AI rice heating and cooking auxiliary solution

Abstract

The invention discloses an AI rice heating and cooking auxiliary solution. According to the method, food pictures are input through an external camera or a file system, picture features are extracted through a deep learning convolutional neural network CNN, simplified multi-modal data are fused through a self-attention mechanism and other modes, the accurate heating time of food on a stove is predicted, and manual estimation is replaced. The method has the beneficial effects that the method can be operated on equipment with limited computing power, such as mobile computing equipment and kitchen appliances, and can accurately predict rice heating and cooking time so as to assist in cooking and improve life efficiency and safety.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence home technology, and in particular to an AI rice heating and cooking auxiliary solution. Background Art

[0002] With the development of artificial intelligence home technology, people's requirements for quality of life and safety are increasing.

[0003] The food heating and cooking temperature control equipment currently used on the market have the following problems: 1. It is not possible to quickly, accurately, and conveniently intelligently estimate the heating time of different types of food to assist in heating and cooking food; 2. Some rely on users to manually estimate and control time and temperature, which makes use inconvenient and the experience unsatisfactory; 3. Some rely on traditional detectors, such as infrared radiation detectors, and some rely on remote servers for auxiliary calculations, which are costly and not conducive to popularization; 4. Some AI estimation algorithms use overly simple or overly complicated technical solutions, are not accurate enough, have low efficiency or have redundant algorithms, and need to be improved; 5. There is no similar artificial intelligence device on the market that can estimate time and assist in heating and cooking.

[0004] Specifically, A method for automatically identifying ingredients and adjusting cooking parameters and a smart cooker, Patent No.: CN118078112A Methods: Use deep learning and machine learning algorithms to extract food features, classify the food by comparing it with the database, and then search the database again to obtain parameters such as cooking methods, cooking time and cooking temperature to cook the food; Disadvantages: The calculation process is cumbersome and requires multiple database searches. The system design is complex, which may lead to unreliable operation. The hardware cost is high, the software development and maintenance cost is high, the database establishment and update are difficult, and the calculation accuracy for niche ingredients and unknown ingredients is insufficient.

[0005] A microwave oven intelligent heating control method and system, patent number: CN118890734A Methods: After collecting food images in the microwave oven, upload them to the remote server, use the machine learning image recognition algorithm on the server to identify the type of food, and retrieve the corresponding heating time equation. At the same time, it is necessary to obtain the weight information of the food, use the algorithm model to calculate the food heating time, and then return it to the client's microwave oven for heating control; Disadvantages: A remote server is used, the process is cumbersome, and the development, maintenance and use costs are high. The food heating time algorithm model used needs to rely on multiple parameter inputs such as food weight, food initial temperature, food specific heat capacity, heating target temperature, microwave oven power, etc., several of which are redundant. The same information can be obtained from the food image, resulting in high software and hardware costs for calculation dependence, cumbersome calculations, and high costs for establishing the data set used to train the algorithm model.

[0006] A vehicle-mounted intelligent lunch box intelligent temperature control method and system, patent number: CN118349048A Method: The camera, temperature sensor, weight sensor in the vehicle-mounted smart lunch box and the remote server are used to classify meat and vegetarian dishes to estimate the heating time of the general type of food in the smart lunch box, and intelligently control the heat preservation or heating; Disadvantages: It uses multiple sensors other than cameras and a remote server to estimate the food heating time, which is costly. The machine learning algorithm that processes food images only roughly classifies food types into meat, vegetables, etc., and cannot be subdivided. It cannot accurately estimate the heating time, and it is only used for heating smart lunch boxes on trucks, so its application range is limited.

[0007] Intelligent temperature control system for efficient electromagnetic cookers, patent number: CN115789717A Method: By monitoring the vibration of the boiling liquid food in the pot, the heating temperature of the stove can be controlled to prevent overflow; Disadvantages: It is only used to heat porridge, soup and other liquid foods that need to be heated to boiling. The applicable scenarios are limited, and the temperature adjustment will only be triggered after the boiling food produces vibrations for a period of time, which is not safe and clean enough.

[0008] Food heating temperature control methods on the market: Some use thermal infrared sensor probes, thermistors, vibration frequency monitors, smart cameras and other sensors to detect several physical properties of food such as temperature and weight to control temperature. Others use the principles of machine learning and deep learning pattern recognition algorithms to calculate food images and combine external sensors to provide additional physical properties of food to control heating and cooking. Disadvantages: Some solutions using sensors are expensive or have a limited lifespan, making them difficult to popularize. Some rely on a few physical data from sensors, resulting in inaccurate and unintelligent temperature control or limited applicable scenarios. Some use machine learning and deep learning estimation algorithms, but the algorithms they use are too simple or too cumbersome to estimate the heating time of food of different types, ingredients and initial temperatures intelligently, accurately and reliably enough, and have limited applicable scenarios. Some estimation algorithms rely on multiple input parameters and additional sensors, making them difficult to implement, costly and inconvenient for users. In addition, some solutions use remote servers, resulting in high software and hardware costs, high user costs, large delays and poor user experience. Summary of the invention

[0009] The purpose of the present invention is to provide an AI rice heating and cooking assistance solution, which can run an optimized deep learning graphic feature extraction algorithm on a mobile computing device, a household appliance or a kitchen appliance with limited computing power, extract graphic features, combine the input of the stove heating power value and the target dining temperature value, fuse these two multimodal features directly or using a self-attention mechanism, and accurately estimate the heating time to replace manual estimation, so as to assist users in heating and cooking rice and provide convenience in life.

[0010] To achieve the above object, the present invention adopts the following technical solutions: An AI-assisted rice cooking solution, including: S1: External input module, accepting food pictures, stove heating power value and target dining temperature value input; S2: a feature extraction module, connected to the external input module, for processing the food picture and extracting image features; S3: a parameter processing module connected to the external input module, used for processing the stove heating power value and the target dining temperature value, and converting them into parameter features that match the image features; S4: a feature fusion module, connected to the feature extraction module and the parameter processing module, for fusing the image feature and the parameter feature; S5: a heating time estimation module, connected to the feature fusion module, for ultimately estimating the food heating time from the fused features; The AI ​​rice heating and cooking assistance solution accepts the food picture input by an external camera, file system, network or other data transmission, uses a pre-trained convolutional neural network CNN or a custom convolutional neural network CNN to extract image features, and processes the input stove heating power value and the target dining temperature value into parameter features that match the image features using one or more fully connected layers, then directly fuses the image features and the parameter features or fuses them using a self-attention mechanism, and predicts the precise heating time that is finally fed back to the user through one or more fully connected layers to assist in heating and cooking rice.

[0011] Furthermore, in step S1, the external input module includes: S101: A picture of food to be heated or cooked inputted from an external camera, file system, network or other data transmission. The picture of food should have good clarity and illumination, because in real life, the human eye can distinguish the composition of the surface of food and its initial temperature range, and can estimate the heating time of food based on this. For example, food may be at room temperature or frozen and have different combinations and proportions of meat, vegetables, soup, rice and noodles, etc. The input picture of food can provide information in this regard; S102: The heating power value of the stove. For example, a microwave oven has power options such as high, medium and low, and these settings need to be referred to to accurately estimate the heating time; S103: Target dining temperature value. When using a stove to defrost, heat rice, boil water or cook noodles, there will be different target temperatures. You also need to refer to this part of the settings to accurately estimate the heating time in order to achieve the best dining experience, ensure a clean environment and safe use.

[0012] Furthermore, in step S2, the feature extraction module includes: S201: Pre-trained CNN or custom CNN, using the food picture described in step S101 as input, to extract image features. If the pre-trained CNN is used, models such as MobileNet series, EfficientNet series, SqueezeNet, ShuffleNet series or ResNet series lightweight version can be used. The advantage is that the model has been pre-trained with a large-scale data set, has fast convergence, high precision, and saves research and development time and cost. If the custom CNN is used, one to several convolutional layers can be used as the main architecture. The convolutional layer can be used in conjunction with a pooling layer to compress data dimensions and improve operating effects and efficiency. It can also be used in conjunction with a Dropout layer to prevent overfitting, enhance robustness, and be quantifiable. The advantage is that the model architecture can be flexibly adjusted to reduce resource consumption and achieve the best balance between resource consumption, calculation accuracy and calculation speed. The pre-trained CNN or the custom CNN can achieve real-time detection or low-latency detection, can run on frameworks such as Tensorflow or PyTorch Mobile version, support mobile, ultra-small devices, household appliances and kitchen appliances with limited computing power, and can reduce equipment costs and energy consumption.

[0013] Furthermore, in step S3, the parameter processing module includes: S301: One to several fully connected layers, using the stove heating power value in step S102 and the target dining temperature value in step S103 as input, and processing through one to several fully connected layers to obtain parameter features that match the image features in step S201. This is an optimization method for multimodality data from different sources to reduce the dimensional difference between the image features and the stove heating power value in step S102 and the target dining temperature value in step S103.

[0014] Furthermore, in step S4, the feature fusion module includes: S401: Direct feature fusion or self-attention mechanism feature fusion, using the image features in step S201 and the parameter features in step S301 as input, and fusing them. If the direct feature fusion is used, the image features and the parameter features can be flattened and connected together, which has the advantages of simple and efficient processing and high real-time calculation. If the self-attention mechanism feature fusion is used, the advantages are strong feature interaction capability, dynamic weight allocation, high interpretability and high accuracy.

[0015] Furthermore, in step S5, the heating time estimation module includes: S501: One to several fully connected layers, using the fused features described in step S401 as input, map to the final predicted value, which is the precise heating time that meets the user's expectations and is suitable for the current stove, and can continue to output this value to the digital stove, other device digital interfaces, or prompt the user in the form of a display screen, voice, etc.

[0016] Furthermore, the solution requires the preparation of a data set for pre-training. The data set should take a certain application scenario of heating and cooking food as a data group. The data group should include the food picture mentioned in step S1, the stove heating power value and the target dining temperature value, and mark the actual heating time in this application scenario, corresponding to the heating time prediction value output in step S5. The data set should be composed of data groups of a certain scale. The food pictures between different data groups should be different, and should respectively reflect a wide range of food types at different angles and illuminations under the influence of the environment to reflect the corresponding application scenarios, so as to reduce abnormal gradient changes in the training and prediction process, reduce overfitting, enhance robustness, and improve prediction accuracy and generalization ability.

[0017] Beneficial effects of the present invention: 1. The computational complexity of the deep learning and other algorithms used does not put much pressure on modern software and hardware, and can be run on devices with limited computing power. It is fast, consumes less energy, and can perform real-time or low-latency recognition. It can be used in mobile devices, ultra-small devices, home appliances or kitchen appliances to achieve low costs and facilitate daily use; 2. The recognition and estimation accuracy is high, which can avoid the errors caused by manual estimation and operation, help users heat and cook food in one go, reduce the time wasted due to manual errors and the need for multiple heating, and improve life efficiency; 3. The system design is simple, which can ensure the reliability and attendance rate of long-term operation; 4. The parameters of the data set used for training the model are extremely simple, and the difficulty of establishment is relatively low, which reduces the development cost and improves feasibility; 5. It can avoid the degradation of dining experience, environmental pollution, and safety and health risks caused by overheating; 6. If combined with remote control and time control system, remote, timed automation and even unmanned operation can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a work flow chart of the present invention; Figure 2 It is a schematic diagram of an application scenario of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0020] In the description of the embodiments of the present invention, it should be noted that if the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, or the orientation or position relationship in which the invented product is usually placed when used. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0021] In addition, the terms "horizontal", "vertical", "overhanging" and the like do not mean that the components must be absolutely horizontal or overhanging, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0022] In the description of the embodiments of the present invention, "plurality" means at least 2.

[0023] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. Example

[0024] An AI-assisted solution for heating and cooking rice.

[0025] System composition: 1. External input module, accepting food pictures, stove heating power value and target dining temperature value input; 2. A feature extraction module, connected to the external input module, for processing the food picture and extracting image features; 3. A parameter processing module, connected to the external input module, for processing the stove heating power value and the target dining temperature value, and converting them into parameter features that match the image features; 4. A feature fusion module, connected to the feature extraction module and the parameter processing module, for fusing the image features and the parameter features; 5. A heating time estimation module, connected to the feature fusion module, is used to ultimately estimate the food heating time from the fused features.

[0026] Working principle: The AI ​​rice heating and cooking assistance solution accepts the food picture input by an external camera, file system, network or other data transmission, uses a pre-trained convolutional neural network CNN or a custom convolutional neural network CNN to extract image features, and processes the input stove heating power value and the target dining temperature value into parameter features that match the image features using one or more fully connected layers, then directly fuses the image features and the parameter features or fuses them using a self-attention mechanism, and predicts the precise heating time that is finally fed back to the user through one or more fully connected layers to assist in heating and cooking rice.

[0027] Specifically: An AI-assisted rice cooking solution, including: S1: External input module, accepting food pictures, stove heating power value and target dining temperature value input; S2: a feature extraction module, connected to the external input module, for processing the food picture and extracting image features; S3: a parameter processing module connected to the external input module, used for processing the stove heating power value and the target dining temperature value, and converting them into parameter features that match the image features; S4: a feature fusion module, connected to the feature extraction module and the parameter processing module, for fusing the image feature and the parameter feature; S5: a heating time estimation module, connected to the feature fusion module, for ultimately estimating the food heating time from the fused features; The AI ​​rice heating and cooking assistance solution accepts the food picture input by an external camera, file system, network or other data transmission, uses a pre-trained convolutional neural network CNN or a custom convolutional neural network CNN to extract image features, and processes the input stove heating power value and the target dining temperature value into parameter features that match the image features using one or more fully connected layers, then directly fuses the image features and the parameter features or fuses them using a self-attention mechanism, and predicts the precise heating time that is finally fed back to the user through one or more fully connected layers to assist in heating and cooking rice.

[0028] In this embodiment, the external input module in step S1 includes: S101: A picture of food to be heated or cooked inputted from an external camera, file system, network or other data transmission. The picture of food should have good clarity and illumination, because in real life, the human eye can distinguish the composition of the surface of food and its initial temperature range, and can estimate the heating time of food based on this. For example, food may be at room temperature or frozen and have different combinations and proportions of meat, vegetables, soup, rice and noodles, etc. The input picture of food can provide information in this regard; S102: The heating power value of the stove. For example, a microwave oven has power options such as high, medium and low, and these settings need to be referred to to accurately estimate the heating time; S103: Target dining temperature value. When using a stove to defrost, heat rice, boil water or cook noodles, there will be different target temperatures. You also need to refer to this part of the settings to accurately estimate the heating time in order to achieve the best dining experience, ensure a clean environment and safe use.

[0029] In this embodiment, the feature extraction module in step S2 includes: S201: Pre-trained CNN or custom CNN, using the food picture described in step S101 as input, to extract image features. If the pre-trained CNN is used, models such as MobileNet series, EfficientNet series, SqueezeNet, ShuffleNet series or ResNet series lightweight version can be used. The advantage is that the model has been pre-trained with a large-scale data set, has fast convergence, high precision, and saves research and development time and cost. If the custom CNN is used, one to several convolutional layers can be used as the main architecture. The convolutional layer can be used in conjunction with a pooling layer to compress data dimensions and improve operating effects and efficiency. It can also be used in conjunction with a Dropout layer to prevent overfitting, enhance robustness, and be quantifiable. The advantage is that the model architecture can be flexibly adjusted to reduce resource consumption and achieve the best balance between resource consumption, calculation accuracy and calculation speed. The pre-trained CNN or the custom CNN can achieve real-time detection or low-latency detection, can run on frameworks such as Tensorflow or PyTorch Mobile version, support mobile, ultra-small devices, household appliances and kitchen appliances with limited computing power, and can reduce equipment costs and energy consumption.

[0030] In this embodiment, the parameter processing module in step S3 includes: S301: One to several fully connected layers, using the stove heating power value in step S102 and the target dining temperature value in step S103 as input, and processing through one to several fully connected layers to obtain parameter features that match the image features in step S201. This is an optimization method for multimodality data from different sources to reduce the dimensional difference between the image features and the stove heating power value in step S102 and the target dining temperature value in step S103.

[0031] In this embodiment, the feature fusion module in step S4 includes: S401: Direct feature fusion or self-attention mechanism feature fusion, using the image features in step S201 and the parameter features in step S301 as input, and fusing them. If the direct feature fusion is used, the image features and the parameter features can be flattened and connected together, which has the advantages of simple and efficient processing and high real-time calculation. If the self-attention mechanism feature fusion is used, the advantages are strong feature interaction capability, dynamic weight allocation, high interpretability and high accuracy.

[0032] In this embodiment, the heating time estimation module in step S5 includes: S501: One to several fully connected layers, using the fused features described in step S401 as input, map to the final predicted value, which is the precise heating time that meets the user's expectations and is suitable for the current stove, and can continue to output this value to the digital stove, other device digital interfaces, or prompt the user in the form of a display screen, voice, etc.

[0033] In this embodiment, the solution needs to prepare a data set for pre-training. The data set should take a certain application scenario of heating and cooking food as a data group. The data group should include the food picture mentioned in step S1, the stove heating power value and the target dining temperature value, and mark the actual heating time in this application scenario, corresponding to the heating time prediction value output in step S5. The data set should be composed of data groups of a certain scale. The food pictures between different data groups should be different, and they should respectively reflect a wide range of food at different angles and illuminations under the influence of the environment to reflect the corresponding application scenarios, so as to reduce abnormal gradient changes in the training and prediction process, reduce overfitting, enhance robustness, and improve prediction accuracy and generalization ability.

[0034] in, The calculation part of algorithms such as image feature extraction can be completed using AI big models in conjunction with prompt words; The advantages are: low development difficulty. You only need to use natural language to write appropriate prompt words with food pictures, and build and call large model services. However, it requires high computing power of the device, and has too high hardware requirements for mobile devices, ultra-small devices, home appliances and kitchen appliances. It is costly, has large computing delays, and consumes a lot of energy. At present, there is no good large model specifically built and optimized for food processing scenarios. Another improvement is that the food pictures can be transmitted to the cloud server by integrating the Internet access module, and the cloud AI big model service can be called to complete the image feature extraction and heating time estimation described in this solution; The advantages are: it does not require high hardware for mobile devices, ultra-small devices, household appliances, and kitchen appliances, and the software development is not difficult. However, it is necessary to add hardware and software modules for Internet access to the devices, which will lead to high manufacturing, development, maintenance and operation costs. It also requires the user's location to have Internet access, and the user needs to manually perform additional configuration, which increases the cost and difficulty of use. There may also be large delays and unstable services. In addition, there will be problems with user data and privacy security. Another improvement is that the user does not need to input the stove heating power value and the target dining temperature value contained in the external input module when using the device, and can directly heat the meal with one click. This is because common electronic stoves on the market, such as microwave ovens, only have a few power options such as high fire, medium fire and low fire, and many people even only use high fire. The target heating temperature can also be predefined with a few options commonly used by users, such as 37 degrees Celsius, 40 degrees, 60 degrees and 100 degrees. The estimation algorithm in this solution can be used to calculate all their combinations at once, and the user can select with one click or reuse the user's historical selection preferences. The benefits are: it can further reduce the complexity of user operations, but it will increase the complexity of user interface design and software system design, and may have problems such as operability and practicality that do not conform to the actual usage habits of users, and it will also limit the freedom of user choice and may also pose safety risks; Another improvement is to continuously collect food images during the food heating process and use deep learning pattern recognition algorithms to classify the food into cooked, boiling, charred, burned, smoking and other states to further assist in temperature control; The advantages are: when heating needs to be stopped, it can be stopped in time, or an alarm will be issued to the user, which can ensure safety and avoid food waste. However, because the recognition range is relatively rough, it cannot be effectively identified before the food temperature changes significantly or the appearance changes significantly, so its practicality is limited.

[0035] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An AI rice heating and cooking auxiliary solution, characterized in that: include: S1: External input module, accepting food pictures, stove heating power value and target dining temperature value input; S2: a feature extraction module, connected to the external input module, for processing the food picture and extracting image features; S3: a parameter processing module connected to the external input module, used for processing the stove heating power value and the target dining temperature value, and converting them into parameter features that match the image features; S4: a feature fusion module, connected to the feature extraction module and the parameter processing module, for fusing the image feature and the parameter feature; S5: a heating time estimation module, connected to the feature fusion module, for ultimately estimating the food heating time from the fused features; The AI ​​rice heating and cooking assistance solution accepts the food picture input by an external camera, file system, network or other data transmission, uses a pre-trained convolutional neural network CNN or a custom convolutional neural network CNN to extract image features, and processes the input stove heating power value and the target dining temperature value into parameter features that match the image features using one or more fully connected layers, then directly fuses the image features and the parameter features or fuses them using a self-attention mechanism, and predicts the precise heating time that is finally fed back to the user through one or more fully connected layers to assist in heating and cooking rice.

2. The AI ​​hot rice cooking auxiliary solution according to claim 1 is characterized in that: In step S1, the external input module includes: S101: A picture of food to be heated or cooked inputted from an external camera, file system, network or other data transmission. The picture of food should have good clarity and illumination, because in real life, the human eye can distinguish the composition of the surface of food and its initial temperature range, and can estimate the heating time of food based on this. For example, food may be at room temperature or frozen and have different combinations and proportions of meat, vegetables, soup, rice and noodles, etc. The input picture of food can provide information in this regard; S102: The heating power value of the stove. For example, a microwave oven has power options such as high, medium and low, and these settings need to be referred to to accurately estimate the heating time; S103: Target dining temperature value. When using a stove to defrost, heat rice, boil water or cook noodles, there will be different target temperatures. You also need to refer to this part of the settings to accurately estimate the heating time in order to achieve the best dining experience, ensure a clean environment and safe use.

3. The AI ​​hot rice cooking auxiliary solution according to claim 1 is characterized in that: In step S2, the feature extraction module includes: S201: Pre-trained CNN or custom CNN, using the food picture described in step S101 as input, to extract image features. If the pre-trained CNN is used, models such as MobileNet series, EfficientNet series, SqueezeNet, ShuffleNet series or ResNet series lightweight version can be used. The advantage is that the model has been pre-trained with a large-scale data set, has fast convergence, high precision, and saves research and development time and cost. If the custom CNN is used, one to several convolutional layers can be used as the main architecture. The convolutional layer can be used in conjunction with a pooling layer to compress data dimensions and improve operating effects and efficiency. It can also be used in conjunction with a Dropout layer to prevent overfitting, enhance robustness, and be quantifiable. The advantage is that the model architecture can be flexibly adjusted to reduce resource consumption and achieve the best balance between resource consumption, calculation accuracy and calculation speed. The pre-trained CNN or the custom CNN can achieve real-time detection or low-latency detection, can run on frameworks such as Tensorflow or PyTorch Mobile version, support mobile, ultra-small devices, household appliances and kitchen appliances with limited computing power, and can reduce equipment costs and energy consumption.

4. The AI ​​hot rice cooking auxiliary solution according to claim 1 is characterized in that: In step S3, the parameter processing module includes: S301: One to several fully connected layers, using the stove heating power value in step S102 and the target dining temperature value in step S103 as input, and processing through one to several fully connected layers to obtain parameter features that match the image features in step S201. This is an optimization method for multimodality data from different sources to reduce the dimensional difference between the image features and the stove heating power value in step S102 and the target dining temperature value in step S103.

5. The AI ​​hot rice cooking auxiliary solution according to claim 1 is characterized in that: In step S4, the feature fusion module includes: S401: Direct feature fusion or self-attention mechanism feature fusion, using the image features in step S201 and the parameter features in step S301 as input, and fusing them. If the direct feature fusion is used, the image features and the parameter features can be flattened and connected together, which has the advantages of simple and efficient processing and high real-time calculation. If the self-attention mechanism feature fusion is used, the advantages are strong feature interaction capability, dynamic weight allocation, high interpretability and high accuracy.

6. The AI ​​hot rice cooking auxiliary solution according to claim 1 is characterized in that: In step S5, the heating time estimation module includes: S501: One to several fully connected layers, using the fused features described in step S401 as input, map to the final predicted value, which is the precise heating time that meets the user's expectations and is suitable for the current stove, and can continue to output this value to the digital stove, other device digital interfaces, or prompt the user in the form of a display screen, voice, etc.

7. The AI ​​hot rice cooking assistance solution according to claim 1 is characterized in that: The solution requires the preparation of a data set for pre-training. The data set should take a certain application scenario of heating and cooking food as a data group. The data group should include the food picture mentioned in step S1, the stove heating power value and the target dining temperature value, and mark the actual heating time in this application scenario, corresponding to the heating time prediction value output in step S5. The data set should be composed of data groups of a certain scale. The food pictures between different data groups should be different, and should respectively reflect a wide range of food types at different angles and illuminations under the influence of the environment to reflect the corresponding application scenarios, so as to reduce abnormal gradient changes in the training and prediction process, reduce overfitting, enhance robustness, and improve prediction accuracy and generalization ability.

Images