CN106874929B - Pearl classification method based on deep learning - Google Patents
Pearl classification method based on deep learning Download PDFInfo
- Publication number
- CN106874929B CN106874929B CN201611234986.4A CN201611234986A CN106874929B CN 106874929 B CN106874929 B CN 106874929B CN 201611234986 A CN201611234986 A CN 201611234986A CN 106874929 B CN106874929 B CN 106874929B
- Authority
- CN
- China
- Prior art keywords
- pearl
- network
- data
- deep
- training data
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000013135 deep learning Methods 0.000 title claims abstract description 23
- 239000011049 pearl Substances 0.000 claims abstract description 88
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000012360 testing method Methods 0.000 claims abstract description 23
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000013507 mapping Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 5
- 230000002146 bilateral effect Effects 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 4
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000005070 sampling Methods 0.000 description 20
- 238000012706 support-vector machine Methods 0.000 description 16
- 238000010586 diagram Methods 0.000 description 9
- 230000009471 action Effects 0.000 description 6
- 230000007246 mechanism Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 239000013505 freshwater Substances 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 239000002932 luster Substances 0.000 description 2
- OYPRJOBELJOOCE-UHFFFAOYSA-N Calcium Chemical compound [Ca] OYPRJOBELJOOCE-UHFFFAOYSA-N 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013529 biological neural network Methods 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A pearl classification method based on deep learning comprises the following steps: 1) obtaining a pearl image as sample data; 2) preprocessing sample data; 3) dividing sample data into training data and test data; 4) training a deep convolutional network using training data; 5) extracting the characteristics of training data and test data by using the trained deep convolutional network; 6) constructing an SVM classifier by using the characteristics of the training data extracted in the step 5), and classifying the test data by using the SVM classifier. The method utilizes a plurality of views of the pearls to carry out deep learning, fully exerts the self-learning advantage of the deep learning, automatically learns good characteristics, avoids the complicated process of manually extracting the characteristics and designing the characteristics, and combines the SVM to classify the pearls, so that whether the pearls have threads or not can be effectively judged.
Description
Technical Field
The invention relates to the technical field of image processing and pattern recognition, in particular to a pearl classification method based on deep learning.
Background
China is a large country for producing fresh water pearls, the yield of the fresh water pearls accounts for 95% of the world yield, Zhejiang province and city are the largest bases for cultivation, processing and sale of fresh water pearls in China, and the total yield accounts for more than half of the total yield of the whole country and is known as 'county of Chinese pearls'. The cultivation area of all fresh water pearls reaches 38 ten thousand mu, and 1500 pearl processing enterprises are owned.
After most pearl enterprises collect a large number of pearls, in order to classify the pearls into different grades, the pearls need to be classified manually, the number of people used is very large, and the requirements on professional literacy of classifiers are high. Because manual classification can be influenced by various factors, especially under the condition that pearls are small in size and large in quantity, the classification result is unstable, and the subjective influence of individuals is large. Therefore, the rapid and accurate classification of pearls into a great number of pearl enterprises by using a machine is urgently needed.
Chinese patent application No. 201210411979.2 discloses an on-line automatic grading device for pearl color and gloss based on monocular multi-view machine vision, which comprises a production line for automatically detecting and classifying pearls, a monocular multi-view machine vision device for shooting images of detected pearls, and a microprocessor for performing image processing, detection, identification, classification and coordination control on the images of the detected pearls and coordinating and controlling the coordination action of each action mechanism on the production line, wherein the production line comprises a feeding action mechanism, a delivery inspection action mechanism, a blanking action mechanism, a grading action mechanism and a grading execution mechanism. The invention detects the color and the luster of the detected pearl, but does not detect the screw thread of the pearl, and the screw thread of the pearl is an important judgment standard for manual classification in the actual manual classification occasion.
The early image classification task solution mainly comprises two steps, wherein one step is to manually design features and classify the designed features by using classifiers such as SVM (support vector machine), and the other step is to construct a shallow learning system.
Support Vector Machine (SVM) is a Machine learning algorithm first proposed by Vapnik et al in 1995. The SVM is based on a VC (virtual component modeling) dimension theory of a statistical learning theory and a structure risk minimization principle, and can seek the best compromise between the complexity (namely the learning precision of a specific training sample) and the learning capability (namely the capability of identifying any sample without errors) of a model according to limited sample information so as to obtain the best classification and identification capability.
Because the number and the position of the threads of different pearls are different, and the clarity of the threads of the pearls is influenced by the color and the luster of the threads when the pearl images are collected, the manual design of the characteristics is not easy, and the designed characteristics are not necessarily suitable for the current classification task.
Disclosure of Invention
In order to solve the problems that the existing pearl classification system based on machine vision is difficult to detect and classify the screw thread of a pearl with high precision, the manual design of characteristics is not easy to extract the screw thread characteristics of the pearl and the like, the invention provides the pearl classification method based on deep learning.
The technical scheme adopted by the invention for realizing the aim of the invention is as follows:
a pearl classification method based on deep learning comprises the following steps:
1) collecting pearl images as sample data, wherein each pearl comprises a set number of images;
2) adjusting the collected pearl image into a set size, and carrying out image preprocessing on the pearl image to remove the noise of the image;
3) dividing sample data into training data and test data;
4) setting initial parameters of each network layer of the deep convolutional network, inputting the training data divided in the step 3) into the deep convolutional network, and training the network;
5) extracting the characteristics of training data and test data by using the trained deep convolutional network;
6) constructing an SVM classifier by using the characteristics of the training data extracted in the step 5), and classifying the test data by using the SVM classifier.
Further, the pearl images obtained in the step 1) comprise two types, namely a pearl image with threads and a pearl image without threads.
Further, bilateral filtering is adopted in the step 2) to remove noise of the image. Of course, other denoising methods may be employed.
Still further, the step 4) comprises the following steps:
4.1) determining the structure of the deep convolutional network;
4.2) initializing parameters to be trained in the network by using different small random numbers;
4.3) inputting the training data divided in the step 3) into the deep convolutional network, calculating the error between the output of the deep convolutional network and the actual class label, and adjusting the weight and the bias term of each layer of the deep convolutional network through an error back propagation algorithm until the network is stable or the set maximum iteration number is reached.
Furthermore, in the step 1), each pearl comprises 5 images which are respectively a top view, a left view, a right view, a rear view and a front view; the deep convolution network constructed in the step 4) is composed of 5 convolution networks, and 5 views of the pearl are respectively used as the input of the 5 convolution networks to extract the thread characteristics.
Preferably, the 5 convolutional networks have the same structure and respectively include 2 convolutional layers, 2 downsampling layers and 1 full-connection layer.
The output characteristics of the 5 convolutional networks need to be accumulated by an ELTWISE layer, combined into 1 eigenvector, input the eigenvector into an FC full connection layer, and input the eigenvector into an SOFTMAX layer after the FC full connection layer maps the eigenvector to two dimensions, wherein the SOFTMAX layer consists of 2 independent neurons and corresponds to a threaded pearl and a non-threaded pearl.
Further, the step 5) of extracting features by using the trained deep convolution network means that sample data is used as the input of the trained deep convolution network, and the output of the elttune layer of the deep convolution network is used as the thread features of the sample pearl.
Further, the step 6) includes the steps of:
6.1) preprocessing the characteristics of the training data extracted in the step 5);
pretreatment: normalizing the obtained feature data, and mapping the value of each dimension of the feature data to [0, 1]]An interval; the conversion function is
Wherein xmaxIs the maximum value of the characteristic data, xminIs the minimum value of the characteristic data;
6.2) constructing an SVM classifier by using the training data after 6.1) normalization;
6.3) normalizing the test data according to the mapping relation during the normalization of the training data;
6.4) classifying the normalized test data by using an SVM classifier, and judging whether the pearl has threads.
The technical conception of the invention is as follows: deep learning is a hot point in recent years, the concept of deep learning is derived from the research of artificial neural networks, and a multi-layer perceptron with multiple hidden layers is a deep learning structure. Deep learning forms a more abstract high-level representation (attribute classes or features) by combining low-level features to find a distributed feature representation of the data. The convolution network is a structure of deep learning, and the convolution network makes a great breakthrough in the problems of image classification and the like. The convolutional network is composed of one or more convolutional layers and a top full-connection layer, and also comprises an associated weight and a down-sampling layer, and the structure enables the convolutional neural network to utilize a two-dimensional structure of input data, and an image can be directly used as the input of the network, so that the complex processes of feature extraction and data reconstruction in the traditional recognition algorithm are avoided. The weight sharing network structure of the method is closer to a biological neural network, so that the complexity of a model is reduced, and the number of weights is reduced. The invention has the beneficial effects that: 1) the thread characteristics of the pearls are extracted by using a deep convolution network, the self-learning advantage of deep learning is fully exerted, good characteristics are automatically learned by a machine, the complex process of manually extracting the characteristics and designing the characteristics is omitted, the defect of manually and manually designing the characteristics is avoided, the process is simplified, and the time is saved; 2) 5 convolution networks are constructed aiming at 5 views of pearls, the image information of the whole pearl is fully utilized, and the defect that the image under a single visual angle has no integral representativeness is avoided; 3) the local features of the images are extracted by the deep convolutional network, and then the features are combined, so that the extracted features have certain translational and rotational invariance, and the influence caused by the randomness of the pearl placement positions when pearl images are collected is avoided to a certain extent; 4) the deep convolutional network is organically combined with a Support Vector Machine (SVM), the deep convolutional network is used for extracting features, and the SVM is used for classification and recognition, so that the classification accuracy is guaranteed.
Drawings
Fig. 1 is a flow chart of the whole classification method.
Fig. 2 is a schematic diagram of the overall structure of the deep convolutional network.
Fig. 3 is a schematic diagram of a specific network structure of a convolutional network.
Detailed Description
The following detailed description of embodiments of the invention is provided in connection with the accompanying drawings.
Referring to fig. 1 to 3, a pearl classification method based on deep learning includes the following steps:
1) acquiring pearl images as sample data, wherein each pearl comprises 5 images which are respectively a top view, a left view, a right view, a rear view and a main view;
the captured images include two types of pearl images, a threaded pearl image and a non-threaded pearl image.
2) Regulating the collected pearl image into 150 x 150 pixels, and carrying out image preprocessing on the pearl image to remove the noise of the image;
removing noise of the image by utilizing bilateral filtering, wherein the collected image is a color image, and the R, G, B three color channels need to be subjected to bilateral filtering respectively; marking the pearl image as I, marking the current pixel point as (x, y), and then marking the brightness value of the R component of the current pixel as Ir(x, y), the brightness value of the G component of the current pixel is denoted as Ig(x, y), the brightness value of the B component of the current pixel is denoted as Ib(x, y), the brightness difference between the current pixel point and the adjacent pixel point (x-i, y-j) is recorded as
The calculation formula of (2) is as follows:
the pixel brightness value after the pixel (x, y) is filtered is denoted as F (x, y), and the calculation formula is:
wherein sigmar、σdIs the smoothing parameter, m and n are the length and width of the filtering window, respectively; the three color channel filtered F are separately computed R, G, B to obtain filtered images.
3) Dividing sample data into training data and test data;
according to the proportion of training data to test data being 8: and 2, dividing sample data, wherein the training data is used for training parameters of the deep convolutional network, and the test data is used for testing the performance of the classifier.
4) Setting initial parameters of each network layer of the deep convolutional network, inputting the training data obtained in the step 3) into the deep convolutional network, and training the network;
4.1) determining the structure of the deep convolutional network;
the structural schematic diagram of the deep convolutional network is shown in FIG. 2, and a top view, a left view, a right view, a rear view and a main view of a preprocessed pearl are respectively used as the input of 5 convolutional networks CNN 1-5;
the ELTWISE layer superposes the outputs of the 5 convolution networks to obtain a feature vector of the pearl image, and the feature vector is used as the input of the FC full-connection layer;
the FC layer maps the feature vector to a two-dimensional vector and takes the two-dimensional vector as the input of the SOFTMAX layer;
the SOFTMAX layer is mainly used for normalizing data after the full connection layer to enable the range of the data to be between [0 and 1], and consists of 2 independent neurons corresponding to threaded pearls and unthreaded pearls;
the 5 convolutional networks CNN 1-5 adopt the same network structure, as shown in FIG. 3, including:
the CONV1 convolutional layer is obtained by multiplying 20 5 × 5 convolutional kernels by corresponding elements of the image sample data, summing the products and adding a bias term to obtain a characteristic diagram of the CONV1 convolutional layer;
POOLING1 downsampling layer, which can utilize the principle of image local correlation to subsample the image and can reduce the data processing amount and retain useful information; sub-sampling the characteristic diagram of the CONV1 convolutional layer by using 20 2 × 2 sampling cores of the POOLING1 downsampling layer to obtain the characteristic diagram of the POOLING1 downsampling layer; the common sampling method of the down-sampling layer comprises average value down-sampling, random down-sampling and maximum value down-sampling, wherein the down-sampling layer adopts maximum value down-sampling. The average value down-sampling refers to outputting the average value of all elements in a sampling window as a sampling result; random down-sampling refers to randomly selecting a value of an element in a sampling window to be output as a sampling result; the maximum value downsampling refers to taking the maximum value of all elements in a sampling window as a sampling result;
the CONV2 convolutional layer is obtained by performing convolution operation on 50 convolution kernels of 5 multiplied by 5 and a feature map of a POOLING1 downsampling layer, summing the convolution kernels and adding a bias term to the sum, and then obtaining a feature map of the CONV2 convolutional layer;
the POOLING2 downsampling layer, and sub-sampling the characteristic diagram of the CONV2 convolutional layer by adopting 50 2 × 2 sampling cores to obtain the characteristic diagram of the POOLING2 downsampling layer;
the FC full-connection layer is used for mapping the characteristic diagram of the POOLING2 to a vector with 500 dimensions;
the RELU active layer adopts an active function max (x,0), when x is greater than 0, x is output, otherwise, 0 is output, and x refers to the numerical value of the output vector of the FC full-connection layer;
4.2) initializing parameters to be trained in the network by using different small random numbers;
4.3) inputting the training data divided in the step 3) into the deep convolutional network, calculating the error between the output of the deep convolutional network and the actual class label, and adjusting the weight and bias items of each layer of the deep convolutional network through an error back propagation algorithm until the network is stable or reaches the set maximum iteration times;
after a deep convolutional network is constructed and parameters to be trained in the network are initialized by different small random numbers, training data divided in the step 3) are used as input of the network, prediction output obtained by the network is compared with real class labels of the training data, errors of the prediction output and the real class labels are propagated reversely, and parameters of the deep convolutional network are updated by a random gradient descent method according to the errors, so that the errors between the prediction output and the real class labels are gradually reduced;
and training a plurality of training data for a plurality of times, updating the parameters of the deep convolutional network every training to continuously reduce the error between the prediction output and the real class label of the training data, and determining the currently learned model parameters as the trained model parameters when the error is less than a preset value or the training iteration times exceed the preset maximum iteration times so as to obtain the trained deep convolutional network.
5) Extracting the characteristics of training data and test data by using the trained deep convolutional network;
taking training data and test data as the input of a trained deep convolution network, and taking the output of an ELTWISE layer of the deep convolution network as the thread characteristics of the training data and the test data;
6) constructing an SVM classifier by using the thread characteristics of the training data extracted in the step 5), and classifying the test data by using the SVM classifier;
6.1) preprocessing the characteristics of the training data extracted in the step 5);
pretreatment: normalizing the obtained feature data, and mapping the value of each dimension of the feature data to [0, 1]]An interval; the conversion function is
Wherein xmaxIs the maximum value of the characteristic data, xminIs the minimum value of the characteristic data;
6.2) constructing an SVM classifier by using the training data after 6.1) normalization;
6.3) normalizing the test data according to the mapping relation during the normalization of the training data;
6.4) classifying the normalized test data by using an SVM classifier, and judging whether the pearl has threads.
Claims (8)
1. A pearl classification method based on deep learning is characterized in that: the method comprises the following steps:
1) collecting pearl images as sample data, wherein each pearl comprises a set number of images;
2) adjusting the collected pearl image into a set size, and carrying out image preprocessing on the pearl image to remove the noise of the image;
3) dividing sample data into training data and test data;
4) setting initial parameters of each network layer of the deep convolutional network, inputting the training data divided in the step 3) into the deep convolutional network, and training the network;
5) extracting the characteristics of training data and test data by using the trained deep convolutional network;
6) constructing an SVM classifier by using the characteristics of the training data extracted in the step 5), and classifying the test data by using the SVM classifier;
in the step 1), each pearl comprises 5 images which are respectively a top view, a left view, a right view, a rear view and a front view; the deep convolution network constructed in the step 4) is composed of 5 convolution networks, and 5 views of the pearl are respectively used as the input of the 5 convolution networks to extract the thread characteristics.
2. The pearl classification method based on deep learning as claimed in claim 1, wherein: the pearl images obtained in the step 1) comprise two types, namely pearl images with threads and pearl images without threads.
3. The pearl classification method based on deep learning as claimed in claim 1, wherein: and 2) removing the noise of the image by adopting bilateral filtering in the step 2).
4. The pearl classification method based on deep learning as claimed in claim 1, wherein: the step 4) comprises the following steps:
4.1) determining the structure of the deep convolutional network;
4.2) initializing parameters to be trained in the network by using different small random numbers;
4.3) inputting the training data divided in the step 3) into the deep convolutional network, calculating the error between the output of the deep convolutional network and the actual class label, and adjusting the weight and the bias term of each layer of the deep convolutional network through an error back propagation algorithm until the network is stable or the set maximum iteration number is reached.
5. The pearl classification method based on deep learning as claimed in one of claims 1 to 4, wherein: the 5 convolutional networks adopt the same structure and respectively comprise 2 convolutional layers, 2 downsampling layers and 1 full-connection layer.
6. The pearl classification method based on deep learning as claimed in one of claims 1 to 4, wherein: the output characteristics of the 5 convolutional networks need to be accumulated by an ELTWISE layer, combined into 1 eigenvector, input the eigenvector into an FC full connection layer, and input the eigenvector into an SOFTMAX layer after the FC full connection layer maps the eigenvector to two dimensions, wherein the SOFTMAX layer consists of 2 independent neurons and corresponds to a threaded pearl and a non-threaded pearl.
7. The pearl classification method based on deep learning as claimed in claim 6, wherein: the step 5) of extracting the features by using the trained deep convolution network means that sample data is used as the input of the trained deep convolution network, and the output of an ELTWISE layer of the deep convolution network is used as the thread features of the sample pearls.
8. The pearl classification method based on deep learning as claimed in one of claims 1 to 4, wherein: the step 6) comprises the following steps:
6.1) preprocessing the characteristics of the training data extracted in the step 5);
pretreatment: normalizing the obtained feature data, and mapping the value of each dimension of the feature data to [0, 1]]An interval; a conversion function of
Is the maximum value of the characteristic data,
is the minimum value of the characteristic data;
6.2) constructing an SVM classifier by using the training data after 6.1) normalization;
6.3) normalizing the test data according to the mapping relation during the normalization of the training data;
6.4) classifying the normalized test data by using an SVM classifier, and judging whether the pearl has threads.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611234986.4A CN106874929B (en) | 2016-12-28 | 2016-12-28 | Pearl classification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611234986.4A CN106874929B (en) | 2016-12-28 | 2016-12-28 | Pearl classification method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106874929A CN106874929A (en) | 2017-06-20 |
| CN106874929B true CN106874929B (en) | 2020-03-17 |
Family
ID=59164374
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611234986.4A Expired - Fee Related CN106874929B (en) | 2016-12-28 | 2016-12-28 | Pearl classification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106874929B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107330446B (en) * | 2017-06-05 | 2020-08-04 | 浙江工业大学 | Image classification-oriented deep convolutional neural network optimization method |
| CN109420622A (en) * | 2017-08-27 | 2019-03-05 | 南京理工大学 | Tobacco leaf method for sorting based on convolutional neural networks |
| CN107665352A (en) * | 2017-09-07 | 2018-02-06 | 浙江工业大学 | A kind of pearl sorting technique based on multichannel residual error network |
| CN107609587B (en) * | 2017-09-11 | 2020-08-18 | 浙江工业大学 | Multi-class multi-view data generation method for generating countermeasure network based on deep convolution |
| CN108038499A (en) * | 2017-12-06 | 2018-05-15 | 浙江大学 | A kind of seeds sorting technique and system based on deep learning |
| CN109304306B (en) * | 2018-09-19 | 2020-08-11 | 广东省智能制造研究所 | Production line object sorting method and system and object sorting system |
| CN113657240A (en) * | 2021-08-12 | 2021-11-16 | 浙江工业大学 | Unbalanced-like pearl classification method based on multi-view image amplification |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN202974886U (en) * | 2012-12-24 | 2013-06-05 | 厦门大学 | Machine-vision-based tank body seam defect detection system |
| CN105095833A (en) * | 2014-05-08 | 2015-11-25 | 中国科学院声学研究所 | Network constructing method for human face identification, identification method and system |
| CN105631480A (en) * | 2015-12-30 | 2016-06-01 | 哈尔滨工业大学 | Hyperspectral data classification method based on multi-layer convolution network and data organization and folding |
| CN105787510A (en) * | 2016-02-26 | 2016-07-20 | 华东理工大学 | System and method for realizing subway scene classification based on deep learning |
| CN106096584A (en) * | 2016-06-29 | 2016-11-09 | 深圳市格视智能科技有限公司 | Prolongable conductor spacer recognition methods based on degree of depth convolutional neural networks |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016097758A1 (en) * | 2014-12-19 | 2016-06-23 | Apical Limited | Sensor noise profile |
-
2016
- 2016-12-28 CN CN201611234986.4A patent/CN106874929B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN202974886U (en) * | 2012-12-24 | 2013-06-05 | 厦门大学 | Machine-vision-based tank body seam defect detection system |
| CN105095833A (en) * | 2014-05-08 | 2015-11-25 | 中国科学院声学研究所 | Network constructing method for human face identification, identification method and system |
| CN105631480A (en) * | 2015-12-30 | 2016-06-01 | 哈尔滨工业大学 | Hyperspectral data classification method based on multi-layer convolution network and data organization and folding |
| CN105787510A (en) * | 2016-02-26 | 2016-07-20 | 华东理工大学 | System and method for realizing subway scene classification based on deep learning |
| CN106096584A (en) * | 2016-06-29 | 2016-11-09 | 深圳市格视智能科技有限公司 | Prolongable conductor spacer recognition methods based on degree of depth convolutional neural networks |
Non-Patent Citations (4)
| Title |
|---|
| Caffe: Convolutional Architecture for Fast Feature Embedding;Yangqing Jia 等;《arXiv》;20140620;第1-4页 * |
| Combining multiple sources of knowledge in deep CNNs for action recognition;Park, E 等;《Applications of Computer Vision》;20160310;第1-8页 * |
| 基于卷积神经网络的人群密度估计;付敏;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160315(第03期);第I138-6479页 * |
| 基于图像处理的珍珠质量检测系统;邓志坚;《中国优秀硕士学位论文全文数据库 信息科技辑》;20080715(第07期);正文第49-53页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106874929A (en) | 2017-06-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106874929B (en) | Pearl classification method based on deep learning | |
| CN109684922B (en) | Multi-model finished dish identification method based on convolutional neural network | |
| CN107016405B (en) | A kind of pest image classification method based on classification prediction convolutional neural networks | |
| CN109509187B (en) | Efficient inspection algorithm for small defects in large-resolution cloth images | |
| CN106295124B (en) | The method of a variety of image detecting technique comprehensive analysis gene subgraph likelihood probability amounts | |
| CN103886344B (en) | A kind of Image Fire Flame recognition methods | |
| CN109101938B (en) | Multi-label age estimation method based on convolutional neural network | |
| CN104331712B (en) | A kind of alga cells classification of images method | |
| CN107977671A (en) | A kind of tongue picture sorting technique based on multitask convolutional neural networks | |
| CN106228137A (en) | A kind of ATM abnormal human face detection based on key point location | |
| CN101441721B (en) | Device and method for counting overlapped circular particulate matter | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN103593670A (en) | Copper sheet and strip surface defect detection method based on-line sequential extreme learning machine | |
| CN112434646A (en) | Finished tea quality identification method based on transfer learning and computer vision technology | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| CN106529568A (en) | Pearl multi-classification method based on BP neural network | |
| CN108932712A (en) | A kind of rotor windings quality detecting system and method | |
| CN107480679B (en) | Road network extraction method based on classification and connected region analysis of convolutional neural network | |
| CN107665352A (en) | A kind of pearl sorting technique based on multichannel residual error network | |
| CN110956212A (en) | Threshing quality detection method based on visual feature fusion | |
| CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
| CN106529441B (en) | Depth motion figure Human bodys' response method based on smeared out boundary fragment | |
| CN109003275A (en) | The dividing method of weld defect image | |
| CN114463843A (en) | Multi-feature fusion fish abnormal behavior detection method based on deep learning | |
| CN110533100A (en) | A method of CME detection and tracking is carried out based on machine learning |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20180202 Address after: 311800 Zhejiang Province, Zhuji city Jiyang Street Puyang Road No. 18 Science Park 801-802 Applicant after: Zhejiang Zhuji Hui Zhi Technology Co.,Ltd. Address before: West Village 311899 Zhejiang Street Tao mountain city of Shaoxing province Zhuji City Zhao No. 121 Applicant before: ZHUJI QIJIAN INTELLIGENT TECHNOLOGY CO.,LTD. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210421 Address after: 310000 No. 38, Zhejiang Road, Hangzhou, Zhejiang, Xihu District Patentee after: Xuan Qi Address before: 311800 Zhejiang Province, Zhuji city Jiyang Street Puyang Road No. 18 Science Park 801-802 Patentee before: Zhejiang Zhuji Hui Zhi Technology Co.,Ltd. |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200317 |