CN110674731A - Road cleanliness quantification method based on deep learning - Google Patents
Road cleanliness quantification method based on deep learning Download PDFInfo
- Publication number
- CN110674731A CN110674731A CN201910896277.XA CN201910896277A CN110674731A CN 110674731 A CN110674731 A CN 110674731A CN 201910896277 A CN201910896277 A CN 201910896277A CN 110674731 A CN110674731 A CN 110674731A
- Authority
- CN
- China
- Prior art keywords
- road
- garbage
- cleanliness
- training
- deep learning
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
-
- 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
Abstract
The invention relates to a road cleanliness quantification method based on deep learning, and belongs to the technical field of artificial intelligence. The quantification method and a road image collected by image collection equipment installed on a vehicle are used for formulating a control strategy through a depth network model; the quantization steps are as follows: classifying and calibrating the image data set to generate a training set required by a training network; training and testing through a deep network model; and freezing part of the layers of the deep network model, judging the road cleanliness by adopting pre-training model parameters, and determining the power of the dust collector by using the road garbage amount. The road cleanliness quantification method based on deep learning provided by the invention applies deep learning to the quantitative detection of road garbage cleanliness, and under the condition that a server (which can be remotely added) is used, the used deep network model has the characteristics of high operation speed and high accuracy, and can meet the quantitative detection task of road cleanliness under real road conditions.
Description
Technical Field
The invention relates to a road cleanliness quantification method based on deep learning, and belongs to the technical field of artificial intelligence.
Background
The continuous development of artificial intelligence promotes the smart city to be improved day by day, and the road cleaning as one of daily works of each city should be more intelligent and efficient. However, the road-sweeping vehicles have a low degree of intelligence, and especially, there is little research on the relationship between the control of the dust collector power and the road cleanliness. How to enable the road cleaning vehicle to consume less fuel and generate less noise on the basis of ensuring the road surface cleaning effect is one of the evaluation criteria for evaluating the intelligence degree of the road cleaning special vehicle.
The road garbage refers to garbage which can be cleaned by the special road cleaning vehicle and comprises leaves, paper scraps, soil and the like. A large number of theoretical argumentations and experiments find that the extraction of the road garbage based on the traditional image processing method has certain limitations, namely: firstly, the existing image processing method can achieve higher extraction accuracy for solid garbage with better distinguishability, but has poorer distinguishability for garbage which is easy to be mixed into road background, such as sandy soil and the like; secondly, the robustness of the traditional image processing algorithm is poor, and for example, the recognition effect is influenced to a large extent by noise such as pavement cracks and damages; thirdly, the traditional image processing algorithm is poor in universality, for example, the difference between a dry road surface and a wet road surface and between a cement road surface and an asphalt road surface can influence the reliability of the algorithm and even can cause the algorithm to fail.
Disclosure of Invention
The invention provides a road cleanliness quantification method based on deep learning aiming at the defects.
The invention adopts the following technical scheme:
the invention relates to a road cleanliness quantification method based on deep learning, which is characterized in that a control strategy is formulated according to an output value of a trained deep network model based on a road image collected by image collection equipment arranged on a vehicle; the quantization steps are as follows:
1) forming an image data set by the road image acquired by the image acquisition equipment, and classifying and calibrating the image data set to generate a training set required by a training network;
2) training and testing the training set in the step 1) through a deep network model built on a server;
3) freezing part of the layers of the deep network model, adopting the parameters of the deep network pre-training model trained on the large-scale data set, and then training the rest layers; obtaining a deep network model K after training is finished;
4) and the obtained depth network model K is used for quantitatively detecting and judging the road cleanliness of the road rubbish, the road image acquired in real time is input into the depth network model K, the output result is fed back to a dust collector control device of the road cleaning vehicle, and the power of a dust collector is determined according to the quantity of the road rubbish.
The road cleanliness quantification method based on deep learning is characterized in that in the step 1), a road image data set is set as that different quantities of garbage image samples on different roads are collected, and the image sample grades are divided into garbage-free, small garbage, medium garbage and large garbage according to the road garbage coverage area ratio grade;
the deep learning network aims at making a no-garbage image sample, a small amount of garbage image samples, a medium amount of garbage image samples and a large amount of garbage image samples into a training set for training a deep network model.
According to the road cleanliness quantification method based on deep learning, the false alarm rate and the missing report rate are reduced by adopting a voting mechanism in the step 4) for quantitatively detecting and judging the road cleanliness of the road rubbish. The voting mechanism adopted by the method is as follows: the algorithm can process a plurality of frames of images every second, so that the result with the highest repetition can be selected and output from a certain number of ranges of image frames as the identification result of the period.
The road cleanliness quantification method based on deep learning comprises the following strategies of determining the power of a dust collector by using the road garbage amount in the step 4): the larger the amount of road refuse, the larger the power of the dust collector.
According to the road cleanliness quantification method based on deep learning, the image acquisition equipment is a vehicle-mounted camera, the vehicle-mounted camera is mounted at the head of a vehicle, and an image acquisition end of the vehicle-mounted camera is vertically downward and perpendicular to the ground.
Advantageous effects
According to the road cleanliness quantification method based on deep learning, provided by the invention, the division standard of a data set is set to be the size of the garbage amount, and the data set is divided into four types of no garbage, a small amount of garbage, a medium amount of garbage and a large amount of garbage, and the deep learning can be used for classification with strong data universality through verification.
The road cleanliness quantification method based on deep learning provided by the invention applies deep learning to the quantitative detection of road garbage cleanliness, and under the condition that a server (which can be remotely added) is used, the used deep network model has the characteristics of high operation speed and high accuracy, and can meet the quantitative detection task of road cleanliness under real road conditions.
According to the road cleanliness quantification method based on deep learning, provided by the invention, a migration learning method is adopted, a freezing layer is designed according to the characteristics of a network and a data set, the size of the required data set is greatly reduced, and the training result of a model meets the requirement.
Drawings
Figure 1 is a flow chart of the road cleanliness quantification method based on deep learning of the present invention,
FIG. 2 is a schematic diagram of the installation position of a vehicle-mounted camera of the road cleaning special vehicle adopted by the road cleanliness quantification method based on deep learning of the invention;
FIG. 3 is an example of road garbage part data classification of the road cleanliness quantification method based on deep learning according to the present invention.
Detailed Description
In order to make the purpose and technical solution of the embodiments of the present invention clearer, the technical solution of the embodiments of the present invention will be clearly and completely described below with reference to the drawings of the embodiments of the present invention. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the described embodiments of the invention without any inventive step, are within the scope of protection of the invention.
As shown in fig. 1: a road cleanliness quantification method based on deep learning comprises the following specific steps:
step one, a vertically downward camera is installed at the head of the road cleaning vehicle and used for collecting road images in real time, and road images obtained by the camera are connected into a vehicle-mounted industrial personal computer and finally connected into a background server.
The installation requirement of the vehicle-mounted camera is as follows: 1. the camera is arranged at the head of the vehicle; 2. the camera is vertically downward and is arranged at the rearview mirror of the vehicle; the installation schematic is shown in fig. 2;
and step two, acquiring a real pavement image data set of the road cleaning vehicle through the vehicle-mounted camera, and classifying and calibrating the data set. Collecting different quantities of garbage image samples on different roads, and classifying the garbage image samples into four types of garbage-free garbage, small garbage, medium garbage and large garbage according to the road garbage coverage area ratio grade; an example of the road trash portion data classification is shown in fig. 3;
and step three, training and testing the deep learning model based on the road garbage quantification data set. Firstly, dividing a data set into a training set and a testing set according to a certain proportion, freezing the number of layers of a deep network model, adopting deep network pre-training model parameters trained on a large-scale data set, and then training the rest of layers; and obtaining a deep network model K after the training is finished.
And step four, after model training is completed, the model is used for quantitative detection of road garbage, and image data acquired by the vehicle-mounted camera in the step 1) is input into the model to judge the road cleanliness. Meanwhile, a 'voting' mechanism is adopted to reduce the false alarm rate and the false negative rate. The 'voting' mechanism is that because the algorithm can process multi-frame images every second, the result with the highest repetition can be selected and output from a certain number of image frames as the identification result of the period; the output of the deep network model is: the road cleanliness can be judged according to the quantity of the garbage output by the deep network.
And step five, feeding the output result back to a dust collector control device of the road cleaning vehicle, and determining the power of the dust collector according to the road garbage amount.
The relationship between the power of the dust collector and the quantity of road garbage is as follows: the larger the amount of road garbage, the larger the power of the dust collector.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (5)
1. A road cleanliness quantification method based on deep learning is characterized in that: the quantification method and the method are based on road images collected by image collection equipment installed on a vehicle, and a control strategy is formulated according to the output value of a trained depth network model; the quantization steps are as follows:
1) forming an image data set by road images acquired by image acquisition equipment, and classifying and calibrating the image data set by a deep learning network to generate a training set required by the training network;
2) training and testing the training set in the step 1) through a deep network model built on a server;
3) freezing part of the layers of the deep network model, adopting the parameters of the deep network pre-training model trained on the large-scale data set, and then training the rest layers; obtaining a deep network model K after training is finished;
4) and the obtained depth network model K is used for quantitatively detecting and judging the road cleanliness of the road rubbish, the road image collected in real time is input into the depth network model K, the output result is fed back to a dust collector control device of the road cleaning vehicle, and the power of a dust collector is determined according to the quantity of the road rubbish.
2. The deep learning-based road cleanliness quantifying method according to claim 1, characterized in that: the road image data set of the step 1) is set as that different quantities of garbage image samples on different roads are collected, and the image samples are classified into garbage-free, small garbage, medium garbage and large garbage according to the road garbage coverage area ratio grade;
the deep learning network aims at making a no-garbage image sample, a small amount of garbage image samples, a medium amount of garbage image samples and a large amount of garbage image samples into a training set for training a deep network model.
3. The deep learning-based road cleanliness quantifying method according to claim 1, characterized in that: in the step 4), the road rubbish quantitative detection is adopted to judge the road cleanliness, and a voting mechanism is adopted to reduce the false alarm rate and the false alarm rate; the voting mechanism is adopted to process a plurality of frames of images per second, and the result with the highest repetition is selected and output from a certain number of image frames as the identification result of the period.
4. The deep learning-based road cleanliness quantifying method according to claim 1, characterized in that: the strategy for determining the power of the dust collector by the road garbage amount in the step 4) is as follows: the larger the amount of road refuse, the larger the power of the dust collector.
5. The deep learning-based road cleanliness quantifying method according to claim 1, characterized in that: the image acquisition equipment is a vehicle-mounted camera, the vehicle-mounted camera is mounted at the head of the vehicle, and an image acquisition end of the vehicle-mounted camera is vertically and downwards perpendicular to the ground.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910896277.XA CN110674731A (en) | 2019-09-22 | 2019-09-22 | Road cleanliness quantification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910896277.XA CN110674731A (en) | 2019-09-22 | 2019-09-22 | Road cleanliness quantification method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110674731A true CN110674731A (en) | 2020-01-10 |
Family
ID=69077126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910896277.XA Pending CN110674731A (en) | 2019-09-22 | 2019-09-22 | Road cleanliness quantification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110674731A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598476A (en) * | 2020-05-22 | 2020-08-28 | 济源职业技术学院 | Smart city environmental sanitation resource scheduling system based on sparse self-coding and SVM |
| CN112257623A (en) * | 2020-10-28 | 2021-01-22 | 长沙立中汽车设计开发股份有限公司 | Road surface cleanliness judging and automatic cleaning method and automatic cleaning environmental sanitation device |
| CN114951160A (en) * | 2022-04-19 | 2022-08-30 | 北京石油化工学院 | Air pipe dust removal device based on roller type |
| WO2023231021A1 (en) * | 2022-06-02 | 2023-12-07 | 深圳市正浩创新科技股份有限公司 | Target object collection method and device, and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824066A (en) * | 2014-03-18 | 2014-05-28 | 厦门翼歌软件科技有限公司 | Video stream-based license plate recognition method |
| CN109024417A (en) * | 2018-07-24 | 2018-12-18 | 长安大学 | A kind of Intelligent road sweeper and its roadway pollutants recognition methods and control method |
| CN109452914A (en) * | 2018-11-01 | 2019-03-12 | 北京石头世纪科技有限公司 | Intelligent cleaning equipment, cleaning mode selection method, computer storage medium |
| CN109614924A (en) * | 2018-12-07 | 2019-04-12 | 江西洪都航空工业集团有限责任公司 | A kind of garbage on water detection method based on deep learning algorithm |
| CN109746226A (en) * | 2019-02-20 | 2019-05-14 | 南京海得电力科技有限公司 | Photovoltaic panel intelligence purging system for photovoltaic plant O&M |
-
2019
- 2019-09-22 CN CN201910896277.XA patent/CN110674731A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824066A (en) * | 2014-03-18 | 2014-05-28 | 厦门翼歌软件科技有限公司 | Video stream-based license plate recognition method |
| CN109024417A (en) * | 2018-07-24 | 2018-12-18 | 长安大学 | A kind of Intelligent road sweeper and its roadway pollutants recognition methods and control method |
| CN109452914A (en) * | 2018-11-01 | 2019-03-12 | 北京石头世纪科技有限公司 | Intelligent cleaning equipment, cleaning mode selection method, computer storage medium |
| CN109614924A (en) * | 2018-12-07 | 2019-04-12 | 江西洪都航空工业集团有限责任公司 | A kind of garbage on water detection method based on deep learning algorithm |
| CN109746226A (en) * | 2019-02-20 | 2019-05-14 | 南京海得电力科技有限公司 | Photovoltaic panel intelligence purging system for photovoltaic plant O&M |
Non-Patent Citations (1)
| Title |
|---|
| 王平: ""基于深度卷积网的迁移学习技术研究"", 《中国优秀硕士学位论文全文数据库-信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598476A (en) * | 2020-05-22 | 2020-08-28 | 济源职业技术学院 | Smart city environmental sanitation resource scheduling system based on sparse self-coding and SVM |
| CN112257623A (en) * | 2020-10-28 | 2021-01-22 | 长沙立中汽车设计开发股份有限公司 | Road surface cleanliness judging and automatic cleaning method and automatic cleaning environmental sanitation device |
| CN114951160A (en) * | 2022-04-19 | 2022-08-30 | 北京石油化工学院 | Air pipe dust removal device based on roller type |
| WO2023231021A1 (en) * | 2022-06-02 | 2023-12-07 | 深圳市正浩创新科技股份有限公司 | Target object collection method and device, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110674731A (en) | Road cleanliness quantification method based on deep learning | |
| CN103714363B (en) | A kind of motor vehicle exhaust smoke video identification system | |
| Akagic et al. | Pavement crack detection using Otsu thresholding for image segmentation | |
| CN108765404B (en) | A kind of road damage testing method and device based on deep learning image classification | |
| CN106845424B (en) | Pavement remnant detection method based on deep convolutional network | |
| CN103413441B (en) | Road weather condition monitoring method | |
| CN107256635B (en) | Vehicle identification method based on distributed optical fiber sensing in intelligent traffic | |
| CN103345842B (en) | A kind of road vehicle classification system and method | |
| WO2021174857A1 (en) | Method for constructing water damage identification model for asphalt pavement, and identification method and system | |
| CN104008550A (en) | Wafer surface defect feature analysis method and system and wafer surface detect feature classification method and system | |
| CN102682601A (en) | Expressway traffic incident detection method based on optimized support vector machine (SVM) | |
| CN110688922A (en) | Deep learning-based traffic jam detection system and detection method | |
| CN112435463B (en) | Vehicle type and vehicle weight classification method based on road internet of things monitoring | |
| CN101814237A (en) | Road condition identifier based on license plate recognition technology and identification method thereof | |
| CN111553902A (en) | Highway road surface safety monitoring system based on big data | |
| CN103679214A (en) | Vehicle detection method based on online area estimation and multi-feature decision fusion | |
| CN103198707B (en) | A kind of vehicle regulate and control method based on traffic flow character dangerous under fine day situation | |
| CN115311229A (en) | Laser radar-based pavement disease detection and classification method and system and storage medium | |
| CN116289444A (en) | Intelligent pavement disease real-time monitoring and predicting device and method | |
| CN111025285B (en) | Asphalt pavement water damage detection method based on map gray scale self-adaptive selection | |
| CN115439643A (en) | Road disease size analysis and management method based on monocular measurement | |
| CN115527364A (en) | Traffic accident tracing method and system based on radar vision data fusion | |
| CN101075377A (en) | Method for automatically inspecting highway traffic event based on offset minimum binary theory | |
| CN113469097B (en) | Multi-camera real-time detection method for water surface floaters based on SSD network | |
| CN116342895B (en) | Method and system for improving sorting efficiency of renewable resources based on AI (advanced technology attachment) processing |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200110 |
|
| WD01 | Invention patent application deemed withdrawn after publication |