CN116630787A - Light-weight detection method and device for overflow of garbage can and storage device - Google Patents
Light-weight detection method and device for overflow of garbage can and storage device Download PDFInfo
- Publication number
- CN116630787A CN116630787A CN202310679679.0A CN202310679679A CN116630787A CN 116630787 A CN116630787 A CN 116630787A CN 202310679679 A CN202310679679 A CN 202310679679A CN 116630787 A CN116630787 A CN 116630787A
- Authority
- CN
- China
- Prior art keywords
- garbage
- module
- layer
- overflow
- lightweight
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000010813 municipal solid waste Substances 0.000 title claims abstract description 179
- 238000001514 detection method Methods 0.000 title claims abstract description 97
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000000034 method Methods 0.000 claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 14
- 238000012795 verification Methods 0.000 claims abstract description 12
- 238000011897 real-time detection Methods 0.000 claims abstract description 7
- 238000002372 labelling Methods 0.000 claims abstract description 4
- 210000000988 bone and bone Anatomy 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 6
- 238000001228 spectrum Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000004140 cleaning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 241000592183 Eidolon Species 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000005574 cross-species transmission Effects 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000001771 impaired effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02W—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO WASTEWATER TREATMENT OR WASTE MANAGEMENT
- Y02W30/00—Technologies for solid waste management
- Y02W30/10—Waste collection, transportation, transfer or storage, e.g. segregated refuse collecting, electric or hybrid propulsion
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method, equipment and storage equipment for detecting overflow of a garbage can in a lightweight way, wherein the method comprises the following steps: constructing a garbage can and a garbage sample data set; labeling a sample data set and dividing the sample data set into a training set, a verification set and a test set according to a proportion; constructing a garbage can overflow detection model, lightening the garbage can overflow detection model, and combining an E-ELAN module to enhance the network learning capacity to obtain a lightweight improved garbage can overflow detection model; training a lightweight improved garbage can overflow detection model to obtain a trained model, and deploying the model to an intelligent robot; the intelligent robot is used for inspecting the area to be detected, and real-time detection is completed; if the garbage can is detected, judging whether the garbage can overflows or not according to a formulated garbage overflow rule, if so, reporting a garbage overflow result in real time; otherwise, no treatment is performed. The invention has the beneficial effects that: the real-time detection and the accuracy can be considered.
Description
Technical Field
The invention relates to the field of target detection, in particular to a method and equipment for detecting overflow of a garbage can in a lightweight manner and storage equipment.
Background
At present, garbage cleaning mainly depends on manual regular fixed-point inspection cleaning, community garbage is related to geographic positions, garbage is easy to generate in remote locations of corners, and is related to traffic flow, and the generation sites and the total amount are very irregular, so that in the development of modern society, the manual garbage cleaning mode is large in workload and low in efficiency. Therefore, the intelligent detection of garbage overflow is realized, the labor cost is reduced, the intelligent environment-friendly working efficiency is improved, and the intelligent garbage overflow detection device is a solution idea of the current problem.
At present, detecting garbage overflow is mainly processed by two modes of intelligent hardware and software. The intelligent hardware is utilized to detect garbage overflow, various sensors are mainly arranged on the garbage can, and when the processing result of the sensors reaches a set threshold value through sensing the height, temperature, humidity, smell and the like of the garbage in the can, the garbage overflow is the garbage overflow, automatic alarm is triggered, and related staff is reminded of processing events. For example, the information technology limited company is focused on building an intelligent sanitation informatization system platform, the designed garbage overflow detection system is combined with the technology of the Internet of things, a sensor is arranged at the center position right above a dustbin, an alarm threshold is set in advance, and when garbage in the dustbin reaches the threshold, the alarm system is triggered to prompt that the garbage is full and is processed in time. The northly-awakened TF series laser radar module automatically detects the garbage amount by using a laser radar, places equipment on a barrel cover in a garbage can, and transmits the garbage amount to a data management system when the garbage amount reaches 80%, so as to prompt garbage clearance. Above detect rubbish through the sensor and spill over, be limited by the fixed point suggestion to aggravate the garbage bin cost, if the sensor is impaired, when losing etc. need change, further improved the cost. The intelligent software is utilized to detect garbage overflow, and a mature garbage detection system is mainly carried in intelligent equipment based on deep learning, and the intelligent equipment is used for inspecting garbage overflow in real time and reporting overflow events.
At present, garbage overflow is detected through a deep learning algorithm, and the garbage overflow detection method can be mounted in intelligent equipment, but is limited by the size of a model, and the real-time performance and accuracy of the existing industrial convolution network model cannot be balanced.
Comprehensive analysis, the existing garbage overflow detection has the following problems:
(1) Detecting garbage overflow through mounting hardware is limited by fixed-point prompt, has no flexibility, and has no full coverage because of cost consideration;
(2) The garbage overflow is detected by deploying a software system, the size of the model needs to be considered, the existing network model is not light enough, the deployment in equipment is not facilitated, and the accuracy and the instantaneity cannot be guaranteed.
Therefore, an effective method and system for detecting garbage overflow should be realized: higher detection accuracy, lower computational complexity, better robustness and easy deployment.
Disclosure of Invention
In order to solve the problem that the detection accuracy, the detection speed and the robustness are difficult to consider in the existing garbage can overflow detection model, the improved convolutional network model YOLOv5 is adopted to detect garbage overflow, and the YOLOv5 model is not light enough and is not easy to deploy in intelligent equipment, so that the invention combines the ideas of grouping convolution and efficient aggregation network, improves the YOLOv5 model, and realizes more light and high-accuracy real-time detection of garbage overflow.
Specifically, the invention provides a lightweight detection method for overflow of a garbage can, which comprises the following steps:
step S101, constructing a garbage can and a garbage sample data set;
step S102, labeling a sample data set and dividing the sample data set into a training set, a verification set and a test set according to a proportion;
step S103, constructing a garbage can overflow detection model, lightening the garbage can overflow detection model, and combining an E-ELAN module to enhance the network learning capability to obtain a lightweight improved garbage can overflow detection model; in step S103, the lightweight modified trash can overflow detection model includes: the device comprises an input module, a back bone module, a Neck module and a Head module; the input module is used for preprocessing an input image; the backstone module is used for extracting deep features of the image; the Neck module is used for improving the feature extraction capability; the Head module is used for evaluating a lightweight improved garbage can overflow detection model;
step S104, training, verifying and detecting the lightweight improved garbage can overflow detection model by utilizing the training set, the verification set and the test set;
step S105, outputting a fully trained garbage can overflow detection model based on a predicted result, and deploying the model to the intelligent robot;
step S106, the intelligent robot is used for inspecting the area to be detected, and real-time detection is carried out by using a trained garbage can overflow detection model;
step S107, if the garbage can is detected, judging whether the garbage can overflows or not according to a formulated garbage overflow rule, if so, reporting a garbage overflow result in real time; otherwise, no treatment is performed.
A storage device stores instructions and data for implementing a lightweight detection method for garbage can overflow.
A dustbin overflow lightweight detection device, comprising: a processor and the storage device; the processor loads and executes the instructions and the data in the storage device to realize a lightweight detection method for the overflow of the garbage can.
The beneficial effects provided by the invention are as follows:
(1) The invention has flexibility and sensitivity, is not limited by places and cost, and has strong practicability;
(2) The improved garbage overflow detection model is easier to be deployed in any intelligent equipment, is lighter, reduces the calculation complexity of the model, improves the detection speed and the detection precision, is easy to detect in real time, and provides more accurate information.
Drawings
FIG. 1 is a schematic flow chart of the method of the present invention;
FIG. 2 is a schematic structural view of a lightweight improved garbage can overflow detection model in the invention;
FIG. 3 is a schematic diagram of the operation of the hardware device of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be further described with reference to the accompanying drawings.
Please refer to fig. 1, which is a flowchart illustrating a method according to the present invention.
The invention provides a lightweight detection method for overflow of a garbage can, which comprises the following steps:
step S101, constructing a garbage can and a garbage sample data set;
the sample data set in step S101 includes a certain number of garbage cans and garbage data, and is divided into positive samples and negative samples. The garbage bin and the garbage data can be acquired according to network pictures in reality.
In this embodiment, 1991 pictures are collected according to network collection and real life, and pictures containing garbage cans and garbage are taken as positive samples, and pictures without garbage cans and garbage are taken as negative samples, so as to form a dataset containing 1991 pictures.
Step S102, labeling a sample data set and dividing the sample data set into a training set, a verification set and a test set according to a proportion;
the step S102 uses a picture marking tool to mark the sample data set, and generates a corresponding mark file, where the mark file includes: file format, file content. The file format may be XML, and the file content is category and position information of the target, wherein the position information of the target comprises upper left corner and lower right corner coordinate information of the real target. Dividing the marked sample data set into a training set, a verification set and a test set according to a preset proportion.
Here, the marking tool may be, but is not limited to, a picture marking tool, such as a wizard marking assistant software; the predetermined ratio is a given value, for example 8:1:1.
In this embodiment, the collected garbage can and garbage picture are respectively marked as a garbage can (TrashCan) and garbage (track), and the position and the corresponding category of the garbage can or garbage in the image are marked by using the eidolon marking assistant software; and dividing the garbage bin, the garbage picture and the corresponding annotation file into the training set, the verification set and the test set according to the ratio of 8:1:1.
Step S103, constructing a garbage can overflow detection model, lightening the garbage can overflow detection model, and combining an E-ELAN module to enhance the network learning capability to obtain a lightweight improved garbage can overflow detection model; in step S103, the lightweight modified trash can overflow detection model includes: the device comprises an input module, a back bone module, a Neck module and a Head module; the input module is used for preprocessing an input image; the backstone module is used for extracting deep features of the image; the Neck module is used for improving the feature extraction capability; the Head module is used for evaluating a lightweight improved garbage can overflow detection model.
The lightweight improved garbage can overflow detection model is built based on a deep learning framework, and the deep learning framework is not limited to a Pytorch framework.
In this embodiment, a Pytorch framework is used for the construction. Referring to fig. 2, fig. 2 is a schematic structural diagram of a lightweight improved garbage can overflow detection model according to the present invention.
It should be noted that, the input module is used for preprocessing the image; specifically, the input module scales each image according to the preset grid size 640×640, and then unifies and normalizes each image. Meanwhile, data enhancement processing, such as a Mosaic method, is adopted, so that the accuracy and the training speed of the model are improved.
It should be noted that, in the construction process of the backhaul module, the grouped convolution IGCV3 and E-ELAN modules replace part of the C3 modules and part of the Conv modules in the traditional backhaul module, so as to form a sixteen-layer network structure of the new backhaul module, as follows:
a first layer: conv module, 32 convolution kernels with the size of 6×6, step length of 2, obtaining data with the characteristic of 320×320×32;
a second layer: the Conv module, 64 convolution kernels with the size of 3×3 and the step length of 2, obtains data with the characteristics of 160×160×64;
third layer: repeating 3C 3 modules, 64 convolution kernels with the size of 3×3 and the step length of 1 to obtain data with the characteristics of 160×160×64;
fourth layer: the Conv module is used for obtaining data with the characteristics of 80 multiplied by 128, wherein the size of the 128 convolution kernels is 3 multiplied by 3, and the step length is 2;
fifth layer: repeating 6C 3 modules, 128 convolution kernels with the size of 3×3 and the step length of 1 to obtain data with the characteristic of 80×80×128;
sixth layer: the IGCV3 module, 128 convolution kernels, the step length of 2 and the spread spectrum coefficient of 2, obtains data with the characteristic of 40 multiplied by 128;
seventh layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 64, wherein 64 convolution kernels with the size of 1 multiplied by 1 and the step length of 1;
eighth layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 64, wherein 64 convolution kernels with the size of 1 multiplied by 1 and the step length of 1;
ninth layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 64, wherein 64 convolution kernels with the size of 3 multiplied by 3 are obtained, and the step length is 1;
tenth layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 64, wherein 64 convolution kernels with the size of 3 multiplied by 3 are obtained, and the step length is 1;
eleventh layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 64, wherein 64 convolution kernels with the size of 3 multiplied by 3 are obtained, and the step length is 1;
twelfth layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 64, wherein 64 convolution kernels with the size of 3 multiplied by 3 are obtained, and the step length is 1;
thirteenth layer: connecting the output of the seventh, eighth, tenth and twelfth layers of the network structure of the new backhaul module, and outputting data with the characteristics of 40×40×256
Fourteenth layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 256, wherein 256 convolution kernels with the size of 1 multiplied by 1 and the step length of 1;
fifteenth layer: the IGCV3 module is used for obtaining data with characteristics of 20 multiplied by 512, wherein 512 convolution kernels are provided, the step length is 2, and the spread spectrum coefficient is 2;
sixteenth layer: SPPF module, 3×3 pooling window, step size 1, yields data featuring 20×20×512.
Here, the SPPF module pools the spatial pyramid, solving the problem of non-uniform input image size.
It can be appreciated that the backbox module is used for extracting features in the lightweight modified garbage can overflow detection model, and the original lightweight modified garbage can overflow detection model is composed of a convolution Conv and a C3 module; the improved lightweight improved garbage can overflow detection model has lighter weight and higher precision, and particularly, the grouping convolution IGCC V3 and E-ELAN modules replace part of modules in the backstone module; the grouping convolution IGCV3 module is used for light network results, and the structure is that 1X 1 point convolution is firstly carried out and divided into two groups for dimension rising; then carrying out 3×3 deep packet convolution, dividing the deep packet convolution into a plurality of groups for extracting features; finally, 1×1 point convolution is performed, and the two groups are used for dimension reduction. The E-ELAN module is used for improving the model performance and improving the network learning capacity, the result is divided into two branches, the first branch is subjected to 1X 1 convolution to change the channel number, the second branch is subjected to 1X 1 convolution to change the channel number, and then is subjected to four 3X 3 convolution modules to extract the characteristics.
Therefore, the invention can improve partial modules of the backbox module in the garbage can overflow detection model based on partial replacement of the grouping convolution IGCC V3 and the E-ELAN module, lighten the network under the condition of ensuring the precision, and has fewer calculation parameters, thereby improving the detection efficiency and sensitivity.
Note that the negk module. Fusing the grouped convolved IGCV3 into the lightweight modified trash can overflow detection model, constructing the lighter weight trash overflow detection model, comprising: and replacing part of C3 modules in the Neck module with the grouping convolution IGCV3 module to form a new Neck module.
Here, the back module follows the back box module, i.e. the final output of the back box module serves as input to the back module. And replacing all C3 modules in the Neck module with the IGCC 3 model to form a new fifteen-layer network structure of the Neck module, wherein the fifteen-layer network structure is as follows:
a first layer: the Conv module is used for obtaining data with the characteristics of 20 multiplied by 256, wherein 256 convolution kernels with the size of 1 multiplied by 1 and the step length of 1;
a second layer: up-sampling by 2 times to obtain data with the characteristics of 40 multiplied by 256;
third layer: connecting the second layer of the new network structure of the Neck module with the output of the fourteenth layer of the new network structure of the Backbone module, and outputting data with the characteristics of 40 multiplied by 512;
fourth layer: the IGCV3 module, 256 convolution kernels, the step length of 1 and the spread spectrum coefficient of 2, obtain data with the characteristics of 40 multiplied by 256;
fifth layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 128, wherein the number of the convolution kernels with the size of 1 multiplied by 1 is 128, the step length is 1, the sign coefficient is 2;
sixth layer: up-sampling by 2 times to obtain data featuring 80×80×128;
seventh layer: connecting the sixth layer of the network structure of the new back module with the output of the fifth layer of the network structure of the new back module, and outputting data with the characteristics of 80 multiplied by 256;
eighth layer: the IGCV3 module, 128 convolution kernels, the step length of 1 and the spread spectrum coefficient of 2, obtain the data with the characteristic of 80 multiplied by 128;
ninth layer: the Conv module is used for obtaining data with the characteristics of 40 multiplied by 128, wherein the size of the 128 convolution kernels is 3 multiplied by 3, and the step length is 2;
tenth layer: connecting the output of the ninth layer and the fifth layer of the network structure of the new Neck module, and outputting data with the characteristics of 40 multiplied by 256;
eleventh layer: the IGCV3 module, 256 convolution kernels, the step length of 1 and the spread spectrum coefficient of 2, obtain data with the characteristics of 40 multiplied by 256;
twelfth layer: the Cnov module obtains data with characteristics of 20 multiplied by 256, wherein the 256 convolution kernels are 3 multiplied by 3, and the step length is 2;
thirteenth layer: connecting the twelfth layer of the network structure of the new Neck module with the output of the first layer, and outputting data with the characteristics of 20 multiplied by 512;
fourteenth layer: the IGCV3 module is used for obtaining data with characteristics of 20 multiplied by 512, wherein 512 convolution kernels are provided, the step length is 1, the spread spectrum coefficient is 2;
fifteenth layer: the Detect module obtains data featuring 80×80, 40×40 and 20×20, respectively, for detecting targets at different sizes.
It can be understood that, by extracting the features of the objects by multi-layer convolution, the feature sizes finally output by the negk module are respectively 80×80, 40×40 and 20×20 data, and different sizes of grid feature data are used for detecting the objects with different sizes, specifically, 20×20 feature maps predict large-sized objects, 20×20 and 40×40 feature maps together predict medium-sized objects, and 20×20, 40×40 and 80×80 feature maps predict small-sized objects.
It can be appreciated that the negk module is configured to enhance the capability of extracting features, and includes SPPF and PANet structures in the lightweight retrofit garbage can overflow detection model; the grouping convolution IGCC 3 module replaces all C3 modules in the Neck module in the lightweight improved garbage can overflow detection model. Therefore, the invention can replace the C3 module in the Neck module in the lightweight improved garbage can overflow detection model based on the grouping convolution IGCC 3 module, reduce the calculated amount while maintaining the model precision, meet the real-time requirement, and is suitable for being deployed in an intelligent robot for real-time detection.
The Head module is used for measuring the quality of the predicted result of the garbage overflow detection model, and comprises a classification loss function, a positioning loss function and a confidence loss function; here, the classification loss function is used to calculate classification losses of the anchor frame and the target frame; the positioning loss function is used for calculating errors of the prediction frame and the target frame, and can determine an average precision mean value (Mean Average Precision, mAP) of the garbage overflow detection model; the confidence loss function is used for calculating the confidence of the garbage overflow detection model.
Step S104, training, verifying and detecting the lightweight improved garbage can overflow detection model by utilizing the training set, the verification set and the test set;
the step S104 includes: training the lightweight modified garbage overflow detection model using the training set; verifying the trained garbage overflow detection model by using the verification set; and detecting the garbage overflow detection model by using a test set. In this embodiment, the specific steps are as follows:
step S104-1, training and verifying the model. Training can set up the round in advance, based on training set, training lightweight improves garbage bin overflow detection model. Typically, 300 rounds or 600 rounds are set to achieve a convergence state. It will be appreciated that after each round of training and verification, the mAP of the current garbage overflow detection model is obtained, so that after the training is finished, the mAP of the final lightweight improved garbage overflow detection model is obtained, and the final mAP is the optimal mAP of the lightweight improved garbage overflow detection model on the verification set.
Step S104-2, testing the model. And predicting the test set to be identified by using the verified lightweight improved garbage can overflow detection model to obtain a prediction result on the test set. And setting an accuracy threshold in advance, and if the accuracy is smaller than a preset accuracy threshold, stopping training and outputting the lightweight improved garbage overflow detection model. Here, the preset accuracy threshold is a given value, for example 80%.
Step S105, outputting a fully trained garbage can overflow detection model based on a predicted result, and deploying the model to the intelligent robot;
the step S105 includes deploying the output trained garbage can overflow detection model to an intelligent robot. And (3) giving a prediction accuracy threshold, and if the final prediction result reaches the threshold, using the model. The model is deployed to the intelligent robot, so that the model can be ensured to run.
Step S106, the intelligent robot is used for inspecting the area to be detected, and real-time detection is carried out by using a trained garbage can overflow detection model;
the step S106 includes: setting a region to be detected in advance, inspecting the region to be detected by the intelligent robot, and detecting in real time by using a trained garbage can overflow detection model; the set area to be detected is a range and can be a community or an intersection.
In this embodiment, a user plans a community as a region to be detected, patrols and examines the intelligent robot in the community, and the model analyzes the video stream in real time and detects each frame in real time to obtain the garbage can and the garbage position.
Step S107, if the garbage can is detected, judging whether the garbage can overflows or not according to a formulated garbage overflow rule, if so, reporting a garbage overflow result in real time; otherwise, no treatment is performed.
The step S107 includes formulating a garbage overflow rule. When the garbage bin and the garbage position are detected, judging whether garbage overflows or not according to a garbage overflow rule, if yes, reporting to a community management center, and informing relevant staff to process in time.
In this embodiment, the garbage overflow rule may be that garbage is detected in a limited area around the garbage can, that is, it is determined that garbage overflows, where the height of the garbage can is h, the width is w, and the limited area is within 1/2h above and below the garbage can and w ranges on the left and right sides of the garbage can.
Referring to fig. 3, fig. 3 is a schematic working diagram of a hardware device according to an embodiment of the present invention, where the hardware device specifically includes: a lightweight garbage can overflow detection device 401, a processor 402 and a storage device 403.
Dustbin overflow lightweight detection device 401: the garbage can overflow lightweight detection device 401 realizes the garbage can overflow lightweight detection method.
Processor 402: the processor 402 loads and executes instructions and data in the storage device 403 to implement the method for lightweight detection of garbage can overflow.
Storage device 403: the storage device 403 stores instructions and data; the storage device 403 is configured to implement the method for detecting the overflow of the garbage can in a lightweight manner.
The beneficial effects of the invention are as follows:
(1) The invention has flexibility and sensitivity, is not limited by places and cost, and has strong practicability;
(2) The improved garbage overflow detection model is easier to be deployed in any intelligent equipment, is lighter, reduces the calculation complexity of the model, improves the detection speed and the detection precision, is easy to detect in real time, and provides more accurate information.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (10)
1. A lightweight detection method for overflow of a garbage can is characterized by comprising the following steps: the method comprises the following steps:
step S101, constructing a garbage can and a garbage sample data set;
step S102, labeling a sample data set and dividing the sample data set into a training set, a verification set and a test set according to a proportion;
step S103, constructing a garbage can overflow detection model, lightening the garbage can overflow detection model, and combining an E-ELAN module to enhance the network learning capability to obtain a lightweight improved garbage can overflow detection model;
in step S103, the lightweight modified trash can overflow detection model includes: the device comprises an input module, a back bone module, a Neck module and a Head module; the input module is used for preprocessing an input image; the backstone module is used for extracting deep features of the image; the Neck module is used for improving the feature extraction capability; the Head module is used for evaluating a lightweight improved garbage can overflow detection model;
step S104, training, verifying and detecting the lightweight improved garbage can overflow detection model by utilizing the training set, the verification set and the test set;
step S105, outputting a fully trained garbage can overflow detection model based on a predicted result, and deploying the model to the intelligent robot;
step S106, the intelligent robot is used for inspecting the area to be detected, and real-time detection is carried out by using a trained garbage can overflow detection model;
step S107, if the garbage can is detected, judging whether the garbage can overflows or not according to a formulated garbage overflow rule, if so, reporting a garbage overflow result in real time; otherwise, no treatment is performed.
2. The lightweight detection method for overflow of a garbage can as claimed in claim 1, wherein: the garbage can overflow detection model in step S103 adopts an improved YOLOv5 model.
3. The lightweight detection method for overflow of garbage can as claimed in claim 2, wherein: in step S103, the garbage can overflow detection model is lightweight, specifically, the grouping convolution and the E-ELAN module are fused to the YOLOv5 model, so that the garbage can overflow detection model is lightweight.
4. The lightweight detection method for overflow of a garbage can as claimed in claim 1, wherein: the construction process of the backup module is as follows: and replacing part of the C3 module and the Conv module in the traditional backbond module by using the grouping convolution IGCC 3 and E-ELAN modules to obtain the backbond module.
5. The lightweight detection method for overflow of garbage can as claimed in claim 4, wherein: the structure of the back bone module comprises: a first layer: a Conv module; a second layer: a Conv module; third layer: repeating 3C 3 modules; fourth layer: a Conv module; fifth layer: repeating 6C 3 modules; sixth layer: IGCV3 modules; seventh layer: a Conv module; eighth layer: a Conv module; ninth layer: a Conv module; tenth layer: a Conv module; eleventh layer: a Conv module; twelfth layer: a Conv module; thirteenth layer: connecting the outputs of the seventh, eighth, tenth and twelfth layers; fourteenth layer: a Conv module; fifteenth layer: IGCV3 modules; sixteenth layer: SPPF module.
6. The lightweight detection method for overflow of a garbage can as claimed in claim 1, wherein: the construction process of the Neck module is as follows: and replacing part of C3 modules in the traditional Neck module with the grouping convolution IGCC 3 module to form the Neck module.
7. The lightweight detection method for overflow of garbage can as claimed in claim 6, wherein: the structure of the Neck module comprises: a first layer: a Conv module; a second layer: up-sampling by 2 times; third layer: connecting the second layer with the output of the fourteenth layer of the network structure of the backhaul module; fourth layer: IGCV3 modules; fifth layer: a Conv module; sixth layer: up-sampling by 2 times; seventh layer: connecting the sixth layer with the output of the fifth layer of the network structure of the backhaul module; eighth layer: IGCV3 modules; ninth layer: a Conv module; tenth layer: connecting outputs of the ninth layer and the fifth layer; eleventh layer: IGCV3 modules; twelfth layer: a Cnov module; thirteenth layer: connecting the twelfth layer with the output of the first layer; fourteenth layer: IGCV3 modules; fifteenth layer: and a Detect module.
8. The lightweight detection method for overflow of a garbage can as claimed in claim 1, wherein: in step S107, the garbage overflow rule specifically includes: detecting garbage in a limited garbage can surrounding area, namely judging that the garbage overflows, wherein the height of the garbage can is h, the width of the garbage can is w, and the limited area is 1/2h above and below the garbage can and within the range of w on the left and right of the garbage can, wherein h and w are preset values.
9. A memory device, characterized by: the storage device stores instructions and data for implementing any one of the garbage can overflow lightweight detection methods described in claims 1-6.
10. The utility model provides a garbage bin overflow lightweight check out test set which characterized in that: comprising the following steps: a processor and a storage device; the processor loads and executes instructions and data in the storage device to implement any one of the garbage can overflow lightweight detection methods of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310679679.0A CN116630787A (en) | 2023-06-08 | 2023-06-08 | Light-weight detection method and device for overflow of garbage can and storage device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310679679.0A CN116630787A (en) | 2023-06-08 | 2023-06-08 | Light-weight detection method and device for overflow of garbage can and storage device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116630787A true CN116630787A (en) | 2023-08-22 |
Family
ID=87616993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310679679.0A Pending CN116630787A (en) | 2023-06-08 | 2023-06-08 | Light-weight detection method and device for overflow of garbage can and storage device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116630787A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407666A (en) * | 2023-12-15 | 2024-01-16 | 深圳市迈睿迈特环境科技有限公司 | Intelligent garbage can parameter analysis and control method and device based on artificial intelligence |
-
2023
- 2023-06-08 CN CN202310679679.0A patent/CN116630787A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407666A (en) * | 2023-12-15 | 2024-01-16 | 深圳市迈睿迈特环境科技有限公司 | Intelligent garbage can parameter analysis and control method and device based on artificial intelligence |
| CN117407666B (en) * | 2023-12-15 | 2024-02-13 | 深圳市迈睿迈特环境科技有限公司 | Intelligent garbage can parameter analysis and control method and device based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108154134B (en) | Pornographic image detection method is broadcast live in internet based on depth convolutional neural networks | |
| CN112560755B (en) | Target detection method for identifying urban exposed garbage | |
| CN111369543B (en) | Rapid pollen particle detection algorithm based on dual self-attention modules | |
| CN106651036A (en) | Air quality forecasting system | |
| CN110826514A (en) | Construction site violation intelligent identification method based on deep learning | |
| CN113240688A (en) | Integrated flood disaster accurate monitoring and early warning method | |
| CN109859163A (en) | A kind of LCD defect inspection method based on feature pyramid convolutional neural networks | |
| CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
| CN111709646B (en) | Air pollution exposure risk evaluation method and system | |
| CN111914634A (en) | Complex-scene-interference-resistant automatic manhole cover type detection method and system | |
| CN116630787A (en) | Light-weight detection method and device for overflow of garbage can and storage device | |
| CN112163367A (en) | Firefly algorithm and cellular automaton fused city expansion simulation prediction method | |
| CN112966925B (en) | Village and town rubbish increment risk analysis system based on remote sensing time sequence change analysis | |
| CN113240208B (en) | Photovoltaic power generation output power prediction method, device and computer readable storage medium | |
| CN117556197B (en) | Typhoon vortex initialization method based on artificial intelligence | |
| CN113033516A (en) | Object identification statistical method and device, electronic equipment and storage medium | |
| CN114882410B (en) | Tunnel dome lamp fault detection method and system based on improved positioning loss function | |
| Floors et al. | Satellite-based estimation of roughness lengths and displacement heights for wind resource modelling | |
| CN116151610A (en) | Method for simulating risk exposure space of disaster-bearing body on underlying surface of heterogeneous city | |
| CN117571056A (en) | Environment protection monitoring method and system based on Internet of things | |
| CN112016845A (en) | DNN and CIM based regional economic benefit evaluation method and system | |
| KR20050063615A (en) | Method for providing surface roughness in geographic information system | |
| CN115830302A (en) | Multi-scale feature extraction and fusion power distribution network equipment positioning identification method | |
| Li et al. | Image recognition and empirical application of desert plant species based on convolutional neural network | |
| CN113344247B (en) | Deep learning-based power facility site selection prediction method and system |
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 |