LU500063B1 - Garbage identification and classification system based on improved efficient net - Google Patents
Garbage identification and classification system based on improved efficient net Download PDFInfo
- Publication number
- LU500063B1 LU500063B1 LU500063A LU500063A LU500063B1 LU 500063 B1 LU500063 B1 LU 500063B1 LU 500063 A LU500063 A LU 500063A LU 500063 A LU500063 A LU 500063A LU 500063 B1 LU500063 B1 LU 500063B1
- Authority
- LU
- Luxembourg
- Prior art keywords
- efficientnet
- garbage
- improved
- learning rate
- model
- Prior art date
Links
Classifications
-
- 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
- G06V10/30—Noise filtering
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/06—Recognition of objects for industrial automation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
In view of the technical problems, the present invention provides a garbage classification system based on an improved EfficientNet, and uses an improved EfficientNet classification algorithm to achieve more accurate garbage classification and faster classification speed. The specific steps include: (1) collecting images of all kinds of garbage through a picture collection device and establishing a database; (2) performing image preprocessing on the collected garbage pictures, wherein common preprocessing methods include gray processing, median filter processing and image enhancement, for enhancing features of interest in the images and suppress features not of interest in the images, thereby effectively improving the quality of the images; then dividing preprocessed data; (3) keeping original weights of the EfficientNet on ImageNet data sets using a transfer learning method, training new weights through fine adjustment, and saving a trained EfficientNet model; and (4) classifying and identifying the preprocessed data sets through the improved EfficientNet model so as to obtain classification results.
Description
: LU500063
BACKGROUND Technical Field The present invention relates to the field of target detection, and in particular relates to a garbage identification and classification system based on an improved EfficientNet.
Related Art With the development of the economy, the living standards of people have improved significantly, but all kinds of domestic garbage that follow have also increased at an alarming rate. If a large amount of domestic garbage is not disposed of in time, it will not only affect the appearance of the city, but also damage people's health. In today's advocating of a conservation and environment-friendly society, it is not only necessary to deal with the increasing amount of garbage, but also to recycle the available resources from the garbage.
Garbage classification is an efficient resource recovery method, and it is also a social issue related to people's livelihood and the sustainable development of society.
At present, the problem about garbage classification has become the focus. More and more people are studying garbage classification through deep learning. Among them, artificial neural networks have made great contributions in the field of garbage classification. The predecessors used various neural networks, such the BP neural network, the improved BP, and the convolutional neural network, for garbage classification. But these neural networks still have some problems: (1) The effect of garbage identification and classification is poor. A variety of neural networks are used to classify garbage, and the application of these neural networks in garbage classification has performed very well. But the final garbage classification effect is not good enough. Some classification systems have very low classification accuracy.
(2) The identification speed of a garbage classification algorithm is slow. Due to the large changes in the shape and color of garbage, it is not easy to manually extract category features, and the amount of data is large. As a result, the classification algorithm is very slow in garbage classification and identification, and cannot meet the real-time requirements of embedded devices on the production site.
Therefore, it is necessary to develop a garbage classification system with high classification effect and fast running speed.
SUMMARY In view of the technical problems, the present invention provides a garbage identification and classification system based on an improved EfficientNet, and uses an improved EfficientNet classification algorithm to achieve more accurate garbage classification and faster classification speed. In order to solve the above technical problems, the technical solution adopted by the present invention 1s that a garbage identification and classification system based on an improved EfficientNet specifically includes the following steps (the related flowchart 1s shown in FIG. 1): (1) collecting images of all kinds of garbage through a picture collection device and establishing a database; (2) performing image preprocessing on the collected garbage pictures, preprocessing the collected garbage images by gray processing, median filter processing, image enhancement, etc., and then dividing preprocessed data sets; (3) keeping original weights of the EfficientNet on ImageNet data sets using a transfer learning method, training new weights through fine adjustment, and saving a trained EfficientNet model; and (4) classifying and identifying the preprocessed data sets using the improved EfficientNet model so as to obtain classification and identification results.
In the garbage identification and classification system based on the improved EfficientNet provided by the present invention, some optimizations are performed on the algorithm of EfficientNet. Specifically, the learning rate of the entire algorithm is adjusted using the cosine learning rate decay method with Warmup. Common ways to change the learning rate include exponential decay, gradual decay, and polynomial decay. Since the weights of the model are initialized randomly at the beginning of training, choosing a larger learning rate at this time may bring about instability of the model. Therefore, at the beginning of training, a smaller learning rate is used to warm up the learning rate. When the model is stable, the learning rate is modified to the preset learning rate for training, so that the model convergence rate is faster and the model effect is better. Then, the learning rate is reduced by a cosine annealing method, and the accuracy is improved. The combination of the two methods improves the identification rate and speed of the system.
The specific process of improvement is as follows: The improvement is mainly to change the computing mode of the learning rate. The new learning rate computing method is mainly divided into two stages: a Warmup stage and a cosine annealing stage.
’ LU500063 Warmup stage: (1) First, a smaller warmup learning rate is selected. The network model is trained. Some epochs or steps (such as 4 epochs, 10,000 steps) are trained. Then each step is increased a little bit. The results of the training are observed.
(2) When the trained model is stable, the learning rate is increased to the preset learning rate.
Cosine annealing stage: The learning rate after the Warmup stage is gradually reduced from the initial value according to the cosine function. The training results are observed.
After the two stages are combined, the specific learning rate is computed as follows: (1) First, the warmup initial learning rate, the warmup basic learning rate, epochs (total number of iterations), Warmup epoch (number of warmup iterations), batch size, number of samples, and number of reserved steps are initialized.
(2) The learning rate of Warmup is computed according to a formula.
(3) After Warmup is over, the cosine annealing learning rate is computed according to a formula.
(4) The cosine annealing learning rate is further computed and judged. If the number of reserved steps is greater than 0, whether the current number of steps is greater than the number of warmup steps plus the number of reserved steps is judged. If yes, the step returns to the learning rate computed above. If not, the basic learning rate after Warmup is used.
(5) When the number of warmup steps is greater than 0, if the current number of steps is less than the number of warmup steps, then the step returns to the current learning rate of Warmup, or otherwise directly returns to the cosine annealing computation of the second step.
(6) If the number of steps currently reached is greater than the total number of steps finally, the learning rate returns to 0, or otherwise the step returns to the currently computed learning rate.
BRIEF DESCRIPTION OF THE DRAWINGS The following describes the implementations of the present invention in detail with reference to accompanying drawings.
FIG. 1 is a schematic diagram of a process of a garbage identification and classification system based on an improved EfficientNet.
FIG. 2 is a flowchart of an improved EfficientNet algorithm.
FIG. 3 is a schematic diagram of cosine annealing with Warmup.
FIG. 4 is a comparison diagram of cosine learning rate decay and gradual learning rate decay.
Claims (5)
1. À garbage identification and classification system based on an improved EfficientNet, specifically comprising the following steps: (1) collecting images of all kinds of garbage through a picture collection device and establishing a database; (2) performing image preprocessing on the collected garbage pictures, preprocessing the collected garbage images by gray processing, median filter processing, image enhancement, etc., and then dividing preprocessed data sets; (3) keeping original weights of the EfficientNet on ImageNet data sets using a transfer learning method, training new weights through fine adjustment, and saving a trained EfficientNet model; and (4) classifying and identifying the preprocessed data sets using the improved EfficientNet model so as to obtain classification and identification results.
2. The garbage identification and classification system based on the improved EfficientNet according to claim 1, wherein in step (1), original images of garbage are collected through a hardware device such as a camera, a pick-up head and a mobile phone, and the pictures are collected to construct a picture database.
3. The garbage identification and classification system based on the improved EfficientNet according to claim 1, wherein in step (2), the pictures in the original database are preprocessed, useless information such as noise in the pictures is eliminated through median filtering and picture enhancement methods, and the picture enhancement methods usually adopt a frequency domain method and a spatial domain method.
4. The garbage identification and classification system based on the improved EfficientNet according to claim 1, wherein in step (3), the EfficientNet trained on the ImageNet data set is transplanted by the transfer learning method, and then the preprocessed pictures are used to retrain the network to remodify the weights of the network to meet the conditions for classifying garbage.
5. The garbage identification and classification system based on the improved EfficientNet according to claim 1, wherein in step (4), on the basis of the transfer learning method, the EfficientNet is improved, and the improved EfficientNet is used to classify garbage; the improvement content is mainly to improve a learning rate of the network, and the learning rate is modified by a cosine learning rate decay method with Warmup, so as to improve identification accuracy and running speed of the EfficientNet model; specifically, the
) LU500063 learning rate of an entire algorithm is adjusted using a cosine annealing learning rate scheduling method with Warmup; since the weights of the model are initialized randomly at the beginning of training, choosing a larger learning rate at this time may bring about instability of the model; therefore, at the beginning of training, a smaller learning rate is used to warm up the learning rate; when the model is stable, the learning rate is modified to a preset learning rate for training, so that a model convergence rate is faster and a model effect is better; then, the learning rate is reduced by a cosine annealing method, and the accuracy is improved; and the combination of the two methods improves the identification rate and speed of the system.
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011303297.0A CN112488167A (en) | 2020-11-19 | 2020-11-19 | Rubbish identification and classification system based on improved EfficientNet network |
Publications (1)
Publication Number | Publication Date |
---|---|
LU500063B1 true LU500063B1 (en) | 2022-05-19 |
Family
ID=74932046
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
LU500063A LU500063B1 (en) | 2020-11-19 | 2021-04-20 | Garbage identification and classification system based on improved efficient net |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN112488167A (en) |
LU (1) | LU500063B1 (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113657143B (en) * | 2021-06-25 | 2023-06-23 | 中国计量大学 | Garbage classification method based on classification and detection combined judgment |
CN113610163B (en) * | 2021-08-09 | 2024-08-09 | 安徽工业大学 | Knowledge distillation-based lightweight apple leaf disease identification method |
Family Cites Families (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110852420B (en) * | 2019-11-11 | 2021-04-13 | 北京智能工场科技有限公司 | Garbage classification method based on artificial intelligence |
CN110884791A (en) * | 2019-11-28 | 2020-03-17 | 石家庄邮电职业技术学院(中国邮政集团公司培训中心) | Vision garbage classification system and classification method based on TensorFlow |
CN110786824B (en) * | 2019-12-02 | 2021-06-15 | 中山大学 | Coarse marking fundus oculi illumination bleeding lesion detection method and system based on bounding box correction network |
CN111079639B (en) * | 2019-12-13 | 2023-09-19 | 中国平安财产保险股份有限公司 | Method, device, equipment and storage medium for constructing garbage image classification model |
CN111611949A (en) * | 2020-05-25 | 2020-09-01 | 陈金山 | Intelligent urban water area garbage density detection method based on CIM and block chain |
CN111783841B (en) * | 2020-06-09 | 2023-08-04 | 中科院成都信息技术股份有限公司 | Garbage classification method, system and medium based on migration learning and model fusion |
CN111723772B (en) * | 2020-06-30 | 2024-03-19 | 平安国际智慧城市科技股份有限公司 | Perishable garbage identification method and device based on image identification and computer equipment |
CN111768406A (en) * | 2020-08-05 | 2020-10-13 | 赛维森(广州)医疗科技服务有限公司 | Cell image processing method, device, equipment and storage medium |
-
2020
- 2020-11-19 CN CN202011303297.0A patent/CN112488167A/en active Pending
-
2021
- 2021-04-20 LU LU500063A patent/LU500063B1/en active IP Right Grant
Also Published As
Publication number | Publication date |
---|---|
CN112488167A (en) | 2021-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110399929B (en) | Fundus image classification method, fundus image classification apparatus, and computer-readable storage medium | |
CN109815826B (en) | Method and device for generating face attribute model | |
WO2018188453A1 (en) | Method for determining human face area, storage medium, and computer device | |
CN104143079B (en) | The method and system of face character identification | |
LU500063B1 (en) | Garbage identification and classification system based on improved efficient net | |
CN100367294C (en) | Method for dividing human body skin area from color digital images and video graphs | |
CN106951856A (en) | Bag extracting method of expressing one's feelings and device | |
CN110458038A (en) | The cross-domain action identification method of small data based on double-strand depth binary-flow network | |
CN111860587B (en) | Detection method for small targets of pictures | |
CN109902558A (en) | A kind of human health deep learning prediction technique based on CNN-LSTM | |
CN111008639B (en) | License plate character recognition method based on attention mechanism | |
CN109902757B (en) | Face model training method based on Center Loss improvement | |
CN113762009B (en) | Crowd counting method based on multi-scale feature fusion and double-attention mechanism | |
CN104881852B (en) | Image partition method based on immune clone and fuzzy kernel clustering | |
CN111723239A (en) | Multi-mode-based video annotation method | |
WO2022127814A1 (en) | Method and apparatus for detecting salient object in image, and device and storage medium | |
CN117392484A (en) | Model training method, device, equipment and storage medium | |
CN110503092B (en) | Improved SSD monitoring video target detection method based on field adaptation | |
CN103903017B (en) | A kind of face identification method based on adaptive soft histogram local binary patterns | |
CN111126155A (en) | Pedestrian re-identification method for generating confrontation network based on semantic constraint | |
CN112488165A (en) | Infrared pedestrian identification method and system based on deep learning model | |
CN117095471A (en) | Face counterfeiting tracing method based on multi-scale characteristics | |
CN112200008A (en) | Face attribute recognition method in community monitoring scene | |
Guo | [Retracted] System Analysis of the Learning Behavior Recognition System for Students in a Law Classroom: Based on the Improved SSD Behavior Recognition Algorithm | |
CN202855297U (en) | Background music control device based on expression |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
FG | Patent granted |
Effective date: 20220519 |