CN114067488B

CN114067488B - Recovery system

Info

Publication number: CN114067488B
Application number: CN202111296590.3A
Authority: CN
Inventors: 钟金华
Original assignee: Shenzhen Black Ant Environmental Protection Technology Co ltd
Current assignee: Shenzhen Black Ant Environmental Protection Technology Co ltd
Priority date: 2021-11-03
Filing date: 2021-11-03
Publication date: 2024-06-11
Anticipated expiration: 2041-11-03
Also published as: CN114067488A

Abstract

The invention discloses a recovery system, comprising: the device comprises a box body, a recovery bin arranged below the inside of the box body, a camera arranged in the recovery bin and an article identifier for identifying and classifying recovered articles, wherein a bottom plate is arranged at the bottom end of the inside of the recovery bin, a weight sensor is arranged at the bottom end of the bottom plate and used for weighing the recovered articles on the bottom plate, and the camera is arranged above the bottom plate, is aligned with the recovered articles on the bottom plate and is used for acquiring RGB data and IR data of the recovered articles; a billing system is also included. The scheme automatically performs category-based pricing on the recyclable item. When more than one type of recyclable item is detected, the price is calculated based on the lowest priced item type and the weight of the recyclable item.

Description

Recovery system

Technical Field

The invention relates to the technical field of recovery, in particular to a recovery system.

Background

At present, along with the increasing importance of people on environmental protection, recycling technology is also developed greatly, and various recycling systems such as a clothes recycling box and a mobile phone recycling station are proposed for recycling. An automated recovered article recovery system is required to identify different types of recovered articles for recovery or to provide the customer with a cost of recovered article treatment, respectively, based on the different types of recovered articles.

The chinese patent publication CN112543680a discloses a method of recovering coins from waste material by capturing image data of each piece in a heterogeneous mixture of materials moving through the camera in a stream, wherein the materials in the heterogeneous mixture of materials have various different shapes including one or more different geometries; classifying materials having one or more specified geometries into a first class; classifying materials that do not have the one or more specified geometries into a second classification; and sorting, by an automatic sorting apparatus, materials classified into the first classification from materials classified into the second classification.

The method solves the recovery problem that the existing recovery box or recovery station cannot intelligently identify a certain type of products to a certain extent. However, most of the current consumer product recycling fields are oriented to electronic equipment such as mobile phones and fabrics, so that the current recycling technology needs cannot be met with maximum efficiency. And the electronic equipment is easy to cause secondary damage due to overlarge drop when being put into the box.

Therefore, there is a need for a recycling system that solves the above problems.

Disclosure of Invention

The invention aims to: in order to overcome the defects in the prior art, the invention provides a recovery system which is reasonable in layout and simple to manufacture.

The technical scheme is as follows: a recovery system, comprising:

The front side of the box body is provided with a delivery door and a recovery door in sequence from top to bottom;

the recovery bin is arranged below the box body and is respectively communicated with the delivery door and the recovery door, a bottom plate is arranged at the bottom end of the recovery bin, and a weight sensor is arranged at the bottom end of the bottom plate and used for weighing recovered objects on the bottom plate;

the camera is arranged above the bottom plate, is aligned with the recovered objects on the bottom plate and is used for acquiring RGB data and IR data of the recovered objects;

An article identifier for performing the following recovery article classification identification method:

Normalizing the RGB data and IR data, respectively;

Inputting the normalized RGB data into an RGB feature extraction network to obtain an RGB feature map, wherein the RGB feature map characterizes the outline and texture of the recovered object;

Inputting the normalized IR data into an IR feature extraction network to obtain an IR feature map, wherein the IR feature map characterizes the material quality of the recovered article, and the IR feature map and the RGB feature map are equal in length and width;

Merging the RGB feature map and the IR feature map into a feature dataset and inputting the feature dataset into a classification network to obtain a classification of the recovered articles, wherein the recovered article classification comprises metal waste, textile waste and glass waste and the metal waste classification comprises at least one sub-classification of iron, aluminum, copper;

the system also comprises a charging system for generating a payment amount according to the classification and the weight of the recovered articles and paying.

Further, the system also comprises a notification system for notifying a cleaning person to recycle the recycled objects according to the fact that the total weight of the recycled objects exceeds a preset value.

Further, the RGB feature extraction network and the IR feature network respectively include a first convolution network, the first convolution network shares parameters with a depth residual network, and the training method of the RGB feature extraction network and the IR feature network specifically includes:

Extracting a first convolution network: training the depth residual error network through a training set; a first convolutional network of parameters discarded by the depth residual network; scoring the first convolutional network by a test set;

And circularly executing the first convolution network extraction on the depth residual error network to obtain a plurality of groups of first convolution networks and scores corresponding to the first convolution networks.

And respectively designating the highest scores in a plurality of groups of first convolution networks as the convolution layers in the RGB feature extraction network or the IR feature network.

Further, the delivery door is arranged to be rotatable open only in the direction of the inside of the recovery compartment.

Further, a first electronic lock used for locking the delivery door is arranged on the delivery door, and a second electronic lock used for locking the recovery door is arranged on the recovery door.

Further, a mobile phone recycling bin for recycling electronic equipment is further arranged in the box body, a mobile phone bin door for opening or closing the mobile phone recycling bin is arranged on the box body, a third electronic lock is arranged on the mobile phone bin door, and a detection device is arranged in the mobile phone recycling bin and used for detecting whether the electronic equipment exists in the mobile phone recycling bin or not and notifying cleaning personnel to recycle through the notification system.

Further, the detection device is a pressure sensor or a data line.

Further, a printer is further arranged on one side of the recycling bin in the box body.

Further, the top is equipped with L type support in the box L type support bottom is equipped with the camera, including IR camera and RGB camera, just the camera is close to retrieve the door setting.

Further, a lamp box is arranged at the top of the box body.

The beneficial effects are that: in the recycling system, a camera is arranged on a traditional recycling box body, RGB data and IR data are respectively acquired for recyclable objects, and an RGB feature map and an IR feature map are respectively extracted, wherein the RGB feature map and the IR feature map reflect the image features of one recyclable object. The RGB feature map contains rich information, can embody the outline and texture of the recyclable item, and is beneficial to identifying specific items of the recyclable item. The IR characteristic map mainly reflects the material property of the recyclable item through reflection, and the IR data is obtained through infrared rays, so that the IR characteristic map can reflect the material property of the recyclable item under various illumination environments. Then, the weight of the recyclable material is calculated by classification, and the recycled material can be converted into money to be paid to personnel delivering the recycled material by a charging system, so that the recycling enthusiasm is improved. The scheme is that the recyclable article is weighed through a weighing table, then corresponding unit price is carried out according to the detected type of the recyclable article, and the recyclable article is priced through a charging system and is output. The scheme automatically performs category-based pricing on the recyclable item. When more than one type of recyclable item is detected, the price is calculated based on the lowest priced item type and the weight of the recyclable item.

Drawings

FIG. 1 is a schematic perspective view of one embodiment of a recovery system of the present application;

FIG. 2 is a schematic cross-sectional plan view of the recovery system of FIG. 1;

FIG. 3 is a flow chart of article sorting by the article sorter of the recycling system of FIG. 1;

FIG. 4 is a flow chart of a training method of the step RGB feature extraction network and IR feature network of FIG. 3;

FIG. 5 is an additional flow chart of the item classifier classification process of FIG. 3;

fig. 6 is a schematic diagram of the embodiment shown in fig. 1.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

It should be noted that the descriptions of "first," "second," "left," "right," etc. in this disclosure are for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions of the embodiments may be combined with each other, but it is necessary to base that the technical solutions can be realized by those skilled in the art, and when the technical solutions are contradictory or cannot be realized, the combination of the technical solutions should be considered to be absent and not within the scope of protection claimed in the present invention.

Referring to fig. 1-6, one embodiment of the recycling system of the present invention includes a housing 1, a recycling bin 2, a camera 3, an item identifier, and a billing system. It should be noted that this embodiment takes textile recycling as an example, but is not limited to recycling materials such as clothing fabrics. The front side of the box body 1 is sequentially provided with a delivery door 21 and a recovery door 22 from top to bottom, the delivery door 21 and the recovery door 22 are arranged up and down, and a user can conveniently deliver recovered articles, so that the user is prevented from squatting or bending down, and the user experience is improved.

The recovery bin 2 is arranged below the inside of the box body 1 and is respectively communicated with the delivery door 21 and the recovery door 22, a bottom plate 23 is arranged at the bottom end in the recovery bin 2, and a weight sensor 24 is arranged at the bottom end of the bottom plate 23 and used for weighing recovery articles on the bottom plate 23.

The camera 3 is arranged above the bottom plate 23 in the recycling bin 2, is aligned with the recycling object on the bottom plate 23, and is used for acquiring RGB data and IR data of the recycling object. Specifically, IR cameras and RGB cameras are included.

The article identifier, see fig. 3, is used to perform the following recovery article classification identification method:

step S100: acquiring RGB data and IR data of a recyclable item, and normalizing the RGB data and IR data, respectively;

step S200: inputting the normalized RGB data into an RGB feature extraction network to obtain an RGB feature map, wherein the RGB feature map characterizes the outline and texture of the recyclable item;

Step S300: inputting the normalized IR data into an IR feature extraction network to obtain an IR feature map, wherein the IR feature map characterizes the material of the recyclable item, the IR feature map and the RGB feature map being equal in length and width;

Step S400: the RGB feature map and the IR feature map are combined into a feature dataset and the feature dataset is input into a classification network to obtain a classification of the recyclable item, wherein the recyclable item classification comprises metal waste, textile waste, and glass waste and the metal waste classification comprises at least one sub-classification of iron, aluminum, copper. In one embodiment, the three sub-classifications of iron, aluminum and copper can be provided at the same time, and the two sub-classifications of iron, aluminum or iron and copper can be provided.

The charging system is used for generating a payment amount according to the classification and the weight of the recovered articles and paying.

Preferably, the object identifier is a hardware or software processor with the above functions, and in this embodiment, may be specifically an android-based management program APP. In other embodiments, it may be divided into the following modules:

A preprocessing module 100 for normalizing the RGB data and the IR data, respectively;

an RGB extraction module 200 for inputting the normalized RGB data into an RGB feature extraction network to obtain an RGB feature map, wherein the RGB feature map characterizes the outline and texture of the recyclable item;

An IR extraction module 300 for inputting normalized IR data into an IR feature extraction network to obtain an IR feature map, wherein the IR feature map characterizes the material of the recyclable item, and the IR feature map and the RGB feature map are equal in length and width;

a classification module 400 for merging the RGB feature map and the IR feature map into feature data sets and inputting the feature data sets into a classification network to obtain a classification of the recyclable item, wherein the classification of the recyclable item comprises metal waste, textile waste, and glass waste, and the classification of the metal waste comprises iron, aluminum, copper sub-classification.

In the recycling system, a camera 3 is arranged on a traditional recycling box body 1, RGB data and IR data are respectively acquired for recyclable objects, and an RGB feature map and an IR feature map are respectively extracted, wherein the RGB feature map and the IR feature map reflect the image features of one recyclable object. The RGB feature map contains rich information, can embody the outline and texture of the recyclable item, and is beneficial to identifying specific items of the recyclable item. The IR characteristic map mainly reflects the material property of the recyclable item through reflection, and the IR data is obtained through infrared rays, so that the IR characteristic map can reflect the material property of the recyclable item under various illumination environments. Then, the weight of the recyclable material is calculated by classification, and the recycled material can be converted into money to be paid to personnel delivering the recycled material by a charging system, so that the recycling enthusiasm is improved. The scheme automatically performs category-based pricing on the recyclable item. When more than one type of recyclable item is detected, the price is calculated based on the lowest priced item type and the weight of the recyclable item.

Specifically, the top is equipped with L type support 31 in the box 1L type support 31 bottom is equipped with the camera 3, just the camera 3 is close to delivery door 21 sets up, and two kinds of cameras 3 of IR and RGB are integrated simultaneously to the camera 3. Meanwhile, as the camera 3 is close to the delivery door 21, the identification can be performed at the moment that the recovered articles are input by the delivery door 21, the identification error caused by rolling and other reasons after the recovered articles fall into the recovery bin 2 is prevented, and the identification efficiency and accuracy are improved.

Further, the system also comprises a notification system for notifying a cleaning person to recycle the recycled objects according to the fact that the total weight of the recycled objects exceeds a preset value. Preferably, the notification can be performed by using instant messaging tool software such as WeChat, payment treasures and the like, and also can be performed by using short messages, telephone notification and the like.

Further referring to fig. 3, the RGB feature extraction network and the IR feature network respectively include a first convolution network, where the first convolution network shares parameters with a depth residual network, and the training method of the RGB feature extraction network and the IR feature network specifically includes:

step S11: extracting a first convolution network: training the depth residual error network through a training set; a first convolutional network of parameters discarded by the depth residual network; scoring the first convolutional network by a test set;

step S12: and circularly executing the first convolution network extraction on the depth residual error network to obtain a plurality of groups of first convolution networks and scores corresponding to the first convolution networks.

Step S13: and respectively designating the highest scoring of a plurality of groups of first convolution networks as the convolution layers in the rgb feature extraction network or the IR feature network.

The RGB feature extraction network and the IR feature extraction network are respectively trained, the steps required by training are the same, the required training set and parameter adjustment are set according to the requirements of the RGB feature extraction network or the IR feature extraction network. Taking RGB feature extraction network as an example, the method trains a depth residual error network through a training set, in the training process, in order to prevent gradient disappearance, a part of parameters of the depth residual error network are randomly discarded, and a first convolution network with smaller parameter scale is formed by combining the discarded part of parameters with a feature data set.

Along with the training, the first convolution network is extracted once in each round of training, and the discarded parameters of the depth residual error network are random, so that the discarded parameters are trained as well, the overall performance of the obtained first convolution network is improved, and a plurality of first convolution networks generated after a plurality of times of training are scored.

For the extracted plurality of first convolution networks, one of the highest scores is extracted as a portion of convolution in the RGB feature extraction network, and similarly the portion of convolution in the IR feature extraction network is also determined by extraction in the same manner. The scheme can ensure the feature extraction precision, reduce the sizes of the RGB feature extraction network and the IR feature extraction network in response on the basis of ensuring the classification precision, and reduce the parameter quantity of the RGB feature extraction network and the IR feature extraction network so as to reduce the calculation cost in the feature extraction process. And the identification efficiency of recyclable articles is improved.

In one embodiment, the feature extraction accuracy of the first network obtained after several training is 87.4%,88.6%,92.7%,94.1%,96.2%,97.5%, respectively, while the feature extraction accuracy of the residual network itself after final training is 98.7%, and the first network with the highest accuracy is visible, and the feature extraction accuracy is already very close to the level of the residual network, and the calculation cost is greatly reduced.

Further, in the depth residual network, the scaling parameter γ in the last BN layer in each residual block is set to 0 to output an all 0 vector.

The parameter gamma is set to 0 to output the all 0 vectors, so that the residual block outputs the all 0 vectors before residual connection, the scale of a residual network output matrix is reduced, and the training efficiency of the first matrix is improved. The scheme can accelerate the speed of deploying the recyclable object identification method in any environment.

Further, the learning rate of the depth residual network is set to exponentially decays, and the lowest learning rate of the depth residual network is set to five percent of the initial learning rate.

According to the scheme, the learning rate of the depth residual error network is kept, the overfitting in the first network training process is prevented, and the richness of the recyclable object features extracted by the first network is improved, so that the accuracy of the recyclable object classification is improved.

Further, in the training process of the depth residual network through the training set, when a parameter in a convolution kernel in the depth residual network is smaller than a preset value, the parameter is set to 0, so that each convolution kernel forms a sparse matrix.

Specifically, the scheme can reduce the number of parameters in the depth residual error network and the first convolution network, for some parameters with values close to 0, obvious characteristics cannot be provided, computer resources are occupied, the calculated amount of the neural network can be greatly reduced by setting the parameters to zero, meanwhile, the characteristics of recyclable articles can be prevented from being lost, and the scheme improves the identification efficiency of the recyclable articles.

Further, the non-zero parameter set in the sparse matrix forms a block.

According to the scheme, the feature extraction can be concentrated in the image, and the recyclable object can provide enough feature positions.

In some embodiments, the delivery door 21 is arranged to be rotatable open only towards the inside of the recovery compartment 2. In particular, this can be achieved by providing a unidirectional spindle, torsion spring, on the inside edge of the delivery door 21 to prevent someone from stealing the fabric in the stowage bin 2 during delivery.

The delivery door 21 is provided with a first electronic lock for locking the delivery door 21, and the recovery door 22 is provided with a second electronic lock for locking the recovery door 22. To prevent someone from stealing the recovered items in the recovery bin 2 in an unused state.

As a further optimization for this embodiment, the box 1 is further provided with a mobile phone recycling bin 4 for recycling electronic devices, the box 1 is provided with a mobile phone bin door 41 for opening or closing the mobile phone recycling bin 4, the mobile phone bin door 41 is provided with a third electronic lock, and the mobile phone recycling bin 4 is internally provided with a detection device for detecting whether the mobile phone recycling bin 4 is provided with electronic devices or not and notifying cleaning personnel of recycling through the notification system. The third electronic lock can prevent someone from stealing the electronic equipment in the mobile phone recycling bin 4.

Specifically, the detection device is a pressure sensor or a data line. The pressure sensor detects the pressure at the bottom of the mobile phone recycling bin 4, so that whether the mobile phone recycling bin 4 contains electronic equipment to be recycled is judged. The data line can be connected manually by a user, so that the electronic equipment can be charged or identified, and the state of the electronic equipment to be recovered in the mobile phone recovery bin 4 can be detected.

In another embodiment, a printer is further disposed on one side of the recycling bin 2 in the case 1, and is used for connecting with a cloud terminal, and printing a cloud terminal photograph or document to improve the entertainment function of the recycling system.

In another embodiment, the top of the box 1 is provided with a lamp box 5, which can broadcast advertisements or other information to improve the entertainment function of the recycling system.

The foregoing is only a preferred embodiment of the invention, it being noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present invention, and such modifications and adaptations are intended to be comprehended within the scope of the invention.

Claims

1. A recycling system, comprising:

Normalizing the RGB data and IR data, respectively;

the RGB feature extraction network and the IR feature network respectively comprise a first convolution network, the first convolution network shares parameters with a depth residual error network, and the training method of the RGB feature extraction network and the IR feature network specifically comprises the following steps:

circularly executing extraction of a first convolution network on the depth residual error network to obtain a plurality of groups of first convolution networks and scores corresponding to the first convolution networks;

respectively designating the highest scores in a plurality of groups of first convolution networks as convolution layers in an RGB feature extraction network or an IR feature network;

2. The recovery system of claim 1, wherein: the system also comprises a notification system for notifying a cleaning person to recycle the recycled objects according to the fact that the total weight of the recycled objects exceeds a preset value.

3. The recovery system of claim 1, wherein: the delivery door is arranged to be rotatable open only in the direction of the inside of the recycling bin.

4. A recovery system as claimed in claim 3, wherein: the delivery door is provided with a first electronic lock used for locking the delivery door, and the recovery door is provided with a second electronic lock used for locking the recovery door.

5. The recovery system of claim 2, wherein: still be equipped with the cell-phone recovery bin that is used for retrieving electronic equipment in the box, be equipped with the cell-phone door that is used for opening or close the cell-phone recovery bin on the box, be equipped with the third electronic lock on the cell-phone door, just be equipped with detection device in the cell-phone recovery bin for detect whether there is electronic equipment in the cell-phone recovery bin and warp notify the system and inform the clearance personnel to retrieve.

6. The recovery system of claim 5, wherein: the detection device is a pressure sensor or a data line.

7. The recovery system of claim 1, wherein: and a printer is further arranged on one side of the recycling bin in the box body.

8. The recovery system of claim 1, wherein: the top is equipped with L type support in the box L type support bottom is equipped with the camera, including IR camera and RGB camera, just the camera is close to retrieve the door setting.

9. The recovery system of claim 1, wherein: the top of the box body is provided with a lamp box.