Artificial intelligence cooking method of cooking equipment

Abstract

The invention belongs to the technical field of kitchen robots, and particularly relates to an artificial intelligence cooking method of cooking equipment. Through an image recognition technology, the shape and color of food materials and a kitchen scene are recognized; according to the cooking equipment, a high-definition camera is used for recording operation of a cook, cooking skill actions are observed and learned through deep learning and action capture technologies, gestures and steps are analyzed through an algorithm, and then the cooking process is reproduced; the cooking equipment simulates human actions, and through repeated practice and algorithm optimization, the equipment gradually masters cooking skills and independently cooks dishes; through observation, learning and reproduction of complex cooking actions and skills and autonomous learning ability, the robot can evolve continuously, master more cooking skills and meet diversified cooking requirements.

Description

Technical Field

[0001] The invention belongs to the technical field of kitchen robots, and in particular relates to an artificial intelligence cooking method of a cooking device. Background Art

[0002] In today's era of rapid technological development, humanoid robots are no longer just a fantasy in science fiction works, but are gradually integrated into real life. Among them, cooking equipment that learns and reproduces cooking skills responds to people's demand for smart homes and smart services, bringing convenience and fun to daily life. Behind the cooking equipment that learns and reproduces cooking skills, a number of cutting-edge technologies are integrated. It uses high-precision sensors, such as visual and tactile sensors, to simulate human chefs' identification and operation of ingredients. At the same time, through deep learning algorithms, the robot can analyze and understand complex recipes, master ingredient matching and cooking skills. In addition, it also has a powerful robotic arm and power system to ensure that various tasks can be completed flexibly and accurately during the cooking process. The combined application of these technologies enables humanoid robots to efficiently and accurately reproduce human cooking skills, bringing revolutionary changes to the cooking industry. Summary of the invention

[0003] The purpose of the present invention is to provide an artificial intelligence cooking method for a cooking device. By observing, learning and reproducing complex cooking movements and techniques, the autonomous learning ability enables the robot to continuously evolve, master more cooking skills, and meet diverse cooking needs.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is as follows:

[0005] An artificial intelligence cooking method for a cooking device, the cooking method comprising the following steps:

[0006] Step 1: The cooking equipment uses image recognition technology to identify the shape and color of ingredients and the kitchen scene;

[0007] Step 2: The cooking equipment uses a high-definition camera to record the chef's actions, observes and learns cooking movements through deep learning and motion capture technology, analyzes gestures and steps through algorithms, and then reproduces the cooking process;

[0008] Step 3: The cooking equipment imitates human movements. Through repeated practice and algorithm optimization, the equipment gradually masters cooking skills and cooks dishes independently.

[0009] The step 1 of identifying the shape and color of food includes the following steps:

[0010] Step 1: Data collection and preprocessing: collect image data of various ingredients, preprocess the images, and ensure the consistency of input data;

[0011] Step 2: Build a CNN model and design a suitable CNN architecture based on task requirements;

[0012] Step 3: Model training, use the preprocessed food image data to train the CNN model;

[0013] Step ④: Model evaluation and optimization: evaluate the performance of the model on an independent test set and tune the model;

[0014] Step 5: Apply the trained CNN model to the actual scene, identify the input food pictures, and output the recognition results.

[0015] The kitchen scene recognition in step 1 includes active scene recognition and passive scene recognition;

[0016] The active scene recognition is achieved by actively scanning the environment through a sensor or actively capturing images through a camera;

[0017] The passive scene recognition is that the cooking device analyzes and recognizes the environmental data after receiving it.

[0018] The cooking device imitating human actions in step 3 includes the following steps:

[0019] Step ①: Motion capture and reproduction;

[0020] Through sensor technology, the robot can capture the cooking movements of human chefs and then accurately reproduce these movements through behavioral reproduction technology;

[0021] Step ②: Posture recognition and optimization;

[0022] Use computer vision technology to recognize the chef’s movements, thereby helping the cooking equipment understand and optimize the execution process of cooking actions;

[0023] Step ③: Hand movement planning;

[0024] Cooking equipment needs to learn how to use hands correctly to perform cooking operations. Behavior replication can help cooking equipment imitate the movement trajectory and operation methods of human hands.

[0025] Step ④: Real-time feedback and adjustment;

[0026] Behavior reproduction combined with real-time feedback mechanism enables cooking equipment to adjust its actions according to the situation during the cooking process to ensure the smooth progress of the cooking process;

[0027] Step ⑤: Multimodal data integration;

[0028] Integrate multimodal data such as vision and voice into the behavior reproduction system to achieve a more comprehensive and intelligent cooking process simulation.

[0029] By observing, learning and reproducing complex cooking movements and techniques, the robot's autonomous learning ability enables it to continuously evolve, master more cooking skills and meet diverse cooking needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention is further illustrated by means of the non-limiting examples given in the accompanying drawings.

[0031] Figure 1 It is a schematic diagram of the DDPG algorithm flow of the present invention;

[0032] Figure 2 It is a schematic diagram of the VGG-16 convolutional network architecture flow of the present invention;

[0033] Figure 3 Identify food ingredients for the process CNN of the present invention;

[0034] Figure 4 It is the process implicit shape model of the present invention. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0036] The cooking equipment consists of the following structures:

[0037] Robot body structure: A humanoid robot needs to have a human structure, including a head, torso, limbs, and arms, in order to imitate human movements.

[0038] Joints and drives: The robot's joints need to be able to move flexibly, so they need to be equipped with drives such as motors, servo motors or hydraulic systems to achieve various movements.

[0039] Camera: used for visual perception and recognition of ingredients, utensils and actions. Camera can help the robot observe and learn the cooking process.

[0040] Sensors: including but not limited to:

[0041] Force sensor: used to control the force of the robot, such as when stirring or cutting food.

[0042] Tactile sensors: used to detect touch and pressure of objects in order to control the robot's movements.

[0043] Position sensors: used to monitor joint positions and robot posture.

[0044] Accelerometers and gyroscopes: Help the robot maintain balance and stability.

[0045] Speech recognition and synthesis: Equipped with microphones and speakers so that the robot can understand commands and communicate with users.

[0046] Control system: used to control the robot's movement, action and learning algorithm, which may include a central processor, control board and corresponding software.

[0047] Power system: Provides the power supply required by the robot, which can be a battery or a wired power supply.

[0048] Screen or display: used to display information, recipes, instructions and interact with the user.

[0049] Wi-Fi / Bluetooth connectivity: Allows the robot to connect to the internet for remote control and data transfer.

[0050] like Figure 1-4 As shown, an artificial intelligence cooking method for a cooking device of the present invention comprises the following steps:

[0051] Step 1: The cooking equipment uses image recognition technology to identify the shape and color of ingredients and the kitchen scene;

[0052] Step 2: The cooking equipment uses a high-definition camera to record the chef's actions, observes and learns cooking movements through deep learning and motion capture technology, analyzes gestures and steps through algorithms, and then reproduces the cooking process;

[0053] Step 3: The cooking equipment imitates human movements. Through repeated practice and algorithm optimization, the equipment gradually masters cooking skills and cooks dishes independently.

[0054] The step 1 of identifying the shape and color of food includes the following steps:

[0055] Step 1: Data collection and preprocessing: collect image data of various ingredients, preprocess the images, and ensure the consistency of input data;

[0056] Step 2: Build a CNN model and design a suitable CNN architecture based on task requirements;

[0057] Step 3: Model training, use the preprocessed food image data to train the CNN model;

[0058] Step ④: Model evaluation and optimization: evaluate the performance of the model on an independent test set and tune the model;

[0059] Step 5: Apply the trained CNN model to the actual scene, identify the input food pictures, and output the recognition results.

[0060] Through the above steps, CNN can effectively extract features from food images and perform classification and identification, providing strong technical support for the food industry.

[0061] In the VGG-16 architecture, there are 5 groups of conv-conv-maxpool combinations, among which ReLU is used as the activation function to calculate the maximum value of 0 and the input value, in order to speed up the training speed of the VGG-16 model. After the convolution layer, there are 3 fully connected layers, each of which contains a dropout layer for regularization during the convolution process, which can play a role in generalization and effectively prevent the occurrence of overfitting. When classifying, softmax is used as a classifier to classify images and output the possibility of the predicted class. Its loss function is:

[0062]

[0063] Among them, (x i ,y i ) represents the labeled training samples, m represents the number of training samples, and h θ (x i ) represents the hypothesis function of logistic regression. The training process of softmax is to find the optimal parameter value that minimizes the loss function.

[0064] The implementation steps of implicit shape model (ISM) in scene recognition can be briefly summarized as follows:

[0065] First, the point cloud data of the scene is preprocessed, including filtering, downsampling, etc., to reduce data redundancy and improve processing efficiency. Then, key point detection is performed, and representative key points are extracted from the point cloud data using methods such as voxel grids.

[0066] Then, feature estimation is performed for each key point, and its local geometric features are extracted using feature descriptors such as FPFH (Fast Point Feature Histogram). These features can describe the shape and neighborhood information of the key point.

[0067] Next, cluster all features using a clustering algorithm (such as K-means) to build a visual dictionary. Each cluster center in the clustering result represents a visual word, and the visual dictionary is a collection of these visual words.

[0068] In the training phase, scene point cloud data of known categories are used to count and model the key point features of each category, and construct implicit shape models of each category. These models contain the location, features and spatial relationships of key points.

[0069] Finally, in the recognition stage, for new scene point cloud data, first extract its key point features, then find the most similar visual words in the visual dictionary, and use the implicit shape model for matching and recognition. By comparing the matching degree of different category models, it can be determined which category the new scene belongs to.

[0070] The above steps show the basic implementation process of implicit shape model in scene recognition. It should be noted that in practical applications, other factors such as noise and occlusion may need to be considered to further improve the accuracy and robustness of recognition.

[0071] Active scene recognition:

[0072] In active scene recognition, the system intentionally seeks or collects data for scene recognition, such as actively scanning the environment through sensors or actively capturing images with cameras. In this case, implicit shape models can be used to help the system infer objects or structures in the scene.

[0073] Applications: Active scene recognition is often used in areas such as robot navigation, intelligent monitoring systems, and autonomous driving, where the system needs to actively perceive and understand the environment.

[0074] The role of implicit shape models: Implicit shape models can help the system infer the shape, structure, and spatial layout of objects in the scene, so as to better understand the environment and make corresponding decisions.

[0075] Passive scene recognition:

[0076] In passive scene recognition, the system typically analyzes and recognizes data after it is received, rather than actively acquiring it. In this case, implicit shape models can also play a role, but the way they are applied may be different.

[0077] Application: Passive scene recognition is common in image and video analysis, security monitoring, smart home and other fields. The system will identify and process the received data.

[0078] The role of implicit shape models: In passive scene recognition, implicit shape models can help the system more accurately analyze the shape, structure, and spatial relationship of objects in the scene, thereby improving recognition accuracy and efficiency.

[0079] In general, implicit shape models can play an important role in both active and passive scene recognition, but the specific application methods and effects will be affected by the characteristics of the scene and the system design. In different scenarios, developers can flexibly adjust the model and algorithm as needed to achieve better scene recognition results.

[0080] In the step 2:

[0081] Meta-learning technology is used to share knowledge across different tasks or environments, accelerating the learning speed and performance of the model in new tasks. Application of meta-learning in humanoid robot cooking: Quickly adapt to new recipes: Humanoid robots can quickly learn and adapt cooking steps and techniques when faced with new recipes through meta-learning methods. Adaptive adjustment: When encountering new kitchen environments or cooking tools, meta-learning can help robots quickly adapt to environmental changes and optimize the cooking process. Task generalization: Through meta-learning, robots can learn a wide range of cooking skills and scenarios, enabling better generalization and transfer between different cooking tasks.

[0082] First, decompose the meta-learning objective function of MAML. The formula of the objective function is as follows:

[0083]

[0084] R(τ′) is the discounted reward value of trajectory τ′, is the relative strategy parameter β, in task A normal policy gradient update on .

[0085] For the convenience of analysis and without loss of generality, we assume that MAML contains only one internal policy update. The meta-learning gradient of

[0086]

[0087]

[0088]

[0089] Among them, P k (π′|θ′) represents the strategy π θ′ The probability of generating trajectory τ′ under is the derivative of the internal policy update, which can be written as:

[0090]

[0091] The second term of the above formula is the internal Hessian matrix, which can be decomposed into the following formula:

[0092]

[0093]

[0094]

[0095] and because Represents the three-level nested summation formula. The new objective function is as follows:

[0096]

[0097] θ old represents the previous strategy parameters, Represents the advantage function. Finally, the optimization objective of approximate meta-reinforcement learning is:

[0098]

[0099]

[0100] in, represents the KL discreteness operator, δ target represents the desired meta-gradient update step size, and η and λ represent different KL regularization coefficients.

[0101] The cooking device imitating human actions in step 3 includes the following steps:

[0102] Step ①: Motion capture and reproduction;

[0103] Through sensor technology, the robot can capture the cooking movements of human chefs and then accurately reproduce these movements through behavioral reproduction technology;

[0104] Step ②: Posture recognition and optimization;

[0105] Use computer vision technology to recognize the chef’s movements, thereby helping the cooking equipment understand and optimize the execution process of cooking actions;

[0106] Step ③: Hand movement planning;

[0107] Cooking equipment needs to learn how to use hands correctly to perform cooking operations. Behavior replication can help cooking equipment imitate the movement trajectory and operation methods of human hands.

[0108] Step ④: Real-time feedback and adjustment;

[0109] Behavior reproduction combined with real-time feedback mechanism enables cooking equipment to adjust its actions according to the situation during the cooking process to ensure the smooth progress of the cooking process;

[0110] Step ⑤: Multimodal data integration;

[0111] Integrate multimodal data such as vision and voice into the behavior reproduction system to achieve a more comprehensive and intelligent cooking process simulation.

[0112] Step ⑥: Security assurance:

[0113] Behavior replication technology can also be used to ensure that robots remain safe during the cooking process to avoid accidents or injuries.

[0114] Step 7: Personalization:

[0115] Based on behavior reproduction technology, the robot can learn and imitate the cooking styles and skills of different chefs to provide personalized cooking services.

[0116] By applying behavior reproduction technology to robotic cooking, the robot's flexibility, accuracy, and efficiency in the cooking process can be improved, thereby achieving a more intelligent and efficient cooking experience.

[0117] Problem identification:

[0118] Problem recognition can play a key role in robotic cooking, helping the robot understand and solve problems and difficulties that may be encountered during the cooking process. Here are the ways in which problem recognition is used in robotic cooking:

[0119] 1. Food identification:

[0120] Through computer vision and image recognition technology, the robot can identify various ingredients used in cooking, including vegetables, meat, seasonings, etc., to ensure that the correct ingredients are used.

[0121] 2. Cooking process monitoring:

[0122] The robot can monitor the temperature, time and cooking status of the cooking process through sensors, detect problems such as burning, boiling over, etc. in time, and take corresponding actions.

[0123] 3. Action recognition:

[0124] By identifying the process of the robot performing cooking actions, errors or anomalies in the execution of actions can be discovered, such as action errors, hand position deviations, etc.

[0125] 4. Recipe analysis:

[0126] The robot can recognize and analyze cooking recipes, including steps and ingredient ratios, to ensure that cooking is carried out in the correct steps and proportions.

[0127] 5. Anomaly Detection:

[0128] Through machine learning technology, the robot can detect abnormal situations in the cooking process, such as missing ingredients, insufficient heat, etc., and provide early warning or correction.

[0129] 6. User interaction feedback:

[0130] After identifying the problem, the robot can interact with the user, present the problem to the user and provide solution suggestions, or adjust the cooking process based on the user's feedback.

[0131] 7. Adaptive adjustment:

[0132] Problem identification can also help robots automatically adjust cooking parameters such as temperature and time to solve problems and ensure the quality of the final dish. Through the application of problem identification technology, robots can more intelligently deal with various problems and challenges during the cooking process, improving cooking accuracy, efficiency and user experience.

[0133] Behavior Cloning (BC)

[0134] Behavioral cloning in robotic cooking refers to the process of allowing robots to imitate and replicate the cooking movements and techniques of human chefs by observing and learning them. This method can be achieved through various sensors and machine learning algorithms, allowing robots to accurately perform various complex movements and steps in the cooking process. The following are the general steps and processes of behavioral cloning in robotic cooking:

[0135] 1. Data collection:

[0136] Use sensor technology (such as visual cameras, depth cameras, etc.) to capture the action and posture data of human chefs during the cooking process. This data can include hand movements, action trajectories, posture changes, etc.

[0137] 2. Data preprocessing:

[0138] The collected data is preprocessed and features extracted so that the robot can better understand and learn the data and convert it into a form that can be used for training.

[0139] 3. Behavioral pattern learning:

[0140] The collected data is trained using machine learning algorithms such as deep learning neural networks to learn the cooking behavior patterns and techniques of human chefs. This helps the robot understand and imitate these movements.

[0141] 4. Behavior Recurrence:

[0142] After training, the robot can use the learned behavior patterns to reproduce the cooking actions of human chefs in real time. The robot will perform corresponding actions and steps based on the learned patterns and skills.

[0143] 5. Real-time adjustment and optimization:

[0144] During the cooking process, the robot can make real-time adjustments and optimizations based on feedback data and environmental conditions to ensure the accuracy and efficiency of the cooking process. Through behavioral cloning technology, the robot can learn and imitate the cooking movements of human chefs to achieve an intelligent cooking process. This approach can help robots better adapt to different cooking tasks and scenarios and improve cooking efficiency and quality.

[0145] Basic concepts:

[0146] Behavior cloning is an imitation learning method that learns strategies by imitating the behavior of human experts. It does not require rewards from environmental feedback, but uses a pre-prepared dataset (containing state and action pairs) for training. Behavior cloning is essentially supervised learning, which optimizes the policy network by minimizing the loss function so that its decisions are close to those of human experts. This method is simple and direct, but may face problems such as dataset diversity and "error accumulation".

[0147] Learning process:

[0148] The behavior cloning learning process is briefly described as follows: First, a large amount of state-action pair data is collected when the expert performs the task; then, a neural network model is trained using this data to learn to imitate the expert's behavior and generate similar action strategies. This process aims to enable the intelligent agent to copy the expert's operations and achieve efficient learning.

[0149] Applications of Control Theory:

[0150] The Stable-BC method enhances the robustness of the robot's policy by analyzing the error dynamics between the current state of the system and the state of the expert dataset and deriving stability conditions from them.

[0151] The concept of stability is also crucial in control theory, where stability usually involves stability analysis and stability design of the system to ensure that the control system maintains the desired performance under various conditions.

[0152] Although the "Stable-BC" method does not have common applications directly in control theory, the concept of stability plays a key role in control system design. In control theory, stability analysis usually involves determining whether a system is stable, asymptotically stable, or boundedly stable. When designing a controller, one usually considers the stability of the system and how to ensure the stability of the system through controller design.

[0153] Therefore, although the "Stable-BC" method is not directly applied in control theory, the extensive application of the stability concept in control theory ensures the stable operation of the system under various working conditions. The stability analysis and design methods of control theory provide an important theoretical basis for ensuring the reliability and performance of the control system.

[0154] definition:

[0155] Ideally, a robot should be in the same environment when it is trained and deployed. However, in reality, the state distribution encountered by the robot during training may be inconsistent with the state distribution when it actually performs the task, which results in covariate shift.

[0156] Cause:

[0157] Covariate shift can be caused by many factors, such as natural changes in the environment, the impact of the robot's own behavior on the environment, sensor noise or errors, and incompleteness or bias in the training dataset.

[0158] Influence:

[0159] Covariate shift can affect the performance of a robot’s policy. If the robot is not exposed to certain states or combinations of states during the training phase, it may not handle these states correctly, leading to incorrect behavior in real-world applications.

[0160] Stable-BC method:

[0161] The "Stable-BC" method usually refers to the "Stable Behavior Cloning" method, which is a method used to solve partial differential equations in the field of numerical computing. Control theory studies how to design controllers to make the system achieve the desired indicators in terms of stability, robustness and performance.

[0162] Stable-BC algorithm implementation:

[0163] Stable-BC (Stable Behavior Cloning) is a method for improving the robustness of robots when learning tasks through behavior cloning (BC). In robot control or cooking tasks, the implementation of the Stable-BC algorithm may involve ensuring that the control actions remain stable under boundary conditions. The following is a general step-by-step guide for implementing the Stable-BC algorithm:

[0164] Implementation steps:

[0165] 1. Define the problem domain:

[0166] Identify a specific problem to solve, such as stability issues in robot control, especially behavior under boundary conditions.

[0167] 2. Choose the appropriate algorithm:

[0168] Choose a stability-enhancing algorithm that is appropriate for the problem, such as Deep Deterministic Policy Gradient (DDPG) or other reinforcement learning algorithms. Optimize the Deep Deterministic Policy Gradient (DDPG) algorithm

[0169] 3. Build the model:

[0170] Design machine learning models, including Actor and Critic networks, to learn action policies and value functions.

[0171] 4. Define the environment:

[0172] Create simulated or actual robotic environments in which models can learn and be tested.

[0173] 5. Define the reward function:

[0174] Define a reward function to provide a reward signal based on the success or failure of the task.

[0175] 6. Training model:

[0176] Using reinforcement learning algorithms, the Actor and Critic networks are trained in the environment to learn stable behavior policies.

[0177] 7. Optimization model:

[0178] Adjust model parameters based on training results to improve performance and stability.

[0179] 8. Testing and evaluation:

[0180] Test the performance of your model in your environment and assess its stability under boundary conditions.

[0181] 9. Adjustment strategy:

[0182] Based on the test results and performance evaluation, the model's strategies and parameters are adjusted to improve stability.

[0183] 10. Deployment and practical application:

[0184] The trained and optimized model is deployed to a real robotic system to perform cooking tasks and ensure stability.

[0185] These steps provide a general framework for implementing the Stable-BC algorithm in robotic control or cooking tasks. The actual implementation may need to be adjusted and customized according to the complexity of the specific problem and the environment requirements. The Stable-BC algorithm (Stable Behavior Cloning) is often used for boundary condition stability in numerical computing. In robotic control or cooking tasks, the Stable-BC algorithm may point to a stability reinforcement learning algorithm to ensure that the robot's motion control remains stable under boundary conditions. Below is a simplified example showing how to use a stability reinforcement learning algorithm (such as Deep Deterministic Policy Gradient, DDPG) to implement the concept of the Stable-BC algorithm.

[0186] The DDPG (Deep Deterministic Policy Gradient) algorithm is a deep reinforcement learning algorithm that is mainly used to solve decision-making problems in continuous action spaces. The process of the DDPG algorithm can be divided into the following three main parts:

[0187] 1. Initialization process:

[0188] Initialize the Actor network and the Critic network, as well as their target networks.

[0189] Initialize the optimizers for the Actor network and the Critic network.

[0190] Copy the parameters of the Actor network and the Critic network to the corresponding target network.

[0191] 2. Sampling process:

[0192] Randomly select an initial state from the environment.

[0193] Input the state into the Actor current network and generate an action (A) according to the current strategy (pi).

[0194] Input the action (A) into the environment Environment, execute the action and observe the new state (S'), reward (R) and end flag (done) of the environment feedback.

[0195] 3. Training and parameter update process:

[0196] Store the sampled (S, A, R, S', done) in the experience replay pool (Replay Buffer).

[0197] A batch of data is randomly selected from the experience replay pool for training.

[0198] Training the Critic network:

[0199] The parameters of the Critic network are updated using gradient descent to minimize the squared difference between the Q value and the target Q value.

[0200] Training the Actor Network:

[0201] Use the gradient provided by the Critic network to update the parameters of the Actor network to maximize the Q value output by the Critic network.

[0202] Update the target network:

[0203] The parameters of the Actor network and the Critic network are copied to the corresponding target network at a certain ratio (such as the soft update coefficient (τ)).

[0204] Through the above process, the DDPG algorithm can continuously learn and optimize strategies to achieve the goals of reinforcement learning tasks.

[0205] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. An artificial intelligence cooking method for a cooking device, comprising the following steps: Step 1: The cooking equipment uses image recognition technology to identify the shape and color of ingredients and the kitchen scene; Step 2: The cooking equipment uses a high-definition camera to record the chef's actions, observes and learns cooking movements through deep learning and motion capture technology, analyzes gestures and steps through algorithms, and then reproduces the cooking process; Step 3: The cooking equipment imitates human movements. Through repeated practice and algorithm optimization, the equipment gradually masters cooking skills and cooks dishes independently.

2. The artificial intelligence cooking method of a cooking device according to claim 1, characterized in that: The step 1 of identifying the shape and color of food includes the following steps: Step 1: Data collection and preprocessing: collect image data of various ingredients, preprocess the images, and ensure the consistency of input data; Step 2: Build a CNN model and design a suitable CNN architecture based on task requirements; Step 3: Model training, use the preprocessed food image data to train the CNN model; Step ④: Model evaluation and optimization: evaluate the performance of the model on an independent test set and tune the model; Step 5: Apply the trained CNN model to the actual scene, identify the input food pictures, and output the recognition results.

3. The artificial intelligence cooking method of a cooking device according to claim 1, characterized in that: The kitchen scene recognition in step 1 includes active scene recognition and passive scene recognition; The active scene recognition is achieved by actively scanning the environment through a sensor or actively capturing images through a camera; The passive scene recognition is that the cooking device analyzes and recognizes the environmental data after receiving it.

4. The artificial intelligence cooking method of a cooking device according to claim 1, characterized in that: The cooking device imitating human actions in step 3 includes the following steps: Step ①: Motion capture and reproduction; Through sensor technology, the robot can capture the cooking movements of human chefs and then accurately reproduce these movements through behavioral reproduction technology; Step ②: Posture recognition and optimization; Use computer vision technology to recognize the chef’s movements, thereby helping the cooking equipment understand and optimize the execution process of cooking actions; Step ③: Hand movement planning; Cooking equipment needs to learn how to use hands correctly to perform cooking operations. Behavior replication can help cooking equipment imitate the movement trajectory and operation methods of human hands. Step ④: Real-time feedback and adjustment; Behavior reproduction combined with real-time feedback mechanism enables cooking equipment to adjust its actions according to the situation during the cooking process to ensure the smooth progress of the cooking process; Step ⑤: Multimodal data integration; Integrate multimodal data such as vision and voice into the behavior reproduction system to achieve a more comprehensive and intelligent cooking process simulation.

Images