CN108224894A - The recognition methods of food materials freshness, device, refrigerator and medium based on deep learning - Google Patents

The recognition methods of food materials freshness, device, refrigerator and medium based on deep learning Download PDF

Info

Publication number
CN108224894A
CN108224894A CN201810015399.9A CN201810015399A CN108224894A CN 108224894 A CN108224894 A CN 108224894A CN 201810015399 A CN201810015399 A CN 201810015399A CN 108224894 A CN108224894 A CN 108224894A
Authority
CN
China
Prior art keywords
food materials
freshness
identification
image information
deep learning
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810015399.9A
Other languages
Chinese (zh)
Other versions
CN108224894B (en
Inventor
唐红强
戴江
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hefei Midea Intelligent Technologies Co Ltd
Original Assignee
Hefei Midea Intelligent Technologies Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hefei Midea Intelligent Technologies Co Ltd filed Critical Hefei Midea Intelligent Technologies Co Ltd
Priority to CN201810015399.9A priority Critical patent/CN108224894B/en
Publication of CN108224894A publication Critical patent/CN108224894A/en
Application granted granted Critical
Publication of CN108224894B publication Critical patent/CN108224894B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • FMECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F25REFRIGERATION OR COOLING; COMBINED HEATING AND REFRIGERATION SYSTEMS; HEAT PUMP SYSTEMS; MANUFACTURE OR STORAGE OF ICE; LIQUEFACTION SOLIDIFICATION OF GASES
    • F25DREFRIGERATORS; COLD ROOMS; ICE-BOXES; COOLING OR FREEZING APPARATUS NOT OTHERWISE PROVIDED FOR
    • F25D29/00Arrangement or mounting of control or safety devices

Abstract

The present invention provides a kind of recognition methods of food materials freshness, device, refrigerator and medium based on deep learning, wherein, method includes:Acquire food materials image information;According to food materials image information and advance trained food materials freshness identification model, food materials freshness recognition result information is determined;Based on deep learning, according to predetermined period using the food materials image information acquired in food materials freshness identification process as sample, training update food materials freshness identification model.By technical scheme of the present invention, recognition accuracy is higher, can realize real-time identification, and promotion is optimized in the application of food materials freshness identification model, improves the scope of application of food materials freshness identification, can be suitable for complex scene.

Description

The recognition methods of food materials freshness, device, refrigerator and medium based on deep learning
Technical field
The present invention relates to intelligent refrigerator technical field, in particular to a kind of food materials freshness based on deep learning Recognition methods, a kind of food materials freshness identification device, a kind of refrigerator and a kind of computer-readable storage medium based on deep learning Matter.
Background technology
With the development of informationization technology, refrigerator not only carries the function of storage food materials, also progressively towards intelligent family It occupies, the direction for providing more intelligent Services to the user is developed, and the identification of food materials freshness is the important information of intelligent Service Offer approach.
In the relevant technologies, food materials freshness identification is generally carried out by images match identification, there are following technological deficiencies:
(1) discrimination is relatively low, it is difficult to cope with the food materials freshness identification under complex scene.
(2) computation complexity is higher, it is difficult to meet the requirement of real time of food materials freshness identification.
(3) food materials freshness recognition result is not supervised effectively, it is difficult to optimize update to existing model.
Invention content
The present invention is directed at least solve one of technical problem present in the prior art or the relevant technologies.
For this purpose, it is an object of the present invention to provide a kind of food materials freshness recognition methods based on deep learning.
It is another object of the present invention to provide a kind of food materials freshness identification devices based on deep learning.
It is yet a further object of the present invention to provide a kind of refrigerators.
A further object of the present invention is to provide a kind of computer readable storage medium.
To achieve these goals, the technical solution of the first aspect of the present invention provides a kind of food based on deep learning Material freshness recognition methods, including:Acquire food materials image information;It is fresh according to food materials image information and advance trained food materials Identification model is spent, determines food materials freshness recognition result information;Based on deep learning, known according to predetermined period with food materials freshness The food materials image information acquired during not is sample, and training updates food materials freshness identification model.
In the technical scheme, by acquiring food materials image information, according to food materials image information and advance trained food Material freshness identification model determines food materials freshness recognition result information, realizes the real-time identification of food materials freshness, Er Qieshi Other accuracy rate is higher, suitable for complex scene, by being based on deep learning, according to predetermined period with food materials freshness identification process The food materials image information of middle acquisition is sample, and training update food materials freshness identification model realizes and food materials freshness is identified The training update of model, optimizes promotion in the application of food materials freshness identification model, is conducive to promote food materials freshness The scope of application of identification further improves the accuracy rate of food materials freshness identification.
It should be noted that deep learning can be convolutional neural networks deep learning or recurrent neural network Deep learning can also be Recognition with Recurrent Neural Network deep learning.
In the above-mentioned technical solutions, it is preferable that food materials freshness recognition result information includes identification freshness and identification is tied Fruit confidence level, based on deep learning, according to the food materials image information that predetermined period is acquired using in food materials freshness identification process as Sample, training update food materials freshness identification model, including:Judge whether recognition result confidence level is less than the first predetermined threshold value; When judging that recognition result confidence level is less than the first predetermined threshold value, the corresponding food materials image information of record recognition result confidence level; Mark the actual fresh degree of the food materials image information recorded in predetermined period;Judge actual fresh degree and corresponding identification freshness It is whether identical;When judging that actual fresh degree is different with identification freshness, it is sample to determine corresponding food materials image information;It is based on Deep learning, according to all samples determined in predetermined period, training update food materials freshness identification model.
In the technical scheme, by judging whether recognition result confidence level is less than the first predetermined threshold value, when judgement identifies As a result when confidence level is less than the first predetermined threshold value, the corresponding food materials image information of record recognition result confidence level is advantageously implemented Confirmation is marked in the food materials image information relatively low to recognition result confidence level, by marking the food materials figure recorded in predetermined period As the actual fresh degree of information, judge whether actual fresh degree and corresponding identification freshness are identical, are conducive to filter out identification The food materials image information of mistake, by when judging that actual fresh degree is different with identification freshness, determining corresponding food materials image Information is sample, based on deep learning, according to all samples determined in predetermined period, training update food materials freshness identification mould Type realizes the supervised learning optimization of food materials freshness identification model, using the food materials image information of identification mistake as sample Training update food materials freshness identification model, while the recognition accuracy for promoting food materials freshness identification model, is also reduced Training burden.
It should be noted that the first predetermined threshold value is 80%.
In any of the above-described technical solution, it is preferable that food materials freshness recognition result information includes recognition failures information, Based on deep learning, according to predetermined period using the food materials image information acquired in food materials freshness identification process as sample, training Food materials freshness identification model is updated, including:Record the corresponding food materials image information of recognition failures information;It determines in predetermined period The food materials type of the food materials image information of record, and correspondence markings actual fresh degree;It is corresponding different practical to increase food materials type Other food materials image informations of freshness, and correspondence markings actual fresh degree;Based on deep learning, with what is recorded in predetermined period Food materials image information and other increased food materials image informations are sample, and training updates food materials freshness identification model.
In the technical scheme, by recording the corresponding food materials image information of recognition failures information, know for food materials freshness The training update of other model provides data foundation, passes through the food materials kind of food materials image information for determining to record in predetermined period Class, and correspondence markings actual fresh degree are conducive to determine that food materials are fresh in the application process of food materials freshness identification model The new food materials type not included in degree identification model is conducive to increase other foods of new food materials type difference actual fresh degree Material image information is trained update food materials freshness identification model, by increasing the corresponding different actual fresh of food materials type Other food materials image informations of degree, and correspondence markings actual fresh degree, are then based on deep learning, with what is recorded in predetermined period Food materials image information and other increased food materials image informations are sample, and training update food materials freshness identification model is conducive to The ability that food materials freshness identification model adapts to new food materials type is promoted, further improves food materials freshness identification model The scope of application.
In any of the above-described technical solution, it is preferable that further include:Acquisition structure food materials image information sets in advance;It is corresponding Food materials image information is marked to concentrate the food materials type and actual fresh degree of each food materials image information;Based on deep learning, with food Material image information concentrates each food materials image information as input sample, training structure food materials freshness identification model.
In the technical scheme, by acquiring structure food materials image information sets in advance, correspondence markings food materials image is believed later Breath concentrates the food materials type and actual fresh degree of each food materials image information, is provided for training structure food materials freshness identification model Data foundations by being based on deep learning, concentrate each food materials image information as input sample, training using food materials image information Food materials freshness identification model is built, the advance training structure of food materials freshness identification model is realized, is advantageously implemented basis Food materials freshness identification model determines food materials freshness recognition result information.
It should be noted that the training structure of food materials freshness identification model carries out in advance, it later will be trained Food materials freshness identification model is stored, and carries out real-time food materials freshness identification, and food materials freshness identification model can deposit It stores up in server, server progress food materials freshness identification, food materials freshness is uploaded to after acquiring food materials image information by terminal Identification model can also be stored in terminal, and food materials freshness identification is directly carried out after acquiring food materials image information by terminal.
In any of the above-described technical solution, it is preferable that based on deep learning, each food materials are concentrated with food materials image information Image information is input sample, training structure food materials freshness identification model, including:Based on deep learning, believed with food materials image Breath concentrates each food materials image information as input sample, successively through convolution layer operation, pond layer operation, normalization operation and entirely Layer operation is connected, exports result information;According to result information and the label information of input sample, counting loss functional value;According to Loss function value, update convolution layer operation, pond layer operation, normalization operation and the full correspondence parameter for connecting layer operation, with Training structure food materials freshness identification model;Judge whether the absolute value of loss function value is less than the second predetermined threshold value;If it is determined that The absolute value of loss function value is greater than or equal to the second predetermined threshold value, then continues to concentrate each food materials image with food materials image information Information is input sample, training structure food materials freshness identification model;If it is determined that loss function value is less than the second predetermined threshold value, then Store trained food materials freshness identification model.
In the technical scheme, by being based on deep learning, using food materials image information concentrate each food materials image information as Input sample connects through convolution layer operation, pond layer operation, normalization operation and full layer operation successively, exports result information, Later according to result information and the label information of input sample, counting loss functional value, whether to need to continue training structure food The update of parameter provides data foundation in material freshness identification model and food materials freshness identification model, by according to loss letter Numerical value, update convolution layer operation, pond layer operation, normalization operation and the full correspondence parameter for connecting layer operation, to train structure Food materials freshness identification model is built, the gradual training structure for realizing food materials freshness identification model is perfect so that trains and Food materials freshness identification model be more bonded reality, recognition accuracy higher, by judge loss function value absolute value whether Less than the second predetermined threshold value, and when the absolute value for judging loss function value is greater than or equal to the second predetermined threshold value, continue to eat Material image information concentrates each food materials image information as input sample, and training structure food materials freshness identification model is damaged in judgement When losing functional value less than the second predetermined threshold value, trained food materials freshness identification model is stored, on the one hand, be conducive to promote food The training validity of material freshness identification model, on the other hand so that train the food materials freshness identification model come and be more bonded It is practical, recognition accuracy higher.
It should be noted that convolution layer operation refers to that checking food materials image information using specified convolution carries out convolution algorithm, Increase biasing, feature extraction obtains characteristic image information, can realize the dimensionality reduction of food materials image information;Pond layer operation refers to Using spatial coherence, the down-sampled polymerization of pondization, the redundancy of reduction characteristic image information;Normalization operation refers to input Parameter again it is regular to mean value be 0, variance be 1 normal distribution, generalization ability, training for promotion speed can be promoted;Full connection Layer operation refers to that comprehensive characteristics image information obtains feature vector.
It should also be noted that, the second predetermined threshold value is the numerical value close to 0, such as 0.001.
The technical solution of the second aspect of the present invention proposes a kind of food materials freshness identification device based on deep learning, Including:Collecting unit, for acquiring food materials image information;Determination unit, for training according to food materials image information and in advance Food materials freshness identification model, determine food materials freshness recognition result information;Training updating unit, for being based on depth It practises, according to predetermined period using the food materials image information acquired in food materials freshness identification process as sample, training update food materials are new Freshness identification model.
In the technical scheme, by acquiring food materials image information, according to food materials image information and advance trained food Material freshness identification model determines food materials freshness recognition result information, realizes the real-time identification of food materials freshness, Er Qieshi Other accuracy rate is higher, suitable for complex scene, by being based on deep learning, according to predetermined period with food materials freshness identification process The food materials image information of middle acquisition is sample, and training update food materials freshness identification model realizes and food materials freshness is identified The training update of model, optimizes promotion in the application of food materials freshness identification model, is conducive to promote food materials freshness The scope of application of identification further improves the accuracy rate of food materials freshness identification.
It should be noted that deep learning can be convolutional neural networks deep learning or recurrent neural network Deep learning can also be Recognition with Recurrent Neural Network deep learning.
In the above-mentioned technical solutions, it is preferable that food materials freshness recognition result information includes identification freshness and identification is tied Fruit confidence level, food materials freshness identification device further include:Judging unit, for judging whether recognition result confidence level is less than first Predetermined threshold value;First recording unit, for when judging that recognition result confidence level is less than the first predetermined threshold value, recording recognition result The corresponding food materials image information of confidence level;Indexing unit, for marking the reality of food materials image information recorded in predetermined period Freshness;Judging unit is additionally operable to:Judge whether actual fresh degree and corresponding identification freshness are identical;Determination unit is also used In:When judging that actual fresh degree is different with identification freshness, it is sample to determine corresponding food materials image information;Training update is single Member is additionally operable to:Based on deep learning, according to all samples determined in predetermined period, training update food materials freshness identification mould Type.
In the technical scheme, by judging whether recognition result confidence level is less than the first predetermined threshold value, when judgement identifies As a result when confidence level is less than the first predetermined threshold value, the corresponding food materials image information of record recognition result confidence level is advantageously implemented Confirmation is marked in the food materials image information relatively low to recognition result confidence level, by marking the food materials figure recorded in predetermined period As the actual fresh degree of information, judge whether actual fresh degree and corresponding identification freshness are identical, are conducive to filter out identification The food materials image information of mistake, by when judging that actual fresh degree is different with identification freshness, determining corresponding food materials image Information is sample, based on deep learning, according to all samples determined in predetermined period, training update food materials freshness identification mould Type realizes the supervised learning optimization of food materials freshness identification model, using the food materials image information of identification mistake as sample Training update food materials freshness identification model, while the recognition accuracy for promoting food materials freshness identification model, is also reduced Training burden.
It should be noted that the first predetermined threshold value is 80%.
In any of the above-described technical solution, it is preferable that food materials freshness recognition result information includes recognition failures information, Food materials freshness identification device further includes:Second recording unit, for recording the corresponding food materials image information of recognition failures information; Determination unit is additionally operable to:Determine the food materials type of the food materials image information recorded in predetermined period, and correspondence markings actual fresh Degree;Food materials freshness identification device further includes:Adding unit, for increase the corresponding different actual fresh degree of food materials type its His food materials image information, and correspondence markings actual fresh degree;Training updating unit is additionally operable to:Based on deep learning, with default week The food materials image information and other increased food materials image informations recorded in phase is sample, and training updates food materials freshness identification mould Type.
In the technical scheme, by recording the corresponding food materials image information of recognition failures information, know for food materials freshness The training update of other model provides data foundation, passes through the food materials kind of food materials image information for determining to record in predetermined period Class, and correspondence markings actual fresh degree are conducive to determine that food materials are fresh in the application process of food materials freshness identification model The new food materials type not included in degree identification model is conducive to increase other foods of new food materials type difference actual fresh degree Material image information is trained update food materials freshness identification model, by increasing the corresponding different actual fresh of food materials type Other food materials image informations of degree, and correspondence markings actual fresh degree, are then based on deep learning, with what is recorded in predetermined period Food materials image information and other increased food materials image informations are sample, and training update food materials freshness identification model is conducive to The ability that food materials freshness identification model adapts to new food materials type is promoted, further improves food materials freshness identification model The scope of application.
In any of the above-described technical solution, it is preferable that further include:Construction unit is acquired, for acquiring structure food in advance Material image information sets;Indexing unit is additionally operable to:Correspondence markings food materials image information concentrates the food materials kind of each food materials image information Class and actual fresh degree;Food materials freshness identification device further includes:Training construction unit, for being based on deep learning, with food materials Image information concentrates each food materials image information as input sample, training structure food materials freshness identification model.
In the technical scheme, by acquiring structure food materials image information sets in advance, correspondence markings food materials image is believed later Breath concentrates the food materials type and actual fresh degree of each food materials image information, is provided for training structure food materials freshness identification model Data foundations by being based on deep learning, concentrate each food materials image information as input sample, training using food materials image information Food materials freshness identification model is built, the advance training structure of food materials freshness identification model is realized, is advantageously implemented basis Food materials freshness identification model determines food materials freshness recognition result information.
It should be noted that the training structure of food materials freshness identification model carries out in advance, it later will be trained Food materials freshness identification model is stored, and carries out real-time food materials freshness identification, and food materials freshness identification model can deposit It stores up in server, server progress food materials freshness identification, food materials freshness is uploaded to after acquiring food materials image information by terminal Identification model can also be stored in terminal, and food materials freshness identification is directly carried out after acquiring food materials image information by terminal.
In any of the above-described technical solution, it is preferable that further include:Output unit, for being based on deep learning, with food materials Image information concentrates each food materials image information as input sample, successively through convolution layer operation, pond layer operation, normalization operation And layer operation is connected entirely, export result information;Computing unit, for the label information according to result information and input sample, Counting loss functional value;Training construction unit is additionally operable to:According to loss function value, update convolution layer operation, is returned at pond layer operation One changes the correspondence parameter of operation and full connection layer operation, and food materials freshness identification model is built with training;Judging unit is also used In:Judge whether the absolute value of loss function value is less than the second predetermined threshold value;Training construction unit is additionally operable to:If it is determined that loss letter The absolute value of numerical value is greater than or equal to the second predetermined threshold value, then continue using food materials image information concentrate each food materials image information as Input sample, training structure food materials freshness identification model;Food materials freshness identification device further includes:Storage unit, if for Judge that loss function value is less than the second predetermined threshold value, then store trained food materials freshness identification model.
In the technical scheme, by being based on deep learning, using food materials image information concentrate each food materials image information as Input sample connects through convolution layer operation, pond layer operation, normalization operation and full layer operation successively, exports result information, Later according to result information and the label information of input sample, counting loss functional value, whether to need to continue training structure food The update of parameter provides data foundation in material freshness identification model and food materials freshness identification model, by according to loss letter Numerical value, update convolution layer operation, pond layer operation, normalization operation and the full correspondence parameter for connecting layer operation, to train structure Food materials freshness identification model is built, the gradual training structure for realizing food materials freshness identification model is perfect so that trains and Food materials freshness identification model be more bonded reality, recognition accuracy higher, by judge loss function value absolute value whether Less than the second predetermined threshold value, and when the absolute value for judging loss function value is greater than or equal to the second predetermined threshold value, continue to eat Material image information concentrates each food materials image information as input sample, and training structure food materials freshness identification model is damaged in judgement When losing functional value less than the second predetermined threshold value, trained food materials freshness identification model is stored, on the one hand, be conducive to promote food The training validity of material freshness identification model, on the other hand so that train the food materials freshness identification model come and be more bonded It is practical, recognition accuracy higher.
It should be noted that convolution layer operation refers to that checking food materials image information using specified convolution carries out convolution algorithm, Increase biasing, feature extraction obtains characteristic image information, can realize the dimensionality reduction of food materials image information;Pond layer operation refers to Using spatial coherence, the down-sampled polymerization of pondization, the redundancy of reduction characteristic image information;Normalization operation refers to input Parameter again it is regular to mean value be 0, variance be 1 normal distribution, generalization ability, training for promotion speed can be promoted;Full connection Layer operation refers to that comprehensive characteristics image information obtains feature vector.
It should also be noted that, the second predetermined threshold value is the numerical value close to 0, such as 0.001.
The technical solution of the third aspect of the present invention proposes a kind of refrigerator, and refrigerator includes the second party such as the invention described above The food materials freshness identification device based on deep learning for any one that the technical solution in face proposes;Image acquisition device, set on ice In case, it is connected with food materials freshness identification device, for acquiring food materials image letter under the control of food materials freshness identification device Breath.
In the technical scheme, refrigerator includes the base of any one that the technical solution of the second aspect of the invention described above proposes In the food materials freshness identification device of deep learning, therefore times that the technical solution of the second aspect with the invention described above proposes Whole advantageous effects of the food materials freshness identification device based on deep learning of one, details are not described herein, by being set on ice The image acquisition device being connected in case and with food materials freshness identification device, can adopt under the control of food materials freshness identification device Collect food materials image information, be conducive to collect accurate food materials image information, and then be conducive to promote the identification of food materials freshness Accuracy.
The technical solution of the fourth aspect of the present invention proposes a kind of computer readable storage medium, is stored thereon with calculating Machine program realizes the base of any one that the technical solution of the first aspect of the present invention proposes when computer program is executed by processor In the food materials freshness recognition methods of deep learning the step of.
In the technical scheme, computer readable storage medium, is stored thereon with computer program, and computer program is located Reason device realizes that the food materials based on deep learning of any one that the technical solution of the first aspect of the present invention proposes are fresh when performing Spend recognition methods the step of, therefore with the invention described above first aspect technical solution propose any one based on depth Whole advantageous effects of the food materials freshness recognition methods of study, details are not described herein.
By above technical scheme, structure and update food materials freshness identification model are trained based on deep learning, passed through Food materials freshness identification model identifies food materials freshness, and recognition accuracy is higher, can realize real-time identification, and new in food materials Promotion is optimized in the application of freshness identification model, improves the scope of application of food materials freshness identification, can be suitable for multiple Miscellaneous scene.
The additional aspect and advantage of the present invention will provide in following description section, will partly become from the following description It obtains significantly or is recognized by the practice of the present invention.
Description of the drawings
The above-mentioned and/or additional aspect and advantage of the present invention will become in the description from combination accompanying drawings below to embodiment Significantly and it is readily appreciated that, wherein:
Fig. 1 shows showing for the food materials freshness recognition methods according to an embodiment of the invention based on deep learning Meaning flow chart;
Fig. 2 shows showing for the food materials freshness identification device according to an embodiment of the invention based on deep learning Meaning block diagram;
Fig. 3 shows the schematic block diagram of refrigerator according to an embodiment of the invention;
Fig. 4 shows the food materials freshness recognition methods based on deep learning according to another embodiment of the invention Schematic flow diagram.
Specific embodiment
It is to better understand the objects, features and advantages of the present invention, below in conjunction with the accompanying drawings and specific real Mode is applied the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application Feature in example and embodiment can be combined with each other.
Many details are elaborated in the following description to facilitate a thorough understanding of the present invention, still, the present invention may be used also To be implemented using other different from other modes described here, therefore, protection scope of the present invention is not by described below Specific embodiment limitation.
Fig. 1 shows showing for the food materials freshness recognition methods according to an embodiment of the invention based on deep learning Meaning flow chart.
As shown in Figure 1, the food materials freshness recognition methods according to an embodiment of the invention based on deep learning, including:
S102 acquires food materials image information;
S104 according to food materials image information and advance trained food materials freshness identification model, determines food materials freshness Recognition result information;
S106 based on deep learning, is believed according to predetermined period with the food materials image acquired in food materials freshness identification process It ceases for sample, training update food materials freshness identification model.
In this embodiment, by acquiring food materials image information, according to food materials image information and advance trained food materials Freshness identification model determines food materials freshness recognition result information, realizes the real-time identification of food materials freshness, and identify Accuracy rate is higher, suitable for complex scene, by being based on deep learning, according to predetermined period in food materials freshness identification process The food materials image information of acquisition is sample, and training update food materials freshness identification model realizes and identifies mould to food materials freshness The training update of type, optimizes promotion in the application of food materials freshness identification model, is conducive to promote the knowledge of food materials freshness Other scope of application further improves the accuracy rate of food materials freshness identification.
It should be noted that deep learning can be convolutional neural networks deep learning or recurrent neural network Deep learning can also be Recognition with Recurrent Neural Network deep learning.
In the above embodiment, it is preferable that food materials freshness recognition result information includes identification freshness and recognition result Confidence level, based on deep learning, according to predetermined period using the food materials image information acquired in food materials freshness identification process as sample This, training update food materials freshness identification model, including:Judge whether recognition result confidence level is less than the first predetermined threshold value;When When judging that recognition result confidence level is less than the first predetermined threshold value, the corresponding food materials image information of record recognition result confidence level;Mark Remember the actual fresh degree of the food materials image information recorded in predetermined period;Judging actual fresh degree and corresponding identification freshness is It is no identical;When judging that actual fresh degree is different with identification freshness, it is sample to determine corresponding food materials image information;Based on depth Degree study, according to all samples determined in predetermined period, training update food materials freshness identification model.
In this embodiment, by judging whether recognition result confidence level is less than the first predetermined threshold value, when judgement identification knot When fruit confidence level is less than the first predetermined threshold value, the corresponding food materials image information of record recognition result confidence level is advantageously implemented pair Confirmation is marked in the relatively low food materials image information of recognition result confidence level, by marking the food materials image recorded in predetermined period The actual fresh degree of information judges whether actual fresh degree and corresponding identification freshness are identical, are conducive to filter out identification mistake Food materials image information accidentally, by when judging that actual fresh degree is different with identification freshness, determining corresponding food materials image letter It ceases for sample, based on deep learning, according to all samples determined in predetermined period, training update food materials freshness identification mould Type realizes the supervised learning optimization of food materials freshness identification model, using the food materials image information of identification mistake as sample Training update food materials freshness identification model, while the recognition accuracy for promoting food materials freshness identification model, is also reduced Training burden.
It should be noted that the first predetermined threshold value is 80%.
In any of the above-described embodiment, it is preferable that food materials freshness recognition result information includes recognition failures information, base In deep learning, according to predetermined period using the food materials image information acquired in food materials freshness identification process as sample, training is more New food materials freshness identification model, including:Record the corresponding food materials image information of recognition failures information;It determines to remember in predetermined period The food materials type of the food materials image information of record, and correspondence markings actual fresh degree;It is corresponding different practical new to increase food materials type Other food materials image informations of freshness, and correspondence markings actual fresh degree;Based on deep learning, with the food recorded in predetermined period Material image information and other increased food materials image informations are sample, and training updates food materials freshness identification model.
In this embodiment, it by recording the corresponding food materials image information of recognition failures information, is identified for food materials freshness The training update of model provides data foundation, by determining the food materials type of food materials image information that records in predetermined period, And correspondence markings actual fresh degree, be conducive to determine that food materials freshness is known in the application process of food materials freshness identification model The new food materials type not included in other model is conducive to increase other food materials figures of new food materials type difference actual fresh degree As information, update food materials freshness identification model is trained, by increasing the corresponding different actual fresh degree of food materials type Other food materials image informations, and correspondence markings actual fresh degree, are then based on deep learning, with the food materials recorded in predetermined period Image information and other increased food materials image informations are sample, and training update food materials freshness identification model is conducive to be promoted Food materials freshness identification model adapts to the ability of new food materials type, further improves being applicable in for food materials freshness identification model Range.
In any of the above-described embodiment, it is preferable that further include:Acquisition structure food materials image information sets in advance;Corresponding mark Remember that food materials image information concentrates the food materials type and actual fresh degree of each food materials image information;Based on deep learning, with food materials Image information concentrates each food materials image information as input sample, training structure food materials freshness identification model.
In this embodiment, food materials image information sets are built by acquiring in advance, later correspondence markings food materials image information The food materials type of each food materials image information and actual fresh degree are concentrated, is provided for training structure food materials freshness identification model Data foundation by being based on deep learning, concentrates each food materials image information as input sample, training structure using food materials image information Food materials freshness identification model is built, the advance training structure of food materials freshness identification model is realized, is advantageously implemented according to food Material freshness identification model determines food materials freshness recognition result information.
It should be noted that the training structure of food materials freshness identification model carries out in advance, it later will be trained Food materials freshness identification model is stored, and carries out real-time food materials freshness identification, and food materials freshness identification model can deposit It stores up in server, server progress food materials freshness identification, food materials freshness is uploaded to after acquiring food materials image information by terminal Identification model can also be stored in terminal, and food materials freshness identification is directly carried out after acquiring food materials image information by terminal.
In any of the above-described embodiment, it is preferable that based on deep learning, each food materials figure is concentrated with food materials image information Picture information is input sample, and training builds food materials freshness identification model, including:Based on deep learning, with food materials image information Each food materials image information is concentrated as input sample, is connected successively through convolution layer operation, pond layer operation, normalization operation and entirely Layer operation is connect, exports result information;According to result information and the label information of input sample, counting loss functional value;According to damage Lose functional value, update convolution layer operation, pond layer operation, normalization operation and the full correspondence parameter for connecting layer operation, with instruction Practice structure food materials freshness identification model;Judge whether the absolute value of loss function value is less than the second predetermined threshold value;If it is determined that damage The absolute value for losing functional value is greater than or equal to the second predetermined threshold value, then continues each food materials image to be concentrated to believe with food materials image information It ceases for input sample, training structure food materials freshness identification model;If it is determined that loss function value is less than the second predetermined threshold value, then deposit Store up trained food materials freshness identification model.
In this embodiment, by being based on deep learning, each food materials image information is concentrated to be defeated using food materials image information Enter sample, connect through convolution layer operation, pond layer operation, normalization operation and full layer operation successively, export result information, it Afterwards according to result information and the label information of input sample, counting loss functional value, whether to need to continue training structure food materials The update of parameter provides data foundation in freshness identification model and food materials freshness identification model, by according to loss function Value, update convolution layer operation, pond layer operation, normalization operation and the full correspondence parameter for connecting layer operation, is built with training Food materials freshness identification model, the gradual training structure for realizing food materials freshness identification model are perfect so that train what is come Food materials freshness identification model is more bonded reality, recognition accuracy higher, and whether the absolute value by judging loss function value is small In the second predetermined threshold value, and when the absolute value for judging loss function value is greater than or equal to the second predetermined threshold value, continue with food materials Image information concentrates each food materials image information as input sample, and training structure food materials freshness identification model is lost in judgement When functional value is less than the second predetermined threshold value, trained food materials freshness identification model is stored, on the one hand, be conducive to promote food materials The training validity of freshness identification model, on the other hand so that train the food materials freshness identification model come and be more bonded reality Border, recognition accuracy higher.
It should be noted that convolution layer operation refers to that checking food materials image information using specified convolution carries out convolution algorithm, Increase biasing, feature extraction obtains characteristic image information, can realize the dimensionality reduction of food materials image information;Pond layer operation refers to Using spatial coherence, the down-sampled polymerization of pondization, the redundancy of reduction characteristic image information;Normalization operation refers to input Parameter again it is regular to mean value be 0, variance be 1 normal distribution, generalization ability, training for promotion speed can be promoted;Full connection Layer operation refers to that comprehensive characteristics image information obtains feature vector.
It should also be noted that, the second predetermined threshold value is the numerical value close to 0, such as 0.001.
Fig. 2 shows the food materials freshness identification devices 200 according to an embodiment of the invention based on deep learning Schematic block diagram.
As shown in Fig. 2, the food materials freshness identification device 200 according to an embodiment of the invention based on deep learning, packet It includes:Collecting unit 202, for acquiring food materials image information;Determination unit 204, for according to food materials image information and advance instruction The food materials freshness identification model perfected determines food materials freshness recognition result information;Training updating unit 206, for being based on Deep learning, according to predetermined period using the food materials image information acquired in food materials freshness identification process as sample, training update Food materials freshness identification model.
In this embodiment, by acquiring food materials image information, according to food materials image information and advance trained food materials Freshness identification model determines food materials freshness recognition result information, realizes the real-time identification of food materials freshness, and identify Accuracy rate is higher, suitable for complex scene, by being based on deep learning, according to predetermined period in food materials freshness identification process The food materials image information of acquisition is sample, and training update food materials freshness identification model realizes and identifies mould to food materials freshness The training update of type, optimizes promotion in the application of food materials freshness identification model, is conducive to promote the knowledge of food materials freshness Other scope of application further improves the accuracy rate of food materials freshness identification.
It should be noted that deep learning can be convolutional neural networks deep learning or recurrent neural network Deep learning can also be Recognition with Recurrent Neural Network deep learning.
In the above embodiment, it is preferable that food materials freshness recognition result information includes identification freshness and recognition result Confidence level, food materials freshness identification device 200 further include:Judging unit 208, for judging whether recognition result confidence level is less than First predetermined threshold value;First recording unit 210, for when judging that recognition result confidence level is less than the first predetermined threshold value, recording The corresponding food materials image information of recognition result confidence level;Indexing unit 212, for marking the food materials image recorded in predetermined period The actual fresh degree of information;Judging unit 208 is additionally operable to:Judge whether actual fresh degree and corresponding identification freshness are identical; Determination unit 204 is additionally operable to:When judging that actual fresh degree is different with identification freshness, determine that corresponding food materials image information is Sample;Training updating unit 206 is additionally operable to:Based on deep learning, according to all samples determined in predetermined period, training update Food materials freshness identification model.
In this embodiment, by judging whether recognition result confidence level is less than the first predetermined threshold value, when judgement identification knot When fruit confidence level is less than the first predetermined threshold value, the corresponding food materials image information of record recognition result confidence level is advantageously implemented pair Confirmation is marked in the relatively low food materials image information of recognition result confidence level, by marking the food materials image recorded in predetermined period The actual fresh degree of information judges whether actual fresh degree and corresponding identification freshness are identical, are conducive to filter out identification mistake Food materials image information accidentally, by when judging that actual fresh degree is different with identification freshness, determining corresponding food materials image letter It ceases for sample, based on deep learning, according to all samples determined in predetermined period, training update food materials freshness identification mould Type realizes the supervised learning optimization of food materials freshness identification model, using the food materials image information of identification mistake as sample Training update food materials freshness identification model, while the recognition accuracy for promoting food materials freshness identification model, is also reduced Training burden.
It should be noted that the first predetermined threshold value is 80%.
In any of the above-described embodiment, it is preferable that food materials freshness recognition result information includes recognition failures information, food Material freshness identification device 200 further includes:Second recording unit 214, for recording the corresponding food materials image of recognition failures information Information;Determination unit 204 is additionally operable to:Determine the food materials type of the food materials image information recorded in predetermined period, and correspondence markings Actual fresh degree;Food materials freshness identification device 200 further includes:Adding unit 216, for increasing the corresponding difference of food materials type Other food materials image informations of actual fresh degree, and correspondence markings actual fresh degree;Training updating unit 206 is additionally operable to:It is based on Deep learning, using the food materials image information and other increased food materials image informations recorded in predetermined period as sample, training is more New food materials freshness identification model.
In this embodiment, it by recording the corresponding food materials image information of recognition failures information, is identified for food materials freshness The training update of model provides data foundation, by determining the food materials type of food materials image information that records in predetermined period, And correspondence markings actual fresh degree, be conducive to determine that food materials freshness is known in the application process of food materials freshness identification model The new food materials type not included in other model is conducive to increase other food materials figures of new food materials type difference actual fresh degree As information, update food materials freshness identification model is trained, by increasing the corresponding different actual fresh degree of food materials type Other food materials image informations, and correspondence markings actual fresh degree, are then based on deep learning, with the food materials recorded in predetermined period Image information and other increased food materials image informations are sample, and training update food materials freshness identification model is conducive to be promoted Food materials freshness identification model adapts to the ability of new food materials type, further improves being applicable in for food materials freshness identification model Range.
In any of the above-described embodiment, it is preferable that further include:Construction unit 218 is acquired, for acquiring structure food in advance Material image information sets;Indexing unit 212 is additionally operable to:Correspondence markings food materials image information concentrates the food materials of each food materials image information Type and actual fresh degree;Food materials freshness identification device 200 further includes:Training construction unit 220, for being based on depth It practises, each food materials image information is concentrated as input sample using food materials image information, training structure food materials freshness identification model.
In this embodiment, food materials image information sets are built by acquiring in advance, later correspondence markings food materials image information The food materials type of each food materials image information and actual fresh degree are concentrated, is provided for training structure food materials freshness identification model Data foundation by being based on deep learning, concentrates each food materials image information as input sample, training structure using food materials image information Food materials freshness identification model is built, the advance training structure of food materials freshness identification model is realized, is advantageously implemented according to food Material freshness identification model determines food materials freshness recognition result information.
It should be noted that the training structure of food materials freshness identification model carries out in advance, it later will be trained Food materials freshness identification model is stored, and carries out real-time food materials freshness identification, and food materials freshness identification model can deposit It stores up in server, server progress food materials freshness identification, food materials freshness is uploaded to after acquiring food materials image information by terminal Identification model can also be stored in terminal, and food materials freshness identification is directly carried out after acquiring food materials image information by terminal.
In any of the above-described embodiment, it is preferable that further include:Output unit 222, for being based on deep learning, with food Material image information concentrates each food materials image information as input sample, is grasped successively through convolution layer operation, pond layer operation, normalization Make and connect layer operation entirely, export result information;Computing unit 224, for the label according to result information and input sample Information, counting loss functional value;Training construction unit 220 is additionally operable to:According to loss function value, update convolution layer operation, Chi Hua The correspondence parameter of layer operation, normalization operation and full connection layer operation, food materials freshness identification model is built with training;Judge Unit 208 is additionally operable to:Judge whether the absolute value of loss function value is less than the second predetermined threshold value;Training construction unit 220 is also used In:If it is determined that the absolute value of loss function value is greater than or equal to the second predetermined threshold value, then continue to concentrate with food materials image information every One food materials image information is input sample, and training builds food materials freshness identification model;Food materials freshness identification device 200 also wraps It includes:Storage unit 226, for if it is determined that loss function value then stores trained food materials freshness less than the second predetermined threshold value Identification model.
In this embodiment, by being based on deep learning, each food materials image information is concentrated to be defeated using food materials image information Enter sample, connect through convolution layer operation, pond layer operation, normalization operation and full layer operation successively, export result information, it Afterwards according to result information and the label information of input sample, counting loss functional value, whether to need to continue training structure food materials The update of parameter provides data foundation in freshness identification model and food materials freshness identification model, by according to loss function Value, update convolution layer operation, pond layer operation, normalization operation and the full correspondence parameter for connecting layer operation, is built with training Food materials freshness identification model, the gradual training structure for realizing food materials freshness identification model are perfect so that train what is come Food materials freshness identification model is more bonded reality, recognition accuracy higher, and whether the absolute value by judging loss function value is small In the second predetermined threshold value, and when the absolute value for judging loss function value is greater than or equal to the second predetermined threshold value, continue with food materials Image information concentrates each food materials image information as input sample, and training structure food materials freshness identification model is lost in judgement When functional value is less than the second predetermined threshold value, trained food materials freshness identification model is stored, on the one hand, be conducive to promote food materials The training validity of freshness identification model, on the other hand so that train the food materials freshness identification model come and be more bonded reality Border, recognition accuracy higher.
It should be noted that convolution layer operation refers to that checking food materials image information using specified convolution carries out convolution algorithm, Increase biasing, feature extraction obtains characteristic image information, can realize the dimensionality reduction of food materials image information;Pond layer operation refers to Using spatial coherence, the down-sampled polymerization of pondization, the redundancy of reduction characteristic image information;Normalization operation refers to input Parameter again it is regular to mean value be 0, variance be 1 normal distribution, generalization ability, training for promotion speed can be promoted;Full connection Layer operation refers to that comprehensive characteristics image information obtains feature vector.
It should also be noted that, the second predetermined threshold value is the numerical value close to 0, such as 0.001.
Fig. 3 shows the schematic block diagram of refrigerator 300 according to an embodiment of the invention.
As shown in figure 3, refrigerator 300 according to an embodiment of the invention, refrigerator 300 is included such as the invention described above, implements The food materials freshness identification device 200 based on deep learning for any one that example proposes;Image acquisition device 302, set on refrigerator 300 It is interior, it is connected with food materials freshness identification device 200, for acquiring food materials figure under the control of food materials freshness identification device 200 As information.
In this embodiment, refrigerator 300 include embodiments of the invention described above propose any one based on deep learning Food materials freshness identification device 200, therefore with embodiments of the invention described above propose any one based on deep learning Food materials freshness identification device 200 whole advantageous effects, details are not described herein, by being set in refrigerator 300 and and food materials The connected image acquisition device 302 of freshness identification device 200, can acquire under the control of food materials freshness identification device 200 Food materials image information is conducive to collect accurate food materials image information, and then is conducive to promote the identification of food materials freshness Accuracy.
Computer readable storage medium according to an embodiment of the invention is stored thereon with computer program, computer journey The food materials freshness based on deep learning of any one that embodiments of the invention described above propose is realized when sequence is executed by processor The step of recognition methods.
In this embodiment, computer readable storage medium, is stored thereon with computer program, and computer program is handled The food materials freshness recognition methods based on deep learning of any one that embodiments of the invention described above propose is realized when device performs The step of, therefore the food materials freshness identification side based on deep learning of any one proposed with embodiments of the invention described above Whole advantageous effects of method, details are not described herein.
Fig. 4 shows the food materials freshness recognition methods based on deep learning according to another embodiment of the invention Schematic flow diagram.
As shown in figure 4, the food materials freshness recognition methods according to an embodiment of the invention based on deep learning, including:
S402 acquires food materials image information;
S404 according to food materials image information and advance trained food materials freshness identification model, determines food materials freshness Recognition result information;
S406, it is that identification freshness and recognition result confidence level or identification are lost to judge food materials freshness recognition result information Lose information;
If food materials freshness recognition result information is identification freshness and recognition result confidence level, S408 is performed, is judged Whether recognition result confidence level is less than the first predetermined threshold value;
If the judgement to S408 is yes, S410 is performed, records the corresponding food materials image information of recognition result confidence level, such as Fruit is no to the judgement of S408, then terminates;
S412, the actual fresh degree of the interior food materials image information recorded of label predetermined period;
S414 judges whether actual fresh degree and corresponding identification freshness are identical;
If the judgement to S414 is yes, terminate, if being no to the judgement of S414, perform S416, determine to correspond to Food materials image information be sample;
S418, based on deep learning, according to all samples determined in predetermined period, training update food materials freshness identification Model;
If food materials freshness recognition result information is recognition failures information in S406, S420 is performed, record identification is lost Lose the corresponding food materials image information of information;
S422 determines the food materials type of the food materials graphical information recorded in predetermined period, and correspondence markings actual fresh degree;
S424 increases other food materials image informations of the corresponding different actual fresh degree of food materials type, and correspondence markings are real Border freshness;
S426, based on deep learning, with the food materials image information recorded in predetermined period and other increased food materials images Information is sample, training update food materials freshness identification model.
Above in association with attached drawing it is described in detail technical scheme of the present invention, the present invention proposes a kind of based on deep learning The recognition methods of food materials freshness, a kind of food materials freshness identification device, a kind of refrigerator and a kind of computer based on deep learning Readable storage medium storing program for executing is trained structure and update food materials freshness identification model based on deep learning, is known by food materials freshness Other Model Identification food materials freshness, recognition accuracy is higher, can realize real-time identification, and in food materials freshness identification model Application in optimize promotion, improve food materials freshness identification the scope of application, complex scene can be suitable for.
Step in the method for the present invention can be sequentially adjusted, combined, and deleted according to actual needs.
Unit in apparatus of the present invention can be combined, divided, and deleted according to actual needs.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is can It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage Medium include read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), programmable read only memory (Programmable Read-only Memory, PROM), erasable programmable is read-only deposits Reservoir (Erasable Programmable Read Only Memory, EPROM), disposable programmable read-only memory (One- Time Programmable Read-Only Memory, OTPROM), the electronics formula of erasing can make carbon copies read-only memory (Electrically-Erasable Programmable Read-Only Memory, EEPROM), CD-ROM (Compact Disc Read-Only Memory, CD-ROM) or other disk storages, magnetic disk storage, magnetic tape storage or can For carrying or storing any other computer-readable medium of data.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, that is made any repaiies Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.

Claims (12)

1. a kind of food materials freshness recognition methods based on deep learning, which is characterized in that including:
Acquire food materials image information;
According to the food materials image information and advance trained food materials freshness identification model, food materials freshness identification knot is determined Fruit information;
Based on deep learning, according to predetermined period using the food materials image information acquired in food materials freshness identification process as sample This, training updates the food materials freshness identification model.
2. the food materials freshness recognition methods according to claim 1 based on deep learning, which is characterized in that the food materials Freshness recognition result information includes identification freshness and recognition result confidence level, described based on deep learning, according to default week Training updated the food materials freshness and knew using the food materials image information acquired in food materials freshness identification process as sample phase Other model, including:
Judge whether the recognition result confidence level is less than the first predetermined threshold value;
When judging that the recognition result confidence level is less than first predetermined threshold value, record the recognition result confidence level and correspond to The food materials image information;
Mark the actual fresh degree of the food materials image information recorded in the predetermined period;
Judge whether the actual fresh degree and the corresponding identification freshness are identical;
When judging the actual fresh degree and the identification freshness difference, it is institute to determine the corresponding food materials image information State sample;
Based on deep learning, according to all samples determined in the predetermined period, training updates the food materials freshness Identification model.
3. the food materials freshness recognition methods according to claim 1 based on deep learning, which is characterized in that the food materials Freshness recognition result information includes recognition failures information, described based on deep learning, according to predetermined period with food materials freshness The food materials image information acquired in identification process is sample, and training updates the food materials freshness identification model, including:
Record the corresponding food materials image information of the recognition failures information;
Determine the food materials type of the food materials image information recorded in the predetermined period, and correspondence markings actual fresh degree;
Increase other food materials image informations of the food materials type corresponding difference actual fresh degree, and described in correspondence markings Actual fresh degree;
Based on deep learning, with the food materials image information recorded in the predetermined period and other increased described food materials figures Picture information is the sample, and training updates the food materials freshness identification model.
4. the food materials freshness recognition methods according to any one of claim 1 to 3 based on deep learning, feature exist In further including:
Acquisition structure food materials image information sets in advance;
Food materials image information described in correspondence markings concentrates the food materials type of each food materials image information and the reality Freshness;
Based on deep learning, each food materials image information is concentrated as input sample using the food materials image information, training structure Build the food materials freshness identification model.
5. the food materials freshness recognition methods according to claim 4 based on deep learning, which is characterized in that described to be based on Deep learning, concentrating each food materials image information using the food materials image information, training builds the food as input sample Material freshness identification model, including:
Based on deep learning, each food materials image information is concentrated as the input sample using the food materials image information, according to It is secondary to connect through convolution layer operation, pond layer operation, normalization operation and full layer operation, export result information;
According to the result information and the label information of the input sample, counting loss functional value;
According to the loss function value, the convolution layer operation, the pond layer operation, the normalization operation and institute are updated The correspondence parameter of full connection layer operation is stated, the food materials freshness identification model is built with training;
Judge whether the absolute value of the loss function value is less than the second predetermined threshold value;
If it is determined that the absolute value of the loss function value is greater than or equal to second predetermined threshold value, then continue with the food materials figure As information concentrates each food materials image information, for the input sample, training builds the food materials freshness identification model;
If it is determined that the loss function value is less than second predetermined threshold value, then the trained food materials freshness identification is stored Model.
6. a kind of food materials freshness identification device based on deep learning, which is characterized in that including:
Collecting unit, for acquiring food materials image information;
Determination unit, for according to the food materials image information and advance trained food materials freshness identification model, determining food Material freshness recognition result information;
Training updating unit, for being based on deep learning, according to institute of the predetermined period to be acquired in food materials freshness identification process Food materials image information is stated as sample, training updates the food materials freshness identification model.
7. the food materials freshness identification device according to claim 6 based on deep learning, which is characterized in that the food materials Freshness recognition result information includes identification freshness and recognition result confidence level, the food materials freshness identification device also wrap It includes:
Judging unit, for judging whether the recognition result confidence level is less than the first predetermined threshold value;
First recording unit, for when judging that the recognition result confidence level is less than first predetermined threshold value, described in record The corresponding food materials image information of recognition result confidence level;
Indexing unit, for marking the actual fresh degree of the food materials image information recorded in the predetermined period;
The judging unit is additionally operable to:Judge whether the actual fresh degree and the corresponding identification freshness are identical;
The determination unit is additionally operable to:When judging the actual fresh degree and the identification freshness difference, determine corresponding The food materials image information is the sample;
The trained updating unit is additionally operable to:Based on deep learning, according to all samples determined in the predetermined period, Training updates the food materials freshness identification model.
8. the food materials freshness identification device according to claim 6 based on deep learning, which is characterized in that the food materials Freshness recognition result information includes recognition failures information, and the food materials freshness identification device further includes:
Second recording unit, for recording the corresponding food materials image information of the recognition failures information;
The determination unit is additionally operable to:Determine the food materials type of the food materials image information recorded in the predetermined period, and Correspondence markings actual fresh degree;
The food materials freshness identification device further includes:
Adding unit, for increasing other food materials image informations of the food materials type corresponding difference actual fresh degree, And actual fresh degree described in correspondence markings;
The trained updating unit is additionally operable to:Based on deep learning, believed with the food materials image recorded in the predetermined period Breath and other increased described food materials image informations are the sample, and training updates the food materials freshness identification model.
9. the food materials freshness identification device based on deep learning according to any one of claim 6 to 8, feature exist In further including:
Construction unit is acquired, for acquiring structure food materials image information sets in advance;
The indexing unit is additionally operable to:Food materials image information described in correspondence markings concentrates the described of each food materials image information Food materials type and the actual fresh degree;
The food materials freshness identification device further includes:
For being based on deep learning, each food materials image information is concentrated with the food materials image information for training construction unit For input sample, training builds the food materials freshness identification model.
10. the food materials freshness identification device according to claim 9 based on deep learning, which is characterized in that further include:
For being based on deep learning, each food materials image information is concentrated as institute using the food materials image information for output unit Input sample is stated, connects through convolution layer operation, pond layer operation, normalization operation and full layer operation successively, output result letter Breath;
Computing unit, for the label information according to the result information and the input sample, counting loss functional value;
The trained construction unit is additionally operable to:According to the loss function value, the convolution layer operation, pond layer behaviour are updated The correspondence parameter of work, the normalization operation and the full connection layer operation, builds the food materials freshness with training and identifies Model;
The judging unit is additionally operable to:Judge whether the absolute value of the loss function value is less than the second predetermined threshold value;
The trained construction unit is additionally operable to:It is preset if it is determined that the absolute value of the loss function value is greater than or equal to described second Threshold value then continues to concentrate each food materials image information as the input sample using the food materials image information, training structure The food materials freshness identification model;
The food materials freshness identification device further includes:
Storage unit, for if it is determined that the loss function value then stores trained described less than second predetermined threshold value Food materials freshness identification model.
11. a kind of refrigerator, which is characterized in that including:
The food materials freshness identification device based on deep learning as described in any one of claim 6 to 10;
Image acquisition device in the refrigerator, is connected with the food materials freshness identification device, for fresh in the food materials It spends and food materials image information is acquired under the control of identification device.
12. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program The food materials freshness identification side based on deep learning as described in any one of claim 1 to 5 is realized when being executed by processor The step of method.
CN201810015399.9A 2018-01-08 2018-01-08 Food material freshness identification method and device based on deep learning, refrigerator and medium Active CN108224894B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810015399.9A CN108224894B (en) 2018-01-08 2018-01-08 Food material freshness identification method and device based on deep learning, refrigerator and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810015399.9A CN108224894B (en) 2018-01-08 2018-01-08 Food material freshness identification method and device based on deep learning, refrigerator and medium

Publications (2)

Publication Number Publication Date
CN108224894A true CN108224894A (en) 2018-06-29
CN108224894B CN108224894B (en) 2020-04-28

Family

ID=62643294

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810015399.9A Active CN108224894B (en) 2018-01-08 2018-01-08 Food material freshness identification method and device based on deep learning, refrigerator and medium

Country Status (1)

Country Link
CN (1) CN108224894B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN109447165A (en) * 2018-11-02 2019-03-08 西安财经学院 A kind of quality of agricultural product state identification method and device
CN109543605A (en) * 2018-11-21 2019-03-29 珠海格力电器股份有限公司 Food spoilage management method based on convolutional neural networks, system, storage facilities
CN109631486A (en) * 2018-12-18 2019-04-16 广东美的白色家电技术创新中心有限公司 A kind of food monitoring method, refrigerator and the device with store function

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101936894A (en) * 2010-08-23 2011-01-05 北京工商大学 Near infrared spectrum and microscopic adipose cell data fusion-based pork freshness non-destructive testing technology
CN104636757A (en) * 2015-02-06 2015-05-20 中国石油大学(华东) Deep learning-based food image identifying method
CN106991427A (en) * 2017-02-10 2017-07-28 海尔优家智能科技(北京)有限公司 The recognition methods of fruits and vegetables freshness and device
US20170219276A1 (en) * 2016-02-03 2017-08-03 Multimedia Image Solution Limited Smart Refrigerator
CN107133650A (en) * 2017-05-10 2017-09-05 合肥华凌股份有限公司 Food recognition methods, device and the refrigerator of refrigerator
CN107330522A (en) * 2017-07-04 2017-11-07 北京百度网讯科技有限公司 Method, apparatus and system for updating deep learning model
CN107480711A (en) * 2017-08-04 2017-12-15 合肥美的智能科技有限公司 Image-recognizing method, device, computer equipment and readable storage medium storing program for executing

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101936894A (en) * 2010-08-23 2011-01-05 北京工商大学 Near infrared spectrum and microscopic adipose cell data fusion-based pork freshness non-destructive testing technology
CN104636757A (en) * 2015-02-06 2015-05-20 中国石油大学(华东) Deep learning-based food image identifying method
US20170219276A1 (en) * 2016-02-03 2017-08-03 Multimedia Image Solution Limited Smart Refrigerator
CN106991427A (en) * 2017-02-10 2017-07-28 海尔优家智能科技(北京)有限公司 The recognition methods of fruits and vegetables freshness and device
CN107133650A (en) * 2017-05-10 2017-09-05 合肥华凌股份有限公司 Food recognition methods, device and the refrigerator of refrigerator
CN107330522A (en) * 2017-07-04 2017-11-07 北京百度网讯科技有限公司 Method, apparatus and system for updating deep learning model
CN107480711A (en) * 2017-08-04 2017-12-15 合肥美的智能科技有限公司 Image-recognizing method, device, computer equipment and readable storage medium storing program for executing

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
周飞燕 等: "卷积神经网络研究综述", 《计算机学报》 *
常亮 等: "图像理解中的卷积神经网络", 《自动化学报》 *
郭礼华: "食材数据库统计与对比实验性能分析", 《中国图像图形学报》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN109447165A (en) * 2018-11-02 2019-03-08 西安财经学院 A kind of quality of agricultural product state identification method and device
CN109543605A (en) * 2018-11-21 2019-03-29 珠海格力电器股份有限公司 Food spoilage management method based on convolutional neural networks, system, storage facilities
CN109631486A (en) * 2018-12-18 2019-04-16 广东美的白色家电技术创新中心有限公司 A kind of food monitoring method, refrigerator and the device with store function
WO2020125456A1 (en) * 2018-12-18 2020-06-25 广东美的白色家电技术创新中心有限公司 Food monitoring method, refrigerator, and apparatus having storage function

Also Published As

Publication number Publication date
CN108224894B (en) 2020-04-28

Similar Documents

Publication Publication Date Title
US10043059B2 (en) Assisted photo-tagging with facial recognition models
Park et al. Meta-tracker: Fast and robust online adaptation for visual object trackers
US20180284777A1 (en) Method, control apparatus, and system for tracking and shooting target
CN106650630B (en) A kind of method for tracking target and electronic equipment
US9736432B2 (en) Identifying popular network video segments
CN104679876B (en) Photograph taking policy recommendation device and method
US7545973B2 (en) Image matching system using 3-dimensional object model, image matching method, and image matching program
US8270806B2 (en) Information processing apparatus and method of controlling same
US7526110B2 (en) Biometric information authentication device, biometric information authentication method, and computer-readable recording medium with biometric information authentication program recorded thereon
US7481226B2 (en) Method and device for analyzing cigarette smoking
US8295611B2 (en) Robust video retrieval utilizing audio and video data
US8379999B2 (en) Methods, circuits, devices, apparatuses and systems for providing image composition rules, analysis and improvement
CN107194318A (en) The scene recognition method of target detection auxiliary
CN101877060B (en) Information processing apparatus and method
CN108288096B (en) Method and device for estimating travel time and training model
CN108492482A (en) Kinds of goods monitoring system and monitoring method
CN106170246A (en) For four-dimensional (4D) stream equipment of nuclear magnetic resonance, method and product
TWI424865B (en) Golf simulation apparatus and method for the same
CN103714094B (en) The apparatus and method of the object in identification video
CN107566781B (en) Video monitoring method and video monitoring equipment
CN103902746B (en) DTC, data flow and the method for freezing frame data preservation and playback
US9471982B2 (en) Information processing apparatus and information processing method for associating an image with related information
KR101864412B1 (en) Data managing method, apparatus and program for machine learning
CN108388876A (en) A kind of image-recognizing method, device and relevant device
CN102945366B (en) A kind of method and device of recognition of face

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
GR01 Patent grant
GR01 Patent grant