CN104809472A - SVM-based food classifying and recognizing method - Google Patents
SVM-based food classifying and recognizing method Download PDFInfo
- Publication number
- CN104809472A CN104809472A CN201510217604.6A CN201510217604A CN104809472A CN 104809472 A CN104809472 A CN 104809472A CN 201510217604 A CN201510217604 A CN 201510217604A CN 104809472 A CN104809472 A CN 104809472A
- Authority
- CN
- China
- Prior art keywords
- food
- svm
- data
- classifying
- user
- 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
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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/68—Food, e.g. fruit or vegetables
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an SVM-based food classifying and recognizing method. The method includes that a user starts a camera of a mobile terminal to record food to be eaten, and after the food is eaten up, remaining food is removed; the food of the user is analyzed through an internal algorithm chip module, displaying is realized through a display terminal of equipment, and eaten calorie can be comprehensively assessed by means of intelligent selection and expert judgment. An improved SVM is adopted, so that food recognition effect under complex condition is improved. A mobile application module provides a unique mechanism to collect diet information, so that burden of a diet recorder can be reduced. Assisting and prompting effect on diabetic patients can be realized. Equipment is compact, easy to carry and convenient to use and can be used anytime and anyplace. Society in future will enter a stage of intelligent cities, much labor can be replaced by intelligent machines. Research and development of the product will lead to a new revolution.
Description
Technical field
The invention belongs to technology of Internet of things field, relate to a kind of food-classifying recognition methods based on support vector machine (SVM), the object of the invention is to use this technology to solve the problem that in actual life, food intake dose is measured.
Background technology
Along with the continuous lifting of quality of life and level, in young man, the quantity of bariatric patients progressively increases is a huge problem.Unfortunately, this lasting rising tendency causes the quantity also sustainable growth of diabetes B patient.In 2008, obesity patient accounted for world adult population's 1/10th, and has risen to 1/6th this numeral in 2012, and is increasing with surprising rapidity.Nearest research shows, fat people more may produce serious health disease, as hypertension, heart disease, diabetes B, high fat of blood, breast cancer, colon cancer and respiratory disorder etc.The main cause of obesity is the imbalance between the absorption of food and energy and consumption.So, in order to lose weight by the mode of health and maintain the healthy weight of normal person, food intake must be measured every day.In fact, the intake that all existing bariatrician technical requirement patients record food every day comes absorption to specific energy and consumption.
Dietary intake, namely defines someone and what is eaten every day.Dietary intake can provide valuable advices for increasing intervention stratege, thus can prevent numerous disease.At nutrition and health field, accurately measuring dietary intake is open research topic.By introducing a new semi-automatic Dietary estimation device, it contributes to nutritionist and monitors daily cost.Accomplish this point, in the different step of food recognition device, have employed various image processing techniques.The mobile device of obesity patient uses this food recognition device, and anyone can monitor the intake of his/her food.
Summary of the invention
The object of this invention is to provide and relate to a kind of food-classifying recognition methods based on SVM.
Above-mentioned object is realized by following technical scheme:
Consider the universal of in young crowd mobile device (such as smart mobile phone and panel computer), and these communication facilities ubiquities, they can load the relevant application module of diet, such as our food identification module.The object of the invention is to use this technology to solve the problem that in actual life, food intake dose is measured.The research in this field current can be divided three classes, i.e. clinical research, support study dies and semi-automatic technique research.In such as 24 hours dietary recalls (24HR) and this kind of clinical research of Food frequency questionnaire, its data are by manual patient's record, calculate caloric intake more afterwards.Data due to these methods are hand-kepts, therefore may produce a large amount of mistakes in whole recording process.Another shortcoming of these methods is that patient is difficult to record to be applied to treatment.Recently, some researchists study semi-automatic method.The present invention will design and develop a simply application module for the semi-automation of practicality, and people can use application module to carry out alternately.In addition, in this case, by using modular method to increase the use chance of mobile device as this kind of in smart mobile phone.Use mobile device as user interface, and send to expert to calculate caloric intake the food image data of oneself by network.Mobile solution module provides a unique mechanism to collect diet information, can reduce the burden of diet keeper thus.
In the stage in early days, user takes image with mobile device, carries out pre-treatment step subsequently.Then, in segmentation step, each image analyzed and extract each food portion.As everyone knows, if not have good Iamge Segmentation mechanism, this device then suitably can not process image.Therefore, this step has been resolved into more step by the present invention.For this reason, we used color segmentation, K mean cluster and Texture Segmentation instrument.For each food products part detected, characteristic extraction procedure must be performed.For each food image detected, characteristic extraction step must be performed.Each attribute of food will be extracted in this step, as size, shape, color and texture.The feature extracted will be sent to classifying step, and classifying step uses SVM scheme to be identified by food.Finally, by estimating the area of food portion, and with reference to nutrition table, caloric content in food is calculated.Wherein, food-classifying identification, present invention employs SVM method, specific as follows:
1, the extraction of feature and classification
By using the framework of four kinds of features, comprise color, texture, size and dimension.For color characteristic, employ rim detection and color K-mean cluster.For textural characteristics, use Gabor filter.In order to size and dimension feature, the area-of-interest pixel computing technique of the edge detecting technology employed and image.Next step is classified to the feature extracted, so that identify often kind of food.SVM algorithm it be one of popular technique for Data classification.A classification task generally includes training data and test data, and data are made up of some examples.Each example in training data comprises a class label and several feature.The target of SVM is framework model, and this model can go out the desired value of this example by means of only the attribute forecast of data instance in test data.
In order to improve accuracy, after SVM module has determined often kind of food type, this device and user have verified food species alternately.Device can demonstrate the image of food to user and explain out the food type that device thinks, as chicken, meat, vegetables etc.Then, user can confirm or change the type of food.Although this change makes device become semi-automation from robotization, but it can increase the accuracy of device.
Use RBF core in this model, it in non-linear mode at more high-dimensional spatial mappings sample.Different from linear kernel, RBF core is very suitable for class mark and attribute is nonlinear situation.
In the present invention, the proper vector of SVM comprises five textural characteristics, ten color characteristics, three shape facilities and six size characteristics.Extract the proper vector of various food in the segmentation stage, then become the training vector of SVM.
2, the support vector machine training stage
Before the SVM using this food recognition system, carry out a training stage, this is to produce SVM model.Fig. 1 is the block diagram of the training stage realized.As shown in the drawing, a series of different feature of often kind of food is the input in this stage.It should be noted, in order to the output providing system best, the title of food should import system into as input.The output in this stage is SVM model.
3, classify
When completing segmentation to food image and extracting characteristic manipulation, next step is Using statistics mode identification technology determination food species.Sorter is used to various pattern-recognition and machine learning aspect, and range of application is from automatic speech recognition to recognition of face.Identifying mainly comprises two steps.In the training stage, model is taken as training data study.This model is the original data of sorter, thus can assess the precision of model at test phase.By training, sorter study how by Feature Mapping to various classification or mark.But training is faulty, sorter can be made mistakes in practical operation, by a wrong subcarrier label assignments to the eigenvector observed.
Training classifier has two targets.Target defines how to the eigenvector distribute labels observed, and another estimates its performance error or nicety of grading.Support vector machine be one group for the manageable learning method of classifying and return.They belong to a serial spherical linear classifier.In other words, support vector machine is a kind of classification and return device, and its uses machine Learning Theory to maximize precision of prediction, can automatically integral data to the classification of the best.Support vector machine can be defined as the system using linear function hypothesis space in a high-dimensional feature space.
SVM to provide in identical task compared with neural net method result more accurately.It is also used to many application, as human face analysis and speech recognition, particularly based on the application of pattern classification and recurrence.The support vector machine past is for solving classification problem, but recently, they have also expanded to and have been applied to solution regression problem.
4, support vector machine is explained
Nerual network technique is that it has good effect for one of supervision and unsupervised study application and identification method.A kind of neural network algorithm is multilayer perceptron (MLP) algorithm.MLP adopts feedforward network and recirculating network.This algorithm comprises general approximation capability and the input and output mode study of Continuous Nonlinear function.
Fig. 2 is the behavior figure of simple neural network and multilayer perceptron.
But MLP also has some problems, therefore encourage it is found that and use additive method.First, neural network has many local minimums and is difficult to obtain neuronic number, therefore needs the learning method of testing other.In addition, even if the neural network solution used has a kind of tendentiousness and convergence, but it is not a unique solution.In the method for neural network, by attempting and testing to find out best solution, so real-time is good not.
Many linear classifiers (lineoid) can be separated data.If go grouped data with a linear lineoid, the data set of one group of local minimum instead of global minimum finally may be drawn.Therefore, largest interval sorter can be released.
Fig. 3 is largest interval sorter Linear SVM figure.
In Linear SVM, largest interval can be used to.A problem that here may occur why uses largest interval? one of them reason is, even if create a little error in the position on border, uses largest interval also can drop to minimum by the probability of mis-classification.Another reason is that of avoiding local minimum and finally obtains a better classification.
The target of SVM carrys out mask data with lineoid and uses geo-nuclear tracin4 to extend to nonlinear limit.The target calculating SVM will carry out correct classification to all data.For mathematical computations, we have following formula.
(1)
(2)
Above-mentioned formula can be combined as following formula.
(3)
In this equation, x is vector point, and w is weight is also a vector.Therefore, see separately that formula (1) should be that perseverance is greater than zero.In all possible lineoid, the distance of the lineoid that support vector machine is selected should be large as much as possible.If training data is good and each test vector is arranged in trained vector radius r, so it will be selected as default value.Now, if selected lineoid is in range data position far away as far as possible, now desirable lineoid makes interval maximum, has also divided the line between the nearest point of two data centralizations equally.
Fig. 4 is the expression figure of lineoid.
Because x is on lineoid, the point on lineoid finds to the minimum distance of initial point by maximizing x.Equally, for the point of another side, we have a similar algorithm.Therefore, by obtaining and deduct this two distances, the distance summation from Optimal Separating Hyperplane to nearest point is obtained.
A unique solution is had to utilize Lagrange's theorem to maximize this amount, as shown in the formula.
(4)
This formula has following constraint.
(5)
Herein,
for Lagrange multiplier.
Non-linear SVM replaces Linear SVM, and wherein trained vector is not linear separability.During with non-linear SVM, trained vector is separated by nonlinear boundary.Non-linear decision boundary determines to other Euclidean spaces by mapping trained vector.
Then, lineoid is used to be separated the trained vector mapped out.Due to trained vector will be mapped to high-dimensional, in order to avoid the increase of complicacy, map
substituted by the kernel function as formula (6).
(6)
Under nonlinear situation, will with lagrange formula (6)
replace with
.In formula (7), the maximization problems of amendment will be produced.
(7)
SVM classifier uses LIBSVM training mode.In addition,
be called as kernel function.SVM uses one of formula (8), (9), these four cores shown in (10) and (11).
Linear:
(8)
Radial basis function (RBF):
(9)
Polynomial expression:
(10)
Contrary flexure:
(11)
In above-mentioned equation,
all nuclear parameter with d.
Use SVM can according to following steps to food-classifying: the form 1) converting data to SVM; 2) simple scalability of data is prepared; 3) RBF core is considered
4) use cross validation best to find
parameter; 5) use
parameter trains whole training set; 6) test;
5, data prediction
1) tagsort
In SVM, all data are all expressed as a vector of real number.Therefore, the value of each eigenvector must be converted to numeric data by us.A kind of mode is that use 0,1 number combinatorics on words shows them.Such as, three kinds of categorical attributes that a picture is red, green, blue, can be expressed as (0,0,1), (0,1,0) and (1,0,0).
1) convergent-divergent
The difficulty of numerical evaluation can be avoided by convergent-divergent.Value due to core depends on the center of proper vector usually, therefore during convergent-divergent preferably in the scope of [-1,1] or [0,1].Note that we must make to use the same method convergent-divergent training and testing data.
3) RBF core
RBF core be one reasonably first-selected, this endorses with mapped sample in nonlinear higher dimensional space, with linear kernel unlike, it can process the situation that class mark and attribute are nonlinear relationships.Second reason is the complicacy that the number of hyper parameter adds Model Selection.The hyper parameter that polynomial kernel has is more than RBF core.Finally, RBF core has less numerical evaluation difficulty.In some cases, RBF core is unaccommodated.Such as, when the quantity of feature is very large, then linear kernel can only be used.
4) cross validation and grid search
RBF core has
two parameters.The target of this step finds
optimum value, to enable the data (i.e. test data) of sorter Accurate Prediction the unknown.The common strategy of of sorter is that data set is divided into two parts, and wherein a part is considered to unknown.Can reflect that sorter is to an independently data set performance of classifying more accurately from the unknown group of precision of prediction obtained.One of this process is improved version and is called as cross validation.
The advantage of cross validation is used to be to prevent over-fitting problem.Fig. 5 is explained by binary classification problems.Filled circles and black triangle are training datas, and open circles and hollow triangle are test datas.In (a) and (b) of Fig. 5, the measuring accuracy of sorter is because of its training data over-fitting, so poorly.If the training data in (a) and (b) of Fig. 5 and test data are used for training and cross validation by us, its precision is also bad.On the other hand, the sorter not overfitting training data in (c) and (d), and better cross validation and accuracy test can be provided.One find and good mode be use " grid search ".
Fig. 5 be over-fitting sorter and better sorter figure (
with
: training data;
with
: test data).
6, support vector machine evaluation stage
Use SVM model, the feature that often kind of food extracts will be admitted to SVM system, then just can obtain the title of food.The step schematic diagram of whole process as shown in Figure 6.As shown in the drawing, first, will be exported with the size of food that counter cluster extracts in the segmentation stage.According to number food being detected, SVM classifier brings into operation.The SVM that the color of often kind of food, size, shape and textural characteristics will be sent to.The feature of the generation in these features and training process can be compared in SVM, and the title of food is provided.The circulation of this process is carried out, until all food is all labeled.
Fig. 6 is the sorting phase figure using SVM.
beneficial effect:
Interface of the present invention can include display and man-machine operation interface, and display is mobile phone, flat-type touch-control or non-touch-control.The food that user identifies by display analysis, determines whether as user is consumed, also can sets itself food content and component.
1) adopt the SVM improved, improve the food recognition effect of complex condition;
2) starting outfit, the food that the video camera that mobile terminal carries will be eaten is recorded, after user eats up food, also to remove leftover, analyzed by the food of internal algorithm chip module to user, shown by the display terminal of equipment again, and the pattern of intelligent selection and expert judgments can carry out comprehensive test to the calorie eaten up.
3) use mobile device as user interface, and send to expert to calculate caloric intake the food image data of oneself by network.Mobile solution module provides a unique mechanism to collect diet information, can reduce the burden of diet keeper thus.The mobile device of obesity patient uses this food recognition device, and anyone can monitor the intake of his/her food.
4) effect that is auxiliary and that remind can also be played to diabetic.
5) equipment is small and exquisite, is easy to carry about with one, easy to use, all can use whenever and wherever possible.
6) intelligent equipment, leads industry tap.Future society will be a wisdom Urban Age, and a lot of work can go to replace with intelligent machine.The research and development of this product, will cause a new change;
embodiment:
Above-mentioned object is realized by following technical scheme:
1) based on a food-classifying recognition methods of SVM, its composition comprises: camera, loads the dsp chip of identification and analytical algorithm, wireless transport module, and USB interface equipment forms, and design achieves a kind of food-classifying recognition methods based on SVM.
2) according to claim 1, a kind of food-classifying recognition methods based on SVM, by using the framework of four kinds of features, comprises color, texture, size and dimension.For color characteristic, employ rim detection and color K-mean cluster.For textural characteristics, use Gabor filter.In order to size and dimension feature, the area-of-interest pixel computing technique of the edge detecting technology employed and image.Next step is classified to the feature extracted, so that identify often kind of food.A classification task generally includes training data and test data, and data are made up of some examples.Each example in training data comprises a class label and several feature.The target of SVM is framework model, and this model can go out the desired value of this example by means of only the attribute forecast of data instance in test data.
3) according to claim 1 and 2, in order to improve accuracy, after SVM module has determined often kind of food type, this device and user have verified food species alternately.Device can demonstrate the image of food to user and explain out the food type that device thinks, as chicken, meat, vegetables etc.Then, user can confirm or change the type of food.Although this change makes device become semi-automation from robotization, but it can increase the accuracy of device.
4) according to claim 1 and 2,3, in this model, use RBF core, it in non-linear mode at more high-dimensional spatial mappings sample.Different from linear kernel, RBF core is very suitable for class mark and attribute is nonlinear situation.RBF core has
two parameters.The target of this step finds
optimum value, to enable the data (i.e. test data) of sorter Accurate Prediction the unknown.The common strategy of of sorter is that data set is divided into two parts, and wherein a part is considered to unknown.Can reflect that sorter is to an independently data set performance of classifying more accurately from the unknown group of precision of prediction obtained.One of this process is improved version and is called as cross validation.The advantage of cross validation is used to be to prevent over-fitting problem.One is found
good mode be use " grid search ".
In the present invention, the proper vector of SVM comprises five textural characteristics, ten color characteristics, three shape facilities and six size characteristics.Extract the proper vector of various food in the segmentation stage, then become the training vector of SVM.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.
Claims (4)
1. based on a food-classifying recognition methods of SVM, its composition comprises: camera, loads the dsp chip of identification and analytical algorithm, wireless transport module, and USB interface equipment forms, and design achieves a kind of food-classifying recognition methods based on SVM.
2. according to claim 1, a kind of food-classifying recognition methods based on SVM, by using the framework of four kinds of features, comprises color, texture, size and dimension; For color characteristic, employ rim detection and color K-mean cluster; For textural characteristics, use Gabor filter; In order to size and dimension feature, the area-of-interest pixel computing technique of the edge detecting technology employed and image; Next step is classified to the feature extracted, so that identify often kind of food; A classification task generally includes training data and test data, and data are made up of some examples; The target of SVM is framework model, and this model can go out the desired value of this example by means of only the attribute forecast of data instance in test data.
3., according to claim 1 and 2, in order to improve accuracy, after SVM module has determined often kind of food type, this device and user have verified food species alternately; Device can demonstrate the image of food to user and explain out the food type that device thinks, as chicken, meat, vegetables etc.; Then, user can confirm or change the type of food; Although this change makes device become semi-automation from robotization, but it can increase the accuracy of device.
4., according to claim 1 and 2,3, in this model, use RBF core, it in non-linear mode at more high-dimensional spatial mappings sample; Different from linear kernel, RBF core is very suitable for class mark and attribute is nonlinear situation; RBF core has
two parameters; The target of this step finds
optimum value, to enable the data (i.e. test data) of sorter Accurate Prediction the unknown; The common strategy of of sorter is that data set is divided into two parts, and wherein a part is considered to unknown; Can reflect that sorter is to an independently data set performance of classifying more accurately from the unknown group of precision of prediction obtained; One of this process is improved version and is called as cross validation; The advantage of cross validation is used to be to prevent over-fitting problem; One is found
good mode be use " grid search "; In the present invention, the proper vector of SVM comprises five textural characteristics, ten color characteristics, three shape facilities and six size characteristics; Extract the proper vector of various food in the segmentation stage, then become the training vector of SVM.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510217604.6A CN104809472A (en) | 2015-05-04 | 2015-05-04 | SVM-based food classifying and recognizing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510217604.6A CN104809472A (en) | 2015-05-04 | 2015-05-04 | SVM-based food classifying and recognizing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104809472A true CN104809472A (en) | 2015-07-29 |
Family
ID=53694283
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510217604.6A Pending CN104809472A (en) | 2015-05-04 | 2015-05-04 | SVM-based food classifying and recognizing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104809472A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203466A (en) * | 2016-06-23 | 2016-12-07 | 珠海市魅族科技有限公司 | The method and apparatus of food identification |
| CN106872513A (en) * | 2017-01-05 | 2017-06-20 | 深圳市金立通信设备有限公司 | A kind of method and terminal for detecting fuel value of food |
| CN109002850A (en) * | 2018-07-06 | 2018-12-14 | 无锡众创未来科技应用有限公司 | The method and device of fuel value of food in a kind of calculating image |
| CN109102413A (en) * | 2018-09-03 | 2018-12-28 | 中国平安人寿保险股份有限公司 | Health index prediction technique, device and storage medium |
| CN109816025A (en) * | 2019-01-29 | 2019-05-28 | 宝鸡文理学院 | A kind of image search method based on image classification |
| CN111768863A (en) * | 2020-06-28 | 2020-10-13 | 暨南大学 | Artificial intelligence-based infant development monitoring system and method |
| CN112215191A (en) * | 2015-11-25 | 2021-01-12 | 三星电子株式会社 | User terminal device and control method thereof |
| CN113449792A (en) * | 2021-06-28 | 2021-09-28 | 四创电子股份有限公司 | Method for nondestructive rapid detection of food quality |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130113933A1 (en) * | 2008-09-05 | 2013-05-09 | Purdue Research Foundation | Dietary Assessment System and Method |
| US20140315161A1 (en) * | 2013-04-18 | 2014-10-23 | Sony Corporation | Information processing apparatus and storage medium |
-
2015
- 2015-05-04 CN CN201510217604.6A patent/CN104809472A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130113933A1 (en) * | 2008-09-05 | 2013-05-09 | Purdue Research Foundation | Dietary Assessment System and Method |
| US20140315161A1 (en) * | 2013-04-18 | 2014-10-23 | Sony Corporation | Information processing apparatus and storage medium |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112215191A (en) * | 2015-11-25 | 2021-01-12 | 三星电子株式会社 | User terminal device and control method thereof |
| US11568981B2 (en) | 2015-11-25 | 2023-01-31 | Samsung Electronics Co., Ltd. | User terminal apparatus and control method thereof |
| CN106203466A (en) * | 2016-06-23 | 2016-12-07 | 珠海市魅族科技有限公司 | The method and apparatus of food identification |
| CN106203466B (en) * | 2016-06-23 | 2020-02-11 | 珠海市魅族科技有限公司 | Food identification method and device |
| CN106872513A (en) * | 2017-01-05 | 2017-06-20 | 深圳市金立通信设备有限公司 | A kind of method and terminal for detecting fuel value of food |
| CN109002850A (en) * | 2018-07-06 | 2018-12-14 | 无锡众创未来科技应用有限公司 | The method and device of fuel value of food in a kind of calculating image |
| CN109102413A (en) * | 2018-09-03 | 2018-12-28 | 中国平安人寿保险股份有限公司 | Health index prediction technique, device and storage medium |
| CN109816025A (en) * | 2019-01-29 | 2019-05-28 | 宝鸡文理学院 | A kind of image search method based on image classification |
| CN111768863A (en) * | 2020-06-28 | 2020-10-13 | 暨南大学 | Artificial intelligence-based infant development monitoring system and method |
| CN113449792A (en) * | 2021-06-28 | 2021-09-28 | 四创电子股份有限公司 | Method for nondestructive rapid detection of food quality |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104809472A (en) | SVM-based food classifying and recognizing method | |
| CN107578060B (en) | Method for classifying dish images based on depth neural network capable of distinguishing areas | |
| CN104778374A (en) | Automatic dietary estimation device based on image processing and recognizing method | |
| CN107591200B (en) | Bone age mark identification and evaluation method and system based on deep learning and image omics | |
| Wang et al. | Area determination of diabetic foot ulcer images using a cascaded two-stage SVM-based classification | |
| Ciocca et al. | Food recognition: a new dataset, experiments, and results | |
| Aguilar et al. | Food recognition using fusion of classifiers based on CNNs | |
| CN109117864A (en) | Coronary heart disease risk prediction technique, model and system based on heterogeneous characteristic fusion | |
| Kaur et al. | Deep neural network for food image classification and nutrient identification: A systematic review | |
| Wang et al. | A review on vision-based analysis for automatic dietary assessment | |
| CN108806792A (en) | Deep learning facial diagnosis system | |
| WO2023159909A1 (en) | Nutritional management method and system using deep learning-based food image recognition model | |
| Ragusa et al. | Food vs non-food classification | |
| CN109977955A (en) | A kind of precancerous lesions of uterine cervix knowledge method for distinguishing based on deep learning | |
| Zhu et al. | Segmentation assisted food classification for dietary assessment | |
| Dalakleidi et al. | Applying image-based food-recognition systems on dietary assessment: a systematic review | |
| Pouladzadeh et al. | Intelligent SVM based food intake measurement system | |
| CN104765980A (en) | Intelligent diet assessment method based on cloud computing | |
| CN114399634A (en) | Three-dimensional image classification method, system, device and medium based on weak supervised learning | |
| Wazumi et al. | Auto-recognition of food images using SPIN feature for Food-Log system | |
| Velvizhy et al. | Automatic food recognition system for diabetic patients | |
| Josephin Shermila et al. | MDEEPFIC: Food item classification with calorie calculation using modified dragon fly deep learning network | |
| Yang | Feature engineering in fine-grained image classification | |
| Izbassar et al. | Image-Based Dietary Assessment: A Healthy Eating Plate Estimation System | |
| CN108229514A (en) | Object detecting method, device and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150729 |
|
| WD01 | Invention patent application deemed withdrawn after publication |