CN113731832A - Garbage sorting treatment method and system for garbage transfer station - Google Patents
Garbage sorting treatment method and system for garbage transfer station Download PDFInfo
- Publication number
- CN113731832A CN113731832A CN202111297530.3A CN202111297530A CN113731832A CN 113731832 A CN113731832 A CN 113731832A CN 202111297530 A CN202111297530 A CN 202111297530A CN 113731832 A CN113731832 A CN 113731832A
- Authority
- CN
- China
- Prior art keywords
- garbage
- convolution
- layer
- module
- stacked
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
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
-
- 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
- B07C5/02—Measures preceding sorting, e.g. arranging articles in a stream orientating
-
- 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
-
- 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
- B07C5/38—Collecting or arranging articles in groups
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Sorting Of Articles (AREA)
Abstract
The invention discloses a garbage sorting treatment method for a garbage transfer station, which comprises the following steps: two layers of conveying belts with height difference are arranged at the outlet of the drum screen, garbage with the diameter larger than 80 mm left by the drum screen is conveyed to the conveying belts, and all garbage on the designated area of the conveying belts is shot; classifying the garbage by adopting a garbage recognition module arranged on an embedded processing terminal; and a corresponding program control servo motor is arranged, a mechanical arm beside the conveying belt is rotated, and the mechanical arm is used for grabbing the garbage and conveying the garbage to a corresponding garbage classification groove. The invention can accept the image input of any size, the lightweight network is suitable for the rapid garbage sorting treatment of the garbage transfer station, can be deployed on an embedded treatment terminal, simultaneously considers the requirements of the precision and the efficiency of the integral sorting treatment of the stacked garbage, and effectively overcomes the defect of sorting treatment of recyclable matters in the garbage transfer station.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a garbage sorting processing method and system for a garbage transfer station.
Background
With the rapid development of social economy, the environmental pollution problem is increasingly serious. The classification management of domestic garbage is implemented in various big cities, and the garbage classification 'mandatory age' is gradually entered. In 2020, the key period for falling on the garbage classification policy in China is provided, and the policy drives the development of the garbage classification industry. In China, when garbage is classified at a garbage transfer station, garbage trucks pour garbage into a stock dump, workers on two sides can sort out large garbage through a conveyor belt, and the rest garbage can enter a drum screen to be sorted for the first time; kitchen waste and other waste are all leaked from the trommel. The garbage can enter a vibrating screen for secondary sorting, organic matters are sent to a composting plant, and other garbage is sent to a landfill plant; most of the garbage left by the drum screen is high-recovery-value garbage with the diameter larger than 80 mm, the small plants are directly sent to an incineration plant, and the large plants can separate out metals and paper for recovery through magnetic separation and air separation. Therefore, the garbage sorting is carried out on the garbage image with high recovery value which can be effectively utilized and is left by the drum screen, and the garbage sorting device has important practical significance for environmental protection work.
In the past few years, researchers have also done a lot of work in classifying spam images under the development of deep learning algorithms. The trashent dataset was created by Yang and Thung of stanford university, which consists of 2527 images, and contains RGB images of six types of spam, where each image contains only one type of spam. Respectively glass, paper, cardboard, plastic, metal and common garbage. The GINI dataset contains 2561 images of spam, of which 956 images were obtained by querying the internet for spam related keywords such as "roadside spam", "market spam", etc. TACO is a data set for garbage classification and detection, including 1500 images and 4784 annotations, on which garbage classification can be performed, although the data set is very small. The aquatrace dataset was developed by comparing the TrashNet dataset and the TACO dataset and found to have some disadvantages, the TrashNet dataset was not annotated, the TACO dataset was not annotated reliably, etc., thus proposing an aquatrace dataset containing 369 images from 4 different categories associated with various waste items. Researchers adopt architectures such as InceptionResnetV2, MobileNet, VGG16 and the like to train, and good effects are achieved. However, the quantity of images of the garbage data sets adopted by the users is small, each image contains a small quantity of garbage objects, the characteristic information is single, the images are not suitable for being used as the data sets of the garbage images with rich identifying characteristics of the garbage transfer stations, and the garbage sorting business requirements of the garbage transfer stations cannot be met no matter from actual operation or final sorting efficiency.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a garbage sorting processing method and a garbage sorting processing system for a garbage transfer station, which can accept image input of any size, are suitable for the rapid garbage sorting processing of the garbage transfer station through a lightweight network, can be deployed on an embedded processing terminal, simultaneously meet the requirements of precision and efficiency during the integral sorting processing of stacked garbage, and effectively overcome the defect of sorting processing of recyclable objects in the garbage transfer station.
In order to achieve the purpose, the invention adopts the following technical scheme:
in a first aspect, an embodiment of the present invention provides a garbage sorting processing method for a garbage transfer station, where the garbage sorting processing method includes the following steps:
s1, installing two layers of conveyor belts with height difference at the outlet of the drum screen, putting the garbage with the diameter larger than 80 mm left by the drum screen into the first layer of conveyor belt, and shooting all garbage on the designated area of the conveyor belt when the garbage is conveyed into the second layer of conveyor belt; sending the shot stacked garbage image to a garbage recognition module installed on the embedded processing terminal to classify the garbage;
s2, sending the garbage classification result into an embedded processor ARM, setting a corresponding program to control a servo motor, rotating a mechanical ARM beside a conveyor belt, and grabbing garbage by a mechanical ARM to send the garbage to a corresponding garbage classification groove;
the garbage identification module is obtained by training a certain amount of similar stacked garbage sample images and comprises a self-adaptive pooling layer, a feature extraction unit, a feature fusion unit, a self-adaptive maximum pooling layer, a global average pooling layer and a full-connection layer which are connected in sequence; the feature extraction unit extracts features based on a lightweight mode, comprises five convolution modules, and adopts a depth separable convolution mode to carry out deep feature extraction on the multi-view feature map with the adjusted size in the self-adaptive pooling layer; wherein, a compression and excitation module is added in the first convolution module, the second convolution module and the third convolution module; the feature fusion unit is used for extracting the outputs of the fourth convolution module and the fifth convolution module and carrying out feature fusion of different sizes; the fused features are subjected to self-adaptive maximum pooling and global average pooling respectively through a self-adaptive maximum pooling layer and a global average pooling layer, and then global features of the stacked garbage images are extracted; and the full connection layer is used for classifying the extracted global features of the stacked garbage images, calculating the probability of the stacked garbage images belonging to each category, and outputting the category information and the corresponding probability of the current stacked garbage images.
Further, the types of the stacked garbage sample images at least comprise four categories of preprocessed kitchen garbage, recyclable matters, other garbage and harmful garbage, and the garbage quantity on each stacked garbage sample image is not less than a preset garbage quantity threshold value;
the pretreatment process comprises the following steps: and enhancing the stacked garbage sample images of each batch, and expanding the data set by adopting various methods of scaling, rotating, transforming and turning.
Further, the preset garbage amount threshold is 3, that is, the amount of garbage on each stacked garbage sample image is not less than 3.
Further, the feature extraction unit comprises five convolution modules, and each convolution module adopts a DepthWise convolution layer and a PointWise convolution layer which are alternately connected: the first convolution module is provided with a set of alternately connected DepthWise convolution layer and PointWise convolution layer; the second convolution module is provided with two groups of DepthWise convolution layers and Pointwise convolution layers which are alternately connected; the third, the fourth and the fifth convolution modules respectively have three groups of alternately connected DepthWise convolution layers and PointWise convolution layers; a BN layer is added after each convolution layer; a zeroPadding layer is added after the convolution layer of each group of convolution modules;
a compression and excitation module in an attention mechanism is added after the zeroPadding layers of the first convolution module, the second convolution module and the third convolution module;
the number of convolution kernels of each convolution module is increased along with the deepening of the network layer number, and the convolution step length is changed to reduce the characteristic diagram.
Further, the fusion process of the feature fusion unit includes the following steps:
extracting 29 × 29 × 512 first output features output by the fourth convolution module and 14 × 14 × 512 second output features output by the fifth convolution module respectively from the fusion features;
for the first output characteristic of the fourth convolution module, performing convolution with convolution kernel size of 3 × 3, kernel number of 256, step length of 1 and filling; then carrying out batch normalization processing, and activating by using a Relu function to obtain a first characteristic value with the output of 28 multiplied by 256; the output first characteristic value reflects that the receptive field of the garbage image is moderate, the representation capability of the geometric detail information of the garbage is strong, and the extraction capability of the medium-sized target in the garbage image is better. Although the resolution is high, the semantic information characterization capability is weak.
For the output characteristics of the fifth convolution module, performing convolution with convolution kernel size of 1 × 1, kernel number of 512, step length of 1 and filling; then, carrying out batch normalization processing, activating by using a Relu function, and scaling the depth features to the same scale through upsampling to obtain a second feature value with the output of 28 x 512; the output second characteristic value reflects that the receptive field of the garbage image is large, the semantic information representation capability of the garbage is strong, the extraction capability of large targets in the garbage image is better, but the resolution of the characteristic image is low, and the representation capability of the geometric information is weak.
And adjusting the first characteristic value and the second characteristic value output by the fourth convolution module and the fifth convolution module to be consistent in size, carrying out Concatenate fusion to form a characteristic diagram of 28 multiplied by 768, fusing the characteristics of different layers in the network, and obtaining the global characteristic of the stacked garbage image.
Further, the training process of the garbage recognition model comprises the following steps:
when training is started, the preprocessed image passes through the self-adaptive pooling layer, so that the resolution of the input image received by the garbage recognition model is kept to be 224 multiplied by 3;
let the initial learning rate be 1 × 10-3When 7 epoch learning rates are iterated and are not reduced, the learning rate is automatically reduced by 10 percent, the training is stopped when 100 epochs are reached, and the batch _ size is set to be 16; using a random gradient descent method, the momentum was set to 0.9 and the weight attenuation coefficient was set to 2X 10-4；
After 100 epochs are trained, saving a current weight file;
and according to the weight file and the test set, evaluating the performance of the trained garbage recognition model in the aspects of parameter quantity, accuracy, precision, recall rate, F1-score and FPS.
Further, the process of calculating the probability of the stacked spam image belonging to each category includes the following steps:
and loading the trained weight file into a model, and predicting the probability value of the category to which the garbage image belongs in the test set by using a prediction function. The calculated confusion matrix is used to evaluate the accuracy of the classification using the API interface function in the scimit-learn packet that calculates the confusion matrix. In the confusion matrix, numeral 1 represents other garbage, numeral 2 represents kitchen garbage, numeral 3 represents recyclables, and numeral 4 represents harmful garbage.
Further, the garbage sorting treatment method further comprises the following steps:
and S3, sending the garbage classification result output by the full connecting layer to an embedded processor ARM, setting a corresponding program according to the garbage classification result to control a servo motor so as to rotate a mechanical ARM beside the conveying belt, grabbing the garbage by using a mechanical ARM, sending the garbage to a garbage classification groove, and sorting the garbage.
Further, the garbage sorting treatment method further comprises the following steps:
s31, dividing the second layer of conveyor belt into n conveying areas according to the conveying direction;
s32, performing garbage category identification on the stacked garbage images on the n conveying areas by adopting a garbage identification module:
if the probability corresponding to the recognized garbage category is higher than a preset probability threshold value, grabbing the stacked garbage on the n conveying areas to the corresponding garbage classifying grooves once by using a mechanical arm, otherwise, recognizing the stacked garbage image on the nearest conveying area again, and grabbing the stacked garbage on the nearest conveying area to the corresponding garbage classifying grooves according to the recognition result;
and S33, repeating the step S32 until all the garbage is sorted.
In a second aspect, an embodiment of the present invention provides a garbage sorting processing system for a garbage transfer station based on the aforementioned garbage sorting processing method, including:
two layers of conveyor belts with height difference are arranged at the outlet of the drum screen, and the garbage with the diameter larger than 80 mm left by the drum screen is placed into the first layer of conveyor belt. Because two-layer conveyer belt has certain difference in height to control first layer conveyer belt transmission rate and be slower than second layer conveyer belt transmission rate, put into the pile-up situation of conveyer belt with effectively alleviating rubbish.
The camera is arranged above the second layer of conveyor belt, and when the second layer of conveyor belt is conveyed in, all the garbage on the designated area of the conveyor belt is shot; and sending the shot stacked garbage image to a garbage recognition module installed on the embedded processing terminal to classify the garbage.
The embedded processor ARM is internally provided with a garbage recognition module and servo motor control software; the garbage recognition module is used for processing the stacked garbage images shot by the camera and recognizing the stacked garbage images to obtain the categories and corresponding probabilities of the stacked garbage images; and the servo motor control software generates a corresponding control program according to the recognition result, rotates the mechanical arm beside the conveyor belt, and adopts the mechanical arm to grab the garbage and send the garbage to the corresponding garbage classification groove.
The invention has the beneficial effects that:
(1) the garbage sorting processing method provided by the invention can accept the image input of any size, and the network adopts a deep separable convolution mode, thereby reducing the parameters of the network. The quantity of the parameters of the method is about 3M, and the quantity of the parameters of other similar methods is more than 10 times of that of the method, so that the method is lighter in network and is beneficial to being deployed on an embedded processing terminal for rapid garbage sorting processing.
(2) And a compression and excitation module in an attention mechanism is added in the first convolution module to the third convolution module, so that the dependency between the garbage image channels is enhanced. The output characteristics of the fourth convolution module and the fifth convolution module are fused, so that the accuracy rate of identifying the garbage image is effectively improved. Aiming at the problem of garbage sorting, the recognition accuracy of the method can reach over 75 percent, and other similar methods are generally about 71 percent. In addition, the garbage sorting treatment method provided by the invention uniformly identifies all garbage in the area, simultaneously considers the requirements of precision and efficiency in the integral sorting treatment of stacked garbage, and effectively overcomes the defect of sorting treatment of recyclable objects in a garbage transfer station.
(3) Adaptive maximum pooling and global average pooling are employed, which is more suitable for convolution structures by enhancing the correspondence between feature maps and classes of spam images. Furthermore, there are no parameters in the global average pool that need to be optimized, thus avoiding overfitting of the layer. Global average pooling sums the spatial information and is therefore more robust to spatial translation of the input.
Drawings
Fig. 1 is an overall process of garbage sorting processing according to an embodiment of the present invention.
FIG. 2 is a diagram of a belt periphery configuration according to an embodiment of the present invention.
FIG. 3 is a schematic structural diagram of LW-GCNet according to an embodiment of the present invention.
Fig. 4 shows a process of weight reduction in the embodiment of the present invention.
FIG. 5 is a schematic diagram of a compression and excitation module of an embodiment of the invention.
FIG. 6 is a schematic diagram of a feature fusion module according to an embodiment of the present invention.
FIG. 7 is a graph comparing the recognition accuracy of LW-GCNet and several other methods according to an embodiment of the present invention.
FIG. 8 is a schematic diagram of a confusion matrix according to an embodiment of the invention; wherein, the number 1 represents other garbage, the number 2 represents kitchen garbage, the number 3 represents recoverable matters, and the number 4 represents harmful garbage.
Detailed Description
The present invention will now be described in further detail with reference to the accompanying drawings.
It should be noted that the terms "upper", "lower", "left", "right", "front", "back", etc. used in the present invention are for clarity of description only, and are not intended to limit the scope of the present invention, and the relative relationship between the terms and the terms is not limited by the technical contents of the essential changes.
Fig. 1 is an overall process of garbage sorting processing according to an embodiment of the present invention. Referring to fig. 1, the garbage sorting method includes the following steps:
s1, installing two layers of conveyor belts with height difference at the outlet of the drum screen, putting the garbage with the diameter larger than 80 mm left by the drum screen into the first layer of conveyor belt, and shooting all garbage on the designated area of the conveyor belt when the garbage is conveyed into the second layer of conveyor belt; and sending the shot stacked garbage image to a garbage recognition module installed on the embedded processing terminal to classify the garbage. FIG. 2 is a diagram of a belt periphery configuration according to an embodiment of the present invention.
And S2, sending the garbage classification result into an embedded processor ARM, setting a corresponding program to control a servo motor, rotating a mechanical ARM beside the conveyor belt, and grabbing the garbage by using a mechanical ARM to send the garbage to a corresponding garbage classification groove.
Referring to fig. 3, the garbage recognition module (LW-GCNet network structure) of this embodiment is obtained by training a certain amount of images of stacked garbage samples of the same type, and includes an adaptive pooling layer, a feature extraction unit, a feature fusion unit, an adaptive maximum pooling layer, a global average pooling layer, and a full connection layer, which are connected in sequence; the feature extraction unit extracts features based on a lightweight mode, comprises five convolution modules, and adopts a depth separable convolution mode to carry out deep feature extraction on the multi-view feature map with the adjusted size in the self-adaptive pooling layer; wherein, a compression and excitation module is added in the first convolution module, the second convolution module and the third convolution module; the feature fusion unit is used for extracting the outputs of the fourth convolution module and the fifth convolution module and carrying out feature fusion of different sizes; the fused features are subjected to self-adaptive maximum pooling and global average pooling respectively through a self-adaptive maximum pooling layer and a global average pooling layer, and then global features of the stacked garbage images are extracted; and the full connection layer is used for classifying the extracted global features of the stacked garbage images, calculating the probability of the stacked garbage images belonging to each category, and outputting the category information and the corresponding probability of the current stacked garbage images.
Sample data set
The sample images adopted during the training of the garbage recognition module are also stacking garbage images, but the difference from the actual garbage images is that the stacking garbage on each sample image belongs to the same category even though the stacking garbage is different. For example, the trash in one of the sample images is all recyclable trash, and the images may include old furniture, paper, cans, and the like. Illustratively, the types of the stacked garbage sample images at least include four categories of preprocessed kitchen garbage, recyclable matters, other garbage and harmful garbage, and the garbage amount on each stacked garbage sample image is not less than a preset garbage amount threshold. It should be understood that in practical applications, the types of the trash may not be limited to the above four types, and specifically, the types of the trash recognizable by the trash recognition module and the types of the selected trash sample images may be set according to the types of the trash sorting grooves. For example, a metal garbage sorting groove, even a pop can sorting groove and the like are added.
And in the data preprocessing part, enhancing the image data of each batch, expanding the size of the data set by adopting various methods of scaling, rotating, transforming and overturning, and enhancing the generalization capability of the model. The data set is divided into a training set and a test set.
Because of the small number of open-source garbage data sets, researchers generally establish data sets by means of self-shooting and internet searching. The garbage data set adopted by the embodiment is composed of 12500 garbage images with different resolutions, four categories of kitchen garbage, recoverable objects, other garbage and harmful garbage are contained in the garbage data set, and a plurality of categories of garbage are contained in each category. The garbage image contains more garbage and rich characteristic information. Since the network model is deployed on the embedded processing terminal, a lightweight network model LW-GCNet is required. And finally, the trained LW-GCNet is applied to the aspect of processing the garbage with high recovery value in the garbage transfer station, so that the garbage classification is closer to the reality.
Network structure of (II) garbage recognition module
Referring to fig. 3 to 6, the lightweight network model LW-GCNet proposed in this embodiment is developed by using Keras V2.3.1, and a lightweight garbage classification network model is trained on a graphics card Nvidia GeForce RTX 2080Ti, a processor Intel Core i7-9700K and a MAG Z390 TOMAHAWK as a motherboard.
The feature extraction unit comprises five convolution modules, wherein each convolution module adopts a DepthWise convolution layer and a PointWise convolution layer which are alternately connected: the first convolution module is provided with a set of alternately connected DepthWise convolution layer and PointWise convolution layer; the second convolution module is provided with two groups of DepthWise convolution layers and Pointwise convolution layers which are alternately connected; the third, the fourth and the fifth convolution modules respectively have three groups of alternately connected DepthWise convolution layers and PointWise convolution layers; a BN layer is added after each convolution layer; a ZeroPadding layer is added after the convolutional layer of each set of convolutional modules.
The zeroPadding layer of the first convolution module, the second convolution module and the third convolution module is added with a compression and excitation module in an attention mechanism.
The number of convolution kernels of each convolution module is increased along with the deepening of the network layer number, and the convolution step length is changed to reduce the characteristic diagram.
The garbage image is sent to an embedded processing terminal and processed by using an LW-GCNet network, and the processing process comprises the following steps:
step one, adopting a self-adaptive pooling layer to solve the problem of receiving images with any size by a network. Through the adaptive pooling layer, the size of the output image is 224 × 224 × 3 as the input image size of the feature extraction module.
And step two, processing the input image by adopting a depth separable convolution mode through five convolution modules in the network so as to reduce the parameter number of the network model, wherein the characteristic extraction structure of the lightweight mode is shown in the table 1.
TABLE 1 light-weight extraction of feature composition Table
And 2-1) the lightweight extraction features comprise five convolution modules, and each convolution module adopts a DepthWise convolution layer and a PointWise convolution layer which are alternately connected. The first convolution module has a set of alternately connected DepthWise and PointWise convolution layers. The second convolution module has two alternately connected DepthWise convolution layer and Pointwise convolution layer. The third, fourth and fifth convolution modules respectively have three groups of alternately connected DepthWise convolution layers and PointWise convolution layers. The number of convolution kernels used in the five convolution modules is respectively 64, 128, 256, 512 and 512, wherein the size of the DepthWise convolution kernel is 3 multiplied by 3, and the size of the PointWise convolution kernel is 1 multiplied by 1.
And 2-2) adding a BN layer after each convolution layer, wherein the number of convolution kernels of each module is increased along with the deepening of the number of network layers, and the convolution step length is changed to reduce the characteristic diagram. After each convolution module has convolved, a zeroPadding layer is added.
Step 2-3) a compression and excitation module in the attention mechanism is also added after the ZeroPadding layer of the first, second and third convolution modules, the compression and excitation module is shown in figure 5,
width, height and channel number of input feature mapIs inputted
Results after compression
And adjusting the size between two full connection layers by using the hyperparameter r, controlling the capacity and the calculation amount of a compression and excitation module, and determining r = 16. Post-excitation output weights
Weighting channel by Scale multiplication and distributing weight to output result
The dependency between the channels is enhanced.
Step three, the fusion characteristics are shown in fig. 6.
And 3-1) in the characteristic extraction in a lightweight mode, when the output of the fourth convolution module is 29 multiplied by 512 and the output of the fifth convolution module is 14 multiplied by 512, respectively extracting the characteristic map.
And 3-2) performing convolution downsampling on the output of the fourth convolution module, performing upsampling on the output of the fifth convolution module, and then performing Concatenate fusion on the image with the adjusted size to obtain the 28 × 28 × 768 characteristics, so that the detailed characteristics of the network are prevented from being lost in the convolution process, and the characteristics extracted by the network are enhanced.
And step four, extracting the global features of the garbage image by adopting self-adaptive maximum pooling and global average pooling on the fused features.
And step five, classifying and identifying the extracted global characteristics layer by using a full connection layer, sending the result into an embedded processor ARM according to the garbage classification result of the full connection layer by layer, setting a corresponding program to control a servo motor, rotating a mechanical ARM beside a conveyor belt, grabbing the garbage by using a mechanical ARM, and sending the garbage to a garbage classification groove to realize the sorting purpose.
FIG. 7 is a graph comparing the recognition accuracy of LW-GCNet of the present invention with several other methods. The LW-GCNet garbage identification method can accept image input of any size, the lightweight network is suitable for quick garbage sorting processing of a garbage transfer station, parameters of a network model are reduced to 3.15M on the premise of ensuring the number of network layers, and meanwhile, the garbage sorting precision is improved.
(III) training process of garbage recognition model
When training is started, the preprocessed image passes through the self-adaptive pooling layer, so that the input received by the model is always 224 multiplied by 3, and the initial learning rate is 1 multiplied by 10-3Then, each time 7 epoch iterations are iterated with the learning rate not dropping, the learning rate will automatically drop by 10%, 100epoch is reached to stop training, and batch _ size is set to 16. Using a random gradient descent method, the momentum was set to 0.9 and the weight attenuation coefficient was set to 2X 10-4。
After LW-GCNet trains 100 epochs, the network stores a weight file.
In the testing stage, the performance of the network model is evaluated from six aspects of parameter quantity, accuracy, precision, recall, F1-score and FPS according to the weight file and the test set. The results of the performance evaluations are detailed in tables 2 and 3.
Table 2 evaluation of performance on test set by other network models
TABLE 3 accuracy of other network models for identifying each type of spam on test set
(IV) mechanical arm sorting process
In this embodiment, the sorting process of the robot arm includes the following steps: the garbage classification result output by the full connecting layer is sent to the embedded processor ARM, and the corresponding program control servo motor is set according to the garbage classification result so as to rotate the mechanical ARM beside the conveying belt, grab the garbage by the mechanical ARM and send the garbage to the garbage classification groove, and sort the garbage.
Illustratively, in order to further improve the sorting efficiency on the premise of meeting the sorting precision, the garbage sorting processing method further comprises the following steps:
s31, dividing the second layer of conveyor belt into n conveying areas according to the conveying direction;
s32, performing garbage category identification on the stacked garbage images on the n conveying areas by adopting a garbage identification module: and if the probability corresponding to the recognized garbage category is higher than a preset probability threshold value, grabbing the stacked garbage on the n conveying areas to the corresponding garbage classifying grooves once by using the mechanical arm, otherwise, recognizing the stacked garbage image on the nearest conveying area again, and grabbing the stacked garbage on the nearest conveying area to the corresponding garbage classifying grooves according to the recognition result.
And S33, repeating the step S32 until all the garbage is sorted.
The embodiment of the invention provides a garbage sorting treatment system for a garbage transfer station, which comprises the following structural components:
(1) two layers of conveyor belts with height difference are arranged at the outlet of the drum screen, and the garbage with the diameter larger than 80 mm left by the drum screen is placed into the first layer of conveyor belt. The two layers of conveyor belts have certain height difference, and the conveying speed of the first layer of conveyor belt is controlled to be slower than that of the second layer of conveyor belt. Thus, the stacking condition of the garbage put into the conveyor belt can be effectively relieved. When the garbage is conveyed into the second layer of conveyor belt, shooting all garbage on a designated area of the conveyor belt; sending the shot stacked garbage image to a garbage recognition module installed on the embedded processing terminal to classify the garbage;
(2) and the camera is arranged above the second layer of conveyor belt and is used for shooting all garbage on the designated area of the conveyor belt.
(3) The embedded processor ARM is internally provided with a garbage recognition module and servo motor control software; the garbage recognition module is used for processing the stacked garbage images shot by the camera and recognizing the stacked garbage images to obtain the categories and corresponding probabilities of the stacked garbage images; and the servo motor control software generates a corresponding control program according to the recognition result, rotates the mechanical arm beside the conveyor belt, and adopts the mechanical arm to grab the garbage and send the garbage to the corresponding garbage classification groove.
Illustratively, if m garbage conveying belt areas are divided at equal intervals according to the length of the conveying belt, m groups of cameras, embedded processing terminals, servo motors, mechanical arms and garbage sorting grooves which can be used for garbage sorting processing can be arranged beside the garbage conveying belt areas, and wired connection is utilized among the cameras, the embedded processing terminals, the servo motors, the mechanical arms and the garbage sorting grooves, so that real-time communication can be ensured, time delay is avoided, and sorting efficiency is improved. Through m synchronous letter sorting of robotic arm, realize higher letter sorting efficiency.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention.
Claims (10)
1. A method for sorting and processing refuse for refuse transfer stations, characterized in that it comprises the following steps:
s1, installing two layers of conveyor belts with height difference at the outlet of the drum screen, putting the garbage with the diameter larger than 80 mm left by the drum screen into the first layer of conveyor belt, and shooting all garbage on the designated area of the conveyor belt when the garbage is conveyed into the second layer of conveyor belt; sending the shot stacked garbage image to a garbage recognition module installed on the embedded processing terminal to classify the garbage;
s2, sending the garbage classification result into an embedded processor ARM, setting a corresponding program to control a servo motor, rotating a mechanical ARM beside a conveyor belt, and grabbing garbage by a mechanical ARM to send the garbage to a corresponding garbage classification groove;
the garbage identification module is obtained by training a plurality of similar stacked garbage sample images and comprises a self-adaptive pooling layer, a feature extraction unit, a feature fusion unit, a self-adaptive maximum pooling layer, a global average pooling layer and a full-connection layer which are connected in sequence; the feature extraction unit extracts features based on a lightweight mode, comprises five convolution modules, and adopts a depth separable convolution mode to carry out deep feature extraction on the multi-view feature map with the adjusted size in the self-adaptive pooling layer; wherein, a compression and excitation module is added in the first convolution module, the second convolution module and the third convolution module; the feature fusion unit is used for extracting the outputs of the fourth convolution module and the fifth convolution module and carrying out feature fusion of different sizes; the fused features are subjected to self-adaptive maximum pooling and global average pooling respectively through a self-adaptive maximum pooling layer and a global average pooling layer, and then global features of the stacked garbage images are extracted; and the full connection layer is used for classifying the extracted global features of the stacked garbage images, calculating the probability of the stacked garbage images belonging to each category, and outputting the category information and the corresponding probability of the current stacked garbage images.
2. The method according to claim 1, wherein the type of the stacked garbage sample image at least includes four categories of pre-processed kitchen garbage, recyclable materials, other garbage and harmful garbage, and the amount of garbage on each stacked garbage sample image is not less than a preset garbage amount threshold;
the pretreatment process comprises the following steps: and enhancing the stacked garbage sample images of each batch, and expanding the data set by adopting various methods of scaling, rotating, transforming and turning.
3. Method for the sorting and processing of rubbishes for a rubbish terminal according to claim 2, characterized in that the preset threshold value of the quantity of rubbishes is 3.
4. The method as claimed in claim 1, wherein the feature extraction unit comprises five convolution modules, each convolution module adopts a DepthWise convolution layer and a PointWise convolution layer which are alternately connected: the first convolution module is provided with a set of alternately connected DepthWise convolution layer and PointWise convolution layer; the second convolution module is provided with two groups of DepthWise convolution layers and Pointwise convolution layers which are alternately connected; the third, the fourth and the fifth convolution modules respectively have three groups of alternately connected DepthWise convolution layers and PointWise convolution layers; a BN layer is added after each convolution layer; a zeroPadding layer is added after the convolution layer of each group of convolution modules;
a compression and excitation module in an attention mechanism is added after the zeroPadding layers of the first convolution module, the second convolution module and the third convolution module;
the number of convolution kernels of each convolution module is increased along with the deepening of the network layer number, and the convolution step length is changed to reduce the characteristic diagram.
5. The method for sorting and processing refuse for refuse transfer stations according to claim 1, characterized in that the fusion process of the feature fusion unit comprises the following steps:
extracting 29 × 29 × 512 first output features output by the fourth convolution module and 14 × 14 × 512 second output features output by the fifth convolution module respectively from the fusion features;
for the first output characteristic of the fourth convolution module, performing convolution with convolution kernel size of 3 × 3, kernel number of 256, step length of 1 and filling; then, carrying out batch normalization processing, and activating by using a Relu function to obtain a first characteristic value with the output of 28 multiplied by 256, wherein the first characteristic value is used for representing the geometric detail information of the garbage and emphasizing the extraction of a medium target of the garbage image;
for the output characteristics of the fifth convolution module, performing convolution with convolution kernel size of 1 × 1, kernel number of 512, step length of 1 and filling; then, carrying out batch normalization processing, activating by using a Relu function, and scaling the depth features to the same scale through up-sampling to obtain a second feature value with the output of 28 multiplied by 512, so as to represent semantic information of garbage and extract medium and large targets of garbage images with emphasis;
and adjusting the first characteristic value and the second characteristic value output by the fourth convolution module and the fifth convolution module to be consistent in size, carrying out Concatenate fusion to form a characteristic diagram of 28 multiplied by 768, fusing the characteristics of different layers in the network, and obtaining the global characteristic of the stacked garbage image.
6. The method for sorting and processing refuse for refuse transfer terminals according to claim 1, characterized in that the training process of the refuse recognition model comprises the following steps:
when training is started, the preprocessed image passes through the self-adaptive pooling layer, so that the resolution of the input image received by the garbage recognition model is kept to be 224 multiplied by 3;
let the initial learning rate be 1 × 10-3When 7 epoch learning rates are iterated and are not reduced, the learning rate is automatically reduced by 10 percent, the training is stopped when 100 epochs are reached, and the batch _ size is set to be 16; using a random gradient descent method, the momentum was set to 0.9 and the weight attenuation coefficient was set to 2X 10-4；
After 100 epochs are trained, saving a current weight file;
and according to the weight file and the test set, evaluating the performance of the trained garbage recognition model in the aspects of parameter quantity, accuracy, precision, recall rate, F1-score and FPS.
7. The method of claim 6, wherein the step of calculating the probability of the stacked garbage image belonging to each category comprises the steps of:
loading the trained weight file into a model, and predicting the probability value of the category to which the garbage image belongs in the test set by using a prediction function; using an API (application programming interface) function for calculating a confusion matrix in the scimit-learn packet, and calculating the confusion matrix to evaluate the classification accuracy;
wherein, the number 1 in the confusion matrix represents other garbage, the number 2 represents kitchen garbage, the number 3 represents recoverable matters, and the number 4 represents harmful garbage.
8. The method of sorting waste for a waste transfer station according to claim 1, further comprising the steps of:
and S3, sending the garbage classification result output by the full connecting layer to an embedded processor ARM, setting a corresponding program according to the garbage classification result to control a servo motor so as to rotate a mechanical ARM beside the conveying belt, grabbing the garbage by using a mechanical ARM, sending the garbage to a garbage classification groove, and sorting the garbage.
9. The method of sorting waste for a waste transfer station according to claim 8, further comprising the steps of:
s31, dividing the second layer of conveyor belt into n conveying areas according to the conveying direction;
s32, performing garbage category identification on the stacked garbage images on the n conveying areas by adopting a garbage identification module:
if the probability corresponding to the recognized garbage category is higher than a preset probability threshold value, grabbing the stacked garbage on the n conveying areas to the corresponding garbage classifying grooves once by using a mechanical arm, otherwise, recognizing the stacked garbage image on the nearest conveying area again, and grabbing the stacked garbage on the nearest conveying area to the corresponding garbage classifying grooves according to the recognition result;
and S33, repeating the step S32 until all the garbage is sorted.
10. A refuse sorting system for refuse transfer terminals based on the refuse sorting method according to any one of claims 1-9, comprising:
two layers of conveyor belts with height difference are arranged at the outlet of the drum screen, garbage with the diameter larger than 80 mm left by the drum screen is placed into the first layer of conveyor belt, and the transmission speed of the first layer of conveyor belt is controlled to be slower than that of the second layer of conveyor belt;
the camera is arranged above the second layer of conveyor belt, when garbage is conveyed to the second layer of conveyor belt, all garbage on the designated area of the conveyor belt is shot, and a shot stacked garbage image is sent to the garbage recognition module arranged on the embedded processing terminal, so that the garbage recognition module classifies the garbage in the image;
the embedded processor ARM is internally provided with a garbage recognition module and servo motor control software; the garbage recognition module is used for processing the stacked garbage images shot by the camera and recognizing the stacked garbage images to obtain the categories and corresponding probabilities of the stacked garbage images; and the servo motor control software generates a corresponding control program according to the recognition result, rotates the mechanical arm beside the conveyor belt, and adopts the mechanical arm to grab the garbage and send the garbage to the corresponding garbage classification groove.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111297530.3A CN113731832B (en) | 2021-11-04 | 2021-11-04 | Garbage sorting treatment method and system for garbage transfer station |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111297530.3A CN113731832B (en) | 2021-11-04 | 2021-11-04 | Garbage sorting treatment method and system for garbage transfer station |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113731832A true CN113731832A (en) | 2021-12-03 |
| CN113731832B CN113731832B (en) | 2022-02-15 |
Family
ID=78727320
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111297530.3A Active CN113731832B (en) | 2021-11-04 | 2021-11-04 | Garbage sorting treatment method and system for garbage transfer station |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113731832B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109201514A (en) * | 2017-06-30 | 2019-01-15 | 京东方科技集团股份有限公司 | Waste sorting recycle method, garbage classification device and classified-refuse recovery system |
| CN110228673A (en) * | 2019-04-17 | 2019-09-13 | 华南师范大学 | A kind of intelligent classification dustbin |
| CN110427896A (en) * | 2019-08-07 | 2019-11-08 | 成都理工大学 | A kind of garbage classification intelligence system based on convolutional neural networks |
| CN110473130A (en) * | 2019-07-30 | 2019-11-19 | 五邑大学 | A kind of garbage classification evaluation method, device and storage medium based on deep learning |
| CN110472115A (en) * | 2019-08-08 | 2019-11-19 | 东北大学 | A kind of social networks text emotion fine grit classification method based on deep learning |
| CN110598800A (en) * | 2019-09-23 | 2019-12-20 | 山东浪潮人工智能研究院有限公司 | Garbage classification and identification method based on artificial intelligence |
| CN110589282A (en) * | 2019-08-16 | 2019-12-20 | 喻鑫童 | Intelligent garbage classification method based on machine learning and automatic garbage sorting device |
| US20200050922A1 (en) * | 2018-08-13 | 2020-02-13 | National Chiao Tung University | Recycling system and method based on deep-learning and computer vision technology |
| CN111056183A (en) * | 2019-11-29 | 2020-04-24 | 华东师范大学 | Real-time intelligent garbage self-classification system based on ZYNQ |
| CN111814860A (en) * | 2020-07-01 | 2020-10-23 | 浙江工业大学 | Multi-target detection method for garbage classification |
| CN112115974A (en) * | 2020-08-18 | 2020-12-22 | 郑州睿如信息技术有限公司 | Intelligent visual detection method for classification treatment of municipal waste |
| CN112633335A (en) * | 2020-12-10 | 2021-04-09 | 长春理工大学 | Garbage classification method and garbage can |
| US20210272073A1 (en) * | 2020-02-28 | 2021-09-02 | Triple Win Technology(Shenzhen) Co.Ltd. | Garbage sorting and recycliing method, system, and computer readable storage medium |
-
2021
- 2021-11-04 CN CN202111297530.3A patent/CN113731832B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109201514A (en) * | 2017-06-30 | 2019-01-15 | 京东方科技集团股份有限公司 | Waste sorting recycle method, garbage classification device and classified-refuse recovery system |
| US20200050922A1 (en) * | 2018-08-13 | 2020-02-13 | National Chiao Tung University | Recycling system and method based on deep-learning and computer vision technology |
| CN110228673A (en) * | 2019-04-17 | 2019-09-13 | 华南师范大学 | A kind of intelligent classification dustbin |
| CN110473130A (en) * | 2019-07-30 | 2019-11-19 | 五邑大学 | A kind of garbage classification evaluation method, device and storage medium based on deep learning |
| CN110427896A (en) * | 2019-08-07 | 2019-11-08 | 成都理工大学 | A kind of garbage classification intelligence system based on convolutional neural networks |
| CN110472115A (en) * | 2019-08-08 | 2019-11-19 | 东北大学 | A kind of social networks text emotion fine grit classification method based on deep learning |
| CN110589282A (en) * | 2019-08-16 | 2019-12-20 | 喻鑫童 | Intelligent garbage classification method based on machine learning and automatic garbage sorting device |
| CN110598800A (en) * | 2019-09-23 | 2019-12-20 | 山东浪潮人工智能研究院有限公司 | Garbage classification and identification method based on artificial intelligence |
| CN111056183A (en) * | 2019-11-29 | 2020-04-24 | 华东师范大学 | Real-time intelligent garbage self-classification system based on ZYNQ |
| US20210272073A1 (en) * | 2020-02-28 | 2021-09-02 | Triple Win Technology(Shenzhen) Co.Ltd. | Garbage sorting and recycliing method, system, and computer readable storage medium |
| CN111814860A (en) * | 2020-07-01 | 2020-10-23 | 浙江工业大学 | Multi-target detection method for garbage classification |
| CN112115974A (en) * | 2020-08-18 | 2020-12-22 | 郑州睿如信息技术有限公司 | Intelligent visual detection method for classification treatment of municipal waste |
| CN112633335A (en) * | 2020-12-10 | 2021-04-09 | 长春理工大学 | Garbage classification method and garbage can |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113731832B (en) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Automatic detection and classification system of domestic waste via multimodel cascaded convolutional neural network | |
| Adedeji et al. | Intelligent waste classification system using deep learning convolutional neural network | |
| Rabano et al. | Common garbage classification using mobilenet | |
| CN110598800A (en) | Garbage classification and identification method based on artificial intelligence | |
| CN110390691A (en) | A kind of ore scale measurement method and application system based on deep learning | |
| CN101604322B (en) | Decision level text automatic classified fusion method | |
| Hulyalkar et al. | Implementation of smartbin using convolutional neural networks | |
| Gyawali et al. | Comparative analysis of multiple deep CNN models for waste classification | |
| CN112733936A (en) | Recyclable garbage classification method based on image recognition | |
| CN110516625A (en) | A kind of method, system, terminal and the storage medium of rubbish identification classification | |
| CN113807347A (en) | Kitchen waste impurity identification method based on target detection technology | |
| CN113592825A (en) | YOLO algorithm-based real-time coal gangue detection method | |
| Huiyu et al. | Automatic classifications and recognition for recycled garbage by utilizing deep learning technology | |
| Song et al. | Optimization of CNN-based garbage classification model | |
| Desai et al. | Waste segregation using machine learning | |
| Sivakumar et al. | IoT incorporated deep learning model combined with SmartBin technology for real‐time solid waste management | |
| CN113731832B (en) | Garbage sorting treatment method and system for garbage transfer station | |
| Patil et al. | Garbage classifying application using deep learning techniques | |
| Aarthi et al. | A vision based approach to localize waste objects and geometric features exaction for robotic manipulation | |
| Sirimewan et al. | Deep learning-based models for environmental management: Recognizing construction, renovation, and demolition waste in-the-wild | |
| CN114782762B (en) | Garbage image detection method and community garbage station | |
| CN115761259A (en) | Kitchen waste target detection method and system based on class balance loss function | |
| CN112620165B (en) | Garbage classification method | |
| Mehedi et al. | A Transfer Learning Approach For Efficient Classification of Waste Materials | |
| CN113610100A (en) | Image recognition-based garbage processing method and device and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |