书籍等)、金属制品(例如，包括易拉罐等)和玻璃制品等。相应地，回收区可以包括塑料制品子回收区、纸制品子回收区、金属制品子回收区以及玻璃制品子回收区等，第一控制信号也可以包括塑料制品子控制信号、纸制品子控制信号、金属制品子控制信号以及玻璃制品子控制信号等。
在这种情况下，垃圾分类回收方法还可以包括以下操作：判断待分类垃圾是否属于至少一种类型的可回收垃圾之一，如果是，则发出相应的子控制信号，以控制将待分类垃圾送入相应的子回收区；如果不是，则发出第二控制信号，以控制将待分类垃圾送入非回收区。
例如，可回收垃圾包括多种类型的可回收垃圾，判断待分类垃圾是否属于多种类型的可回收垃圾之一可以包括：首先，获取与标识号对应的多个不同的训练参数；然后，利用多个不同的训练参数中的每一个单独处理第一检测图像和/或第二检测图像，从而获得多个第一检测图像的处理结果和/或多个第二检测图像的处理结果；最后，结合多个第一检测图像的处理结果和/或多个第二检测图像的处理结果，判断待分类垃圾是否属于多种类型的可回收垃圾之一。需 要说明的是，与标识号对应的多个不同的训练参数的数量可以与多种类型的可回收垃圾的数量相同。
例如，垃圾分类设备的标识号可以为多种形式，且可以包括不同类型的信息。例如，该标识号可以为识别码(例如字符串)，通过该识别码再到相应的数据库获取训练参数、特征模型、可回收垃圾的种类等一种或多种信息；又例如，该标识信息可以为符合码，例如既包括识别码又包括垃圾分类设备的地理位置(经度、纬度)信息等。
例如，标识号可以集中存储在一个数据库中，且部署在一个或多个服务器上以供查询，本公开的实施例对此不作限制。
例如，在一个示例中，该垃圾分类回收方法还包括以下操作：在确定待分类垃圾属于可回收垃圾的情况下，统计回收数量；在回收数量超过回收区的预定承载数量的情况下，发出回收控制信号，以提示回收中心回收该回收区的垃圾。
例如，预定承载数量可以根据回收区的大小以及可回收垃圾的类型等预先设置。
例如，垃圾分类设备可以设置有计数器或计数程序。当垃圾分类设备接收到第一控制信号并将待回收垃圾送入回收区后，即，在确定待分类垃圾属于可回收垃圾的情况下，计数器或计数程序统计回收数量，在计数器或计数程序所统计的回收数量超过回收区的预定承载数量的情况下，服务器端(或云端)发出回收控制信号，以提示回收中心回收该回收区的垃圾。需要说明的是，当垃圾分类设备接收到第二控制信号并将待回收垃圾送入非回收区后，即，在确定待分类垃圾不属于可回收垃圾的情况下，计数器或计数程序可以统计非回收区的垃圾数量，当非回收区的垃圾数量超过非回收区的预定承载数量的情况下，服务器端(或云端)可以发出垃圾收取控制信号，以提示回收中心或垃圾处理中心收取非回收区的垃圾。
例如，服务器端(或云端)发出回收控制信号后可以对计数器进行清零操作。或者，回收中心的工作人员回收该回收区的垃圾之后，由回收中心的工作人员通过服务器端(或云端)向垃圾分类设备发出结束回收信号，以控制计数器进行清零操作。
需要说明的是，计数器或计数程序也可以设置在服务器端(或云端)。本公开对此不作限制。
例如，回收控制信号可以包括可回收垃圾的种类、垃圾分类设备的位置等信息。
例如，在一个示例中，判断待分类垃圾是否属于可回收垃圾包括以下操作：计算待分类垃圾与可回收垃圾之间的匹配率；判断匹配率是否超过第一预设匹配率阈值，如果是，确定待分类垃圾属于可回收垃圾；如果不是，确定待分类垃圾不属于可回收垃圾。
例如，在确定待分类垃圾属于可回收垃圾的情况下，还可以存储检测图像以作为后续深度学习训练的样本。例如，发出第一控制信号的同时存储检测图像。
例如，匹配率可以为深度学习神经网络输出的检测结果。
例如，第一预设匹配率阈值可以为90％，即，当待分类垃圾与可回收垃圾的匹配率超过90％时，则可以确定待分类垃圾属于可回收垃圾。
例如，在一个示例中，垃圾分类回收方法还包括以下操作：在匹配率低于第一预设匹配率阈值的情况下，判断匹配率是否超过第二预设匹配率阈值，如果是，存储检测图像；如果不是，删除检测图像。
例如，第二预设匹配率阈值小于第一预设匹配率阈值。第二预设匹配率阈值可以为80％，即，当待分类垃圾与可回收垃圾的匹配率超过80％但低于90％时，则可以存储检测图像，以作为后续深度学习训练的样本。
需要说明的是，第一预设匹配率阈值和第二预设匹配率阈值还可以为其他值，只要保证第二预设匹配率阈值小于第一预设匹配率阈值即可。
例如，在一个示例中，垃圾分类回收方法还包括以下操作：再次判断存储的检测图像中的待分类垃圾是否属于可回收垃圾，如果是，将存储的检测图像加入深度学习神经网络的样本图像库；如果不是，删除存储的检测图像。
例如，再次判断存储的检测图像中的待分类垃圾是否属于可回收垃圾可以防止误判，并增加深度学习的训练样本，从而实时动态调整深度学习神经网络的训练参数。
例如，再次判断待分类垃圾是否属于可回收垃圾的方法可以与前次判断的方法不同，例如可以采用统计法(即决策理论法)、句法识别法、神经网络法、模板匹配法或几何变换法中的一种或多种的结合对检测图像进行重新检测和识别，以判断存储的检测图像中的待分类垃圾是否属于可回收垃圾。又例如，还可以由后台用户人工定时查看存储的检测图像中的待分类垃圾是否属于可 回收垃圾，根据用户的输入指令来控制将存储的检测图像加入样本图像库或删除存储的检测图像。
例如，在一个示例中，垃圾分类回收方法还包括以下操作：在将检测图像加入样本图像库后，利用样本图像库中的训练图像重新训练深度学习神经网络，并根据训练结果修正深度学习神经网络的训练参数。该示例提供的垃圾分类回收方法可以及时扩充样本图像库，循环训练样本图像库中的样本图像，修正训练参数，从而进一步提高识别准确率、降低误判率。
本公开实施例还提供一种垃圾分类设备。图3A示出了本公开实施例提供的一种垃圾分类设备的示意性框图，图3B示出了本公开实施例提供的一种垃圾分类设备的结构示意图。
例如，如图3A和图3B所示，垃圾分类设备10可以包括图像采集装置11、箱体12、分类结构13、图像采集区14以及终端控制器15。
例如，图像采集装置11可以包括一个或多个摄像头。终端控制器15可以通过硬件、软件、固件以及它们的任意可行的组合来实现。
例如，图像采集区14被配置为放置待分类垃圾，图像采集装置11被配置为采集待分类垃圾的检测图像。例如，检测图像可以包括多张从不同角度拍摄的图像。在一个示例中，检测图像包括第一检测图像和第二检测图像，且第一检测图像和第二检测图像的拍摄角度不同。如图3B所示，垃圾分类设备10可以包括两个图像采集装置11且分别设置在图像采集区14的顶部和侧面，位于顶部的图像采集装置11可以采集第一检测图像，即第一检测图像为从X方向拍摄的图像；位于侧面的图像采集装置11可以采集第二检测图像，即第二检测图像为从Y方向拍摄的图像，X方向和Y方向可以相互垂直。例如，X方向可以为竖直方向，Y方向可以为水平方向。
例如，图像采集装置11可以为网络摄像机、数字摄像机、彩色半球摄像机、红外摄像机或一体化摄像机等，以对图像采集区14实时摄像，然后定时从图像采集装置11拍摄的视频图像中采集检测图像。又例如，图像采集装置11也可以包括照相机，以对图像采集区14定时拍照，从而采集检测图像。检测图像例如可以存储在图像采集装置11中以根据需要供垃圾分类设备10中的其他组件(例如，终端控制器15等)使用。
例如，在一个示例中，垃圾分类设备10还可以包括定时器或定时程序，从而定时触发图像采集装置11采集待分类垃圾的检测图像。定时器或定时程 序的工作模式可以参见垃圾分类回收方法的实施例中的相关部分，重复之处在此不再赘述。
例如，定时器可以为脉冲型定时器、接通延时型定时器、断开延时型定时器等。又例如，终端控制器15可以存储有定时程序，当需要进行定时操作时，终端控制器15可以直接运行该定时程序，以实现定时功能。
又例如，在一个示例中，垃圾分类设备10还可以包括传感器。传感器用于感测图像采集区14是否存在待分类垃圾，如果是，则图像采集装置11采集待分类垃圾的检测图像，然后对检测图像进行后续操作；如果不是，则图像采集装置11不进行任何操作，以节省功耗。
例如，终端控制器15被配置为发送检测图像至服务器端(或云端)，且还被配置为从服务器端(或云端)接收控制信号并根据控制信号控制分类结构13。如图3B所示，终端控制器15可以设置在图像采集区14的侧面。例如，检测图像可以被传输至服务器端(或云端)，服务器端(或云端)对检测图像进行处理，并根据处理结果生成控制信号，该控制信号例如可以被传输到终端控制器15，终端控制器15根据该控制信号控制控制分类结构13将待分类垃圾送入指定区域。
例如，终端控制器15还被配置为获取垃圾分类设备10的标识号，并发送该标识号至服务器端(或云端)。关于标识号的说明可以参考垃圾分类回收方法的实施例中的相关描述，在此不再赘述。
例如，如图3A所示，终端控制器15可以包括终端处理器、通信装置、电源模块等组件。
例如，终端处理器可以为微控制单元(MCU)等。电源模块可以为终端控制器15中的各个部件提供稳定的电源，还可以为图像采集装置11提供稳定的电源。电源模块可以为外接的直流或交流电源，或者可以为电池，例如一次电池或二次电池。例如，通信装置可以包括有线网络接口等，即其采用双绞线、同轴电缆或光纤等有线传输方式进行信息传输；通信装置也可以包括蓝牙模块、无线网卡(即，WiFi模块)等，即其采用3G/4G/5G移动通信网络、蓝牙、Zigbee或者WiFi等无线传输方式进行信息传输。
例如，箱体12可以包括回收区120和非回收区121。回收区120用于存储可回收垃圾，非回收区121用于存储不可回收垃圾。分类结构13被配置为在控制信号的控制下将待分类垃圾送入回收区120或非回收区121。
例如，如图3A和图3B所示，分类结构13包括电机130和挡板131。终端处理器可以根据控制信号控制电机130的转动方向，电机130可以驱动挡板131向至少两个方向转动，从而将待分类垃圾送入指定区域(例如，回收区120、非回收区121)，完成待分类垃圾的分类。
例如，回收区120、非回收区121和图像采集区14具有一定容纳空间。例如，回收区120、非回收区121和图像采集区14可以为多面体、圆柱体、球体等。回收区120和非回收区121的形状可以相同，也可以不同。
本公开实施例还提供一种垃圾分类回收系统。图4示出了本公开实施例提供的一种垃圾分类回收系统的示意性框图，图5示出了本公开实施例提供的一种垃圾分类回收系统的示意图。
例如，如图4和图5所示，该垃圾分类回收系统包括控制装置20和上述任一实施例所述的垃圾分类设备10。在本实施例中，控制装置20设置在服务器端(或云端)，也就是说，控制装置20为垃圾分类设备10的远程控制装置。但不限于此，控制装置20也可以设置在垃圾分类设备10处。
例如，如图4和图5所示，垃圾分类设备10可以包括图像采集装置11、箱体12、分类结构13、图像采集区14以及终端控制器15。箱体12可以包括回收区120和非回收区121。控制装置20可以包括至少一个处理器21和至少一个存储器22。图像采集装置11、终端控制器15、处理器21和存储器22等组件通过总线系统和/或其它形式的连接机构(未示出)互连。应当注意，图4所示的垃圾分类回收系统的组件和结构只是示例性的，而非限制性的，根据需要，该垃圾分类回收系统也可以具有其他组件和结构。
例如，垃圾分类设备10和控制装置20之间可以通过有线或无线网络信号进行通信，即通过有线或无线网络进行信息传输。
例如，处理器21可以是中央处理单元(CPU)或者具有数据处理能力和/或程序执行能力的其它形式的处理单元，例如图像处理单元(GPU)、现场可编程门阵列(FPGA)或张量处理单元(TPU)等，处理器21可以控制服务器端中的其它组件以执行期望的功能。又例如，中央处理器(CPU)可以为X86或ARM架构等。
例如，存储器22可以包括一个或多个计算机程序产品的任意组合，计算机程序产品可以包括各种形式的计算机可读存储介质，例如易失性存储器和/或非易失性存储器。易失性存储器例如可以包括随机存取存储器(RAM)和/ 或高速缓冲存储器(cache)等。非易失性存储器例如可以包括只读存储器(ROM)、硬盘、可擦除可编程只读存储器(EPROM)、便携式紧致盘只读存储器(CD-ROM)、USB存储器、闪存等。在计算机可读存储介质上可以存储一个或多个计算机程序，处理器21可以运行计算机程序，以实现各种功能。在计算机可读存储介质中还可以存储各种应用程序和各种数据，例如深度学习神经网络的训练参数、样本图像库、以及应用程序使用和/或产生的各种数据等。
例如，控制装置20还可以包括显示器23。显示器23用于显示第一检测图像和/或第二检测图像等。显示器23例如可以为液晶显示器、有机发光二极管显示器等。
需要说明的是，关于垃圾分类设备10中各个组件的说明可以参见垃圾分类设备的实施例的相关部分，在此不再赘述。
例如，计算机程序可以由处理器21运行以执行如下步骤：获取检测图像；使用深度学习神经网络处理检测图像以判断待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号至垃圾分类设备10，以控制将待分类垃圾送入回收区120；如果不是，则发出第二控制信号至垃圾分类设备10，以控制将待分类垃圾送入非回收区121。
例如，在一个示例中，检测图像包括多张从不同角度拍摄的图像。计算机程序还可以由处理器21运行以执行如下步骤：使用深度学习神经网络处理检测图像并结合不同角度拍摄的图像的处理结果，以判断待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号；如果不是，则发出第二控制信号。
例如，在一个示例中，计算机程序还可以由处理器21运行以执行如下步骤：获取与采集检测图像相对应的垃圾分类设备10的标识号；根据标识号选择与标识号相对应的深度学习神经网络的训练参数。
例如，在一个示例中，计算机程序还可以由处理器21运行以执行如下步骤：在确定待分类垃圾属于可回收垃圾的情况下，统计回收数量；在回收数量超过回收区120的预定承载数量的情况下，发出回收控制信号，以提示回收中心30回收该回收区120的垃圾。
例如，在一个示例中，垃圾分类设备10或控制装置20可以设置有计数器或计数程序。计数器或计数程序的工作模式(计数和清零等操作)可以参见垃圾分类回收方法的实施例中的相关部分的说明，重复之处在此不再赘述。
例如，计数器可以为加法计数器、可逆计数器等。
例如，回收中心30的工作人员回收该回收区120的垃圾后，还可以通过控制装置20发出结束回收信号，以控制计数器或计数程序进行清零操作。
例如，每个垃圾分类设备10可以回收一种类型的可回收垃圾，也可以回收多种不同类型的可回收垃圾。
例如，在一个示例中，可回收垃圾包括至少一种类型的可回收垃圾，相应地，每个垃圾分类设备10的回收区120包括至少一个子回收区，第一控制信号包括至少一个子控制信号。在这种情况下，计算机程序还可以由处理器21运行以执行如下步骤：判断待分类垃圾是否属于至少一种类型的可回收垃圾之一，如果是，则发出相应的子控制信号，以控制将待分类垃圾送入相应的子回收区；如果不是，则发出第二控制信号，以控制将待分类垃圾送入非回收区。
例如，在一个示例中，计算机程序还可以由处理器21运行以执行如下步骤：计算待分类垃圾与可回收垃圾之间的匹配率；判断匹配率是否超过第一预设匹配率阈值，如果是，确定待分类垃圾属于可回收垃圾；如果不是，确定待分类垃圾不属于可回收垃圾。
例如，在确定待分类垃圾属于可回收垃圾的情况下，计算机程序还可以由处理器21运行以执行如下步骤：存储检测图像。
例如，在一个示例中，计算机程序还可以由处理器21运行以执行如下步骤：在匹配率低于第一预设匹配率阈值的情况下，判断匹配率是否超过第二预设匹配率阈值，如果是，存储检测图像；如果不是，删除检测图像。
例如，第二预设匹配率阈值小于第一预设匹配率阈值。
例如，在一个示例中，计算机程序还可以由处理器21运行以执行如下步骤：再次判断存储的检测图像中的待分类垃圾是否属于可回收垃圾，如果是，将存储的检测图像加入深度学习神经网络的样本图像库；如果不是，删除存储的检测图像。
例如，可以采用统计法(即决策理论法)、句法识别法、神经网络法、模板匹配法或几何变换法等方法重新判断存储的检测图像中的待分类垃圾是否属于可回收垃圾。或者，还可以由后台用户人工定时查看存储的检测图像中的待分类垃圾是否属于可回收垃圾。例如，可以在显示器23上显示存储的检测图像，以便于后台用户查看。
例如，在一个示例中，计算机程序还可以由处理器21运行以执行如下步骤：在将检测图像加入样本图像库后，利用样本图像库中的训练图像重新训练 深度学习神经网络，并根据训练结果修正深度学习神经网络的训练参数。
需要说明的是，关于标识号、预定承载数量、深度学习神经网络及其训练参数、匹配率、第一预设匹配率阈值、第二预设匹配率阈值、样本图像库、子回收区、子控制信号、可回收垃圾的类型等的说明可以参考垃圾分类回收方法的实施例中的相关描述，重复之处在此不再赘述。
本公开实施例还提供一种存储介质，该存储介质存储有适于由处理器运行的计算机程序。
例如，在本实施例的一个示例中，该存储介质可以应用于上述任一实施例所述的垃圾分类回收系统中，例如，其可以为垃圾分类回收系统中的控制装置中的存储器。
例如，该计算机程序可以由处理器运行以执行如下步骤：获取检测图像；使用深度学习神经网络处理检测图像以判断待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号，以控制将待分类垃圾送入回收区；如果不是，则发出第二控制信号，以控制将待分类垃圾送入非回收区。
例如，关于存储介质的说明可以参考垃圾分类回收系统的实施例中对于存储器的描述，重复之处不再赘述。
对于本公开，还有以下几点需要说明：
(1)本公开实施例附图只涉及到与本公开实施例涉及到的结构，其他结构可参考通常设计。
(2)在不冲突的情况下，本公开的实施例及实施例中的特征可以相互组合以得到新的实施例。
以上所述仅为本公开的具体实施方式，但本公开的保护范围并不局限于此，本公开的保护范围应以所述权利要求的保护范围为准。

Claims (22)

  1. 一种垃圾分类回收方法，包括：
    获取待分类垃圾的检测图像；
    使用深度学习神经网络处理所述检测图像以判断所述待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号，以控制将所述待分类垃圾送入回收区；如果不是，则发出第二控制信号，以控制将所述待分类垃圾送入非回收区。
  2. 根据权利要求1所述的垃圾分类回收方法，其中，所述检测图像包括多张从不同角度拍摄的图像，
    所述垃圾分类回收方法还包括：
    使用所述深度学习神经网络处理所述检测图像并结合不同角度拍摄的图像的处理结果，以判断所述待分类垃圾是否属于所述可回收垃圾，如果是，则发出所述第一控制信号，以控制将所述待分类垃圾送入所述回收区；如果不是，则发出所述第二控制信号，以控制将所述待分类垃圾送入所述非回收区。
  3. 根据权利要求1或2所述的垃圾分类回收方法，还包括：
    获取与采集所述检测图像相对应的垃圾分类设备的标识号；
    根据所述标识号选择与所述标识号相对应的所述深度学习神经网络的训练参数。
  4. 根据权利要求1-3任一项所述的垃圾分类回收方法，还包括：
    在确定所述待分类垃圾属于所述可回收垃圾的情况下，统计回收数量；
    在所述回收数量超过所述回收区的预定承载数量的情况下，发出回收控制信号，以提示回收中心回收所述回收区的垃圾。
  5. 根据权利要求1-4任一项所述的垃圾分类回收方法，其中，所述可回收垃圾包括至少一种类型的可回收垃圾，所述回收区包括至少一个子回收区，所述第一控制信号包括至少一个子控制信号，
    其中，所述垃圾分类回收方法还包括：
    判断所述待分类垃圾是否属于所述至少一种类型的可回收垃圾之一，如果是，则发出相应的所述子控制信号，以控制将所述待分类垃圾送入相应的所述子回收区；如果不是，则发出所述第二控制信号，以控制将所述待分类垃圾送入所述非回收区。
  6. 根据权利要求1-5任一项所述的垃圾分类回收方法，其中，判断所述待分类垃圾是否属于所述可回收垃圾包括：
    计算所述待分类垃圾与所述可回收垃圾之间的匹配率；
    判断所述匹配率是否超过第一预设匹配率阈值，如果是，确定所述待分类垃圾属于所述可回收垃圾；如果不是，确定所述待分类垃圾不属于所述可回收垃圾。
  7. 根据权利要求6所述的垃圾分类回收方法，还包括：
    在所述匹配率低于所述第一预设匹配率阈值的情况下，判断所述匹配率是否超过第二预设匹配率阈值，如果是，存储所述检测图像；如果不是，删除所述检测图像，其中，所述第二预设匹配率阈值小于所述第一预设匹配率阈值。
  8. 根据权利要求7所述的垃圾分类回收方法，还包括：
    再次判断存储的所述检测图像中的所述待分类垃圾是否属于所述可回收垃圾，如果是，将存储的所述检测图像加入所述深度学习神经网络的样本图像库；如果不是，删除存储的所述检测图像。
  9. 根据权利要求8所述的垃圾分类回收方法，还包括：
    在将存储的所述检测图像加入所述样本图像库后，利用所述样本图像库中的训练图像重新训练所述深度学习神经网络，并根据训练结果修正所述深度学习神经网络的训练参数。
  10. 根据权利要求1-9任一项所述的垃圾分类回收方法，其中，所述深度学习神经网络为卷积神经网络。
  11. 一种垃圾分类设备，包括：分类结构、图像采集装置以及终端控制器，其中，
    所述图像采集装置被配置为采集待分类垃圾的检测图像；
    所述终端控制器被配置为发送所述检测图像，且还被配置为接收控制信号并根据所述控制信号控制所述分类结构。
  12. 根据权利要求11所述的垃圾分类设备，其中，所述分类结构包括电机和挡板，
    所述终端控制器用于根据所述控制信号控制所述电机的转动方向，以驱动所述挡板向至少两个方向转动。
  13. 根据权利要求11或12所述的垃圾分类设备，还包括箱体，
    其中，所述箱体包括回收区和非回收区，
    所述分类结构被配置为在所述控制信号的控制下将所述待分类垃圾送入所述回收区或所述非回收区。
  14. 一种垃圾分类回收系统，包括：控制装置和权利要求11-13任一项所述的垃圾分类设备，
    其中，所述控制装置包括：
    处理器和存储器，所述存储器存储有适于由所述处理器运行的计算机程序，所述计算机程序由所述处理器运行以执行如下步骤：
    获取所述检测图像；
    使用深度学习神经网络处理所述检测图像以判断所述待分类垃圾是否属于可回收垃圾，如果是，则发出第一控制信号至所述垃圾分类设备，以控制将所述待分类垃圾送入回收区；如果不是，则发出第二控制信号至所述垃圾分类设备，以控制将所述待分类垃圾送入非回收区。
  15. 根据权利要求14所述的垃圾分类回收系统，其中，所述检测图像包括多张从不同角度拍摄的图像，
    在所述计算机程序由所述处理器运行时还执行如下步骤：
    使用所述深度学习神经网络处理所述检测图像并结合不同角度拍摄的图像的处理结果，以判断所述待分类垃圾是否属于所述可回收垃圾，如果是，则发出所述第一控制信号；如果不是，则发出所述第二控制信号。
  16. 根据权利要求14或15所述的垃圾分类回收系统，其中，在所述计算机程序由所述处理器运行时还执行如下步骤：
    获取与采集所述检测图像对应的所述垃圾分类设备的标识号；
    根据所述标识号选择与所述标识号相对应的所述深度学习神经网络的训练参数。
  17. 根据权利要求14-16任一项所述的垃圾分类回收系统，其中，在所述计算机程序由所述处理器运行时还执行如下步骤：
    在确定所述待分类垃圾属于所述可回收垃圾的情况下，统计回收数量；
    在所述回收数量超过所述回收区的预定承载数量的情况下，发出回收控制信号，以提示回收中心回收所述回收区的垃圾。
  18. 根据权利要求14-17任一项所述的垃圾分类回收系统，其中，在所述计算机程序由所述处理器运行时还执行如下步骤：
    计算所述待分类垃圾与所述可回收垃圾之间的匹配率；
    判断所述匹配率是否超过第一预设匹配率阈值，如果是，确定所述待分类垃圾属于所述可回收垃圾；如果不是，确定所述待分类垃圾不属于所述可回收垃圾。
  19. 根据权利要求18所述的垃圾分类回收系统，其中，在所述计算机程序由所述处理器运行时还执行如下步骤：
    在所述匹配率低于所述第一预设匹配率阈值的情况下，判断所述匹配率是否超过第二预设匹配率阈值，如果是，存储所述检测图像；如果不是，删除所述检测图像，其中，所述第二预设匹配率阈值小于所述第一预设匹配率阈值。
  20. 根据权利要求19所述的垃圾分类回收系统，其中，在所述计算机程序由所述处理器运行时还执行如下步骤：
    再次判断存储的所述检测图像中的所述待分类垃圾是否属于所述可回收垃圾，如果是，将存储的所述检测图像加入所述深度学习神经网络的样本图像库；如果不是，删除存储的所述检测图像。
  21. 根据权利要求20所述的垃圾分类回收系统，其中，在所述计算机程序由所述处理器运行时还执行如下步骤：
    在将存储的所述检测图像加入所述样本图像库后，利用所述样本图像库中的训练图像重新训练所述深度学习神经网络，并根据训练结果修正所述深度学习神经网络的训练参数。
  22. 根据权利要求14-21任一项所述的垃圾分类回收系统，其中，所述控制装置为所述垃圾分类设备的远程控制装置，且所述垃圾分类设备和所述控制装置通过有线或无线网络进行通信。
PCT/CN2018/073809 2017-06-30 2018-01-23 垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统 WO2019000929A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/076,451 US11446706B2 (en) 2017-06-30 2018-01-23 Trash sorting and recycling method, trash sorting device, and trash sorting and recycling system

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710558594.1 2017-06-30
CN201710558594.1A CN109201514B (zh) 2017-06-30 2017-06-30 垃圾分类回收方法、垃圾分类装置以及垃圾分类回收系统

Publications (1)

Publication Number Publication Date
WO2019000929A1 true WO2019000929A1 (zh) 2019-01-03

Family

ID=64741001

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/073809 WO2019000929A1 (zh) 2017-06-30 2018-01-23 垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统

Country Status (3)

Country Link
US (1) US11446706B2 (zh)
CN (1) CN109201514B (zh)
WO (1) WO2019000929A1 (zh)

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110210635A (zh) * 2019-06-05 2019-09-06 周皓冉 一种能识别废弃物的智能分类回收系统
CN110472565A (zh) * 2019-08-14 2019-11-19 六安智华工业自动化技术有限公司 一种基于动作捕捉的垃圾分类能力的判定方法
CN111037554A (zh) * 2019-12-12 2020-04-21 杭州翼兔网络科技有限公司 一种基于机器学习的垃圾清理方法、装置、设备及介质
CN111079548A (zh) * 2019-11-22 2020-04-28 华侨大学 基于目标高度信息和色彩信息的固废在线识别方法
CN111160186A (zh) * 2019-12-20 2020-05-15 上海寒武纪信息科技有限公司 智能垃圾分类处理方法及相关产品
CN111169848A (zh) * 2020-02-27 2020-05-19 东北大学秦皇岛分校 一种智能分类回收垃圾的装置及方法
CN111414926A (zh) * 2019-01-04 2020-07-14 卓望数码技术（深圳）有限公司 一种智能垃圾分类方法、装置、存储介质和机器人
CN111582033A (zh) * 2020-04-07 2020-08-25 苏宁云计算有限公司 垃圾分类的识别方法、系统及计算机可读存储介质
CN111783841A (zh) * 2020-06-09 2020-10-16 中科院成都信息技术股份有限公司 基于迁移学习和模型融合的垃圾分类方法、系统及介质
CN111797872A (zh) * 2019-04-09 2020-10-20 深圳市家家分类科技有限公司 控制方法、电子装置、计算机可读存储介质及降解设备
CN111994513A (zh) * 2020-08-06 2020-11-27 苏州中科先进技术研究院有限公司 智能分类垃圾桶中垃圾种类实时监测方法及装置
CN112364944A (zh) * 2020-12-18 2021-02-12 福州大学 一种基于深度学习的生活垃圾分类方法
CN112651318A (zh) * 2020-12-19 2021-04-13 重庆市信息通信咨询设计院有限公司 基于图像识别的垃圾分类方法、装置和系统
CN113705729A (zh) * 2021-09-27 2021-11-26 中原动力智能机器人有限公司 垃圾分类模型建模方法、垃圾分类方法、装置及介质
CN113899434A (zh) * 2021-09-29 2022-01-07 杭州越歌科技有限公司 一种垃圾称重方法、系统、存储介质及智能终端
CN114821264A (zh) * 2022-03-28 2022-07-29 慧之安信息技术股份有限公司 一种基于神经网络的算法效率提升方法
WO2022217596A1 (zh) * 2021-04-16 2022-10-20 厦门市天诚峰益科技有限公司 一种垃圾处理方法及系统
US11948130B2 (en) 2020-01-30 2024-04-02 BlueOwl, LLC Systems and methods for waste management using recurrent convolution neural network with stereo video input
GB2627923A (en) * 2023-03-06 2024-09-11 Primal Acad Innovations Ltd Waste sorting device

Families Citing this family (41)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109704234A (zh) * 2019-02-25 2019-05-03 齐鲁工业大学 一种医疗垃圾桶识别判断抓取系统及方法
CN109940024A (zh) * 2019-03-22 2019-06-28 深兰科技（上海）有限公司 一种对废弃物进行分类的方法和回收装置
CN110276300B (zh) * 2019-06-24 2021-12-28 北京百度网讯科技有限公司 用于识别垃圾品质的方法和装置
CN110569874A (zh) * 2019-08-05 2019-12-13 深圳大学 一种垃圾分类方法、装置、智能终端及存储介质
CN110653169B (zh) * 2019-08-06 2022-01-14 中国长城科技集团股份有限公司 一种垃圾处理方法、装置及终端设备
CN110440222A (zh) * 2019-08-07 2019-11-12 广州怡港环保科技股份有限公司 一种智慧路灯立杆
CN110550347B (zh) * 2019-08-23 2020-08-04 珠海格力电器股份有限公司 一种垃圾分类回收的方法、用户终端以及垃圾桶
CN110626674A (zh) * 2019-08-26 2019-12-31 江汉大学 垃圾分类装置和垃圾分类方法
CN110436974A (zh) * 2019-09-05 2019-11-12 湖南人文科技学院 一种园林绿化垃圾的无公害处理方法及系统
CN110738131A (zh) * 2019-09-20 2020-01-31 广州游艺云物联网技术有限公司 基于深度学习神经网络的垃圾分类管理方法及装置
CN112570287B (zh) * 2019-09-27 2023-05-30 北京京东尚科信息技术有限公司 一种垃圾分类方法和装置
CN110902210B (zh) * 2019-12-16 2022-04-08 何雨庭 一种小型家用垃圾自动分类机
CN111222949B (zh) * 2020-01-03 2021-06-04 重庆特斯联智慧科技股份有限公司 一种基于深度学习的社区废旧资源共享方法和系统
US11335086B2 (en) * 2020-03-21 2022-05-17 Fidelity Ag, Inc. Methods and electronic devices for automated waste management
CN111498326A (zh) * 2020-04-03 2020-08-07 成都禧来科技有限公司 一种基于二类识别模型的垃圾自动分类机
CN111579524A (zh) * 2020-04-20 2020-08-25 广东轻工职业技术学院 垃圾含量分类的检测系统及其检测方法
CN113751332A (zh) * 2020-06-03 2021-12-07 泰连服务有限公司 视觉检查系统和检查零件的方法
CN113191176A (zh) * 2020-06-08 2021-07-30 阿里巴巴集团控股有限公司 数据处理方法和装置
CN111844042A (zh) * 2020-07-27 2020-10-30 苏州索亚机器人技术有限公司 基于视觉的垃圾分拣机器人
CN112232246A (zh) * 2020-10-22 2021-01-15 深兰人工智能(深圳)有限公司 基于深度学习的垃圾检测分类方法和装置
CN113051989A (zh) * 2020-11-04 2021-06-29 泰州镭昇光电科技有限公司 桶内目标类型鉴定系统
CN112949509A (zh) * 2021-03-08 2021-06-11 三一智造(深圳)有限公司 一种基于人工智能的垃圾分类方法
US20230002150A1 (en) * 2021-07-05 2023-01-05 Carlos Meneses Recycling identification container
CN113743470B (zh) * 2021-08-04 2022-08-23 浙江联运环境工程股份有限公司 自动破袋分类箱基于ai算法垃圾识别精度提升方法
CN113680685A (zh) * 2021-08-24 2021-11-23 安徽自然美环境科技有限公司 一种基于可回收的垃圾分类处理系统
CN113680695A (zh) * 2021-08-24 2021-11-23 武昌工学院 基于机器人的机器视觉垃圾分拣系统
CN113859801B (zh) * 2021-08-24 2023-03-31 青岛海尔科技有限公司 家庭垃圾的处理方法和装置、存储介质及电子装置
CN114067488B (zh) * 2021-11-03 2024-06-11 深圳黑蚂蚁环保科技有限公司 回收系统
CN113731832B (zh) * 2021-11-04 2022-02-15 南京信息工程大学 一种用于垃圾转运站的垃圾分拣处理方法和系统
CN114219010A (zh) * 2021-11-24 2022-03-22 北京华能新锐控制技术有限公司 一种基于先验引导对抗性表征学习的垃圾分类方法
CN114399187A (zh) * 2022-01-07 2022-04-26 西交利物浦大学 基于区块链的建筑垃圾管理方法、装置、设备及存储介质
CN114572579A (zh) * 2022-03-09 2022-06-03 南京南部新城城市物业管理有限公司 一种城市智慧垃圾桶自动识别分类系统
CN114723999B (zh) * 2022-05-24 2022-08-23 航天宏图信息技术股份有限公司 一种基于无人机正射影像的垃圾识别方法及装置
CN115072211A (zh) * 2022-07-30 2022-09-20 青岛理工大学 一种基于机器视觉识别的垃圾自动分类方法
CN115331166B (zh) * 2022-08-12 2023-10-27 广州市净水有限公司 一种格栅清污机智能监控预警方法和系统
CN116430769A (zh) * 2023-03-17 2023-07-14 安徽大学 一种面向大规模管理的智能垃圾分类系统及方法
CN116189099B (zh) * 2023-04-25 2023-10-10 南京华苏科技有限公司 基于改进yolov8的暴露垃圾检测及堆放监控的方法
CN116342895B (zh) * 2023-05-31 2023-08-11 浙江联运知慧科技有限公司 基于ai处理的提升再生资源分拣效率的方法及系统
JP7513167B1 (ja) 2023-07-19 2024-07-09 Ｔｏｐｐａｎホールディングス株式会社 回収判定システム、回収判定装置、回収判定方法、およびプログラム
CN117575543B (zh) * 2024-01-15 2024-04-30 西安卓越软件开发有限公司 一种智慧物业管理方法及装置
CN117923024B (zh) * 2024-03-25 2024-07-09 江西众加利高科技股份有限公司 提升微波检测精度的干湿垃圾分类方法及相关装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103552788A (zh) * 2013-11-07 2014-02-05 邱泽国 集群式垃圾自动识别总成及其自动分类垃圾桶
CN204507883U (zh) * 2015-04-05 2015-07-29 西安航空学院 一种垃圾分类系统
US20160078414A1 (en) * 2013-05-03 2016-03-17 Ecowastehub Corp. Solid waste identification and segregation system
CN105631482A (zh) * 2016-03-03 2016-06-01 中国民航大学 一种基于卷积神经网络模型的危险物品图像分类方法
CN105787506A (zh) * 2016-01-26 2016-07-20 耿春茂 基于图像识别与二维码识别技术的垃圾分类评价方法
CN105772407A (zh) * 2016-01-26 2016-07-20 耿春茂 一种基于图像识别技术的垃圾分类机器人

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7975853B2 (en) * 2006-07-03 2011-07-12 Gregory William Furniss Method and apparatus for sorting small food items for softness
CN102556547A (zh) * 2011-12-15 2012-07-11 沈阳化工大学 一种无线环保监测的垃圾箱
CN104495139B (zh) * 2014-12-31 2017-02-22 东南大学 垃圾投放箱及垃圾分类投放方法
EP3078611A1 (en) * 2015-04-09 2016-10-12 Fomento de Construcciones y Contratas, S.A. Solid waste classification system
CN205023252U (zh) * 2015-09-24 2016-02-10 武汉锦星科技有限公司 一种垃圾桶及基于互联网的垃圾回收管理系统
CN105564864B (zh) * 2015-12-18 2019-01-18 美的集团股份有限公司 垃圾桶、垃圾桶的垃圾分类方法和系统
CN105836340A (zh) * 2016-04-01 2016-08-10 苏州玄禾物联网科技有限公司 一种垃圾桶监控方法
WO2018013903A1 (en) * 2016-07-15 2018-01-18 CleanRobotics, Inc. Automatic sorting of waste
CN106874977A (zh) * 2017-03-22 2017-06-20 北京虹巢环保科技有限公司 垃圾分拣系统和方法
US11135620B2 (en) * 2018-01-12 2021-10-05 Emerging Acquisitions, Llc Autonomous data collection and system control for material recovery facilities
CA3095894A1 (en) * 2018-04-02 2019-10-10 Avangard Innovative, Lp System and method for handling recyclable materials

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160078414A1 (en) * 2013-05-03 2016-03-17 Ecowastehub Corp. Solid waste identification and segregation system
CN103552788A (zh) * 2013-11-07 2014-02-05 邱泽国 集群式垃圾自动识别总成及其自动分类垃圾桶
CN204507883U (zh) * 2015-04-05 2015-07-29 西安航空学院 一种垃圾分类系统
CN105787506A (zh) * 2016-01-26 2016-07-20 耿春茂 基于图像识别与二维码识别技术的垃圾分类评价方法
CN105772407A (zh) * 2016-01-26 2016-07-20 耿春茂 一种基于图像识别技术的垃圾分类机器人
CN105631482A (zh) * 2016-03-03 2016-06-01 中国民航大学 一种基于卷积神经网络模型的危险物品图像分类方法

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111414926A (zh) * 2019-01-04 2020-07-14 卓望数码技术（深圳）有限公司 一种智能垃圾分类方法、装置、存储介质和机器人
CN111797872B (zh) * 2019-04-09 2023-08-01 深圳市家家分类科技有限公司 控制方法、电子装置、计算机可读存储介质及降解设备
CN111797872A (zh) * 2019-04-09 2020-10-20 深圳市家家分类科技有限公司 控制方法、电子装置、计算机可读存储介质及降解设备
CN110210635A (zh) * 2019-06-05 2019-09-06 周皓冉 一种能识别废弃物的智能分类回收系统
CN110472565A (zh) * 2019-08-14 2019-11-19 六安智华工业自动化技术有限公司 一种基于动作捕捉的垃圾分类能力的判定方法
CN110472565B (zh) * 2019-08-14 2023-07-11 广东绿润环境科技有限公司 一种基于动作捕捉的垃圾分类能力的判定方法
CN111079548B (zh) * 2019-11-22 2023-04-07 华侨大学 基于目标高度信息和色彩信息的固废在线识别方法
CN111079548A (zh) * 2019-11-22 2020-04-28 华侨大学 基于目标高度信息和色彩信息的固废在线识别方法
CN111037554A (zh) * 2019-12-12 2020-04-21 杭州翼兔网络科技有限公司 一种基于机器学习的垃圾清理方法、装置、设备及介质
CN111160186A (zh) * 2019-12-20 2020-05-15 上海寒武纪信息科技有限公司 智能垃圾分类处理方法及相关产品
CN111160186B (zh) * 2019-12-20 2023-05-19 上海寒武纪信息科技有限公司 智能垃圾分类处理方法及相关产品
US11948130B2 (en) 2020-01-30 2024-04-02 BlueOwl, LLC Systems and methods for waste management using recurrent convolution neural network with stereo video input
CN111169848A (zh) * 2020-02-27 2020-05-19 东北大学秦皇岛分校 一种智能分类回收垃圾的装置及方法
CN111582033A (zh) * 2020-04-07 2020-08-25 苏宁云计算有限公司 垃圾分类的识别方法、系统及计算机可读存储介质
CN111783841A (zh) * 2020-06-09 2020-10-16 中科院成都信息技术股份有限公司 基于迁移学习和模型融合的垃圾分类方法、系统及介质
CN111783841B (zh) * 2020-06-09 2023-08-04 中科院成都信息技术股份有限公司 基于迁移学习和模型融合的垃圾分类方法、系统及介质
CN111994513A (zh) * 2020-08-06 2020-11-27 苏州中科先进技术研究院有限公司 智能分类垃圾桶中垃圾种类实时监测方法及装置
CN112364944A (zh) * 2020-12-18 2021-02-12 福州大学 一种基于深度学习的生活垃圾分类方法
CN112364944B (zh) * 2020-12-18 2022-07-05 福州大学 一种基于深度学习的生活垃圾分类方法
CN112651318A (zh) * 2020-12-19 2021-04-13 重庆市信息通信咨询设计院有限公司 基于图像识别的垃圾分类方法、装置和系统
WO2022217596A1 (zh) * 2021-04-16 2022-10-20 厦门市天诚峰益科技有限公司 一种垃圾处理方法及系统
CN113705729A (zh) * 2021-09-27 2021-11-26 中原动力智能机器人有限公司 垃圾分类模型建模方法、垃圾分类方法、装置及介质
CN113899434A (zh) * 2021-09-29 2022-01-07 杭州越歌科技有限公司 一种垃圾称重方法、系统、存储介质及智能终端
CN113899434B (zh) * 2021-09-29 2023-11-21 杭州越歌科技有限公司 一种垃圾称重方法、系统、存储介质及智能终端
CN114821264A (zh) * 2022-03-28 2022-07-29 慧之安信息技术股份有限公司 一种基于神经网络的算法效率提升方法
GB2627923A (en) * 2023-03-06 2024-09-11 Primal Acad Innovations Ltd Waste sorting device

Also Published As

Publication number Publication date
CN109201514B (zh) 2019-11-08
US11446706B2 (en) 2022-09-20
CN109201514A (zh) 2019-01-15
US20210178432A1 (en) 2021-06-17

Similar Documents

Publication Publication Date Title
WO2019000929A1 (zh) 垃圾分类回收方法、垃圾分类设备以及垃圾分类回收系统
CN108090458B (zh) 人体跌倒检测方法和装置
CN109684979B (zh) 一种基于图像识别技术的垃圾分类方法、装置及电子设备
CN110210635A (zh) 一种能识别废弃物的智能分类回收系统
CN111178182A (zh) 一种丢垃圾行为的实时检测方法
GB2583253A (en) Method, device and system for adaptive training of machine learning models via detected in-field contextual sensor events and associated located and retrieved
US11922317B2 (en) Learning data generation apparatus, learning model generation system, learning data generation method, and non-transitory storage medium
Gyawali et al. Comparative analysis of multiple deep CNN models for waste classification
CN111985621A (zh) 口罩佩戴实时检测的神经网络模型的搭建方法及实施系统
KR20220156603A (ko) 자동화된 폐기물 관리를 위한 방법 및 전자 장치
CN113673298B (zh) 一种基于时序门控循环单元模型的溢出监测方法
CN111611970A (zh) 一种基于城管监控视频的乱扔垃圾行为检测方法
CN106898173A (zh) 学生作业在线实时审阅系统
CN113657143B (zh) 一种基于分类和检测联合判决的垃圾分类方法
CN104463232A (zh) 一种基于hog特征和颜色直方图特征的密度人群计数的方法
CN110738131A (zh) 基于深度学习神经网络的垃圾分类管理方法及装置
CN112487877B (zh) 一种厨余垃圾规范操作的监控方法、系统、装置及介质
CN113213016A (zh) 一种基于ViT的垃圾识别分类算法及其装置和控制方法
CN113003054A (zh) 垃圾分类方法
CN112183460A (zh) 一种智能识别环境卫生的方法及装置
Hin et al. An Intelligent Smart Bin for Waste Management
CN113673406B (zh) 一种幕墙玻璃爆裂检测方法、系统、电子设备及存储介质
CN111217062A (zh) 一种基于边缘计算和深度学习的垃圾箱垃圾识别方法
KR102321130B1 (ko) 이미지 분석 기반 해양쓰레기 모니터링 시스템 및 이를 이용한 해양쓰레기 모니터링 방법
CN110163081A (zh) 基于ssd的实时区域入侵检测方法、系统及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18825204

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18825204

Country of ref document: EP

Kind code of ref document: A1