CN111461000A - Intelligent office garbage classification method based on CNN and wavelet analysis - Google Patents
Intelligent office garbage classification method based on CNN and wavelet analysis Download PDFInfo
- Publication number
- CN111461000A CN111461000A CN202010246052.2A CN202010246052A CN111461000A CN 111461000 A CN111461000 A CN 111461000A CN 202010246052 A CN202010246052 A CN 202010246052A CN 111461000 A CN111461000 A CN 111461000A
- Authority
- CN
- China
- Prior art keywords
- garbage
- classification
- image
- cnn
- wavelet analysis
- 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 136
- 238000000034 method Methods 0.000 title claims abstract description 68
- 238000004458 analytical method Methods 0.000 title claims abstract description 39
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 30
- 239000013598 vector Substances 0.000 claims abstract description 27
- 238000007637 random forest analysis Methods 0.000 claims abstract description 26
- 238000011176 pooling Methods 0.000 claims abstract description 13
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims abstract description 8
- 230000005540 biological transmission Effects 0.000 claims abstract description 6
- 238000013145 classification model Methods 0.000 claims abstract description 4
- 239000002699 waste material Substances 0.000 claims description 10
- 230000001133 acceleration Effects 0.000 claims description 7
- 238000000513 principal component analysis Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000003066 decision tree Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 3
- 238000010187 selection method Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 12
- 238000013527 convolutional neural network Methods 0.000 description 32
- 238000005457 optimization Methods 0.000 description 8
- 230000004927 fusion Effects 0.000 description 7
- 238000000605 extraction Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000005484 gravity Effects 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000004064 recycling Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002285 radioactive effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention aims to provide an office garbage intelligent classification method based on CNN and wavelet analysis aiming at special scenes such as offices, which comprises the following implementation processes: collecting garbage image data, performing convolution and pooling processing by using an image preprocessing method considering low brightness and high contrast, and extracting image features; collecting garbage vibration frequency signals, performing wavelet decomposition on the vibration frequency signals based on wavelet analysis, performing denoising processing on the signals, reserving effective signal characteristics, and outputting characteristic vectors; based on a random forest algorithm, establishing a garbage classification model by using the feature vectors, and outputting garbage classification signals; transmitting the classification signal to a classification garbage box door control device; feeding back the successfully classified garbage data to the identification and classification system, and updating the sample data set; and transmitting the garbage data failed in classification to the cloud end through wireless transmission, and after the garbage data is artificially marked, feeding the garbage data back to the identification and classification system to update the sample data set.
Description
Technical Field
The invention relates to the technical field of garbage classification, in particular to an intelligent office garbage classification method based on CNN and wavelet analysis.
Background
In modern society, people spend most of the time in daily household and office life, and waste of enterprises and public institutions such as scrapped printing paper, garbage generated in business activities, pollution or radioactive garbage and the like can be generated in daily work of offices. When the garbage classification is about to be popularized, it is important to reduce the time spent on the garbage classification by people. Therefore, the invention provides a method capable of realizing intelligent garbage identification and classification.
With the rise and development of deep learning technology, the computer vision research based on deep learning, such as object classification, face recognition and other related fields, has made a breakthrough progress. Different from the traditional image classification method, the deep learning does not need to artificially extract the features, and the feature learning can be intelligently completed according to the input image. A Convolutional Neural Network (CNN) is an artificial intelligence algorithm of a deep learning technology, and the CNN has become a research hotspot in the field of image processing and pattern recognition at present. In the field of computer vision, methods based on deep learning mostly adopt CNN to process the problem of image classification.
Wavelet analysis is a new signal processing method developed in recent years, and this method is derived from fourier analysis. Wavelets (wavelets), i.e. waves of a small area, have non-zero values only for a very limited section of the interval, rather than being endless as sine and cosine waves do. The wavelet can be translated back and forth along a time axis, and can also be stretched and compressed in proportion to obtain low-frequency and high-frequency wavelets, and the constructed wavelet function can be used for filtering or compressing signals, so that effective signal characteristics in the signals containing noise can be extracted. Therefore, the method combining the CNN and the wavelet analysis has important research value and good development prospect in the field of garbage classification.
The random forest is an important Bagging-based integrated learning method and can be used for classifying, regressing and other problems. Random forests have many advantages as classifiers: in all current algorithms, the method has excellent accuracy; can operate efficiently on large data sets; input samples with high dimensional characteristics can be processed without dimension reduction; the importance of each feature on the classification problem can be evaluated; in the generation process, an unbiased estimation of an internal generation error can be obtained; good results can be obtained for the default value problem.
Disclosure of Invention
The invention aims to provide an office rubbish intelligent classification method based on CNN and wavelet analysis, which can intelligently identify sample rubbish without manually extracting features, realize accurate identification and classification of rubbish by using a dual sensor, feed back identified data to a system and a cloud end to realize automatic update of a sample data set, and filter the edge part of an original signal to eliminate noise information and reconstruct a clearer frequency signal feature graph on the premise of not losing important information components of the frequency signal by using the characteristic of high-low frequency separation of wavelet transformation, thereby improving the accuracy of rubbish identification. The garbage classification and recycling system provided by the invention conforms to the current green concept of garbage classification and recycling treatment, not only effectively improves the classification accuracy, but also can realize machine learning, reduces the labor cost, and has a general popularization value.
In order to achieve the purpose, the invention provides an office rubbish intelligent classification method based on CNN and wavelet analysis, and the technical scheme adopted by the invention is as follows: the method comprises the following steps of collecting image data and vibration frequency data of thrown office garbage by adopting double sensors, carrying out feature extraction on the data based on CNN and wavelet analysis, integrating feature vectors, identifying and classifying by adopting a random forest algorithm, feeding back the identified data to a system and a cloud end, realizing automatic updating of a sample data set, and realizing intelligent garbage classified throwing, wherein the method comprises the following steps:
(1) collecting garbage image data through an image collecting device;
(2) an image preprocessing method is provided, and the acquired garbage image is preprocessed in consideration of low vividness (color and brightness) and high contrast;
(3) performing convolution and pooling on the preprocessed garbage image based on CNN, extracting image features, and outputting feature vectors;
(4) collecting garbage vibration frequency signals through a vibration generator and a vibration data collector;
(5) performing wavelet decomposition on the vibration frequency signal based on a wavelet analysis model, performing denoising processing on the signal, simultaneously reserving effective signal characteristics, and outputting a characteristic vector;
(6) based on a random forest algorithm, establishing a garbage classification model by using the integrated image characteristic vector and the frequency characteristic vector, and outputting a garbage classification signal;
(7) transmitting the classification signal to a classification garbage bin door control device to enable the garbage to fall into a corresponding subarea;
(8) feeding back corresponding feature vector data of successfully classified garbage to the identification and classification system, and updating a sample data set;
(9) and transmitting corresponding image data of the garbage failed to be classified to the cloud end through the wireless transmission module, and after a user or a maintainer marks the image data, feeding the marked image data of the garbage sample back to the recognition and classification system to update the image sample data set.
The invention is further configured to: the image preprocessing system preprocesses the acquired garbage image, and the method provides a garbage region selection algorithm based on the image brightness and contrast. In the method of classifying images by manually extracting features, researchers mostly use color histograms to obtain color features of images. The most common and most effective of these is the HSV color model. The method converts the RGB original image into HSV image to calculate the brightness map. The definition of the luminance map is:
L(x,y)＝0.3×IR(x,y)+0.6×IG(x,y)+0.1×IB(x,y) (1-1)
wherein, IR(x,y)、IG(x,y)、IBAnd (x, y) are red, green and blue components of the RGB original image at (x, y) respectively.
The invention is further configured to: the image preprocessing system preprocesses the acquired garbage image, and based on CNN, the method provides a garbage region selection algorithm based on image brightness and contrast. The main idea is as follows: firstly, an image is divided into various regions, and then a hierarchical value is allocated to each region, so that a contrast map based on the regions is formed. The contrast value of a region is calculated from a global hierarchical value measured by the spatial distance of the current region relative to other regions.
Through above-mentioned technical scheme, its advantage lies in: when classifying the garbage images, the method not only considers the brightness characteristics (color, brightness and the like) of the garbage images, but also considers the contrast ratio of the garbage images, thereby obtaining better classification effect than directly utilizing CNN.
The invention is further configured to: the CNN model adopted by the method comprises the following steps: input layer, convolution layer, pooling layer, full-link layer and output layer. The network first takes an image as input, then uses the interleaved convolutional layer and pooling layer to extract image features, and finally transfers the extracted features to the fully-connected layer.
By the technical scheme, the preprocessed garbage image is subjected to convolution and pooling treatment, and then the characteristic image is extracted; the CNN model constructed by the method has a complete structure, and the recognition rate and the stability are improved.
The invention is further configured to: the vibration generator and the vibration data collector are composed of a force sensor and a piezoelectric acceleration sensor, and acceleration values are obtained by measuring the vibration frequency of the gravity borne by the garbage according to Newton's second law. The realization process is as follows: the garbage falls on the piezoelectric acceleration sensor due to gravity to generate vibration, and vibration frequency data are collected.
The invention is further configured to: based on wavelet analysis, denoising optimization is carried out on vibration frequency data, after a frequency signal is subjected to wavelet transformation, a wavelet coefficient generated by the signal contains important information of the signal, after wavelet decomposition, the wavelet coefficient of an effective signal is large, the wavelet coefficient of noise is small, the wavelet coefficient of the noise is smaller than that of the effective signal, an appropriate threshold value is selected, the wavelet coefficient larger than the threshold value is considered to be the effective signal and should be reserved, and the wavelet coefficient smaller than the threshold value is considered to be the noise generated and is set to be zero, so that the purpose of denoising optimization is achieved.
By the technical scheme, the wavelet denoising optimization is carried out on the original vibration frequency data, and the effective signal characteristics can be kept after the denoising, so that the method is superior to the traditional low-pass filter.
The invention is further configured to: the method realizes the fusion of two characteristics by using a Principal Component Analysis (PCA) method for a high-dimensional characteristic vector after convolution processing of image information and a characteristic vector of garbage vibration after wavelet analysis. The characteristic fusion technology not only fuses the most representative information of a plurality of characteristics, but also can remove some unnecessary or similar information, thereby greatly improving the operation speed and the real-time performance.
The random forest classification algorithm used in the method extracts the generation result of the full-link layer in the CNN algorithm, classifies the generation result and the vibration signal combination after wavelet analysis and denoising, sets the number of random forest decision trees, and obtains the optimal feature set dimension through a sequence forward selection method. Due to the difference of the size of the feature dimension, the selection of the parameters and the feature normalization method, the classification performance of different classifiers is different for different data, so that the best identification result cannot be predicted by using a single classifier. The scheme adopts a random forest to realize the fusion of multiple classifiers. The random forest is an algorithm for gathering a plurality of decision trees together through an integrated learning idea, N classification results can be obtained in the random forest every time a sample image is input, and finally the category with the largest voting times is selected as a final recognition result.
Through the technical scheme, due to the fact that image data of the garbage samples are limited, the samples provided for CNN training are not sufficient, and an overfitting phenomenon is easily caused. And the random forest classification algorithm has good classification precision and classification generalization capability under the conditions of small samples and high dimension. A single prediction problem is solved by combining multiple classifiers. It is able to learn and make predictions independently of each other by generating multiple classifiers. These predictions are finally combined into a single prediction, so the prediction result is better than any single classifier. Therefore, the method uses a random forest classification algorithm to replace a Soft-Max layer in the CNN to train features, and a final recognition and classification system is obtained. The samples are sampled in a database, and the samples are divided into two equal parts, one part is used for training, and the other part is randomly drawn for result testing.
The invention is further configured to: the control device of the classification garbage can door consists of a steering engine and a connecting piece, and the device starts to operate after receiving a classification signal transmitted by the identification and classification system.
Through the technical scheme, the identification and classification system transmits the classification signal to the classification garbage box door control device, so that the garbage falls into the corresponding subarea. The process from garbage putting to intelligent garbage identification and classification recycling is completed.
The invention is further configured to: and after the garbage identification and classification are successfully completed, feeding back corresponding feature vector data of the successfully classified garbage to the identification and classification system, and updating the image sample data set.
The invention is further configured to: after the garbage identification and classification fails, transmitting corresponding image data of the garbage which is failed in classification to a cloud end through a wireless transmission module, and after a user or maintenance personnel marks the image data of the marked garbage sample, feeding the image data of the marked garbage sample back to the identification and classification system to update the image sample data set.
By the technical scheme, the sample image data set is automatically updated by the identification and classification system, the garbage sample images which are successfully identified and unsuccessfully identified are accumulated, the garbage identification type is updated, an autonomous cyclic learning process is formed, the identification precision is improved, and the labor cost is reduced.
According to the invention, the image data and the vibration frequency data of the garbage are acquired by adopting the double sensors, and the data are extracted and optimized based on CNN and wavelet analysis, so that the operation speed and the identification precision of garbage classification are effectively improved; and a classification result feedback mechanism is adopted, so that the machine forms an autonomous cycle learning process, and the labor cost is reduced.
Drawings
FIG. 1 is a flow chart of intelligent garbage classification based on CNN and wavelet analysis provided by the present invention;
FIG. 2 is a flow chart of a garbage image collection and feature extraction system provided by the present invention;
FIG. 3 is a flow chart of the garbage vibration frequency acquisition and feature extraction system provided by the present invention;
FIG. 4 is a flow chart of the garbage recognition and classification system provided by the present invention.
FIG. 5 is a flow chart of the autonomous learning system for garbage recognition provided by the present invention.
Detailed Description
The following detailed description of the present invention is provided in conjunction with the accompanying drawings so that the advantages and features of the present invention will be readily understood by those skilled in the art, and the scope of the present invention will be clearly and clearly defined.
Referring to fig. 1, an embodiment of the present invention includes:
the invention aims to provide an office rubbish intelligent classification method based on CNN and wavelet analysis aiming at special scenes such as offices, wherein a system flow chart is shown in figure 1, the method comprises the steps of carrying out feature extraction on rubbish image data to be classified and vibration frequency data by adopting CNN and wavelet analysis, integrating feature vectors, training, identifying and classifying and releasing by adopting a random forest algorithm, feeding back identified data to a system and a cloud end, and realizing automatic updating of a sample data set.
Fig. 2 is a flowchart of a garbage image collection and feature extraction system, which includes an image collection device, an image preprocessing system, and a CNN model. The realization process is as follows:
collecting garbage image data through an image collecting device; providing a garbage area selection algorithm based on low vividness (color and brightness) and high contrast, and preprocessing the acquired garbage image; and (3) performing convolution and pooling treatment on the preprocessed garbage image based on the CNN, extracting image characteristics and outputting a characteristic vector.
Based on CNN, the method provides a garbage area selection algorithm based on image sharpness. In the method of classifying images by manually extracting features, researchers mostly use color histograms to obtain color features of images. The most common and most effective of these is the HSV color model. The method converts the RGB original image into HSV image to calculate the brightness map. The definition of the luminance map is:
L(x,y)＝0.3×IR(x,y)+0.6×IG(x,y)+0.1×IB(x,y) (1-1)
wherein, IR(x,y)、IG(x,y)、IB(x, y) are the red, green, and blue components of the RGB image at (x, y), respectively.
Based on CNN, the method provides a garbage area selection algorithm based on contrast. The main idea is as follows: the image is first segmented into regions and each region is assigned a contrast value, thereby forming a region-based contrast map. The contrast value of a region is calculated from a global hierarchical value measured by the spatial distance of the current region relative to other regions.
When classifying the garbage images, the method not only considers the brightness characteristics (color, brightness and the like) of the garbage images, but also considers the contrast ratio of the garbage images, thereby obtaining better classification effect than directly utilizing CNN.
And (4) performing convolution and pooling treatment on the preprocessed garbage image based on the CNN, and extracting the characteristic image. The CNN model adopted by the method comprises the following steps: input layer, convolution layer, pooling layer, full-link layer and output layer. The network firstly takes an image as input, then uses the interlaced convolution layer and the pooling layer to extract image characteristics, and finally transmits the extracted characteristics to the full-connection layer for output. The garbage collection device has the advantages that the recognition rate is improved, and the garbage collection device has better stability for recognizing garbage.
FIG. 3 is a flow chart of a garbage vibration frequency acquisition and feature extraction system, which includes a vibration generator, a vibration data acquisition unit and a wavelet de-noising optimization model. The realization process is as follows:
the vibration frequency generator and the vibration frequency acquisition device are arranged below the garbage can throwing opening and are composed of a force sensor and a piezoelectric acceleration sensor, and the acceleration value is obtained by measuring the vibration frequency of the gravity borne by the garbage according to the Newton's second law. The garbage falls on the piezoelectric acceleration sensor due to gravity to generate vibration, and vibration frequency data are collected.
Based on wavelet analysis algorithm, denoising optimization is carried out on vibration frequency data, after a frequency signal is subjected to wavelet transformation, a wavelet coefficient generated by the signal contains important information of the signal, after wavelet decomposition, the wavelet coefficient of an effective signal is large, the wavelet coefficient of noise is small, and the wavelet coefficient of noise is smaller than that of the effective signal. Wavelet analysis is the process of transforming a frequency signal S into wavelet coefficients W, where W ═ Wa,Wb]Including an approximation coefficient WaAnd a detail coefficient Wb. Approximation coefficient WaIs the mean component (low frequency), detail coefficient WbIs a varying component (high frequency). The wavelet original signal decomposition process is as follows: the frequency signal S can be decomposed into the sum of wavelet approximation a and wavelet detail b, S ═ a + b. Then the wavelet coefficient W ═ Wa,Wb]Multiplying by the basis function to form a wavelet decomposition: wavelet approximation coefficient WaBasis function a is approximated decomposition a (average); wavelet detail coefficient WbBasis function D is the approximate decomposition D (variation).
The wavelet analysis algorithm is adopted, the wavelet basis function is utilized to decompose the vibration frequency signal, denoising optimization is carried out, effective signal characteristics can be reserved after denoising, and the method is superior to a traditional low-pass filter.
FIG. 4 is a flow chart of a garbage recognition and classification system. And establishing a garbage classification model based on a random forest algorithm, identifying and classifying the integrated feature vectors, and outputting garbage classification signals. The realization process is as follows:
and (3) carrying out convolution processing on the feature vector of the image information and the feature vector of the garbage vibration after wavelet analysis, and realizing fusion of the two features by adopting a Principal Component Analysis (PCA) method. The characteristic fusion technology not only fuses the most representative information of a plurality of characteristics, but also can remove some unnecessary or similar information, thereby greatly improving the operation speed and the real-time performance.
Before fusion, the raw data is generally normalized, and the processed features are more beneficial to analysis and calculation. The variance is associated with the characteristic variable, and the dispersion degree of the value of the characteristic variable is determined by the variance. The scheme adopts 0-mean normalization, and the normalization formula is as follows:
wherein, mu and sigma are respectively the mean value and variance of the feature vector.
And extracting the generation result of the full-link layer in the CNN algorithm, classifying the generation result and the vibration signal combination after wavelet analysis denoising optimization, setting the number of random forest decision trees, and obtaining the optimal feature set dimension by a sequence forward selection method. Due to the difference of the size of the feature dimension, the selection of the parameters and the feature normalization method, the classification performance of different classifiers is different for different data, so that the best identification result cannot be predicted by using a single classifier. The scheme adopts a random forest to realize the fusion of multiple classifiers. The random forest is an algorithm for gathering a plurality of decision trees together through an integrated learning idea, N classification results can be obtained in the random forest every time a sample image is input, and finally the category with the largest voting times is selected as a final recognition result. If the classified things set can be divided into a plurality of categories, the information of a certain category (xi) can be defined as follows:
I(X＝xi)＝-log2P(xi) (3-1)
where I (x) is information used to represent random variables and p (xi) is the probability when xi occurs.
Due to the limited image data of the garbage samples, the samples provided for CNN training are not sufficient enough, and the overfitting phenomenon is easily caused. And the random forest classification algorithm has good classification precision and classification generalization capability under the conditions of small samples and high dimension. A single prediction problem is solved by combining multiple classifiers. It learns and makes predictions independently from each other by generating multiple classifiers. These predictions are finally combined into a single prediction, so the prediction result is better than any single classifier. Therefore, the method uses a random forest classification algorithm to replace a Soft-Max layer in the CNN to train features, and a final recognition and classification system is obtained. The samples are sampled in a database, and the samples are divided into two equal parts, one part is used for training, and the other part is randomly drawn for result testing.
Fig. 5 is a flow chart of the autonomous learning system for garbage recognition. The intelligent garbage classification throwing system consists of a box door control device. After the classification network model identifies the characteristic images, the classification signals are transmitted to a classification garbage box door control device, and the device starts to operate, so that garbage falls into corresponding subareas.
The identification and classification system judges whether classified garbage is matched with corresponding garbage in a recovery area or not, if the classification is successful, the garbage is put into a corresponding classification box and image and vibration frequency characteristic data are fed back, if the identification is failed, the garbage is put into other classification boxes, the image data are uploaded to a cloud end through the wireless transmission module to be manually marked, the marked data are fed back to the identification and classification system, and self-updating of a database is achieved. The method has the advantages that the characteristics do not need to be extracted manually, the input garbage can be identified and classified intelligently, the garbage can be classified accurately by using a random forest algorithm, the identified data is fed back to the system and the cloud, and self-updating and optimization of a garbage identification model are realized.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all equivalent structures or equivalent flow transformations made by using the contents of the specification and the drawings, or applied directly or indirectly to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. An office rubbish intelligent classification method based on CNN and wavelet analysis is characterized in that: the method comprises the following steps of collecting image data and vibration frequency data of thrown office garbage by adopting double sensors, extracting and integrating features of the data based on CNN and wavelet analysis, identifying and classifying by adopting a random forest algorithm, feeding back the identified data to a system and a cloud end, realizing automatic updating of a sample data set, and realizing intelligent garbage classified throwing, wherein the method comprises the following steps:
(1) collecting garbage image data through an image collecting device;
(2) an image preprocessing method is provided, and the acquired garbage image is preprocessed in consideration of low vividness (color and brightness) and high contrast;
(3) performing convolution and pooling on the preprocessed garbage image based on CNN, extracting image features, and outputting feature vectors;
(4) collecting garbage vibration frequency signals through a vibration generator and a vibration data collector;
(5) performing wavelet decomposition on the vibration frequency signal based on a wavelet analysis model, performing denoising processing on the signal, simultaneously reserving effective signal characteristics, and outputting a vibration frequency characteristic vector;
(6) based on a random forest algorithm, establishing a garbage classification model by using the integrated image characteristic vector and the frequency characteristic vector, and outputting a garbage classification signal;
(7) transmitting the classification signal to a classification garbage bin door control device to enable the garbage to fall into a corresponding subarea;
(8) feeding back corresponding feature vector data of successfully classified garbage to the identification and classification system, and updating a sample data set;
(9) and transmitting corresponding image data of the garbage failed to be classified to the cloud end through the wireless transmission module, and after a user or a maintainer marks the image data, feeding the marked image data of the garbage sample back to the recognition and classification system to update the image sample data set.
2. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: install image acquisition device above the garbage bin input port, image identification device comprises camera module and MCU, gathers rubbish image through the high definition collection equipment including the camera.
3. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: the image preprocessing system preprocesses the acquired garbage image, and the method provides a garbage region selection algorithm based on the image brightness and contrast. The main idea is as follows: firstly, an image is divided into various regions, and then a hierarchical value is allocated to each region, so that a contrast map based on the regions is formed. The contrast value of a region is calculated from a global hierarchical value measured by the spatial distance of the current region relative to other regions.
4. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: and (4) performing convolution and pooling treatment on the preprocessed garbage image based on the CNN, and extracting the image characteristics. The adopted CNN model comprises an input layer, a convolutional layer, a pooling layer, a full-link layer and an output layer. The network firstly takes an image as input, then uses a convolution layer and a pooling layer which are staggered to extract image characteristics, extracts characteristic vectors and outputs the characteristic vectors.
5. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: a vibration generator and a vibration data collector are arranged below the garbage can throwing opening, and the vibration generator and the vibration data collector are composed of a force sensor and a piezoelectric acceleration sensor and are used for collecting a oscillogram of garbage vibration frequency.
6. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: and performing wavelet decomposition on the vibration frequency signal based on a wavelet analysis algorithm, performing denoising processing on the signal, simultaneously reserving effective signal characteristics, and outputting a characteristic vector.
7. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: based on a random forest algorithm, the two features are fused by using a Principal Component Analysis (PCA) method. And extracting the generation result of the full-link layer in the CNN algorithm, classifying the generation result and the vibration signal combination after wavelet analysis and denoising, setting the number of random forest decision trees, and obtaining the optimal feature set dimension by a sequence forward selection method. And establishing a random forest classification model by using the integrated image characteristic vector and frequency characteristic vector to output a garbage classification signal, so as to obtain a final recognition and classification system. The samples are sampled in a database, and the samples are divided into two equal parts, one part is used for training, and the other part is randomly drawn for result testing.
8. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: the identification and classification system transmits the classification signal to the classification garbage box door control device to enable the garbage to fall into the corresponding subareas, the classification garbage box door control device is composed of a steering engine and a connecting piece, and the device starts to operate after receiving the classification signal transmitted by the identification and classification system.
9. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: and feeding back image characteristics and vibration frequency characteristics corresponding to the successfully classified garbage to the identification and classification system, and updating the sample data set.
10. The intelligent classification method of office waste based on CNN and wavelet analysis according to claim 1, characterized in that: and transmitting the corresponding images of the garbage which fails to be classified to the cloud end through the wireless transmission module, and after a user or a maintainer marks the images, feeding back the marked garbage image sample characteristics to the identification and classification system to update the sample data set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010246052.2A CN111461000A (en) | 2020-03-31 | 2020-03-31 | Intelligent office garbage classification method based on CNN and wavelet analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010246052.2A CN111461000A (en) | 2020-03-31 | 2020-03-31 | Intelligent office garbage classification method based on CNN and wavelet analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111461000A true CN111461000A (en) | 2020-07-28 |
Family
ID=71680952
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010246052.2A Pending CN111461000A (en) | 2020-03-31 | 2020-03-31 | Intelligent office garbage classification method based on CNN and wavelet analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111461000A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112044808A (en) * | 2020-08-24 | 2020-12-08 | 华侨大学 | Household garbage recognition system |
| CN112364727A (en) * | 2020-10-28 | 2021-02-12 | 中标慧安信息技术股份有限公司 | Image recognition-based junk information acquisition system |
| CN114266369A (en) * | 2021-12-22 | 2022-04-01 | 安徽自然美环境科技有限公司 | Household garbage resource recovery system |
| CN114511830A (en) * | 2022-01-10 | 2022-05-17 | 上海应用技术大学 | Unmanned vehicle vision identification method for garbage classification |
| CN117181763A (en) * | 2023-09-08 | 2023-12-08 | 深圳市小绿人网络信息技术有限公司 | Renewable resource recycling inspection system based on machine vision recognition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180357542A1 (en) * | 2018-06-08 | 2018-12-13 | University Of Electronic Science And Technology Of China | 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method |
| CN109389161A (en) * | 2018-09-28 | 2019-02-26 | 广州大学 | Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning |
| CN110116415A (en) * | 2019-06-12 | 2019-08-13 | 中北大学 | A kind of Bottle & Can class rubbish identification sorting machine people based on deep learning |
-
2020
- 2020-03-31 CN CN202010246052.2A patent/CN111461000A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180357542A1 (en) * | 2018-06-08 | 2018-12-13 | University Of Electronic Science And Technology Of China | 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method |
| CN109389161A (en) * | 2018-09-28 | 2019-02-26 | 广州大学 | Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning |
| CN110116415A (en) * | 2019-06-12 | 2019-08-13 | 中北大学 | A kind of Bottle & Can class rubbish identification sorting machine people based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 秦斌斌等: "基于卷积神经网络的垃圾分类研究", 《无线通信技术》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112044808A (en) * | 2020-08-24 | 2020-12-08 | 华侨大学 | Household garbage recognition system |
| CN112364727A (en) * | 2020-10-28 | 2021-02-12 | 中标慧安信息技术股份有限公司 | Image recognition-based junk information acquisition system |
| CN112364727B (en) * | 2020-10-28 | 2021-07-06 | 中标慧安信息技术股份有限公司 | Image recognition-based junk information acquisition system |
| CN114266369A (en) * | 2021-12-22 | 2022-04-01 | 安徽自然美环境科技有限公司 | Household garbage resource recovery system |
| CN114511830A (en) * | 2022-01-10 | 2022-05-17 | 上海应用技术大学 | Unmanned vehicle vision identification method for garbage classification |
| CN114511830B (en) * | 2022-01-10 | 2024-05-17 | 上海应用技术大学 | Unmanned vehicle visual recognition method for garbage classification |
| CN117181763A (en) * | 2023-09-08 | 2023-12-08 | 深圳市小绿人网络信息技术有限公司 | Renewable resource recycling inspection system based on machine vision recognition |
| CN117181763B (en) * | 2023-09-08 | 2024-06-18 | 深圳市小绿人网络信息技术有限公司 | Renewable resource recycling inspection system based on machine vision recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111461000A (en) | Intelligent office garbage classification method based on CNN and wavelet analysis | |
| CN106845510B (en) | Chinese traditional visual culture symbol recognition method based on depth level feature fusion | |
| CN108460356B (en) | Face image automatic processing system based on monitoring system | |
| CN104933444B (en) | A kind of design method of the multi-level clustering syncretizing mechanism towards multidimensional property data | |
| CN102880873B (en) | Personnel behavior identification implementation system and method based on image segmentation and semantic extraction | |
| CN106919920A (en) | Scene recognition method based on convolution feature and spatial vision bag of words | |
| CN107480643B (en) | Intelligent garbage classification processing robot | |
| CN107133612A (en) | Based on image procossing and the intelligent ward of speech recognition technology and its operation method | |
| CN106126585B (en) | The unmanned plane image search method combined based on quality grading with perceived hash characteristics | |
| CN109359697A (en) | Graph image recognition methods and inspection system used in a kind of power equipment inspection | |
| CN105718889A (en) | Human face identity recognition method based on GB(2D)2PCANet depth convolution model | |
| CN108830254B (en) | Fine-grained vehicle type detection and identification method based on data balance strategy and intensive attention network | |
| CN107092876A (en) | The low-light (level) model recognizing method combined based on Retinex with S SIFT features | |
| CN109684922A (en) | A kind of recognition methods based on the multi-model of convolutional neural networks to finished product dish | |
| CN104850859A (en) | Multi-scale analysis based image feature bag constructing method | |
| CN110610172A (en) | Myoelectric gesture recognition method based on RNN-CNN architecture | |
| CN110046544A (en) | Digital gesture identification method based on convolutional neural networks | |
| CN115131325A (en) | Breaker fault operation and maintenance monitoring method and system based on image recognition and analysis | |
| CN108375729B (en) | Degenerate state appraisal procedure is clustered based on the motor SOM that Fisher differentiates | |
| Siraj et al. | Flower image classification modeling using neural network | |
| CN115147703B (en) | Garbage segmentation method and system based on GinTrans network | |
| CN103984965A (en) | Pedestrian detection method based on multi-resolution character association | |
| Özyurt et al. | A new method for classification of images using convolutional neural network based on Dwt-Svd perceptual hash function | |
| Xie et al. | Tobacco plant recognizing and counting based on svm | |
| CN109784389A (en) | A kind of coal petrography recognition detection method based on Adaboost and Gabor algorithm |
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: 20200728 |
|
| WD01 | Invention patent application deemed withdrawn after publication |