WO2020147345A1 - 获取米粒的垩白度的方法、装置和烹饪器具 - Google Patents

获取米粒的垩白度的方法、装置和烹饪器具 Download PDF

Info

Publication number
WO2020147345A1
WO2020147345A1 PCT/CN2019/109970 CN2019109970W WO2020147345A1 WO 2020147345 A1 WO2020147345 A1 WO 2020147345A1 CN 2019109970 W CN2019109970 W CN 2019109970W WO 2020147345 A1 WO2020147345 A1 WO 2020147345A1
Authority
WO
WIPO (PCT)
Prior art keywords
rice
neural network
chalkiness
image
rice grain
Prior art date
Application number
PCT/CN2019/109970
Other languages
English (en)
French (fr)
Inventor
邓灿赏
陈翀
魏文应
尹彦斌
Original Assignee
珠海格力电器股份有限公司
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 珠海格力电器股份有限公司 filed Critical 珠海格力电器股份有限公司
Publication of WO2020147345A1 publication Critical patent/WO2020147345A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Definitions

  • This application relates to the field of smart small household appliances, and specifically to a method, device and cooking appliance for obtaining the chalkiness of rice grains.
  • Chalkiness refers to the white opaque part formed by loose tissue in rice endosperm.
  • the chalkiness of rice refers to the ratio of the chalky area of rice grains to the total area of rice grains. Chalkiness is one of the important indicators to measure the quality of rice. It not only relates to the appearance, but also has a great influence on the processing quality and cooking quality. The higher the chalkiness, the worse the quality and taste of the rice grains.
  • the chalkiness detection of rice grains requires the help of a professional rice grain appearance detector.
  • the rice grains are laid flat on the scanning table, the rice grains are scanned by the machine for image collection, and special sensors and special processing software are used to detect the appearance of the rice grains.
  • the above methods are not practical in the scene of smart cooking appliances. On the one hand, it is not realistic for families to purchase professional detection equipment, which is expensive and complicated to operate. On the other hand, it is unrealistic to embed appearance detectors in cooking appliances, causing the prices of cooking appliances to rise sharply.
  • the embodiments of the present application provide a method, a device, and a cooking appliance for obtaining the chalkiness of rice grains, so as to at least solve the problem of detecting the chalkiness of rice grains by using a rice grain appearance detector in the prior art, resulting in low detection efficiency and high cost. problem.
  • a method for obtaining the chalkiness of rice grains includes: collecting multiple rice grain sample images, where the rice grain sample images are images loaded with annotation information, and the annotation information includes at least rice grains.
  • the chalkiness of the rice grains in the sample image uses the convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model; automatically detect the chalkiness of the rice grains based on the trained neural network model.
  • the source of the image of the rice grain sample includes at least one of the following: the image of the rice grain uploaded by the cooking appliance used by the user, the image of the rice grain observed by the appearance detector, the image of the rice grain taken by the photographing device, and the image of the rice grain on the network platform , Where the rice sample image is the training data.
  • the user uploads the chalkiness of the rice sample image by operating cooking appliances, photographing equipment or a network platform, and loads the uploaded chalkiness into the rice sample image as annotation information
  • a convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model, including: scanning the rice grain sample image to obtain a convolutional feature map; pooling the convolutional feature map, Obtain the de-redundant feature map; activate the de-redundant feature map, and classify the activation results; use the BP backpropagation algorithm to train the classified feature map to obtain a trained neural network model.
  • each convolution kernel in the convolution layer of the convolutional neural network is used to scan the rice sample image to obtain a convolution feature map, where the convolution feature map is used to characterize the feature layer of the rice grain in the rice sample image.
  • pooling is performed on the convolution feature map through a pooling layer to obtain a de-redundant feature map.
  • the de-redundant feature map is activated through at least one fully connected layer of the convolutional neural network.
  • the method further includes: saving the neural network model in a local device or uploading it to a server.
  • automatically detecting the chalkiness of rice grains based on the neural network model obtained by training includes: if an image of rice grains to be detected is received, uploading the image of rice grains to be detected to the server; receiving the chalkiness returned by the server, where, The server uses the neural network model to detect the chalkiness of the rice grain image to be detected.
  • a device for acquiring the chalkiness of rice grains including: a collection module configured to collect multiple rice grain sample images, wherein the rice grain sample images are images loaded with annotation information, The labeling information includes at least the chalkiness of the rice grains in the rice grain sample image; the training module is set to use convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model; the detection module is set to be based on training The obtained neural network model automatically detects the chalkiness of rice grains.
  • the collection module includes at least one of the following: a first collection module configured to collect an image of a rice grain sample through a cooking appliance used by the user; a second collection module configured to collect an image of a rice grain sample through an appearance detector; and third
  • the acquisition module is configured to collect the rice sample image through the shooting device; the fourth acquisition module is configured to acquire the rice sample image through the network platform; wherein the rice sample image is the training data.
  • the collection module further includes a loading module configured to load the uploaded chalkiness into the rice sample image as annotation information after uploading the chalkiness of the rice sample image through the collection module.
  • a loading module configured to load the uploaded chalkiness into the rice sample image as annotation information after uploading the chalkiness of the rice sample image through the collection module.
  • the training module includes: a scanning module, configured to scan the rice sample image to obtain a convolutional feature map; pooling module, configured to perform pooling processing on the convolutional feature map to obtain a de-redundant feature map; activation The module is set to activate the de-redundant feature map and classify the activation result; the training sub-module is set to use the BP backpropagation algorithm to train the classified feature map to obtain a trained neural network model.
  • the scanning module includes: a scanning sub-module configured to scan the rice grain sample image through each convolution kernel in the convolutional layer of the convolutional neural network to obtain a convolution feature map, where the convolution feature map is used to represent the rice grain The feature layer of the rice grains in the sample image.
  • the pooling module includes a pooling sub-module configured to perform pooling processing on the convolutional feature map through the pooling layer to obtain a de-redundant feature map.
  • the activation module includes an activation sub-module configured to activate the de-redundant feature map through at least one fully connected layer of the convolutional neural network.
  • the device further includes: a saving module configured to save the neural network model in a local device or upload it to server.
  • the detection module includes: an upload module, which is configured to upload the image of rice grains to be detected to the server if the image of rice grains to be detected is received; the receiving module is configured to receive the chalkiness returned by the server, where the server uses neural The network model detects the chalkiness of the rice grain image to be detected.
  • a storage medium includes a stored program, and when the program is running, the device where the storage medium is located is controlled to execute any one of the aforementioned methods for obtaining the chalkiness of rice grains.
  • a processor which is configured to run a program, wherein any one of the above methods for obtaining the chalkiness of rice grains is executed when the program is running.
  • a cooking appliance including: a collection device, configured to collect multiple rice grain sample images; a processor, configured to run a program, wherein when the program is running, the output from the collection device.
  • the following processing steps are performed on the data of the rice grain sample image: the labeling information includes at least the chalkiness of the rice grain sample image; the convolutional neural network is used to train the collected rice grain sample image to obtain the trained nerve Network model; automatically detects the chalkiness of rice grains based on the trained neural network model.
  • a method for obtaining the chalkiness of rice grains including: receiving an image of rice grains to be detected, uploading the image of rice grains to be detected to a server; receiving the chalkiness returned by the server Among them, the server collects multiple rice grain sample images, and uses the convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model.
  • the server uses the trained neural network model to detect the rice grain image to be detected
  • the chalkiness of the rice grain sample image is an image loaded with annotation information, and the annotation information includes at least the chalkiness of the rice grain in the rice grain sample image.
  • multiple rice grain sample images are collected, where the rice grain sample image is an image loaded with annotation information, and the annotation information includes at least the chalkiness of the rice grain in the rice grain sample image.
  • a convolutional neural network is used to collect The obtained rice grain sample images are trained to obtain a trained neural network model, and the chalkiness of the rice grains is automatically detected based on the trained neural network model.
  • the above solution is based on artificial intelligence technology. It is convenient and quick to directly detect the chalkiness of rice through the recognition of the image of rice grains, which solves the problem that the existing technology uses a rice grain appearance detector to detect the chalkiness of rice grains, resulting in low detection efficiency and high cost. Technical issues.
  • Fig. 1 is a flowchart of an optional method for obtaining the chalkiness of rice grains according to an embodiment of the present application
  • Figure 2 is a schematic diagram of an optional fully connected neural network according to an embodiment of the present application.
  • Fig. 3 is a schematic diagram of an optional convolutional neural network algorithm according to an embodiment of the present application.
  • FIG. 4 is a flowchart of an optional intelligent detection method for obtaining the chalkiness of rice grains according to an embodiment of the present application.
  • Fig. 5 is a schematic diagram of an optional device for obtaining the chalkiness of rice grains according to an embodiment of the present application.
  • an embodiment of a method for obtaining the chalkiness of rice grains is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions And, although the logical sequence is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than here.
  • Fig. 1 is a flowchart of an optional method for obtaining the chalkiness of rice grains according to an embodiment of the present application. As shown in Fig. 1, the method includes the following steps:
  • Step S102 Collect multiple rice grain sample images, where the rice grain sample image is an image loaded with annotation information, and the annotation information includes at least the chalkiness of the rice grain in the rice grain sample image.
  • the above-mentioned rice grains may be rice.
  • the aforementioned rice grain sample image may be a tiled rice grain image without overlap.
  • the above-mentioned labeling information may be the chalkiness of the rice grain, the chalky area of the rice grain and the total area of the rice grain, or the projected area of the chalky part of the rice grain and the total projected area of the rice grain.
  • Step S104 Use the convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model.
  • Convolutional neural network is a supervised learning algorithm, which is a special case of deep neural network. Compared with deep artificial neural network, it has the advantages of fewer weights and fast training speed.
  • Convolutional neural network is mainly composed of three parts, namely input layer, hidden layer and output layer. Among them, the input layer and output layer have only one layer, while the hidden layer can have multiple layers.
  • a deep neural network is a neural network with many hidden layers.
  • Figure 2 shows a schematic diagram of a fully connected neural network with two hidden layers. As shown in Figure 2, in the input layer, each neuron x represents an input feature, that is, rice sample images loaded with different chalkiness labels, and b is a feature-independent bias value. After the input feature is converted by the activation function, it enters hidden layer 1. Similarly, the result of hidden layer 1 is converted again and then enters hidden layer 2, and finally reaches the output layer to obtain various types of y probability values, and then output different rice grains. Whiteness information.
  • step S106 the chalkiness of rice grains is automatically detected based on the neural network model obtained by training.
  • the trained neural network model is used to comprehensively analyze the image data of rice grains to be detected to intelligently detect the chalkiness of the rice grains.
  • multiple rice images with no overlap and clear views are collected, divided into training sample sets and test sample sets, and input into the convolutional neural network model for training and testing. Get the trained convolutional neural network model.
  • the user In actual detection, the user only needs to input the image of rice grains to be detected, and the model can automatically output the chalkiness value of the rice to achieve the purpose of quickly detecting the chalkiness of the rice.
  • the rice grain sample image is an image loaded with annotation information, and the annotation information includes at least the chalkiness of the rice grain in the rice grain sample image; using convolutional nerve
  • the network trains the collected rice grain sample images to obtain a trained neural network model; automatically detects the chalkiness of the rice grains based on the trained neural network model.
  • the above solution is based on artificial intelligence technology. It is convenient and quick to directly detect the chalkiness of rice through the recognition of the image of rice grains, which solves the problem that the existing technology uses a rice grain appearance detector to detect the chalkiness of rice grains, resulting in low detection efficiency and high cost. technical problem.
  • the source of the image of the rice grain sample includes at least one of the following: the image of the rice grain uploaded by the cooking appliance used by the user, the image of the rice grain observed by the appearance detector, the image of the rice grain taken by the photographing device, and the image of the rice grain on the network platform , Where the rice sample image is the training data.
  • the image of the rice sample can be obtained by a cooking appliance used by the user, wherein a collection device such as an image sensor can be built into the top of the cooking appliance.
  • the image of the rice grain sample can be obtained by observation with an appearance detector.
  • the image of the rice grain sample can be obtained by a camera or video camera held by the user.
  • the rice grain sample image can be obtained by directly downloading the rice grain image on the network platform.
  • the user uploads the chalkiness of the rice sample image by operating a cooking appliance, a photographing device or a network platform, and the uploaded chalkiness is loaded into the rice sample image as annotation information.
  • the above-mentioned rice grain sample image is an image loaded with annotation information
  • the annotation information at least includes the chalkiness of the rice grains in the rice grain sample image.
  • the chalkiness of the rice grains can be obtained by manual identification or appearance instrument detection. Upload the chalkiness of the rice sample image by operating cooking appliances, shooting equipment or network platforms, load it into the rice sample image as annotation information, and then use the rice sample image loaded with the annotation information as the training data of the convolutional neural network , Used to train and test neural network models.
  • a convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model, including: scanning the rice grain sample image to obtain a convolutional feature map; pooling the convolutional feature map, Obtain the de-redundant feature map; activate the de-redundant feature map, and classify the activation results; use the BP backpropagation algorithm to train the classified feature map to obtain a trained neural network model.
  • the softmax function can be used to classify the activation result.
  • the collected original rice grain sample images are preprocessed, so that all the input set data are in the same order of magnitude, and the influence of big data is avoided.
  • the sample matrix is input to the convolutional neural network for training.
  • Convolutional neural network is essentially composed of multiple BP neural networks, so when training neural networks, BP backpropagation algorithm is used for training.
  • the activation function of the neuron selects the ReLU linear correction unit. Each time the error loss between the actual output and the expected output is calculated, back propagation is performed according to the loss result, and the weight matrix W of each layer is updated. When the algorithm converges, a stable weight W is finally obtained, the algorithm training is completed, and the final convolutional neural network model is saved for automatic detection of the chalkiness of rice grains.
  • each convolution kernel in the convolution layer of the convolutional neural network is used to scan the rice sample image to obtain a convolution feature map, where the convolution feature map is used to characterize the feature layer of the rice grain in the rice sample image.
  • pooling is performed on the convolution feature map through a pooling layer to obtain a de-redundant feature map.
  • Convolutional layer and pooling layer can have many different combinations, and the specific layer and network depth can be selected according to actual needs.
  • the de-redundant feature map is activated through at least one fully connected layer of the convolutional neural network.
  • the fully connected layer can also have multiple layers, and the specific layers are still selected according to actual needs. The more layers, the more accurate the recognition result and the more complex the network.
  • Fig. 3 is a schematic diagram of an optional convolutional neural network algorithm according to an embodiment of the present application.
  • the input rice grain sample image first passes through the first convolution layer, and the convolution is obtained by scanning the rice grain sample image Feature map, and then pool the convolution feature map.
  • the purpose of pooling is to reduce dimensionality, reduce redundant features, and obtain a de-redundant feature map.
  • the obtained image features are input to the fully connected layer for activation.
  • the fully connected layer can also have multiple layers, and finally the softmax layer outputs the classification results.
  • the above method further includes: saving the neural network model in a local device or uploading it to a server.
  • the above-mentioned local device may be a device that implements the method for obtaining the chalkiness of rice grains, such as smart terminal devices such as cooking appliances, rice cookers, and mobile phones.
  • automatically detecting the chalkiness of rice grains based on the neural network model obtained by training includes: if an image of rice grains to be detected is received, uploading the image of rice grains to be detected to the server; receiving the chalkiness returned by the server, where, The server uses the neural network model to detect the chalkiness of the rice grain image to be detected.
  • the above-mentioned rice grain image to be detected may be obtained by an image sensor, and the above-mentioned server may be a cloud server, or may be implemented by a local device smart chip.
  • Fig. 4 is a flowchart of an optional intelligent detection method for obtaining the chalkiness of rice grains according to an embodiment of the present application.
  • the user uses image acquisition devices such as image sensors, cameras, and video cameras to collect the tiled rice grain images to be detected in real time for preprocessing, and then input the images to the cloud server or the convolutional neural network model in the smart chip of the local device for image analysis.
  • the cloud server or the smart chip of the local device outputs the analysis result to the user, and feeds back the detection result to the user.
  • the user directly photographs the rice, so that the chalkiness detection result can be quickly obtained.
  • the above method for obtaining the chalkiness of rice grains is applied to a cooking appliance, which may include a heating component, a timing module, a decision-making module, a display module, a communication module, and an alarm module.
  • a cooking appliance which may include a heating component, a timing module, a decision-making module, a display module, a communication module, and an alarm module.
  • the aforementioned decision-making module determines the cooking data according to the set cooking data and the type of food, where the cooking data includes at least: heating data of a heating resistor, exhaust time of an exhaust valve, and heating temperatures in different cooking stages; The cooking appliance is controlled to cook the food based on the cooking data.
  • the decision-making module has an executive mechanism, such as a heating resistor, a timing module, etc.
  • the cooking data set above may be the cooking data preset by the user, such as taste and preference (soft, moderate, hard, porridge, soup, etc. modes).
  • the decision-making module automatically selects the cooking method according to the type of food output by the convolutional neural network model, combined with user preferences, such as soaking time, heating temperature of heating resistor, heating time, exhaust time of exhaust valve, exhaust valve Opening degree and holding time, etc., in order to obtain the best cooking method, to ensure the taste of the food, and to ensure that no nutrition is lost.
  • the method further includes: the decision-making module receives the type of food in the food image transmitted by the communication module, and receives the information received by the external interactive interface Cooking data.
  • the aforementioned communication module may be a wired communication module or a wireless communication module, such as a wifi module.
  • the above-mentioned external interaction interface may be a display panel arranged on the outer surface of the cooking appliance, or may be a remote control.
  • the communication module is also used to receive an update instruction transmitted by the remote server, where the update instruction is used to upgrade the function of the cooking appliance.
  • the server can transmit the new version of the operating program to the rice cooker through the communication module to realize remote update and make the service effect more ideal.
  • the rice sample image is an image loaded with annotation information, and the annotation information includes at least the chalkiness of the rice grain in the rice sample image; the convolutional neural network is used to analyze the collected rice sample The image is trained to obtain a trained neural network model; based on the trained neural network model, the chalkiness of rice grains is automatically detected.
  • this application detects rice grain chalkiness based on artificial intelligence technology, provides users with a convenient and intelligent method for detecting rice grain chalkiness, achieves the purpose of automatically detecting rice grain chalkiness, and enables users to Convenient, fast and low-cost detection of rice quality improves user experience.
  • FIG. 5 is a schematic diagram of the device for obtaining the chalkiness of rice grains according to an embodiment of the present application. As shown in FIG. 5, the device 500 includes:
  • the collection module 502 is configured to collect multiple rice grain sample images, where the rice grain sample image is an image loaded with annotation information, and the annotation information at least includes the chalkiness of the rice grain in the rice grain sample image.
  • the training module 504 is configured to use a convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model.
  • the detection module 506 is configured to automatically detect the chalkiness of rice grains based on the neural network model obtained by training.
  • the collection module 502 includes at least one of the following: a first collection module configured to collect an image of a rice grain sample through a cooking appliance used by the user; a second collection module configured to collect an image of a rice grain sample through an appearance detector; The third collection module is configured to collect the rice sample image through the photographing device; the fourth collection module is configured to obtain the rice sample image through the network platform; wherein the rice sample image is the training data.
  • the collection module 502 further includes a loading module, configured to load the chalkiness into the rice grain image as annotation information after uploading the chalkiness of the rice grain sample image through the collection module.
  • a loading module configured to load the chalkiness into the rice grain image as annotation information after uploading the chalkiness of the rice grain sample image through the collection module.
  • the training module 504 includes: a scanning module, configured to scan the rice grain sample image to obtain a convolutional feature map; and a pooling module, configured to perform pooling processing on the convolutional feature map to obtain a de-redundant feature map;
  • the activation module is set to activate the de-redundant feature maps and classify the activation results;
  • the training sub-module is set to use the BP backpropagation algorithm to train the classified feature maps to obtain a trained neural network model.
  • the scanning module includes: a scanning sub-module configured to scan the rice grain sample image through each convolution kernel in the convolutional layer of the convolutional neural network to obtain a convolution feature map, where the convolution feature map is used to represent the rice grain The feature layer of the rice grains in the sample image.
  • the pooling module includes a pooling sub-module configured to perform pooling processing on the convolutional feature map through the pooling layer to obtain a de-redundant feature map.
  • the activation module includes an activation sub-module configured to activate the de-redundant feature map through at least one fully connected layer of the convolutional neural network.
  • the device further includes: a saving module configured to save the neural network model in a local device or upload it to server.
  • the detection module 506 includes: an upload module configured to upload the image of rice grains to be detected to the server if the image of rice grains to be detected is received; the receiving module is configured to receive the chalkiness returned by the server, where the server uses The neural network model detects the chalkiness of the rice grain image to be detected.
  • a storage medium includes a stored program.
  • the program is running, the device where the storage medium is located is controlled to execute the method for obtaining the chalkiness of rice grains in Embodiment 1.
  • a processor is provided, the processor is set to run a program, and the method for obtaining the chalkiness of rice grains in Embodiment 1 is executed when the program is running.
  • a cooking appliance including: a collection device configured to collect multiple rice grain sample images; a processor configured to run a program, wherein the data output from the collection device is executed as follows when the program is running Processing steps: load the annotation information on the rice sample image, the annotation information includes at least the chalkiness of the rice grain in the rice sample image; use the convolutional neural network to train the collected rice sample image to obtain a trained neural network model; The trained neural network model automatically detects the chalkiness of rice grains.
  • a method for obtaining the chalkiness of rice grains includes: receiving an image of rice grains to be detected and uploading the image of the rice grains to be detected to a server; receiving the chalkiness returned by the server, where the server Collect multiple rice grain sample images, and use convolutional neural network to train the collected rice grain sample images to obtain a trained neural network model.
  • the server uses the trained neural network model to detect the chalkiness of the rice grain image to be detected
  • the rice sample image is an image loaded with annotation information, and the annotation information at least includes the chalkiness of the rice grain in the rice sample image.
  • the disclosed technical content may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of units may be a logical function division.
  • multiple units or components may be combined or integrated into Another system, or some features can be ignored, or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, units or modules, and may be in electrical or other forms.
  • the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or software function unit.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • the technical solution of this application essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , Including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the devices in the embodiments of the present application.
  • the aforementioned storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program code .
  • the method, device and cooking appliance for obtaining the chalkiness of rice grains can be applied to the detection of the chalkiness of rice grains by collecting multiple rice grain sample images, wherein the rice grain sample image is The image loaded with the annotation information.
  • the annotation information includes at least the chalkiness of the rice grains in the rice grain sample image.
  • the convolutional neural network is used to train the collected rice grain sample images to obtain a trained neural network model based on the trained nerves
  • the network model automatically detects the chalkiness of rice grains. Based on artificial intelligence technology, it can directly detect the chalkiness of rice grains through the recognition of rice grain images, which is convenient and quick. It solves the problem of using rice grain appearance detector to detect the chalkiness of rice grains in the prior art , Which leads to technical problems of low detection efficiency and high cost, and realizes the effect of quickly identifying the chalkiness of rice grains.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)

Abstract

一种获取米粒的垩白度的方法、装置和烹饪器具。其中，该方法包括：采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度（S102），使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型（S104），基于训练得到的神经网络模型自动检测米粒的垩白度（S106）。其解决了现有技术通过使用米粒外观检测仪检测米粒的垩白度，导致检测效率低、成本高的技术问题。

Description

获取米粒的垩白度的方法、装置和烹饪器具
本申请要求于2019年01月14日提交至中国国家知识产权局、申请号为201910033070.X、发明名称为“获取米粒的垩白度的方法、装置和烹饪器具”的专利申请的优先权。
技术领域
本申请涉及智能小家电领域，具体而言，涉及一种获取米粒的垩白度的方法、装置和烹饪器具。
背景技术
垩白是指稻米胚乳中组织疏松而形成的白色不透明的部分。大米的垩白度是指米粒的垩白面积与米粒总面积之比。垩白度是衡量稻米品质的重要指标之一，不仅关系到外观好看与否，而且对加工品质、蒸煮品质有很大影响。垩白度越高，米粒的品质、口感越差。
目前，米粒的垩白度检测需要借助专业的米粒外观检测仪，将米粒平铺在扫描台上，通过机器扫描米粒进行图像采集，借助特殊的传感器以及专用的处理软件进行米粒的外观检测，从而获得米粒的外观垩白度指标。然而，上述方式在智能烹饪器具的场景中并不实用，一方面对于家庭来说购置专业的检测设备并不现实，价格昂贵且操作复杂。另一方面，在烹饪器具中内嵌外观检测仪也不现实，造成烹饪器具的价格大幅上涨。
针对现有技术通过使用米粒外观检测仪检测米粒的垩白度，导致检测效率低、成本高的问题，目前尚未提出有效的解决方案。
发明内容
本申请实施例提供了一种获取米粒的垩白度的方法、装置和烹饪器具，以至少解决现有技术通过使用米粒外观检测仪检测米粒的垩白度，导致检测效率低、成本高的技术问题。
根据本申请实施例的一个方面，提供了一种获取米粒的垩白度的方法，包括：采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度；使用卷积神经网络对采集到的米粒样本图像进 行训练，得到训练好的神经网络模型；基于训练得到的神经网络模型自动检测米粒的垩白度。
可选地，米粒样本图像的来源包括如下至少之一：通过用户使用的烹饪器具来上传的米粒图像、外观检测仪观测得到的米粒图像、拍摄设备拍摄得到的米粒图像和网络平台上的米粒图像，其中，米粒样本图像为训练数据。
可选地，用户通过操作烹饪器具、拍摄设备或网络平台来上传米粒样本图像的垩白度，将上传的垩白度作为标注信息加载到米粒样本图像中
可选地，使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，包括：扫描米粒样本图像，得到卷积特征图；对卷积特征图进行池化处理，得到去冗余后的特征图；激活去冗余后的特征图，并对激活结果进行分类；使用BP反向传播算法对分类后的特征图进行训练，得到训练好的神经网络模型。
可选地，通过卷积神经网络的卷积层中的每个卷积核扫描米粒样本图像，得到卷积特征图，其中卷积特征图用于表征米粒样本图像中的米粒的特征图层。
可选地，通过池化层对卷积特征图进行池化处理，得到去冗余后的特征图。
可选地，通过卷积神经网络的至少一个全连接层激活去冗余后的特征图。
可选地，在使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型之后，方法还包括：将神经网络模型保存在本地设备，或上传至服务器。
可选地，基于训练得到的神经网络模型自动检测米粒的垩白度，包括：如果接收到待检测的米粒图像，将待检测的米粒图像上传到服务器；接收服务器返回的垩白度，其中，服务器使用神经网络模型检测得到待检测的米粒图像的垩白度。
根据本申请实施例的一个方面，还提供了一种获取米粒的垩白度的装置，包括：采集模块，设置为采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度；训练模块，设置为使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型；检测模块，设置为基于训练得到的神经网络模型自动检测米粒的垩白度。
可选地，采集模块包括如下至少之一：第一采集模块，设置为通过用户使用的烹饪器具来采集米粒样本图像；第二采集模块，设置为通过外观检测仪来采集米粒样本图像；第三采集模块，设置为通过拍摄设备来采集米粒样本图像；第四采集模块，设置为通过网络平台来获取米粒样本图像；其中，米粒样本图像为训练数据。
可选地，采集模块还包括加载模块，设置为在通过采集模块上传米粒样本图像的垩白度后，将上传的垩白度作为标注信息加载到米粒样本图像中。
可选地，训练模块包括：扫描模块，设置为扫描米粒样本图像，得到卷积特征图；池化模块，设置为对卷积特征图进行池化处理，得到去冗余后的特征图；激活模块，设置为激活去冗余后的特征图，并对激活结果进行分类；训练子模块，设置为使用BP反向传播算法对分类后的特征图进行训练，得到训练好的神经网络模型。
可选地，扫描模块包括：扫描子模块，设置为通过卷积神经网络的卷积层中的每个卷积核扫描米粒样本图像，得到卷积特征图，其中卷积特征图用于表征米粒样本图像中的米粒的特征图层。
可选地，池化模块包括池化子模块，设置为通过池化层对卷积特征图进行池化处理，得到去冗余后的特征图。
可选地，激活模块包括激活子模块，设置为通过卷积神经网络的至少一个全连接层激活去冗余后的特征图。
可选地，在使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型之后，装置还包括：保存模块，设置为将神经网络模型保存在本地设备，或上传至服务器。
可选地，检测模块包括：上传模块，设置为如果接收到待检测的米粒图像，将待检测的米粒图像上传到服务器；接收模块，设置为接收服务器返回的垩白度，其中，服务器使用神经网络模型检测得到待检测的米粒图像的垩白度。
根据本申请实施例的一个方面，还提供了一种存储介质，存储介质包括存储的程序，在程序运行时控制存储介质所在设备执行上述任意一种获取米粒的垩白度的方法。
根据本申请实施例的一个方面，还提供了一种处理器，处理器设置为运行程序，其中，在程序运行时执行上述任意一种获取米粒的垩白度的方法。
根据本申请实施例的一个方面，还提供了一种烹饪器具，包括：采集装置，设置为采集多张米粒样本图像；处理器，设置为运行程序，其中，在程序运行时对于从采集装置输出的数据执行如下处理步骤：对米粒样本图像加载标注信息，标注信息至少包括了米粒样本图像中米粒的垩白度；使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型；基于训练得到的神经网络模型自动检测米粒的垩白度。
根据本申请实施例的一个方面，还提供了一种获取米粒的垩白度的方法，包括：接收到待检测的米粒图像，将待检测的米粒图像上传到服务器；接收服务器返回的垩白度，其中，服务器采集多张米粒样本图像，并使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，服务器使用训练好的神经网络模型检测得到待检测的米粒图像的垩白度，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度。
在本申请实施例中，通过采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度，使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，基于训练得到的神经网络模型自动检测米粒的垩白度。上述方案基于人工智能技术，通过对米粒图像的识别，直接检测大米垩白度，方便快捷，进而解决了现有技术通过使用米粒外观检测仪检测米粒的垩白度，导致检测效率低、成本高的技术问题。
附图说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1是根据本申请实施例的一种可选的获取米粒的垩白度的方法流程图；
图2是根据本申请实施例的一种可选的全连接神经网络示意图；
图3是根据本申请实施例的一种可选的卷积神经网络算法示意图；
图4是根据本申请实施例的一种可选的获取米粒的垩白度的智能检测方法流程图；以及
图5是根据本申请实施例的一种可选的获取米粒的垩白度的装置示意图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分的实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本申请保护的范围。
需要说明的是，本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、装置、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、装置、产品或设备固有的其它步骤或单元。
实施例1
根据本申请实施例，提供了一种获取米粒的垩白度的方法实施例，需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
图1是根据本申请实施例的一种可选的获取米粒的垩白度的方法流程图，如图1所示，该方法包括如下步骤：
步骤S102，采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度。
一种可选方案中，上述米粒可以为稻米。上述米粒样本图像可以为平铺无重叠的米粒图像。上述标注信息可以为米粒的垩白度，也可以为米粒的垩白面积和米粒总面积，也可以为米粒的垩白部分投影面积和米粒总投影面积。
步骤S104，使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型。
卷积神经网络属于有监督的学习算法，是深度神经网络中的一种特殊情况，它相比于深度人工神经网络具有权值数量少、训练速度快等优点。
卷积神经网络主要由三部分组成，分别是输入层、隐藏层和输出层。其中，输入层和输出层只有一层，而隐藏层可以有多层，深度神经网络就是有很多个隐藏层的神经网络。图2所示的是有两个隐藏层的全连接神经网络示意图。如图2所示，在输入层，每一个神经元x代表一个输入特征，即加载了不同垩白度标签的米粒样本图像，b是一个与特征无关的偏置值。输入的特征经过激活函数转换后，进入隐藏层1，同样的，隐藏层1的结果经过再次转换后进入隐藏层2，最后到达输出层，得到各类y的概率值，进而输出不同米粒的垩白度信息。
步骤S106，基于训练得到的神经网络模型自动检测米粒的垩白度。
在上述步骤中，利用训练好的神经网络模型对待检测的米粒图像数据进行综合分析，智能检测米粒的垩白度。
在一种可选的实施例中，采集多张平铺无重叠且视图清晰的大米图像，将其分为训练样本集，测试样本集，输入到卷积神经网络模型中进行训练和测试，以得到训练好的卷积神经网络模型。实际检测时，用户只需要输入待检测的米粒图像，模型即可自动输出大米的垩白度值，达到快速检测大米垩白度的目的。
基于本申请上述实施例提供的方案，通过采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度；使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型；基于训练得到的神经网络模型自动检测米粒的垩白度。上述方案基于人工智能技术，通过对米粒图像的识别，直接检测大米垩白度，方便快捷，进而解决了现有技术通过使用米粒外观检测仪检测米粒的垩白度，导致检测效率低、成本高技术问题。
可选地，米粒样本图像的来源包括如下至少之一：通过用户使用的烹饪器具来上传的米粒图像、外观检测仪观测得到的米粒图像、拍摄设备拍摄得到的米粒图像和网络平台上的米粒图像，其中，米粒样本图像为训练数据。
一种可选方案中，米粒样本图像可以通过用户使用的烹饪器具来获得，其中，烹饪器具的顶部可以内置图像传感器等采集装置。
另一种可选方案中，米粒样本图像可以通过外观检测仪观测获得。
另一种可选方案中，米粒样本图像可以通过用户手持的照相机、摄像机等拍摄设备获得。
另一种可选方案中，米粒样本图像可以通过直接下载网络平台上的米粒图像获得。
可选地，用户通过操作烹饪器具、拍摄设备或网络平台来上传米粒样本图像的垩白度，上传的垩白度作为标注信息加载到米粒样本图像中。
需要说明的是，上述米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度，米粒的垩白度可以采用人工鉴定或外观仪器检测的方式获得，用户通过操作烹饪器具、拍摄设备或网络平台来上传米粒样本图像的垩白度，将其作为标注信息加载到米粒样本图像中，然后将加载了标注信息的米粒样本图像作为卷积神经网络的训练数据，用于训练和测试神经网络模型。
可选地，使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，包括：扫描米粒样本图像，得到卷积特征图；对卷积特征图进行池化处理，得到去冗余后的特征图；激活去冗余后的特征图，并对激活结果进行分类；使用BP反向传播算法对分类后的特征图进行训练，得到训练好的神经网络模型。
一种可选方案中，对激活结果进行分类可以采用softmax函数。
在上述步骤中，对收集到的原始米粒样本图像进行预处理，从而使所有的输入集数据在同一个数量级上，避免大数据起决定性作用的影响。图2中，将样本矩阵输入卷积神经网络，进行训练。卷积神经网络本质上是由多个BP神经网络组成的，因此在训练神经网络时，使用BP反向传播算法进行训练。具体来说，输入的样本向量可以表示为X＝(x1,x2,x3,…,xn)，对应的类别为Y＝(y1,y2,…,ym)，初始化的权重矩阵为W。由于ReLU函数求梯度简单，收敛较快，所以利用BP反向传播算法进行训练神经网络时，神经元的激活函数选择ReLU线性修正单元。每一次计算实际输出与期望输出的误差损失，根据损失结果进行反向传播，更新每一层的权重矩阵W。当算法收敛时，最终得到稳定的权重W，算法训练完成，保存最终的卷积神经网络模型，用于自动检测米粒的垩白度。
可选地，通过卷积神经网络的卷积层中的每个卷积核扫描米粒样本图像，得到卷积特征图，其中卷积特征图用于表征米粒样本图像中的米粒的特征图层。
可选地，通过池化层对卷积特征图进行池化处理，得到去冗余后的特征图。
卷积层和池化层可以有多种不同的组合，具体的层次和网络深度可以根据实际需要进行选取。
可选地，通过卷积神经网络的至少一个全连接层激活去冗余后的特征图。
全连接层也可以有多层，具体的层次仍是根据实际需要进行选取。层数越多，识别结果越准确，网络越复杂。
图3是根据本申请实施例的一种可选的卷积神经网络算法示意图，如图3所示，输入的米粒样本图像首先经过第一个卷积层，通过扫描米粒样本图像，得到卷积特征图，然后对卷积特征图进行池化操作。池化的目的是进行降维度，减少冗余特征，得到去冗余后的特征图。经过多层卷积和池化操作，得到的图像特征输入到全连接层，进行激活，全连接层也可以有多层，最后由softmax层输出分类结果。
可选地，在使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型之后，上述方法还包括：将神经网络模型保存在本地设备，或上传至服 务器。
一种可选方案中，上述本地设备可以为实施获取米粒的垩白度的方法的设备，例如烹饪器具、电饭锅、手机等智能终端设备。
将训练好的卷积神经网络模型写入本地设备的芯片中，或者部署于云端服务器，作为数据实时分析处理的中枢。
可选地，基于训练得到的神经网络模型自动检测米粒的垩白度，包括：如果接收到待检测的米粒图像，将待检测的米粒图像上传到服务器；接收服务器返回的垩白度，其中，服务器使用神经网络模型检测得到待检测的米粒图像的垩白度。
一种可选方案中，上述待检测的米粒图像可以由图像传感器获得，上述服务器可以为云端服务器，也可以由本地设备智能芯片实现。
图4是根据本申请实施例的一种可选的获取米粒的垩白度的智能检测方法流程图。用户采用图像传感器、照相机、摄像机等图像采集设备实时采集平铺的待检测米粒图像作预处理，然后将图像输入到云端服务器或本地设备智能芯片中的卷积神经网络模型，进行图像分析。云端服务器或本地设备智能芯片将分析得到的结果输出给用户端，向用户反馈检测结果。本实施例通过用户直接对大米进行拍摄，可以快速获得垩白度检测结果。
可选地，将上述获取米粒的垩白度的方法应用于烹饪器具中，上述烹饪器具可以包括加热部件、计时模块、决策模块、显示模块、通信模块和报警模块。
可选地，上述决策模块根据设定的烹饪数据和食物的类型，确定烹饪数据，其中，烹饪数据至少包括：加热电阻的加热数据、排气阀的排气时间和不同烹饪阶段的加热温度；基于烹饪数据控制烹饪器具对食物进行烹饪。
一种可选方案中，决策模块内部具有执行机构，例如加热电阻、计时模块等。上述设定的烹饪数据可以为用户预先设定的烹饪数据，例如口感与偏好(偏软、适中、偏硬、煮粥、煲汤等模式)。
决策模块根据卷积神经网络模型输出的食物的类型，结合用户喜好设定，自动选择烹饪方式，例如泡米时间、加热电阻的加热温度、加热时间、排气阀的排气时间、排气阀开度和保温时间等，以便获得最佳的烹饪方式，保证食物的口感，同时确保营养不会流失。
可选地，在决策模块根据设定的烹饪数据和食物的类型，确定烹饪数据之前，方 法还包括：决策模块接收通信模块传输的食物图像中食物的类型，并接收到外部交互界面接收到的烹饪数据。
一种可选方案中，上述通信模块可以为有线通信模块或无线通信模块，例如wifi模块。上述外部交互界面可以为设置于烹饪器具外表面的显示面板，也可以为遥控器。
可选地，通信模块还用于接收远程服务器传输的更新指令，其中，更新指令用于对烹饪器具的功能进行升级。
功能不变的烹饪器具势必满足不了需求不断增加的用户，当有新的功能开发出来后，服务器可以通过通信模块将新版本的运行程序传输给电饭煲，实现远程更新，使服务效果更理想。
通过上述方案，采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度；使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型；基于训练得到的神经网络模型自动检测米粒的垩白度。与现有技术相比，本申请基于人工智能技术检测米粒垩白度，为用户提供了一种方便、智能的米粒垩白度检测方法，达到了自动检测米粒垩白度的目的，使得用户能够方便快捷、低成本的检测大米品质，提高了用户体验。
实施例2
根据本申请实施例，提供了一种获取米粒的垩白度的装置，图5是根据本申请实施例的获取米粒的垩白度的装置示意图，如图5所示，该装置500包括：
采集模块502，设置为采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度。
训练模块504，设置为使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型。
检测模块506，设置为基于训练得到的神经网络模型自动检测米粒的垩白度。
可选地，采集模块502包括如下至少之一：第一采集模块，设置为通过用户使用的烹饪器具来采集米粒样本图像；第二采集模块，设置为通过外观检测仪来采集米粒样本图像；第三采集模块，设置为通过拍摄设备来采集米粒样本图像；第四采集模块，设置为通过网络平台来获取米粒样本图像；其中，米粒样本图像为训练数据。
可选地，采集模块502还包括加载模块，设置为在通过采集模块上传米粒样本图像的垩白度后，将垩白度作为标注信息加载到米粒图像中。
可选地，训练模块504包括：扫描模块，设置为扫描米粒样本图像，得到卷积特征图；池化模块，设置为对卷积特征图进行池化处理，得到去冗余后的特征图；激活模块，设置为激活去冗余后的特征图，并对激活结果进行分类；训练子模块，设置为使用BP反向传播算法对分类后的特征图进行训练，得到训练好的神经网络模型。
可选地，扫描模块包括：扫描子模块，设置为通过卷积神经网络的卷积层中的每个卷积核扫描米粒样本图像，得到卷积特征图，其中卷积特征图用于表征米粒样本图像中的米粒的特征图层。
可选地，池化模块包括池化子模块，设置为通过池化层对卷积特征图进行池化处理，得到去冗余后的特征图。
可选地，激活模块包括激活子模块，设置为通过卷积神经网络的至少一个全连接层激活去冗余后的特征图。
可选地，在使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型之后，装置还包括：保存模块，设置为将神经网络模型保存在本地设备，或上传至服务器。
可选地，检测模块506包括：上传模块，设置为如果接收到待检测的米粒图像，将待检测的米粒图像上传到服务器；接收模块，设置为接收服务器返回的垩白度，其中，服务器使用神经网络模型检测得到待检测的米粒图像的垩白度。
需要说明的是，本实施例的可选或优选实施方式可以参见实施例1中的相关描述，但不仅限于实施例1所公开的内容，在此不再赘述。
实施例3
根据本申请实施例，提供了一种存储介质，存储介质包括存储的程序，在程序运行时控制存储介质所在设备执行实施例1中的获取米粒的垩白度的方法。
实施例4
根据本申请实施例，提供了一种处理器，处理器设置为运行程序，在程序运行时执行实施例1中的获取米粒的垩白度的方法。
实施例5
根据本申请实施例，提供了一种烹饪器具，包括：采集装置，设置为采集多张米粒样本图像；处理器，设置为运行程序，其中，在程序运行时对于从采集装置输出的数据执行如下处理步骤：对米粒样本图像加载标注信息，标注信息至少包括了米粒样 本图像中米粒的垩白度；使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型；基于训练得到的神经网络模型自动检测米粒的垩白度。
需要说明的是，本实施例的可选或优选实施方式可以参见实施例1中的相关描述，但不仅限于实施例1所公开的内容，在此不再赘述。
实施例6
根据本申请实施例，提供了一种获取米粒的垩白度的方法，包括：接收到待检测的米粒图像，将待检测的米粒图像上传到服务器；接收服务器返回的垩白度，其中，服务器采集多张米粒样本图像，并使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，服务器使用训练好的神经网络模型检测得到待检测的米粒图像的垩白度，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度。
需要说明的是，本实施例的可选或优选实施方式可以参见实施例1中的相关描述，但不仅限于实施例1所公开的内容，在此不再赘述。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
在本申请的上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中，应该理解到，所揭露的技术内容，可通过其它的方式实现。其中，以上所描述的装置实施例仅仅是示意性的，例如单元的划分，可以为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，单元或模块的间接耦合或通信连接，可以是电性或其它的形式。
作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可 以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本申请各个实施例装置的全部或部分步骤。而前述的存储介质包括：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上仅是本申请的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本申请的保护范围。
工业实用性
如上所述，本申请至少部分实施例提供的获取米粒的垩白度的方法、装置和烹饪器具可以应用在米粒的垩白度检测中，通过采集多张米粒样本图像，其中，米粒样本图像为加载了标注信息的图像，标注信息至少包括了米粒样本图像中米粒的垩白度，使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，基于训练得到的神经网络模型自动检测米粒的垩白度，从而基于人工智能技术，通过对米粒图像的识别，直接检测大米垩白度，方便快捷，解决了现有技术通过使用米粒外观检测仪检测米粒的垩白度，导致检测效率低、成本高的技术问题，实现了快速对米粒的垩白度进行识别的效果。

Claims (22)

  1. 一种获取米粒的垩白度的方法，包括：
    采集多张米粒样本图像，其中，所述米粒样本图像为加载了标注信息的图像，所述标注信息至少包括了所述米粒样本图像中米粒的垩白度；
    使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型；
    基于训练得到的神经网络模型自动检测米粒的垩白度。
  2. 根据权利要求1所述的方法，其中，所述米粒样本图像的来源包括如下至少之一：通过用户使用的烹饪器具来上传的米粒图像、外观检测仪观测得到的米粒图像、拍摄设备拍摄得到的米粒图像和网络平台上的米粒图像，其中，所述米粒样本图像为训练数据。
  3. 根据权利要求2所述的方法，其中，用户通过操作烹饪器具、拍摄设备或网络平台来上传米粒样本图像的垩白度，将上传的垩白度作为所述标注信息加载到所述米粒样本图像中。
  4. 根据权利要求1至3中任意一项所述的方法，其中，使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，包括：
    扫描所述米粒样本图像，得到卷积特征图；
    对所述卷积特征图进行池化处理，得到去冗余后的特征图；
    激活所述去冗余后的特征图，并对激活结果进行分类；
    使用BP反向传播算法对分类后的特征图进行训练，得到所述训练好的神经网络模型。
  5. 根据权利要求4所述的方法，其中，通过所述卷积神经网络的卷积层中的每个卷积核扫描所述米粒样本图像，得到所述卷积特征图，其中，所述卷积特征图用于表征所述米粒样本图像中的米粒的特征图层。
  6. 根据权利要求4所述的方法，其中，通过池化层对所述卷积特征图进行池化处理，得到去冗余后的特征图。
  7. 根据权利要求4所述的方法，其中，通过所述卷积神经网络的至少一个全连接层激活所述去冗余后的特征图。
  8. 根据权利要求1所述的方法，其中，在使用卷积神经网络对采集到的米粒样本图 像进行训练，得到训练好的神经网络模型之后，所述方法还包括：
    将所述神经网络模型保存在本地设备，或上传至服务器。
  9. 根据权利要求1所述的方法，其中，基于训练得到的神经网络模型自动检测米粒的垩白度，包括：
    如果接收到待检测的米粒图像，将所述待检测的米粒图像上传到服务器；
    接收所述服务器返回的垩白度，其中，所述服务器使用所述神经网络模型检测得到所述待检测的米粒图像的垩白度。
  10. 一种获取米粒的垩白度的装置，包括：
    采集模块，设置为采集多张米粒样本图像，其中，所述米粒样本图像为加载了标注信息的图像，所述标注信息至少包括了所述米粒样本图像中米粒的垩白度；
    训练模块，设置为使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型；
    检测模块，设置为基于训练得到的神经网络模型自动检测米粒的垩白度。
  11. 根据权利要求10所述的装置，其中，所述采集模块包括如下至少之一：
    第一采集模块，设置为通过用户使用的烹饪器具来采集米粒样本图像；
    第二采集模块，设置为通过外观检测仪来采集米粒样本图像；
    第三采集模块，设置为通过拍摄设备来采集米粒样本图像；
    第四采集模块，设置为通过网络平台来获取米粒样本图像；
    其中，所述米粒样本图像为训练数据。
  12. 根据权利要求11所述的装置，其中，所述采集模块还包括加载模块，设置为在通过所述采集模块上传米粒样本图像的垩白度后，将上传所述垩白度作为所述标注信息加载到所述米粒样本图像中。
  13. 根据权利要求10至12中任意一项所述的装置，其中，所述训练模块包括：
    扫描模块，设置为扫描所述米粒样本图像，得到卷积特征图；
    池化模块，设置为对所述卷积特征图进行池化处理，得到去冗余后的特征图；
    激活模块，设置为激活所述去冗余后的特征图，并对激活结果进行分类；
    训练子模块，设置为使用BP反向传播算法对分类后的特征图进行训练，得到所述训练好的神经网络模型。
  14. 根据权利要求13所述的装置，其中，所述扫描模块包括：
    扫描子模块，设置为通过所述卷积神经网络的卷积层中的每个卷积核扫描所述米粒样本图像，得到所述卷积特征图，其中所述卷积特征图用于表征所述米粒样本图像中的米粒的特征图层。
  15. 根据权利要求13所述的装置，其中，所述池化模块包括池化子模块，设置为通过所述池化层对所述卷积特征图进行池化处理，得到去冗余后的特征图。
  16. 根据权利要求13所述的装置，其中，所述激活模块包括激活子模块，设置为通过所述卷积神经网络的至少一个全连接层激活所述去冗余后的特征图。
  17. 根据权利要求10所述的装置，其中，在使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型之后，所述装置还包括：
    保存模块，设置为将所述神经网络模型保存在本地设备，或上传至服务器。
  18. 根据权利要求10所述的装置，其中，所述检测模块包括：
    上传模块，设置为如果接收到待检测的米粒图像，将所述待检测的米粒图像上传到服务器；
    接收模块，设置为接收所述服务器返回的垩白度，其中，服务器使用所述神经网络模型检测得到所述待检测的米粒图像的垩白度。
  19. 一种存储介质，所述存储介质包括存储的程序，其中，在所述程序运行时控制所述存储介质所在设备执行权利要求1至9中任意一项所述的获取米粒的垩白度的方法。
  20. 一种处理器，所述处理器设置为运行程序，其中，在所述程序运行时执行权利要求1至9中任意一项所述的获取米粒的垩白度的方法。
  21. 一种烹饪器具，包括:
    采集装置，设置为采集多张米粒样本图像；
    处理器，所述处理器设置为运行程序，其中，在所述程序运行时对于从所述采集装置输出的数据执行如下处理步骤：对所述米粒样本图像加载标注信息，所述标注信息至少包括了所述米粒样本图像中米粒的垩白度；使用卷积神经网络对 采集到的米粒样本图像进行训练，得到训练好的神经网络模型；基于训练得到的神经网络模型自动检测米粒的垩白度。
  22. 一种获取米粒的垩白度的方法，包括：
    接收到待检测的米粒图像，将所述待检测的米粒图像上传到服务器；
    接收所述服务器返回的垩白度，其中，所述服务器采集多张米粒样本图像，并使用卷积神经网络对采集到的米粒样本图像进行训练，得到训练好的神经网络模型，所述服务器使用训练好的神经网络模型检测得到所述待检测的米粒图像的垩白度，所述米粒样本图像为加载了标注信息的图像，所述标注信息至少包括了所述米粒样本图像中米粒的垩白度。
PCT/CN2019/109970 2019-01-14 2019-10-08 获取米粒的垩白度的方法、装置和烹饪器具 WO2020147345A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910033070.XA CN111435541A (zh) 2019-01-14 2019-01-14 获取米粒的垩白度的方法、装置和烹饪器具
CN201910033070.X 2019-01-14

Publications (1)

Publication Number Publication Date
WO2020147345A1 true WO2020147345A1 (zh) 2020-07-23

Family

ID=71579943

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/109970 WO2020147345A1 (zh) 2019-01-14 2019-10-08 获取米粒的垩白度的方法、装置和烹饪器具

Country Status (2)

Country Link
CN (1) CN111435541A (zh)
WO (1) WO2020147345A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11618858B1 (en) 2021-12-06 2023-04-04 Saudi Arabian Oil Company Hydrodearylation catalysts for aromatic bottoms oil, method for producing hydrodearylation catalysts, and method for hydrodearylating aromatic bottoms oil with hydrodearylation catalysts

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114066887B (zh) * 2022-01-11 2022-04-22 安徽高哲信息技术有限公司 大米垩白区域检测方法、装置、设备及存储介质
CN114689527A (zh) * 2022-05-31 2022-07-01 合肥安杰特光电科技有限公司 一种大米垩白检测方法和系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102495068A (zh) * 2011-12-01 2012-06-13 浙江工商大学 大米垩白米在线检测方法和系统
CN104215584A (zh) * 2014-08-29 2014-12-17 华南理工大学 一种基于高光谱图像技术区分大米产地的检测方法
US20160180162A1 (en) * 2014-12-22 2016-06-23 Yahoo! Inc. Generating preference indices for image content
CN107066995A (zh) * 2017-05-25 2017-08-18 中国矿业大学 一种基于卷积神经网络的遥感图像桥梁检测方法
CN207067721U (zh) * 2017-07-10 2018-03-02 昆明理工大学 一种用于远程监控的电饭煲装置

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102279976A (zh) * 2011-09-22 2011-12-14 河南工业大学 不同糙米籽粒识别的bp神经网络构建及识别方法
CN108090517A (zh) * 2017-12-29 2018-05-29 美的集团股份有限公司 一种谷物识别方法、装置和计算机存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102495068A (zh) * 2011-12-01 2012-06-13 浙江工商大学 大米垩白米在线检测方法和系统
CN104215584A (zh) * 2014-08-29 2014-12-17 华南理工大学 一种基于高光谱图像技术区分大米产地的检测方法
US20160180162A1 (en) * 2014-12-22 2016-06-23 Yahoo! Inc. Generating preference indices for image content
CN107066995A (zh) * 2017-05-25 2017-08-18 中国矿业大学 一种基于卷积神经网络的遥感图像桥梁检测方法
CN207067721U (zh) * 2017-07-10 2018-03-02 昆明理工大学 一种用于远程监控的电饭煲装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11618858B1 (en) 2021-12-06 2023-04-04 Saudi Arabian Oil Company Hydrodearylation catalysts for aromatic bottoms oil, method for producing hydrodearylation catalysts, and method for hydrodearylating aromatic bottoms oil with hydrodearylation catalysts

Also Published As

Publication number Publication date
CN111435541A (zh) 2020-07-21

Similar Documents

Publication Publication Date Title
WO2020147345A1 (zh) 获取米粒的垩白度的方法、装置和烹饪器具
US10402980B2 (en) Imaging system object recognition and assessment
CN105142408B (zh) 热处理监控系统
CN108829723A (zh) 基于复杂网络和深度学习的可交互智能冰箱健康服务终端
US20190042873A1 (en) Metric-based recognition, systems and methods
CN107341350A (zh) 食材智能管理方法和装置、服务器、智能冰箱、存储介质
CN110490688B (zh) 商品推荐方法和装置
IL223687A (en) A website that provides a grooming and nutrition plan based on color images
KR20190108049A (ko) 냉장고, 그 제어 방법 및 시스템
CN108197635B (zh) 烹饪方式的展示方法及装置、抽油烟机
US20180330224A1 (en) Diet information recommendation system and diet information recommendation method
CN108663331A (zh) 检测冰箱内食物新鲜度的方法与冰箱
CN112464013A (zh) 信息的推送方法和装置、电子设备和存储介质
TW201719146A (zh) 作物之生理表徵檢測方法、可攜式檢測裝置與檢測系統
CN114898405B (zh) 基于边缘计算的便携式肉鸡异常监测系统
Arunachalam et al. Real-time plant phenomics under robotic farming setup: A vision-based platform for complex plant phenotyping tasks
KR102532651B1 (ko) 온실 작물 관리 시스템
KR101878359B1 (ko) 정보기술을 이용한 다중지능 검사 장치 및 방법
CN109166054A (zh) 烹饪鱼肉类时碳化时点管理服务提供系统
CN114556444A (zh) 联合模型的训练方法和对象信息处理方法、装置及系统
CN117115532B (zh) 一种基于物联网的展览台智能控制方法及系统
JP4625962B2 (ja) カテゴリカル色知覚システム
Vishnu et al. Fruit Recognition System for Calorie Management
CN107421194A (zh) 一种冰箱、饮食方案推送系统以及冰箱业务实现方法
CN107392693B (zh) 一种基于物联网的农作物采购方法及系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19910802

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19910802

Country of ref document: EP

Kind code of ref document: A1